id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
62ab9c204483988df478b293 | 16 | These two enzymes both contain active site cysteines that interact with substrates (ethanolamine phosphate and glutathione respectively) . Alternatively, these proteins could have lower abundance in response to propachlor treatment, though CEPT1 and MGST3 are both fairly long-lived proteins and so would have to be actively degraded to be depleted on the time-scale of the experiment. |
62ab9c204483988df478b293 | 17 | Despite its lack of HSR induction, alachlor exposure increases DNAJB8 affinity of many more proteins than acetochlor exposure (Figure and Figure ). 764 proteins show significantly (q-value < 0.05) increased DNAJB8 affinity following alachlor exposure, as opposed to 81 proteins following acetochlor exposure. Selectively targeted proteins included Microtubule Associated Serine/Threonine Kinase Like (MASTL) (fold change = 2.02, q value = 5 x 10 -5 ), a kinase involved in mitosis , and Zinc Finger Protein 24 (ZNF24), a tumor suppressor . The higher impact of alachlor as opposed to acetachlor is in some ways surprising, as the two molecules are isomers differing only by a methyl group. However, regioisomers can have substantially different lipophilicities and cellular uptake . Alachlor has a log Kow of 3.5, as opposed to 3.0 for acetachlor. |
62ab9c204483988df478b293 | 18 | for other electrophilic series . The stronger protein destabilization response to alachlor treatment could also reflect the differential metabolism of the compounds. Compared to alachlor, acetochlor metabolizes faster into 2-chloro-N-(2,6-diethylphenyl)acetamide (CMEPA) and 2-methyl-6-ethylaniline (MEA) . While microsomal pathways are not available in the HEK293T cells, other chloroacetanilide decomposition pathways could be available. |
62ab9c204483988df478b293 | 19 | Propachlor treatment has the strongest effect on the DNAJB8 H31Q -associated proteome (Figure and Figure ). The two most prominent targets that are unique to propachlor are glyceraldehyde-3-phosphate dehydrogenase (GAPDH; fold change = 5.97, q-value = 6.09 x10 ) and Parkinson's disease protein 7 (PARK7/DJ-1, fold change = 3.8, q-value = 5.56 x 10 -6 ). GAPDH is an enzyme that canonically uses a susceptible active-site cysteine to bind and reduce nicotinamide adenine dinucleotide (NAD) in glycolysis , but consistent with its high abundance also engages in extensive moonlighting activities . Due to its high abundance and pK-perturbed active site cysteines, GAPDH is a frequent conjugation target of electrophilic molecules . |
62ab9c204483988df478b293 | 20 | GAPDH is found in Parkinson's Disease (PD)-associated aggregates, and its aggregation is promoted by electrophilic conjugation . Given its abundance and fold change in DNAJB8 co-IP recovery, it is possible that GAPDH is the feature observed by silver stain at 37 kDa (Figure ). Thermodynamically destabilized PARK7 variants are linked to familial Parkinson's disease , and PARK7 overexpression protects against chemical induction of Parkinson's phenotypes . Although a wide variety of mechanisms have been ascribed to PARK7, including chaperoning and proteolytic activities, the evidence is strong that it serves as an oxidative stress sensor that protects against cysteine oxidation , as well as a cellular deglycase preventing electrophilic protein damage. |
62ab9c204483988df478b293 | 21 | We expected that proteins with the highest reactivities to electrophilic species would also be the most destabilized, while proteins whose Hps40 affinities are less impacted by propachlor treatment would be enriched in less reactive proteins. We compared the ranked SSMDs for all identified proteins against reactivity profiles reported in Kuljanin et al. , with the expectation that our strongest hits for each chloroacetanilide would reflect proteins that are generally more active nucleophiles. |
62ab9c204483988df478b293 | 22 | Proteins were considered to be iodoacetamide-reactive if they were directly modified by the iodoacetamide probe, and were considered chloroacetamide-reactive if any of the 158 tested chloroacetamides could outcompete iodoacetamide reactivity by a factor of 4-fold (Figure ). For each chloroacetanilide herbicide, there is no enrichment of more reported reactive proteins at higher Hsp40 affinity follow herbicide treatment. About 12% of our identified proteins have reported chloroacetamide reactivity, independent of chloroacetanilide-induced destabilization. This finding strongly agrees with a central conclusion from the prior work: chloroacetamide reactivity is sparse, with most molecules showing substantial specificity to individual proteins. In keeping with the greater reactivity of iodoacetamides, about 70% of our identified proteins have reported iodoacetamide reactivity, again independent of chloroacetanilide-induced destabilization. This finding is more surprising, but could reflect that Hsp40 affinity is not the same as reactivity; adduct formation can have different effects on protein stability for different proteins. Indeed, there is no correlation between the promiscuity of a protein for chloroacetamide modification and it's destabilization by any of the three chloroacetanilides that we profiled (Figure ). Nevertheless, our most affected proteins are iodoacetamide reactive, consistent with the primary mechanism of protein destabilization due to cellular chloroacetanilide exposure being mediated through direct conjugation at reactive cysteines (Figure ). |
62ab9c204483988df478b293 | 23 | Including PARK7 and GAPDH, 78 proteins have substantially increased (fold change > 2 and q-value < 0.1) affinity for DNAJB8 following propachlor treatment (Figure ). This is more than the twice the combined number of proteins destabilized under the same criteria after alachlor and acetochlor treatments. The higher susceptibility of the proteome to propachlor could be based on substitution reaction reactivity. Kinetic studies between propachlor and alachlor reactivity found a 2-fold increase in the substitution of propachlor against several nucleophiles and a lower Gibbs free energy required for substitution reactions of propachlor with nucleophilic thiols . All three treatments cause marked destabilization of TYMS and ACAT1, and apparent stabilization (decreased DNAJB8 affinity) of CEPT1 and MGST3 (Figure ). |
62ab9c204483988df478b293 | 24 | Outside of those, the proteins that are most affected by each individual treatment are unique. This selectivity is particularly salient given the high concentrations (1 mM) used for the treatments. The unique profiles of proteins affected by each herbicide is consistent with the high selectivity of protein reactivity with even strongly reactive warheads . We searched for cysteine-chloroacetanilide adducts from each run, finding 32/8637, 40/16029, and 102/15694 modified/total peptides following acetochlor, alachlor, and propachlor treatments respectively. It can be challenging to identify peptide modifications without enriching for the modified sites, as stoichiometry on a per peptide basis for post-translational modifications are often low . Furthermore, when peptide identifications are filtered, the filtration thresholds are set for each search such that the false discovery rate based on a decoy set is <1% of peptides. While that intended threshold is likely reasonably close to the true false discovery rate in the context of the entire proteome , for a modification found on only a small number of modified peptides it introduces a much higher practical risk of misidentification . With these caveats in mind, we include the list of identified chloroacetanilide conjugates in Table . It is also worth noting that we consistently see a DNAJB8 modification at C70. While this is in the DNAJB8 J-domain , which is dispensable for client binding, we cannot rule out the possibility that this modification itself could impact DNAJB8 recognition. |
62ab9c204483988df478b293 | 25 | Despite both aggregation and Hsp40 affinity increasing across a majority of the observed proteome after cellular propachlor exposure, there is no meaningful correlation between these two factors. This is consistent with a previous study that found that stress-dependent protein aggregation does not correlate with stability across diverse stresses . Generally, different measures of protein stabilization are poorly correlated with each other, as protein misfolding involves multiple processes that will be probed differently by different assays . The vesicular trafficking proteins SMAP2, GAK, CLTA, CLTB, CLTC, CLTC1, AP1B1, AP1M1, EPS15, and CLINT1 all substantially lose solubility in response to propachlor treatment, but this network only modestly increases its DNAJB8 affinity. These proteins rely on Hsp70 for clathrin disaggregation, but GAK serves as a dedicated Hsp40 co-chaperone outside of neuronal cells . These proteins may not be well-surveyed by a promiscuous Hsp40 such as DNAJB8, even under conditions that lead to their destabilization. Alternatively, they may have lower thresholds for aggregation as compared to proteins that show greater differential DNAJB8 affinity following propachlor treatment . GAPDH is highly abundant in the human proteome with many functions beyond its canonical role in glycolysis . These functions are readily perturbed by a diverse range of post-translational modifications which can lead to toxic aggregation , making its destabilization following propachlor exposure particularly hazardous for cellular proteostasis. Hence, we looked for the presence of GAPDH-propachlor adducts. |
62ab9c204483988df478b293 | 26 | GAPDH is susceptible to modification by a wide range of electrophiles, including 4hydroxynonenal and methylglyoxal . Cysteines in the NAD+ binding site are particularly subject to electrophilic modification 100-102 . To determine whether GAPDH is modified during propachlor treatment, we immunoprecipitated Flag GAPDH from lysates following cellular propachlor treatment. GAPDH interactions are interrupted by our stringent RIPA washes, so that Flag GAPDH is prepared with minimal contamination by other proteins (Figure ). In the absence of treatment, we see GAPDH primarily present as the N-terminally acetylated protein, with a smaller population of the glutathione conjugate (Figure ). After propachlor treatment, we see a new base peak at +176 Da, consistent with a single propachlor adduct (Figure ). No evidence of multiple adducts is observed. We further investigated immunoprecipitated Flag GAPDH by digestion and shotgun proteomics, followed by an open adduct search . |
62ab9c204483988df478b293 | 27 | The propachlor modification is clearly localized to C152 based on MS2 spectra (46 total spectral counts) (Figure ). C152 is in the NAD + binding site, and is necessary for catalytic activity . Across two biological duplicates, we find a 26±7% drop in the unmodified C152 intensity, implying that about a quarter of GAPDH is modified. No evidence for modification at C156 was observed. |
62ab9c204483988df478b293 | 28 | Destabilized protein domains are more extended and thus more susceptible to cleavage by a promiscuous protease, such as thermolysin or proteinase K (PK) . LiP involves brief treatment of lysate to protease followed by shotgun proteomics to characterize the yield of cleavage events . Loss of a tryptic peptide indicates a protein conformational change in the vicinity of that peptide sequence 107 . This peptide is modified at the C152 position with an adduct that corresponds to propachlor thiocarbamate. C156 is carbamidoylated by iodoacetamide. B) PRM chromatograms demonstrating the dependence of the adduct on propachlor treatment. |
62ab9c204483988df478b293 | 29 | We selected peptides from GAPDH for LiP to assess structural changes after propachlor treatment. We found several GAPDH peptides to be more proteolytically sensitive to proteinase K after propachlor treatment (Figure and Figure ), including the active site peptides LVINGNPITIFQER, LISWYDNEFGYSNR, VGVNGFGR. Hence, propachlor induces a more extended conformation in GAPDH, consistent with destabilization. Destabilization of GAPDH could also affect proteinprotein interactions. One of the destabilized peptides, VPTANVSVVDLTCR, is involved in the dimer interface of GAPDH 108 . Mutations in this peptide are associated with conformational changes at the dimeric interface and a loss of tetrameric stability . |
62ab9c204483988df478b293 | 30 | Destabilization of VPTANVSVVDLTCR in GAPDH by propachlor exposure may inhibit the active conformation and affect binding partners. No meaningful change in proteolytic susceptibility was observed for IISNASCTNCLAPLAK, which encompasses the propachlor adduct site. Since this peptide can only be observed in GAPDH that has not been directly modified by propachlor, this implies that the stability of unmodified GAPDH is not generally perturbed by the treatment. We also attempted LiP on PARK7 peptides. |
62ab9c204483988df478b293 | 31 | Only three peptides proved suitable for LiP (Table ), and none showed evidence of differential proteolytic susceptibility following propachlor treatment (Figure ). While this indicates that the protein as a whole is not destabilized, it does not exclude the possibility that unprofiled regions of the protein might be affected. Protein destabilization can lead to both gain-of-function (toxic conformations) and loss-of-function. GAPDH activity has previously been shown to decrease in response to methylglyoxal and copper exposures, presumably due to conjugate and oxidation respectively . GAPDH modification can further lead to misfolding and aggregation . We evaluated GAPDH activity in cells treated with propachlor. |
62ab9c204483988df478b293 | 32 | GAPDH activity in lysates was measured using a colorimetric assay for NAD + reduction in the presence of substrate. Treating HEK293T cells with 1 mM propachlor for 30 minutes decreased GAPDH activity by 25% (Figure ). This decrease is consistent with the amount of C152 adducts that we detect by mass spectrometry, but low considering the strong negative cooperativity between the two catalytic sites on the GAPDH tetramer 113 . Our proteomic characterization of propachlor-induced aggregation found an increase in GAPDH aggregation induced by propachlor treatment (FC in aggregates = 4.9, q-value = 0.008; Figure ). Similar results were obtained from Western Blot analysis assessing the levels of GAPDH in the pellet fraction after ultracentrifugation (Figure ), however there is no significant depletion of total GAPDH (Figure ). From this we can conclude that although GAPDH destabilization following propachlor treatment does lead to an increase in the aggregated fraction, the total burden of GAPDH aggregation on the cell remains small compared to the high levels of the soluble protein. |
62ab9c204483988df478b293 | 33 | PARK7 is a chaperone-like peptidase that can repair proteins damaged by a series of aldehyde products, including methylglyoxal and glyoxal. PARK7 specifically protects GAPDH from cysteine and lysine adducts, including glycerate damage caused by metabolic products generated by GAPDH itself . PARK7 is also significantly destabilized after propachlor treatment and thus could be inactive. A cellular assay designed to quantify PARK7 ability to deglycate glyoxal modified proteins in HEK293T cells has been previously established . We measured the ability of endogenous PARK7 to degylcate glyoxal-modified proteins after incubation with 1 mM propachlor for 30 minutes (Figure and Figure ). In the presence of propachlor, the intensity of proteins converted to carboxy-methyl-lysine after glyoxal treatment increased significantly in comparison with the control experiment (DMSO vehicle treatment). |
62ab9c204483988df478b293 | 34 | Cellular exposure to propachlor inhibits PARK7's ability to deglycate damaged proteins, offering an alternative mechanism by which propachlor exposure can induce protein misfolding beyond direct modification. We speculate that it could be beneficial to the cell that GAPDH is inhibited in concert with PARK7, preventing accumulation of glycating equivalents when the detoxification mechanism is also inhibited. Due to poor reproducibility for profiling the active C106 in PARK7, we were not able to compare whether PARK7 C106 and GAPDH C152 are similarly modified across the range of chloroacetamides investigated in the reported high throughput screen . For the three chloroacetanilides in our present study, however, the relationship holds (Figure ). |
62ab9c204483988df478b293 | 35 | In summary, we present profiles of destabilized proteomes in response to cellular exposure to three chloroacetanilide herbicides. While some proteins are destabilized by each treatment, the overall profiles from each herbicide exposure are unique. About 70% of targeted proteins are known to be subject to haloacetamide conjugation at cysteine, consistent with adducts being the primary mechanism of destabilization, but the extent of destabilization does not correlate with haloacetamides reactivity, reflecting the distinction between conjugation and stability. Hsp40 affinity profiling is an effective assay for determining the effect of environmental toxicants on the cellular proteome, both distinct from and complementary to existing technologies. |
65edbbba66c1381729ae3ed3 | 0 | Despite substantial advancements in small molecule mass spectrometry (MS) in recent decades, the field faces a persistent challenge. Although an untargeted MS study can acquire thousands to millions of MS/MS spectra, current state-of-the-art methods are only able to confidently annotate around 5 % to 10 % of these spectra on average. This limitation underscores a significant knowledge gap, constraining our capacity to accurately determine the molecular structures present in such studies and, by extension, diminishing the potential impact of numerous biological investigations. |
65edbbba66c1381729ae3ed3 | 1 | To address this challenge, Goldman et al. [2] have recently introduced the Metabolite Inference with Spectrum Transformers (MIST) tool to annotate MS/MS spectra. MIST incorporates a novel "chemical formula transformer" aimed at integrating domain-specific knowledge into the architecture of deep neural networks. This approach deviates from traditional methods that represent peaks in MS/MS spectra as discrete, binned values to process them with neural networks. Instead, MIST represents spectra through the chemical formulas associated with their peaks, which are then processed by a transformer neural network. MIST further enriches its model by encoding neutral loss relationships between fragment ions and incorporating substructure prediction as an auxiliary training objective. To address the challenge of limited large-scale training datasets in the domain of small molecule MS, MIST employs a strategy to simulate spectra, thereby expanding the size and diversity of its training corpus. Additionally, rather than directly predicting full-length fingerprints, MIST adopts a progressive inference methodology by first predicting lower-resolution fingerprints, which are subsequently refined into their full-resolution counterparts. |
65edbbba66c1381729ae3ed3 | 2 | The fingerprints predicted by MIST can be directly utilized for spectrum annotation, employing nearest neighbor searching to match the MS/MS-derived fingerprints against those extracted from molecular structures within chemical databases, such as PubChem. MIST also harnesses a metric learning approach, utilizing a contrastive model to transform both spectral and chemical fingerprints into embeddings, which are then subjected to nearest neighbor searching. Initial findings by Goldman et al. [2] indicate that the fingerprints predicted by MIST from MS/MS spectra surpass those generated by CSI:FingerID 5 in terms of accuracy. However, it is only with the application of contrastive fine-tuning that MIST demonstrates a superior performance in database retrieval tasks. To understand these results, this reusability report aims to critically assess the MIST reproducibility and evaluate its novel neural network architecture contributions. Specifically, we describe the extent to which the originally reported results can be replicated and apply MIST to an external dataset from the CASMI 2022 challenge in order to assess its generalizability and potential broader impact on the field. |
65edbbba66c1381729ae3ed3 | 3 | To evaluate the MIST reproducibility, we followed the training and testing instructions as delineated in the documentation on its GitHub repository (). It is crucial to note that the version corresponding to the published paper by Goldman et al. [2] is MIST v1.0.1. The latest version on GitHub, MIST v2.0.0, introduces a few enhancements, including the elimination of the dependency on SIRIUS for annotating MS/MS spectrum peaks with their corresponding sub-formulas. For our reproducibility analysis, we focused on MIST v1.0.1 to ensure alignment with the version detailed in the publication by Goldman et al. [2]. |
65edbbba66c1381729ae3ed3 | 4 | Next, we retrained MIST's various components as outlined in the original study: the fingerprint prediction model, a baseline feed-forward neural network model, the contrastive model for embedding generation, and the forward model for simulating MS/MS data. In addition to public data from GNPS and MoNA, the primary MIST model was also trained on proprietary data from the commercial NIST20 spectral library. As this model is not publicly available, instead we utilized only data coming from public sources to retrain the models. While weights for the fingerprint prediction model and the contrastive model corresponding to the public data subset have also been provided on Zenodo by Goldman et al. [2], the baseline and simulation model were not provided. |
65edbbba66c1381729ae3ed3 | 5 | Goldman et al. [2] mention that the training process was undertaken using a three-fold structure-disjoint crossvalidation methodology, although they did not specify which cross-validation split was employed or whether an integration strategy was used to obtain the reported results. For our analysis, we consistently utilized the first cross-validation split. Training the MIST model and the baseline model was successful, using the provided data excluding simulated spectra. Retraining the contrastive model required initial preparation of a decoy chemical database and the resolution of several code discrepancies. |
65edbbba66c1381729ae3ed3 | 6 | When including the forward simulation model, we encountered several challenges due to sparse documentation and hard-coded file paths in the corresponding scripts. After these issues were addressed, incorporating simulated spectra during training of the fingerprint prediction model resulted in a reduction of the training time until convergence from approximately four hours to approximately three hours, indicating potential efficiency gains through data augmentation. |
65edbbba66c1381729ae3ed3 | 7 | Upon completion of the training phase, we conducted an evaluation using the test set of the first cross-validation split, consisting of 819 MS/MS spectra, comparing the performance of our retrained models against the pretrained models that are shared publicly. This revealed a high degree of similarity and correlation in the fingerprints predicted by both the original and retrained models (figure |
65edbbba66c1381729ae3ed3 | 8 | To evaluate the generalizability of MIST, we applied it to predict fingerprints and perform database annotation for a set of novel MS/MS spectra from the CASMI 2022 challenge. This dataset, released subsequent to the training of the published version of MIST, offers a stringent benchmark for evaluating MIST's performance on previously unseen and independent data. |
65edbbba66c1381729ae3ed3 | 9 | Applying MIST to this new dataset presented certain challenges, primarily due to a scarcity of documentation on how to preprocess and integrate new data into MIST's workflow. This included a reliance on SIRIUS for data preprocessing-including the creation of its input files-and a retrieval database, necessitating an examination of the source code to deduce the requisite steps. Although not considered in this analysis, it is worth noting that the most recent MIST v2.0.0 has removed the dependency on SIRIUS, streamlining the process. Additionally, modifications were required to address hard-coded file paths in the codebase. |
65edbbba66c1381729ae3ed3 | 10 | We employed both the publicly available MIST model and our retrained version to predict fingerprints for 170 protonated, singly charged spectra from CASMI 2022. The results from both models were highly comparable (figure ), further underscoring MIST's reproducibility, albeit with the caveat that certain assumptions regarding the architecture and hyperparameters of the pretrained model had to be made due to the lack of explicit documentation. Our retrained model incorporated all optional features, including pairwise neutral loss, MAGMa 8 substructure annotation, data augmentation, and unfolding. |
65edbbba66c1381729ae3ed3 | 11 | An interesting observation is a significant decrease in performance of both the public and retrained MIST models on the CASMI 2022 dataset (figure ), however, which is in contrast to the performance levels reported in the original MIST publication and on the provided test set (figure ). To delve deeper into this discrepancy, we also assessed the full MIST model described by Goldman et al. [2], which was trained on both public and commercial data, and which was generously made available by the authors. Although this model demonstrates an enhanced performance on the original MIST test data, its efficacy on the CASMI 2022 dataset mirrored the substantial decline observed with the other models. |
65edbbba66c1381729ae3ed3 | 12 | Next, we evaluated the MIST database retrieval performance, including a comparison with the benchmark ) Ablation results for MIST without MAGMa substructure prediction, without simulated data, without neutral loss featurization (pairwise), and without fingerprint unfolding. Each model was trained five times with different random seeds. standard, CSI:FingerID. Initial comparisons among the public and retrained MIST versions reveal that their performances are highly similar (figure ), indicating consistent spectrum annotation capabilities across versions. Specifically, when utilizing predicted fingerprints for database retrieval using the cosine distance, the pretrained MIST model achieves a top-1 annotation accuracy of 8.9 %, in contrast to 11.0 % for the retrained model. When employing contrastive embeddings for retrieval, both models exhibit identical top-1 annotation accuracy of 10.0 %. |
65edbbba66c1381729ae3ed3 | 13 | Compared to MIST trained exclusively on public data, the full MIST model, which includes both public and proprietary data in its training, achieves a significantly stronger performance (figure ). However, it is noteworthy that across the CASMI 2022 dataset, the MIST database retrieval performance does not match the higher levels previously reported by Goldman et al. [2] on a different test set. This observation is anticipated, considering the lower precision of fingerprint predictions on this novel dataset. Furthermore, CSI:FingerID significantly outperforms all MIST models on the CASMI 2022 dataset, contrary to the similar performance between MIST and CSI:FingerID reported by Goldman et al. [2]. |
65edbbba66c1381729ae3ed3 | 14 | Building on the initial ablation study to evaluate the impact of various MIST features (pairwise neutral loss, MAGMa 8 substructure annotation, data augmentation, and unfolding) on fingerprint prediction accuracy, we extended this analysis to the database retrieval level. Fingerprint prediction is only an intermediate step during database annotation, and as previously reported, 2 improvements during this step might not necessarily correlate with enhanced performance on the full task. Given that certain features of MIST introduce considerable algorithmic complexity with only marginal improvements in fingerprint prediction accuracy, we sought to understand how this translates to changes in performance during database retrieval. |
65edbbba66c1381729ae3ed3 | 15 | Our findings from evaluating the top-5 annotation accuracy across all models (figure ) suggest a uniform performance. The absence of MAGMa substructure labeling had only a minimal impact on retrieval accuracy, whereas a somewhat more notable reduction in annotation accuracy can be observed when the fingerprint unfolding and neutral loss featurization components are removed. This consistent performance pattern persisted in the evaluation of top-20 annotation accuracy, revealing negligible differences between the models (figure ). |
65edbbba66c1381729ae3ed3 | 16 | Statistical analysis using Cochran's Q test to compare the efficacy of multiple classifiers yielded p-values of 0.564 and 0.308 for top-5 and top-20 annotation accuracy levels, respectively, which do not meet the threshold for statistical significance to reject the null hypothesis that all models perform equally well. These results suggest that the added computational complexity introduced by various MIST features, including dependencies on external tools, may not provide a commensurate improvement in database retrieval performance, indicating the need for a critical reassessment of these components within the model's architecture. |
65edbbba66c1381729ae3ed3 | 17 | In this study, we have meticulously assessed the reproducibility and reusability of MIST for the prediction of chemical fingerprints from MS data. Our endeavors to replicate the training and fingerprint prediction of MIST were largely successful, albeit not without encountering several minor challenges. These challenges, including small bugs in the source code and occasional gaps in the documentation, underscore the inherent complexities involved in developing advanced bioinformatics tools. Despite these hurdles, the overall good quality of MIST's documentation and source code facilitated the resolution of these issues, albeit requiring expert effort. This experience demonstrates the significant disparity between the development of a research prototype and the refinement of a tool to a production-ready state, emphasizing the necessity for robustness to ensure successful application across varied scientific environments. The absence of tests within the MIST code repository further highlights this gap, suggesting that the integration of software engineering best practices could greatly enhance the development process, ensuring the production of high-quality scientific software while saving time in the long term. |
65edbbba66c1381729ae3ed3 | 18 | Our findings corroborate previous results by Goldman et al. [2] regarding the discrepancy between MIST's fingerprint prediction accuracy and its database retrieval performance. This observation, particularly highlighted through MIST's underperformance on an external test set from the CASMI 2022 challenge, raises questions about the model's generalizability and the nuanced challenges of model evaluation in computational biology. |
65edbbba66c1381729ae3ed3 | 19 | However, it is imperative to view these findings not as a critique of MIST but rather as a catalyst for broader community engagement in the development and benchmarking of computational tools. The initiative by Goldman et al. [2] to release their training datasets, cross-validation splits, and model configurations sets a commendable precedent for transparency and reproducibility in computational biology and machine learning. This open sharing of resources is crucial for fostering a community-wide effort towards robust benchmarking standards. |
65edbbba66c1381729ae3ed3 | 20 | Data availability presents another critical challenge, as highlighted by the partial reliance of the main MIST model on a combination of public and proprietary datasets, which restricts the full replicability of its reported performance. The increasing reliance of state-of-the-art machine learning models on large-scale datasets underscores the pivotal role of data availability in driving progress within the field of metabolomics and beyond. This scenario suggests that future advancements may hinge not only on algorithmic innovation but also significantly on the accessibility of comprehensive and accurate datasets. |
65edbbba66c1381729ae3ed3 | 21 | Through our ablation study aimed at discerning the impact of these various features on database retrieval performance, it became apparent that the additional complexity introduced by certain features may not substantively enhance model performance. Notably, the reliance on external tools, such as MAGMa, and the adjustments made in MIST v2.0.0 to remove its dependency on SIRIUS, 6 highlight an important area for continued development. |
65edbbba66c1381729ae3ed3 | 22 | Retraining the contrastive model required the construction of a decoy database of isomers, derived from Pub-Chem. During this step, we had to address several minor coding errors and manage the extensive computational resources required for processing the PubChem database. After fixing these issues, training durations were approximately 12 and 35 minutes, achieving contrastive losses of 1.9430 and 1.7349 on the test set without and with data augmentation, respectively. |
65edbbba66c1381729ae3ed3 | 23 | The Critical Assessment of Small Molecule Identification (CASMI) 2022 challenge was an international competition aimed at evaluating and improving the methods used for identifying small molecules from MS data (). While only the raw data and the ground truth compound information is available from the official CASMI 2022 website, we retrieved a Pandas 10 dataframe with annotated spectra from Young et al. [9], who matched compound annotations to the corresponding MS/MS spectra. Next, as MIST has currently only been trained on protonated, singly charged spectra, we filtered the CASMI 2022 data for spectra with adduct "[M+H]+", resulting in a dataset of 170 spectra. |
65edbbba66c1381729ae3ed3 | 24 | Applying MIST to the CASMI 2022 dataset necessitated preprocessing with SIRIUS, a step that was somewhat obfuscated by the lack of explicit documentation and required investigating the code to determine the required data types and directory structure. Note that the recently updated MIST v2.0.0 has simplified this process by eliminating the SIRIUS preprocessing dependency. These challenges underscore that MIST, in its current form, functions more as a research prototype than a fully fledged tool designed for end-user application. |
65edbbba66c1381729ae3ed3 | 25 | MIST offers two database retrieval methods: fingerprint retrieval and contrastive retrieval. Fingerprint retrieval uses the raw predicted fingerprints to rank all isomers within the PubChem database (version April 2022) that correspond to the chemical formula of the precursor compound, as derived by SIRIUS. This ranking is determined by calculating the cosine distance between the predicted spectral fingerprint and the candidate chemical fingerprints, with the compound having the lowest rank being selected. |
672c882ef9980725cf6d6ca9 | 0 | = 2.3 to 3.2, y = 1.2 to 2.0, z = 0.3 to 0.9) demonstrated that above ca. 46% of B-site magnetic cations, the m = 5 structure first rearranges into a mixed-phase material based on m = 5 and sixlayered (m = 6) structures and eventually evolves into an m = 6 phase with 54% magnetic cations at the B-site. It is postulated that increasing the number of perovskite layers by forming the m = 6 structure facilitates the accommodation of additional magnetic cations at a lower average manganese oxidation state (+3.3) compared with an equivalent m = 5 stoichiometry (+4.0). While the minor out-of-plane ferroelectric response decreases as expected with increasing structural reorganization towards the m = 6 phase, the predominant in-plane piezoresponse remains unaffected by magnetic cation substitution. This work shows that higher-layered Aurivillius homologues can be synthesized using aliovalent substitution, without requiring epitaxial growth or kinetically constrained methods. |
672c882ef9980725cf6d6ca9 | 1 | Data storage is a critical enabler to all aspects of modern life. It was estimated that in 2023 digital data storage stood at 64 zettabytes, roughly doubling every two years. Contemporary computer storage technologies use either electric or magnetic polarization to store single bits of information; a power-hungry technology that will soon struggle to keep-up with the ever-increasing demands. Masanet et al. pointed-out that in 2018, data centers used 205 TWh of electricity, ca 1% of global consumption. Interestingly, due to major increases in energy efficiency this was only a 6% increase on the 2010 figure, despite a 550% increase in digital data storage over the same period. This emphasizes the importance for the sector of maintaining the energy-efficiency trajectory. Technologies that simultaneously combine ferroelectric and ferromagnetic/ferrimagnetic properties can use energy-efficient electric fields for the reading and writing of data and in-principle could allow an eight-times increase in data storage capacity per threshold 17 is a well-known strategy for enhancing long-range magnetic ordering and magnetization (Supplementary Section SI1). Additionally, computational studies on the Aurivillius phases by Birenbaum et al. have demonstrated that increasing the concentration of magnetic ions raises the ferrimagnetic Curie temperature. Thus, increasing the magnetic ion concentration at the B-site of the B6TFMO structure beyond the current level of ca. 40% is a critical factor in further improving its magnetic properties. |
672c882ef9980725cf6d6ca9 | 2 | However, the Aurivillius phase structure is expected to have a limit on the concentration of Fe and Mn it can accommodate. One constraint stems from differences in cation radii: when coordinated octahedrally by oxygen, Ti⁴⁺ has a radius of 0.605 Å, whereas both Fe³⁺ and Mn³⁺ (in their high-spin states) have a radii of 0.645 Å. (NB: The high-spin state is favored for transition metals in Aurivillius phase structures due to structural and electronic factors. These include the large size of bismuth resulting in weakened ligand fields, distortion in the coordination environment due to neighboring layers, and incomplete hybridization between metal d orbitals and ligand orbitals. ) Another limitation arises because these cations have a nominal valence of 3+, and they replace Ti⁴⁺, which has a valence of 4+. Substituting Ti 4+ by Mn 3+ /Fe 3+ requires charge compensation, for which several mechanisms can be postulated. One possibility for charge compensation is through changes in the oxidation states (o.s.) of the Mn or Fe cations. Alternatively, exceeding the solubility limit may lead to the segregation of impurity species or the formation of secondary phases at grain boundaries or surfaces. Another potential mechanism is ionic compensation through the introduction of ionic defects such as oxygen vacancies. However, in Aurivillius phase materials, the [Bi2O2] 2+ layers are recognized for their important role in space charge compensation and prevention of oxygen vacancies, which in turn reduces leakage currents and enhances fatigue resistance. Given these considerations, it is important to understand how increasing the concentration of magnetic ions in m = 5 Aurivillius phases affects the crystal structure. To investigate this, we systematically varied the concentrations of the three B-site atoms (Ti, Fe and Mn) in the B6TFMO system, Bi6TixFeyMnzO18 (x = 2.3 to 3.2, y = 1.2 to 2.0, z = 0.3 to 0.9). This approach enabled us to determine the solubility limit of magnetic ion inclusion in the m = 5 Aurivillius phase structure and assess its impact on the superlattice layering. We consider the above size and charge compensation mechanisms and demonstrate that structural reorganization into higher m phases is a further option available to the versatile Aurivillius phase structures to accommodate a greater magnetic ion fraction while preserving in-plane ferroelectric properties. Furthermore, we show that aliovalent substitution using transition metal cations, through simple chemical solution deposition processes, can overcome thermodynamic challenges of synthesizing higher-layered Aurivillius homologues without requiring epitaxial substrates or kinetically limited growth processes. |
672c882ef9980725cf6d6ca9 | 3 | X-ray diffraction measurements (XRD) were performed using a Philips X'pert PW3719 MPD diffractometer, Cu Kα radiation, 40 kV, 35 mA and 2θ range of 6 ˚ to 40 ˚. The m = 5 Aurivillius phase was indexed from calculated XRD patterns generated using VESTA and Crystal Diffract 7 software (Crystal Maker Software Ltd.) using data from García-Guaderrama et al. The m = 6 phase was indexed from calculated XRD patterns using data from Krzhizhanovskaya et al. Atomic Force Microscopy (AFM) and Piezo Force Microscopy (PFM) measurements and ferroelectric lithography experiments were performed using an MFP-3D TM Asylum Research instrument (see Supplementary Section SI2 for details). |
672c882ef9980725cf6d6ca9 | 4 | Scanning electron microscopy (SEM) was performed on a Zeiss Supra 40 instrument and samples were treated with a 3 to 6 nm thickness gold coating before measurement to prevent charging during imaging. Lamella cross sections of the films were prepared using a FEI DualBeam Helios NanoLab 600i Focused Ion Beam instrument. High resolution transmission electron microscopy (HR-TEM) was performed using a Jeol 2100 transmission electron microscope; 200 kV; double tilt holder. Note that normally ~10% error should be accounted for when calculating distances from TEM data due to the electron optics of the instrumentation. Composition analysis was conducted using high-resolution scanning transmission electron microscopy (STEM) and energy dispersive x-ray spectroscopy (EDX) equipped with an X-Max 80 detector and AZTecanalysis software (Oxford Instruments, Abingdon, UK). |
672c882ef9980725cf6d6ca9 | 5 | The concentrations of two of the three types of B-site cations were systematically varied within the B6TFMO Series 1, 2 and 3, while the third cation concentration was kept constant. XRD analysis (Fig. (a) to (c)) of Series 1, 2 and 3 was overall consistent with an m = 5 Aurivillius phase structure. In this section we will examine what factors influence deviations from this. |
672c882ef9980725cf6d6ca9 | 6 | For Series 1 (Bi6Ti3Fe1.6-αMn0.4+αO18 (α = 0, 0.1, 0.2, 0.3 and 0.4)), the concentrations of Fe and Mn are varied, while Ti concentration remains constant. In other words, the Ti : Mn/Fe stoichiometric ratio of 3 : 2 does not change across Series 1. This was considered isovalent substitution since both Fe and Mn are nominally in the 3+ oxidation state (o.s.) in a neutral B6TFMO Aurivillius stoichiometry. They also have similar ionic radii as noted above, although it is acknowledged that the Jahn-Teller effect can tend to elongate Mn 3+ O6 octahedra. The XRD plots of sample Series 1 shown in Fig. show that the peak positions, heights and widths correlate well with a computed XRD diffraction pattern for a refinement of a reference m = 5 structure shown in Supplementary Information Section SI3.1, Fig. . The reference pattern is computed with an assumed 90% preferred crystallite (00l) orientation, a nominal crystal size of 30 nm and an isotropic strain of 0.1%, to give a line breadth similar to that observed in the m = 5 phase in Fig. . The 30 nm crystallite size inferred from XRD analysis is between 0.5 to 0.3 that of the observed crystallite thickness in cross section TEM as shown in Figures ) and (f), an observation that is probably indicative that the crystallite coherence due to structural disorder is significantly smaller than the film thickness. As the concentration of Mn increases and Fe decreases, there is only a slight shift (~0.4 °) in 2θ peak positions accompanied by peak broadening. |
672c882ef9980725cf6d6ca9 | 7 | The Aurivillius phase structure is highly anisotropic, and the films show strongly-preferred caxis orientation, as is evidenced by the majority of XRD peaks displaying (00l) reflections. TEM images (Fig. ) of samples from Series 1 confirm the layered m = 5 structure and verify the preferential growth of grains in directions normal to the stacking axis (c-axis) of the layers. The grains crystallize with plate-like morphology with in-plane grain size ranging from 10 nm to 2 μm (Fig. ). The crystallites overlap one another at grain boundaries (Fig. ), meaning that thickness variation (typically 89 nm ± 33 nm) is observed across the samples, with RMS roughness ranging from 16 to 27 nm. |
672c882ef9980725cf6d6ca9 | 8 | It is evident from XRD analysis of Series 2 (Fig. ), that as Ti content decreases and Mn content increases, there is a clear gradual shift in peak position away from the m = 5 structure and towards the m = 6 Aurivillius phase. As β increases from 0 to 0.6, the 2θ position of the (008) reflection shifts from 13.6° to 14.4° and the position of the (0010) reflection shifts from 17.6° to 18.0°. While the m = 5 structure is maintained, these shifts correspond to movements towards the expected positions for the (0010) and (0012) for an m = 6 structure. A similar trend in asymmetric peak shift on aliovalent substitution is observed in Series 3 (Bi6Ti3.2-γFe1.3+γMn0.5O18 (γ = 0, 0.1, 0.2, 0.3, 0.4, 0.5 and 0.6, Fig. )), where the B-site concentration of Ti is decreased, and the concentration of Fe is increased. For example, as increases from 0 to 0.6, the 2θ position of the (008) reflection moves from 13.4° to 14.5° and the position of the (0010) reflection moves from 17.5° to 18.1°. |
672c882ef9980725cf6d6ca9 | 9 | In Series 4 (Bi6TixFeyMnzO18 (x = 2.3 to 2.8, y = 1.52 to 2.00, z = 0.68 to 0.75)), the concentration of Ti was further reduced. The compositions were selected based on a compositional survey of 55 individual Aurivillius phase grains from a previous study of a mixed m phase B6TFMO sample. This series explores aliovalent substitution of Ti by both Mn and Ti, allowing |
672c882ef9980725cf6d6ca9 | 10 | clearly demonstrate a shift from a pure m = 5 phase (in Bi6Ti2.7Fe1.62Mn0.68O18 (see The XRD pattern for Bi6Ti2.3Fe2.0Mn0.7O18, with the lowest Ti content of the samples investigated, predominantly exhibits m = 6 characteristics, closely matching a simulated pattern for a refined reference model, as discussed in Supplementary Information Section SI3.2 and shown in Figure . A summary of the results from the two-phase XRD modeling results is provided in Table , showing a clear trend toward the structural evolution of higher-layered Aurivillius phases to accommodate increased magnetic cation substitution. |
672c882ef9980725cf6d6ca9 | 11 | The full width half maximum (FWHM) values of the XRD peaks for the m = 5 phase increase as Ti concentration decreases across Series 4 (see Table ). While particle size effects could contribute to peak broadening, the consistent deposition method applied throughout the series should limit significant variations in film thickness and grain size. Additionally, XRD pattern modeling (see Supplementary Information Section SI3) indicates that no single crystallite size can fully account for the broadening observed in all reflections, as the peaks widen independently of crystal size. Increased structural stacking disorder, such as stacking faults, OPB defects and small regions of phases with different m values (evident in the TEM data in Fig. ), almost certainly contributes to the broadening of the XRD peaks, which is reflected in the FWHM values. The increase in peak broadening for the m = 5 phase implies a more disordered periodic structure as x values decrease. Furthermore, peak broadening for the m = 6 phase is over twice that of the m = 5 reflections at the same 2 theta value, indicating that the m = 6 phase is more defective than the m = 5 phase. |
672c882ef9980725cf6d6ca9 | 12 | The shift to the higher m = 6 Aurivillius phase is interesting, given that access to larger layering periods is constrained by thermodynamics. The difference in formation enthalpy between different m-phases of similar composition is often too subtle to stabilize one phase relative to another, therefore it appears that the chemical modifications are responsible for stabilizing the formation of the higher m-Aurivillius phase. |
672c882ef9980725cf6d6ca9 | 13 | TEM analysis (Fig. ) confirmed the increased structural rearrangement to m = 6 layered units as the Ti content within the films decreases. The thickness of a unit cell along the growth (c) axis of the Aurivillius structure can be estimated using the formula: c/2 = f + h, where h = m*p.f refers to the thickness of the [Bi2O2] 2+ interlayer, which is estimated to be 4.08 Å. The average thickness of the perovskite blocks (p) is ≈ 4.11 Å and m refers to the number of perovskite blocks per halfunit cell of the Aurivillius phase structure. Thus, c is calculated to be 41.04 Å for an m = 4 structure, c is 49.26 Å for an m = 5 structure and c is 57.48 Å for an m = 6 structure. This correlates well with the c-axis lattice parameters in the literature physically determined by x-ray and powder neutron diffraction data. TEM imaging supports the XRD findings that the Bi6Ti2.8Fe1.52Mn0.68O18 composition is primarily structured as the m = 5 phase, with a measured c-axis length of 4.6 ± 0.1 nm, which is within 10% instrumental error of the expected 4.9 nm (Fig. |
672c882ef9980725cf6d6ca9 | 14 | Aurivillius phase materials are ferroelectrics in which the polarization is predominantly along the a-axis, in-plane direction. We employed lateral piezoresponse force microscopy (PFM) to confirm in-plane piezoelectricity for samples with three different compositions: However, the vertical piezoresponse decreased to 3.6 pm/V for Bi6Ti2.50Fe1.77Mn0.73O18 (50% Ti at B-site, Fig. ). Furthermore, hysteresis shape is not concave for the Bi6Ti2.50Fe1.77Mn0.73O18 |
672c882ef9980725cf6d6ca9 | 15 | (50% Ti at B-site) composition, resembling characteristics of a lossy dielectric. For the Fecontaining Bi4Bim-3Fem-3Ti3O3m+3 phases, it was previously shown that the leakage current increased as m increased from 4 to 6. The presence of non-ferroelectric secondary-phase impurities within the Bi6Ti2.50Fe1.77Mn0.73O18 sample, constituting a volume fraction of 2.7 vol.% (See Supplementary Information section SI4.3), could also be a factor influencing domain wall pinning, the reduced polarization value, the increased coercive field and an altered hysteresis loop shape. Furthermore, the presence of trace levels of secondary phase impurities would complicate attempts to measure a magnetic signal intrinsic to the main Aurivillius phase, therefore magnetic measurements of the new compositions could not be reliably performed in this work. |
672c882ef9980725cf6d6ca9 | 16 | Ferroelectric lithography investigations (Fig. )) confirm the trend that higher DC fields are required for vertical domain switching as the B-site Ti concentration decreases and the presence of the m = 6 phases increases. Applying a vertical DC bias of 50 V to specific regions on the thin film surface during the "write" step resulted in ferroelectric polarization reversal in both the Bi6Ti2.80Fe1.52Mn0.68O18 (56% Ti at B-site) and Bi6Ti2.50Fe1.77Mn0.73O18 (50% Ti at B-site) samples, as detected by the subsequent PFM scan during the "read" step ((Fig. (f), (g), (j), (k)). However, in the Bi6Ti230Fe1.95Mn0.75O18 (46% Ti at B-site) sample, a higher DC bias of 90 V was necessary for vertical domain switching (Fig. ). |
672c882ef9980725cf6d6ca9 | 17 | Newnham observed that "substantial substitution" of B-site metals "is obtained for only a narrow size range of octahedral ions": 0.58 A° to 0.65 A°. It is therefore expected that there will be a limit to how much Mn 4+ (octahedral site ionic radius 0.53 Å) can be accepted in the m = 5 B6TFMO structure. While the precipitation of trace levels (2.7 vol.%) of secondary phases has been observed for the Bi6Ti2.3Fe1.95Mn0.75O18 composition and may assist with compensating some charge/size effects (Supplementary Section SI4.3), overall, XRD analysis (Fig. and Fig. ) and TEM imaging (Fig. ) clearly demonstrates that the m = 5 Aurivillius structure adapts to a higher m = 6 layered homologue to accommodate increased levels (>46%) of magnetic cations. |
672c882ef9980725cf6d6ca9 | 18 | Given that bismuth is used in excess during the synthesis, structural and compositional rearrangement from an m = 5 phase to an m = 6 phase is permitted based on the concentrations of cations available. To correlate the charge balancing requirements with the compositions where structural adaptation occurs, we tabulate the nominal B-site proportion of Ti, Fe and Mn for the B6TFMO samples from a spread of compositions with decreasing Ti concentration in Supplementary Table . Next, we assume that Ti has a +4 valence, Fe has a +3 valence and we let Mn have a valence of either +3 or +4. We then assume that the proportion of Mn 4+ present is proportional to the concentration necessary to maintain a charge balanced m = 5 Aurivillius structure (see Supplementary Information SI5). Note that the displayed oxidation states are theoretical values based on nominal, assumed phase-pure compositions and are not taken from experimental measurements. The table is arranged in order of decreasing Ti concentration, and we correlate the compositions with the observations (from XRD and TEM data) of whether the sample shows an overall m = 5, m = 6 or mixed m = 5/6 Aurivillius phase. Supplementary Table follows the same trend as Table and summarizes that the appearance of the m = 6 phase (as observed in the XRD and TEM analysis) follows a systematic trend with decreasing Ti concentration. More precisely, the structural rearrangement occurs when the combined concentration of Fe and Mn at the B-site of the perovskite units is above 46% (see Supplementary Information Table ). Furthermore, when the nominal concentration of Mn 4+ is greater than 6.6%, meaning that the theoretical average oxidation state of Mn in the m = 5 structure is larger than ≈ 3.4, the appearance of the m = 6 phase begins to occur. It appears that the cation radius disparity restricts the maximum concentration of Mn 4+ that can be accommodated by the m = 5 structure, with a threshold Mn 4+ solubility limit of between 6.6 to 8.0% (or 46 to 48% magnetic ion content at the B-site). We note that this is a similar average oxidation state threshold to that previously observed in other Aurivillius phases systems by Zurbuchen et al. (Mn 3.2+ ) and |
672c882ef9980725cf6d6ca9 | 19 | With direct evidence from TEM and XRD analysis showing the evolution of the m = 6 phase as the concentration of Ti decreases and the levels of Fe and Mn increase, we next explore how the structural rearrangement of the Aurivillius phase to higher m values accommodates the increased Mn content. Theoretical scenarios are outlined in Table . For example, consider the chemical composition of the m = 5 Bi6Ti2.30Fe2.00Mn0.70O18 phase. When rearranged into the m = 6 structure, it assumes the composition Bi7Ti2.76Fe2.40Mn0.84O21. In Table , for the m = 5 phase (composition Bi6Ti2.30Fe2.00Mn0.70O18), the average oxidation state of Mn is calculated to be +4.00, assuming oxidation states of +4 for Ti and +3 for Fe. When the same ratio of Ti, Fe and Mn cations is applied to the m = 6 structure (composition Bi7Ti2.76Fe2.40Mn0.84O21), the average oxidation state of Mn is calculated to be +3.29. The value falls within the range reported for stable Aurivillius phase compositions by Zurbuchen et al. (Mn 3.2+ ) and McCabe and Greaves 21 (Mn 3.4+ ). It is important to note that while the relative proportions of the B-site atoms remain the same for both arrangements (as shown in Table ), the proportion of Mn⁴⁺ required to achieve charge-balanced stoichiometry is significantly lower for the m = 6 phase (4.06% Mn⁴⁺) compared to the m = 5 phase (14.00% Mn⁴⁺). This explains how the formation of the higher m = 6 Aurivillius phase allows for the accommodation of a higher concentration of magnetic cations while maintaining balanced stoichiometry. |
672c882ef9980725cf6d6ca9 | 20 | We propose that the inability of the m = 5 Aurivillius structure to accommodate Mn⁴⁺ concentrations above 6.6% at the B-site likely drives the structural rearrangement observed, leading to the formation of a higher m-phase, as evidenced by XRD and TEM analysis. This rearrangement, along with an increased number of A-site cations, is posited to enhance the tolerance factor of the perovskite unit (see Supplementary Information, Section SI7), as shown in Table . The structural adaptation to incorporate higher magnetic cation content in the B6TFMO |
672c882ef9980725cf6d6ca9 | 21 | Aurivillius phases is summarized in the composition map in Fig. , where the Mn concentration varies from 0.60 to 0.75, complementing the modelled XRD data in Table . This observation of structural reorganization into a higher-layered m = 6 Aurivillius phase to accommodate increased magnetic fraction is particularly interesting. At large values of m, the free energy of formation in Aurivillius phases becomes thermodynamically degenerate, making the synthesis of higher layered (m > 5) homologues challenging. These phases are typically difficult to achieve without the support of epitaxial substrates or kinetically controlled growth conditions. This work progresses previous studies by Sun et al. The m = 7 Sr4Bi4Ti7O24 phases were synthesized in the kinetic limit using pulsed laser deposition on epitaxial (SrTiO3 (001)), LaAlO3 (001)c, and SrRuO3 (001)c/SrTiO3 (001)) substrates. Alternatively, the work presented in this contribution demonstrates that using transition metal cations for aliovalent substitution in straightforward chemical solution deposition processes can overcome thermodynamic challenges in synthesizing higher-layered Aurivillius homologues. This approach has eliminated the need for epitaxial substrates or kinetically limited growth processes for this Bi/Ti/Mn/Fe-based material system. |
672c882ef9980725cf6d6ca9 | 22 | This study contributes a comprehensive understanding of structural adaptation to compensate for charge and size differences when accommodating higher concentrations of magnetic cations in multiferroic Aurivillius phases. The initial investigation focused on determining the solubility limit for magnetic Fe and Mn cations within the m = 5 Aurivillius phase structure. XRD and TEM analyses of Bi6TixFeyMnzO18 (x = 2.3 to 3.2, y = 1.2 to 2.0, z = 0.3 to 0.9) compositions revealed that when the Ti 4+ content was lowered below 54%, a discernible transformation occurs, leading to the formation of a mixed-phase sample consisting of both the main m = 5 Aurivillius phase and the m = 6 phase. The prevalence of the m = 6 phase within the m = 5 phase matrix increased with decreasing Ti 4+ content. Upon elevating the nominal Mn 4+ content to 14%, the m = 5 structure transforms into a single-phase m = 6 structure, with each perovskite block now containing nominal 4%Mn 4+ . The transition to the higher-layered Aurivillius homologue is interesting, especially considering the thermodynamic constraints on expanding layering periods. It suggests that chemical modifications are pivotal in stabilizing the formation of the m = 6 Aurivillius phase. This transition to the higher m = 6 phase reduces the relative fraction of Mn 4+ required for charge neutrality in the material (refer to Table ). Consequently, the structure becomes more tolerant of higher magnetic cation content. Correspondingly, the average calculated oxidation state for Mn also decreases, falling within the range of values reported previously for stable Aurivillius phase compositions. |
672c882ef9980725cf6d6ca9 | 23 | Moreover, the percolation fraction, which is crucial for long-range magnetic ordering, also increases upon phase transformation. For instance, in the m = 6 composition Bi7Ti2.76Fe2.34Mn0.90O21, the magnetic cation content reaches 54% of B-site cations, which is 14% higher than the m = 5 multiferroic Bi6Ti2.99Fe1.46Mn0.55O18 composition that exhibits an MS value of 215 emu/cm 3 . 11 As expected from crystal symmetry, the minor out-of-plane ferroelectric response diminishes with increasing structural reorganization towards the m = 6 phase. Notably, the predominant in-plane piezoresponse of these materials is unaffected by magnetic cation substitution. |
672c882ef9980725cf6d6ca9 | 24 | This study provides valuable insights into the limiting factors that govern magnetic ion substitution at the B-site of Aurivillius materials, which should be considered during the development of new multiferroic materials. Furthermore, it is shown how aliovalent substitution can overcome the thermodynamic challenges in synthesizing higher-layered Aurivillius homologues, eliminating the requirement for epitaxial growth or kinetically constrained methods. |
638dbc1f44ccbc40080eb749 | 0 | In an organic light-emitting diode (OLED), electrons and holes, injected in the device from opposite electrodes, combine to form excitons. According to spin statistics, 25% of the electrically generated excitons are in a singlet state and 75% are in a triplet state. If the excitons form on a fluorescent molecule, emission will only occur from the singlet excitons, effectively limiting the device Internal Quantum Efficiency (IQE). |
638dbc1f44ccbc40080eb749 | 1 | Phosphorescent dyes harvest both singlet and triplet excitons to emit light from the triplet excited state resulting in an IQE of up to 100%. Most phosphorescent emitters, however, contain noble metals such as platinum or iridium, among the scarcest elements on Earth. Thermally activated delayed fluorescence (TADF) offers a different yet equally appealing strategy to triplet harvesting where 100% IQE is also possible in the device. In TADF emitters, dark triplet excitons are thermally upconverted into emissive singlets via reverse intersystem crossing (RISC). RISC is possible when the energy gap, DEST, between the lowest lying singlet and triplet excited states is of the order of the thermal energy (ca. < 0.02 eV), provided that spin-orbit coupling (SOC) between the two states is nonnegligible. To minimize ΔEST, the overlap between the HOMO and LUMO of the molecule must be reduced, localizing the two orbitals in separate electron-donating (for the HOMO) and electron-accepting (for the LUMO) parts of the molecule. This separation is most often obtained by enforcing a large dihedral angle between the electron donor and acceptor moieties. This strategy, however, leads to a reduction of the fluorescence efficiency and, according El Sayed rule, to a reduction of the spin-orbit coupling between the singlet and triplet states as well. |
638dbc1f44ccbc40080eb749 | 2 | Two of the most common moieties employed as donors and acceptors, respectively, are 9,9-dimethy-9,10-dihydroacridine (DMAC) and 2,4,6-triphenyl-1,3,5-triazine (TRZ), and the combination of the two produces the emitter DMAC-TRZ, first reported by Tsai et al. DMAC-TRZ shows a very high photoluminescence quantum yield, ΦPL, of 90%, at lPL of 495 nm, as an 8 wt% doped film in mCPCN [9-(3-(9H-carbazol-9-yl)phenyl)-9Hcarbazole-3-carbonitrile]. In DMAC-TRZ, the DEST amounts to a few tens of meV, depending on the host matrix, with a delayed fluorescence lifetime of 1.9 µs in mCBPBN, in line with an efficient RISC process. The OLED shows a high maximum external quantum yield, EQEmax, of 26.5% at lEL of 500 nm, while devices prepared from neat DMAC-TRZ films show a comparable EQEmax of 20.0%. The modest reduction of the EQEmax at high concentration can be understood in terms of the orthogonal conformation of the emitter, which effectively prevents aggregation. Both the doped and non-doped devices show a relatively small efficiency roll-off, with an EQE100 of 25.1% and 18.9%, respectively. |
638dbc1f44ccbc40080eb749 | 3 | The emitters a-DMAC-TRZ and MA-TA are derivatives of DMAC-TRZ where adamantane groups are incorporated into the structure (Figure ). In a-DMAC-TRZ, the adamantane-functionalization of the donor moiety leads to a deformed structure that results in an increased optical gap and thus a bluer emission. Dual fluorescence is observed from two different conformers, but overall, the electroluminescence (EL) performance remains similar to the DMAC-TRZ OLED, with an EQEmax of 28.9%, yet with a lEL of 488 nm. In the report by Wada et al., the replacement of the distal phenyl moieties on the TRZ with adamantyl groups results in a weaker acceptor, leading to a blue-shifted electroluminescence compared to DMAC-TRZ. The adamantyl substitution also reduces the non-radiative decay leading to a ΦPL of 99%. The blue solution-processed device (lEL of 475 nm) shows an EQEmax at 22.1%. Conversely, replacement of the distal phenyl rings in TRZ by electron-withdrawing pyrimidines (DMAC-bPmT) results in a redshifted emission (lPL of 520 nm vs 500 nm for DMAC-TRZ in toluene). The delayed emission lifetime of DMAC-bPmT is 3.3 µs in toluene, which is shorter than that of DMAC-TRZ at 8.8 µs in the same medium. The RISC rate constant, kRISC, of DMAC-bPmT is 8.8×10 5 s -1 , is three times faster than that of DMAC-TRZ (2.9×10 5 s -1 ). However, its ΦPL of 70% in toluene is reduced compared to that of DMAC-TRZ (FPL = 93% in toluene). |
638dbc1f44ccbc40080eb749 | 4 | Rajamalli et al., and Dos Santos et al., showed that the introduction of a heteroaromatic bridge in sulfone-based D-A TADF emitters can prevent structural relaxation, enhance FPL and improve the color purity due to the narrower emission, and demonstrated an improvement in the efficiency of the devices compared to that with the reference emitter pDTCz-DPS. The materials in the study of pDTCz-2DPyS, and pDTCz-3DPyS, both show good FPL of ca. 60%. The blue OLEDs, at lEL of 466 nm and 452 nm for the devices with pDTCz-2DPyS, and pDTCz-3Dpy, respectively, showed EQEmax ~12-13%, which are considerably improved over the parent device with pDTCz-DPS (EQEmax of 4.7%). Dos Santos et al., showed that the addition of second nitrogen atom within the bridging heterocycle in pDTCz-DPzS, and pDTCz-DPmS contributed to a further enhancement of the EQEmax to 18% and 14%, respectively, at lEL of 522 nm and 461 nm for the devices with pDTCz-DPzS, and pDTCz-DPmS, respectively. PXZ-BOO shows TADF, relevant devices having EQEmax of 19.4% at λEL of 528 nm. In THF solution, PXZ-PPO is present as a mixture of two conformers, a more planar structure with N-H interaction and a twisted structure, where hydrogen bonding is not present. The planar conformation emits in the deep blue at 420 nm but shows no TADF, while the twisted conformer shows green TADF (610 nm) but with a very short lifetime of 170 ns. |
638dbc1f44ccbc40080eb749 | 5 | The planar conformer is dominant in the crystalline phase, while in solution the twisted conformer is largely responsible for the observed photophysics, giving rise to TADF. The PXZ-PPO-based device, with EQEmax of 14.1% at λEL of 528 nm, is slightly inferior to the OLED based on PXZ-BOO, which has an EQEmax of 19.4% at λEL of 528 nm. |
638dbc1f44ccbc40080eb749 | 6 | The presence of two conformers of a TADF emitter has also been documented by Shi et al. in the TRZ derivative compounds TP2P-PXZ and TP5P-PXZ. Phenoxazine and a central pyridine bridge were used to promote the formation of an intramolecular hydrogen bond in TP2P-PXZ, while in the control compound TP5P-PXZ this interaction is absent. The presence of quasi-equatorial (QE) and quasi-axial (QA) conformations of TP2P-PXZ led to a self-doped system, where the QA conformer effectively acts as the host material. This led to an efficient OLED with an EQEmax of 25.4% at λEL of 548 nm while the performance of the device with TP5P-PXZ was somewhat attenuated with an EQEmax of 14.6% at a slightly red-shifted λEL of 560 nm. |
638dbc1f44ccbc40080eb749 | 7 | These examples show that modification of the nature of the aromatic bridge in a TADF emitter can lead to significant changes in the molecular geometry, with the co-existence of different conformers that show distinctive photophysics. In this work, we introduce the emitter DMAC-py-TRZ, where the phenylene bridge in DMAC-TRZ is replaced by a 2pyridyl bridge. DMAC-py-TRZ emits at lPL of 539 nm and has a FPL of 58% in toluene solution while as a 10 wt% doped mCP film the lPL is 496 nm and the FPL is 57%. Its crystal structure (Figure ) documents a small dihedral angle between the DMAC and pyridyl bridge of 19.7(2)° and a V-shaped or bent structure of the DMAC donor, with an associated bending angle (deviation from a planar conformation) of 45°. This behaviour is in line with that observed by Shi et al. We present an in-depth computational study and an extensive optoelectronic and photophysical characterization that showcases the impact that conformational changes in the excited state and not just in the ground state have on the photophysics of the compound. |
638dbc1f44ccbc40080eb749 | 8 | The ground-state geometries of DMAC-TRZ and DMAC-py-TRZ were optimized in the gas phase using density functional theory (DFT) at the M062X/6-31G(d) level of theory. The excited-state energies were calculated using time-dependent density functional theory (TD-DFT) within the Tamm-Dancoff approximation at the same level of theory (TDA-DFT). We employ the term "orthogonal" to describe the structure with a dihedral angle between the DMAC and the bridge that is close to 90 o and we dub as "bent" the structure with the small dihedral angle and V-shaped geometry of the DMAC. |
638dbc1f44ccbc40080eb749 | 9 | In a recent publication, an extensive computational analysis of DMAC-TRZ set the basis for a few-state model that was carefully validated against spectroscopic properties in solution. Then the same model was exploited to calculate ISC and RISC rate constants, also accounting for environmental effects including dielectric and conformational disorder. In the ground-state equilibrium geometry of DMAC-TRZ, the DMAC and TRZ moieties are mutually orthogonal (Figure ), in line with the crystal structure. In this orthogonal geometry, the S1 and T1 states each have a pure charge transfer (CT) character. In other terms, the HOMO and LUMO have negligible overlap so that the singlettriplet gap is almost closed, with DEST = 0.01 eV. The close similarity between the orbitals involved in S1 and T1 states implies a vanishing SOC, according to El Sayed's rule 6 , as to hinder direct RISC from T1 to S1. |
638dbc1f44ccbc40080eb749 | 10 | As for the excited states of DMAC-TRZ, S1 (with a strong 1 CT character) retains an orthogonal geometry, while T1 ( 3 CT) undergoes a large conformational deformation where the dihedral becomes ~60 °. At this angle, ΔEST increases as does SOC. A rigid scan of the dihedral angle of DMAC-TRZ (Figure ) is informative. Specifically, starting from the optimized ground-state geometry, we calculated the ground and excited state energies upon gradual rotation of the DMAC unit about the phenylene bridge without allowing for any additional molecular relaxation (the dihedral angle for the scan is defined in Figure ). The resulting S0, S1, T2 and T3 potential energy surfaces (PES) all show a flat minimum for the orthogonal geometry, while T1 shows a double minimum around (90±30) ° angle. The molecular orbitals (MOs) and natural transition orbitals (NTOs) |
638dbc1f44ccbc40080eb749 | 11 | In contrast with DMAC-TRZ, the ground-state optimized geometry of DMAC-py-TRZ has a bent structure (Figure ), as also observed in the crystal structure (Figure ). To better understand the structural differences between DMAC-TRZ and DMAC-py-TRZ, a rigid dihedral angle scan of DMAC-py-TRZ starting from an analogous orthogonal conformation to that of DMAC-TRZ has been performed. The rigid scan leads to a qualitatively similar picture for the two compounds (Figure ), with relevant MOs and NTOs in Figures and, showing the HOMO and LUMO localized on the donor and acceptor moieties, respectively. The presence of the nitrogen atom in the pyridine bridge effectively increases the electron-withdrawing strength of the acceptor, resulting in a stabilized LUMO and a smaller HOMO-LUMO gap in DMAC-py-TRZ (EHOMO -LUMO = 4.78 eV for the orthogonal structure) vs DMAC-TRZ (EHOMO -LUMO = 4.99 eV for the orthogonal structure). Accordingly, the S1 and T1 excitations occur at lower energy in DMAC-py-TRZ than in DMAC-TRZ and both the S1 and T1 states are stabilized compared to those of and). The rigid energy scan, however, points to a large increase of the ground-state energy when the dihedral angle deviates significantly from orthogonality, so that non-orthogonal conformations are hardly accessible. To address the bent conformer, we performed a relaxed scan of the dihedral angle, relevant results being shown in Figure . The relaxed scans show that for each of the emitters, two minima are present, corresponding to the orthogonal and bent structures. For DMAC-TRZ, the energy difference between the two conformers amounts to 0.04 eV, slightly larger than thermal energy at room temperature. The energy barrier for the interconversion between the two conformers, 0.22 eV (21.2 kcal/mol), is, however, much larger than thermal energy so that only the orthogonal geometry is expected to be significantly populated at room temperature. The situation is very different for DMAC-py-TRZ where the bent conformer (dihedral angle ~10 o ) is lower in energy than the orthogonal conformer by 0.20 eV and the energy barrier for the interconversion between the bent and orthogonal conformers is 0.20 eV (19.3 kcal/mol). Thus, at room temperature only the bent conformer is populated. The MOs and NTOs calculated for the bent structure (Figures and) |
638dbc1f44ccbc40080eb749 | 12 | For DMAC-TRZ, the rigid (Figure ) and the relaxed scans (Figure ) lead to the same picture: the S1 state maintains the same orthogonal conformation as the ground state, while the T1 state is stabilized and adopts a twisted structure (dihedral angle: ~60°). Full optimisations of S1 and T1 confirm this result. In the orthogonal geometry, the S1 state is an almost pure CT state and, hence, has a negligible oscillator strength. The scenario is much more interesting for DMAC-py-TRZ. In the bent geometry (the energy minimum, Figure ), the vertical excitation energy to S1 amounts to 4.1 eV and the DEST is 0.78 eV, which is far too large for TADF to be operational at ambient temperature. Moreover, the oscillator strength for the S0→S1 is large (1.27, Figure ) in this geometry due to the significant overlap of the orbitals involved in the transition (Figure ). However, the bent geometry is not the equilibrium geometry for S1 (Figure ) and a huge structural deformation is predicted in the S1 state from the bent to the orthogonal structure. In other terms, in DMAC-py-TRZ the absorption occurs from the bent geometry and the lowest energy transition is both high in energy and has a large oscillator strength. By contrast, fluorescence occurs from the orthogonal structure at a much lower energy (3.17 eV) and with negligible oscillator strength (as per the non-overlapping orbitals, in Figure ). In this orthogonal geometry the DEST reduces to only 8.2 meV), making TADF possible. |
638dbc1f44ccbc40080eb749 | 13 | In DMAC-py-TRZ, the T1 state is predicted to exist as a twisted geometry in both the rigid (Figure ) and relaxed scans (Figure ). The full optimization of the excited state geometry, however, yields conflicting results. For S1 the situation is clear: taking either the bent or the orthogonal geometry as starting points for the excited state optimization, the S1 geometry (Figure ) always converges to the orthogonal conformation, supporting the results from the relaxed scan analysis. For T1, instead, two different structures are reached depending on the starting geometry (Figure ), with slightly different ΔE !"#$" values (0.93 eV where the dihedral angle is 30 o and 0.41 eV where the dihedral angle is 60 o ). The energy of the two triplet conformations is similar (Δ𝐸 %" !"°$%"°= 0.07 eV), so that a firm conclusion about the equilibrium geometry for T1 cannot be reached. The CT band of DMAC-py-TRZ at 370 nm is slightly blue-shifted with respect to DMAC-TRZ at 382 nm. The most striking difference is, however, in the much larger intensity of the band measured for DMAC-py-TRZ (e =43,800 M -1 cm -1 ) vs DMAC-TRZ (e = 2100 M -1 cm -1 ). This is a direct consequence of the different conformations adopted by the two compounds in the ground state: the orthogonal conformation of DMAC-TRZ (observed in the crystal structure and predicted by DFT) hinders an effective conjugation and suppresses the intensity of the low-energy CT transition. On the other hand, the bent conformation of DMAC-py-TRZ (observed in the crystal structure and predicted by DFT) promotes an efficient conjugation of the two moieties, perfectly in line with the oscillator strength calculated with TD-DFT (oscillator strengths are reported in Figure , together with NTOs). |
638dbc1f44ccbc40080eb749 | 14 | show a strong positive solvatochromism akin to that observed for DMAC-TRZ (Figure and Ref. 20) The large positive PL solvatochromism observed for both compounds suggests that the emissive excited state has a large permanent dipole moment, thus confirming the CT character of this state in both compounds. The progressive broadening of the emission band in the solvatochromism study is a result of polarityinduced inhomogeneous broadening. The well-resolved vibronic structure of the emission band in non-polar solvents is often considered an indication of a local nature of the relevant excitation. Therefore, to prove the CT nature of the lowest transition in DMAC-py-TRZ in all solvent, including non-polar ones, Figure shows spectra collected for the two molecular fragments, DMAC and py-TRZ. For py-TRZ, only absorption spectra are shown since the species is not emissive. The spectroscopic features of both molecular fragments are located at higher energies than the lowest energy feature seen in DMAC-py-TRZ, confirming that this specific feature is related to a CT state. An important and unusual result is recognized in the large Stokes shift observed for DMAC-py-TRZ in non-polar solvents (Table ): in methylcyclohexane, the absorption band is located at 370 nm, while the emission is seen at 472 nm, amounting to a Stokes shift of ~0.7 eV. This large Stokes shift can only be explained in terms of a very large molecular relaxation upon photoexcitation, well in line with the TD-DFT results that predict the relaxation of the S1 state from the bent to the orthogonal geometry. |
638dbc1f44ccbc40080eb749 | 15 | In degassed toluene, DMAC-TRZ and DMAC-py-TRZ have similar ΦPL of 67% and 58%, respectively, in line with emission originating in both compounds from a similar orthogonal geometry. The FPL decrease in air (ΦPL = 22% and 17%, respectively), indicating the presence of accessible triplet excited states. The prompt and delayed lifetimes, tp and td, for DMAC-TRZ in degassed toluene are of 20.8 ns (1.1%) and 5.2 µs (98.9%), in line with those previously reported, while the tp and td for DMAC-py-TRZ are 44.0 ns (8.1%) and 1.5 µs (91.9%), respectively (Figure ). |
638dbc1f44ccbc40080eb749 | 16 | However, DMAC-py-TRZ undergoes a significant geometric reorganization from the bent geometry in the ground state to the orthogonal geometry in the excited state. Spectra collected in a glassy 2-MeTHF matrix at 77 K (Figure ) shed further light on the geometrical relaxation of DMAC-py-TRZ upon photoexcitation. In the frozen matrix, the emission peaks at 404 nm, blue-shifted compared to that in liquid 2-MeTHF at ambient conditions (lPL = 596 nm, Table and Figure ). Apparently, in the frozen matrix the excited compound cannot relax, so that emission occurs from the bent structure and hence peaks at much higher energy than in the (non-polar) liquid solvent. The gated signal collected in the glassy matrix (red line in Figure ) is ascribed to phosphorescence, suggesting a large DEST for this conformer under these conditions, again in line with that calculated for the bent structure. |
638dbc1f44ccbc40080eb749 | 17 | The PL behavior of DMAC-py-TRZ was also characterized in polyTHF, a viscous solvent where conformational relaxation is hindered. Interestingly, two emission bands are observed in this viscous medium (Figure ). The first emission band at 410 nm is similar to the one observed in the glassy matrix at 77 K while the second emission at 600 nm is similar to the emission observed in DMSO at room temperature. Apparently, at ambient temperature, the excited state relaxation, fully hindered in glassy matrices at low temperature, is only partially hindered in the viscous polyTHF. Accordingly, the presence of the two emission bands is evidence of the simultaneous presence of the unrelaxed (bent) emissive species (as in the glassy matrix) as well as of relaxed (orthogonal) species (as in the liquid solvent). |
638dbc1f44ccbc40080eb749 | 18 | Having observed that the S1 relaxation of DMAC-py-TRZ in frozen glassy matrices at low temperature is fully hindered while it is only partially hindered in viscous solvents at ambient condition, we next transitioned to an investigation of the behavior of this compound in amorphous matrices where large geometric reorganization is also likely to be hindered. Spin-coated thin films of DMAC-py-TRZ doped into PMMA at 10 wt% were first prepared (Figure ). Emission at lPL of 516 nm was observed with a ΦPL of 63.8% under a N2 atmosphere, which decreased to 58.0% upon exposure to oxygen. Biexponential decay kinetics were observed in the time-resolved PL measurements, with τp of 26.0 ns and an average τd of 4.7 µs [τ1=1.0 µs (32.6%), τ2=7.4 µs (67.4%)], |
638dbc1f44ccbc40080eb749 | 19 | respectively. The presence of a delayed fluorescence suggests that at least some emitter molecules adopt an orthogonal conformation, as to allow for TADF. Compared to the data obtained in toluene (tp of 44.0 ns and td of 1.5 µs), DMAC-py-TRZ possesses a shorterlived prompt component and a slightly longer-lived delayed component. We then investigated the photophysics in mCP (1,3-bis(N-carbazolyl)benzene) as the host matrix, a suitable high triplet energy host for both compounds that would be relevant for OLEDs. |
638dbc1f44ccbc40080eb749 | 20 | Figure shows results at a 10 wt% doping concentration. The emission in mCP is blueshifted at lPL of 496 nm, compared to that of the doped PMMA film. The ΦPL of the doped film in mCP is 57.4% under N2, which decreased to 53.5% in air. The τp = 24.9 ns and the average τd = 5.3 µs [τ1=1.4 µs (46.8%), τ2=7.7 µs (53.2%)]. In both PMMA and mCP matrices at 10% doping the delayed emission is thermally activated (Figure ). However, extracting detailed information from such highly doped matrices is dangerous because of spurious phenomena, including homo energy-transfer and inner filter effects (selfabsorption). To minimize spurious concentration effects, low-concentration (down to 1 wt%) films were fabricated, relevant spectra being shown in Figure . In these films both the high frequency emission originating from the bent structure, and the low-frequency emission from the orthogonal structure, are present, suggesting that both conformers are present in all films. The high frequency emission decays much more rapidly (Figure ), again confirming that it originates from the bent conformer. Upon increasing concentration, the high frequency emission progressively weakens and disappears for doping concentrations above 3 wt%. Two phenomena may exist to explain this observation, both related to the large transition dipole moment (large oscillator strength) of the S0→S1 transition in the bent geometry: (1) self-absorption; and (2) energy transfer from the bent to the orthogonal structure. Both phenomena are expected to become more efficient upon increasing the concentration of the emitter in the host matrix. |
638dbc1f44ccbc40080eb749 | 21 | The synthesis of a new TADF emitter, DMAC-py-TRZ, is presented, together with an extensive computational analysis and experimental characterization. The chemical structure of DMAC-py-TRZ only marginally differs from that of the parent DMAC-TRZ compound. However, this minor change of the bridging moiety between the donor and acceptor has an enormous impact on the conformation and hence on the photophysics of the compound. Specifically, while DMAC-TRZ maintains the same orthogonal geometry in both the ground and the first excited singlet states, DMAC-py-TRZ assumes a bent geometry in the ground state, as highlighted by the large oscillator strength measured in solution for this dye. However, upon excitation to the S1 state, the system undergoes a large geometrical rearrangement to the orthogonal structure. This large relaxation is confirmed by the very large Stokes shift measured in non-polar solvents. In frozen 2-MeTHF glass at very dilute conditions, the relaxation is hindered and only a blue-shifted emission is seen from the unrelaxed bent conformer, without any hint of emission from the orthogonal structure. In mCP films a distribution of conformers exists and at low concentrations dual emission is observed, originating both from both the bent and orthogonal structures. However, upon increasing the doping concentration, the emission from the orthogonal conformer dominates. While self-absorption can be partly responsible for the phenomenon, we conclude that an efficient energy transfer from one conformer to the other also contributes to the spectral change. Indeed, TADF is not expected nor observed in the bent structure, due to a too large DEST. The good TADF efficiency of DMAC-py-TRZ in solution, similar as for DMAC-TRZ, is in line with the very fast molecular relaxation from the bent (TADF-silent) to the orthogonal geometry (TADFactive) in solution. The situation is more delicate in matrices where the host rigidity hinders a large molecular rearrangement. The observed good efficiency of TADF in matrices then suggests efficient energy transfer of excitons created on the bent (and TADF silent) structures towards molecules in the orthogonal (and TADF-active) structure as to retrieve all photogenerated singlet states for TADF activity. Most probably, efficient triplet-to-triplet energy transfer is also required to explain the good efficiency of DMACpy-TRZ OLEDs, but this will be subject of a subsequent study. |
6171b5d58acf7ece28c6e7e5 | 0 | Aromatic diimides, also known as bis(dicarboximide)s, are the linchpin of a diversity of organic materials encompassing vivid pigments and dyes, thermally robust polymers, and ntype electronic materials. The electron-withdrawing nature of the cyclic imides lends itself to the creation of n-type semiconducting materials. The annulation of aromatic rings with cyclic imides tends to lead to more significant LUMO-lowering effects than HOMO-lowering effects, thus resulting in red-shifted bandgaps and absorption profiles. Aromatic diimides derived from benzene (PMDI), naphthalene (NDI), and pyrene (PDI) (Figure ) have received significant research attention because of their facile syntheses and amenability to derivatization at both the core and imide N positions. While changes in Nfunctionalization are often exploited for tuning solid-state packing behavior and solubility profiles, core modification through substitution and metal-mediated cross-coupling reactions of halogenated aromatic diimides results in fine control over energy levels and electronic structure. In recent years, researchers have developed new strategies for incorporating cyclic imides onto a growing number of aromatic scaffolds, resulting in interesting redox activity and near-IR absorptions. We recently began exploring the ortho-diimide structural isomer of the well-known pyromellitic diimide, which is known as mellophanic diimide (MDI), and have discovered that it has significant potential as a building block in the construction of compounds with properties of interest to organic materials chemists (Figure ). Core dichlorinated N,N'-dihexyl MDI (Cl2-MDI-Hex), for example, readily undergoes both SNAr and Pd-catalyzed substitutions with aromatic ortho dinucleophiles to lead to a range of highly chromophoric and electron-accepting hetero-and azaacene structures. The development of MDI as a building block is further attractive because it is derived from 1,2,3,4-tetramethylbenzene, which is a byproduct of durene synthesis and a constituent of petroleum extract that has no significant industrial use. With these observations in mind, we set out to establish generalizable synthetic methods for obtaining differently halogenated MDI derivatives. Results and Discussion. Conventionally, aromatic diimides are synthesized by condensation between the relevant aromatic cyclic dianhydride and an amine, a process which proceeds through an amide-carboxylic (amic) acid intermediate. Extending this approach to the synthesis of ortho aromatic diimides such as mellophanic diimide, however, is complicated by the possibility of incorrect cyclizations to yield 3,6-dicarboxyphthalimide byproducts. Two strategies were identified by Fang et. al for overcoming this challenge: 1) room-temperature amic acid formation followed by acetic anhydride-mediated dehydration and 2) high-temperature equilibration of the reaction mixture to reach the MDI thermodynamic product (Fig ). While both of these methods were successful for synthesizing N,N'-diaryl MDI compounds, we found that they could not be reliably extended toward either N,N'-dialkyl-or corechlorinated MDIs because of competing nucleophilic aromatic substitution reactions and significant 3,6-dicarboxyphthalimide formation. Inspired by the mellitic triimide synthesis developed by Rose et. al, we were previously able to obtain N,N'-dihexyl-4,5-dichloro-MDI (Cl2-MDI-Hex) after the 3-day solidstate dehydration of an ammonium carboxylate salt (Fig. ). In further explorations, however, we found that this solid-state method could not be consistently extrapolated to differently halogenated benzene tetracarboxylic acids, and furthermore was not successful with more sterically demanding amines. As a result of further synthetic exploration, here we report our findings that the direct solution-phase reaction between benzene-1,2,3,4tetracarboxylic acids and primary amines is a widely generalizable method for synthesizing MDI derivatives of a variety of N substitutions and core halogenations (Fig. ). Synthesis. The commercially available 1,2,3,4-tetramethylbenzene was chlorinated, 20 brominated, or iodinated following literature procedures to yield intermediates 1-X (X = Cl, Br, or I) which were then subjected to exhaustive oxidation by 10 eq. of KMnO4 in tBuOH/H2O (1/1:v/v) to provide the halogenated benzene-1,2,3,4-tetracarboxylic acids. Although many KMnO4 methylarene oxidation procedures use pyridine as a co-solvent to improve reactant solubility, we have found that the use of tBuOH co-solvent reduces the equivalents of KMnO4 required to achieve full oxidation and furthermore is less prone to exotherm during the addition of KMnO4. Complete removal of reaction solvent prior to acidification of the carboxylate intermediate is important for avoiding the formation of t-butyl ester impurities. Our largest scale oxidation (20.0 g, 98.5 mmol of 1-Cl) proceeded smoothly to provide the tetraacid 2-Cl in 85% isolated yield. |
6171b5d58acf7ece28c6e7e5 | 1 | Although we initially followed our previously developed solid-state dehydration protocol for synthesizing 3-X-R, we were motivated by inconsistent yields and long reaction times to investigate a solution-phase method. Unexpectedly, simply heating the tetraacids 2-X with primary amines in acetic acid solvent followed, if necessary, by the precipitation of products with the addition of water or MeOH to the reaction mixture, provided MDIs 3-X-R in up to 93% isolated yield. In some cases, a small amount of additional product can be recovered by extraction of the aqueous filtrate with CH2Cl2. Although the reaction byproducts are often polar and soluble enough in AcOH to be removed during the filtration process, additional purification can be easily achieved by passing the crude product mixture through a SiO2 column. This method is successful with a variety of amines and there is no discernible trend between the identity of the halogen and the reaction yield. Hexylamine, aniline, and benzylamine react smoothly with 2-X (X = Cl, Br, and I) in good yields to produce a suite of 3-X-R compounds. Single crystal X-ray data for 3-Br-Ph confirms the MDI constitution of the products (Figure ). It is notable, as will be discussed below, that yields tend to be higher for the halogenated 3-X-R compounds than the nonhalogenated ones. Amino acid methyl esters can also be readily converted into MDIs, which suggests they may be interesting building blocks for self-assembling small molecules. Attempts to perform imidization with 6-amino-1-hexanol, however, resulted in complex mixtures as a consequence of acetate ester formation with the free alcohol. With these results in hand, we attempted to install more sterically demanding Ngroups such as branched alkyl chains and 2,6-dialkylaryl groups since these are commonly used by organic materials chemists to manipulate crystal packing and solubility. Although 2ethylhexylamine reacted readily with 2-H to yield 3-H-EtHex, imidization of 2-H with the more sterically hindered nucleophiles such as (S)-1-phenylethan-1-amine and 2,6-diisopropylaniline proved to be more recalcitrant. We were unable to isolate any products after applying our standard reaction method, and heating (S)-1-phenylethan-1-amine with 2-H for 3 days at 110 °C resulted in only 11% formation of 3-H-1PhEt. These low yields for sterically hindered amines were also found when attempting these same reactions using our older solid-state approach (Figure ). |
6171b5d58acf7ece28c6e7e5 | 2 | To overcome this slow reaction rate, we turned to microwave reaction conditions. Gratifyingly, reacting 2-Cl and (S)-1-phenylethan-1-amine at 200 °C for 24 hours led to 58% isolated yield for 3-H-1PhEt. Under these forcing conditions, an important consideration is whether nucleophilic aromatic substitution (SNAr) reactions at the aryl halides start to take place when X = halogen. In our trials, we did not observe significant levels of SNAr reactivity when 2.1 equiv of amine was reacted with tetraacid 2-Cl. However, when 12 equivalents of (S)-1-phenylethan-1-amine were reacted with 2-Cl, we did observe core-diaminosubstituted derivatives in the resulting product mixture. Mechanistic Insights. Our first attempted synthesis of 3-H-1PhEt under microwave conditions was performed at 200 °C for 2 hr of reaction time and resulted in 11% isolated yield. Both running the reactions for longer (200 °C for 24 hours) or with more equivalents of nucleophile (12 equiv. amine, 200 °C for 2 or 24 hours) improved the isolated yield, to 58% and 46%, respectively. It is worth noting that the excess amine approach is not feasible under microwave conditions when X = halogen because of the competitive SNAr reactions. From a mechanistic perspective, these observations suggest that in AcOH solvent, the reaction is likely taking place under overall equilibrating conditions. To gain more insight into the reaction process, we performed a 1 H NMR time-course study of the reaction between 2-H and (S)-1-phenylethan-1-amine in d4-acetic acid at 110 °C. Upon dissolving only 2-H, the 1 H NMR spectrum reflects the presence of a complex mixture of anhydrides corresponding to intermediates with or without symmetry around the central benzene ring. Although we cannot rule out cyclic anhydride formation, we believe that mixed acetic anhydride formation is more likely based on the solubility of the reaction intermediates. Upon addition of the amine and heating for 5 minutes, a number of different intermediates are detected by 1 H NMR spectroscopy and formation of the desired MDI product is first observable after 20 minutes of reaction time. After 19 hours, the mixture resolves itself into being primarily four species, three with symmetry around the benzene core and one without. After 11 days of reaction time, the desired product constitutes only 18% of the product mixture, which highlights the value of microwave reaction conditions for achieving higher yields on a reasonable time scale. The NMR spectrum of the reaction after a total of 31.5 days at 110 °C shows that the product distribution reaches a roughly 1:1:1 ratio of 3-H-1PhEt, 4-H-1PhEt, and 5-H-1PhEt (Figure ). |
6171b5d58acf7ece28c6e7e5 | 3 | We were able to tentatively assign the structure of the reaction intermediates by performing column chromatography on incomplete reaction mixtures and analyzing the filtrates collected during reaction workups. Of these, the most notable intermediates are dicarboxyphthalimides 4-X-R and 5-X-R, which correspond to the asymmetric and symmetric possibilities, respectively, for monophthalimide formation between the tetraacids 2-X and an amine. It is again worth noting that during the reaction, however, it is difficult to rule out whether mixed acetic acid anhydride derivatives are the dominant species. Liquid chromatography mass spectrometry analysis of crude reaction mixtures corroborates the 1 H NMR and preparatory observations of 4-X-R and 5-X-R as the primary reaction byproducts, but also reveal one other identifiable product corresponding to a diamidophthalimide species 6-X-R. We believe this isomer is more likely because the analogous diamido derivative of 4-X-R should be more likely to proceed to cyclize into 3-X-R. From a purely statistical perspective, intermediates 4-X-R and 5-X-R should be formed in a 3 to 1 ratio because both the 1-and 2-carboxamide derivatives of 2-X can dehydrate into intermediate 4-X-R, while only the 2-carboxamide can cyclize into intermediate 5-X-R. To add context to our understanding, we performed density functional theory calculations (M062X/6-31G(d)) with implicit acetic acid solvent to evaluate the energy landscape of the reaction for the reaction between methylamine and either 2-H or 2-Cl. In both cases, the formation of the less symmetric 4-X-Me is thermodynamically favorable compared to the formation of 5-X-Me. Interestingly, the transformation of 4-H-Me into 3-H-Me is found to be an uphill process by 3.5 kcal/mol, while the analogous conversion of 4-Cl-Me into 3-Cl-Me costs only 0.3 kcal/mol. Although these calculations do not account for the added complexities of mixed anhydride formation or R group identity, they do correlate with our experimental findings that 3-X-R formation is higher yielding when X = halogen. Taken together, it is reasonable to posit that the statistical advantage of monophthalimide formation balances against substrate-specific energetic differences to influence overall conversion into MDI products. |
6171b5d58acf7ece28c6e7e5 | 4 | In conclusion, we have developed methods for the preparation of a wide range of 5,6-dihalobenzene-1,2,3,4-bis(dicarboximides) from the direct solution-phase condensation of benzene-1,2,3,4-tetracarboxylic acids and primary amines in acetic acid solvent. The preparation of compounds with sterically demanding groups at the N atoms of the imides can be achieved readily under microwave conditions. Experimental and computational studies suggest that the reaction can take place under equilibrium conditions that are favored both statistically and thermodynamically to yield the desired ortho-diimide compounds instead of the symmetric dicarboxyphthalimide. Importantly, it is possible to avoid perturbing the aryl halide positions under these reaction conditions, which sets the stage for exploring a broad range of chemistries available for producing imide-decorated aromatic compounds with tailored form and function. |
664c278821291e5d1dd5ecc1 | 0 | Accurate binding affinity prediction to given protein-ligand complexes is of the greatest importance for scoring function in structure-based virtual screening (SBVS) method. The recent progress of machine learning (ML) has enabled great advancements in ML-based scoring function development. Such data-driven binding affinity prediction approaches heavily rely on data scale and quality of complex crystal structures, 5,6 limiting their performance and applicability since the current crystal structure databases contain insufficient chemical and protein structure spaces (i.g., 22,920 protein-ligand complexes in PDBbind 2024 7 ). A more general field -compound-protein interaction (CPI) prediction -has gradually gained researchers' attentions since it focuses on directly predicting binding affinities or bioactivities for protein-ligand pairs without complex structure required. Such structure-free CPI methods are much flexible than ML-based scoring functions with substantially larger-scale CPI data available for model training (i.g, ~1.6M assays in ChEMBL 8 and 1.1M binding measurements in BindingDB 9 ). Some recent studies demonstrated that well-trained CPI models can beat traditional protein-ligand binding affinity prediction methods (e.g., Vina, 10 Gnina, 11 RFScore, 12 ∆VinaRF, 13 etc.) regarding both prediction accuracy and screening power. However, the availability of a vast amount of bioactivity data for training CPI models presents both opportunities and challenges. |
664c278821291e5d1dd5ecc1 | 1 | From data heterogeneity aspect, a critical issue is the comparability and uncertainty of bioassay endpoints with different activity types, which can potentially limit the effectiveness of data-driven CPI methods in practical virtual screening. For instance, IC 50 values are influenced by the reaction types, concentrations of enzymes, inhibitors, substrates, and other experimental conditions, while K i values are intrinsic thermodynamic quantities that depend solely on the interactions between enzymes and inhibitors. It is generally accepted that thermodynamic affinity properties, such as K i and K d , offer the highest data quality and consistency. In contrast, other bioactivity measures like EC 50 , IC 50 , and percentage inhibition tend to be less reliable. Given these inconsistencies, although directly convert all bioactivity values into identitcal types is impractical, such highly-correlated bioactivity values cross different types may still benefit for learning a robust affinity predictor given the superior power of ML models in fitting linear relationships. However, blindly combining bioactivity data from different assays could introduce significant noise, thereby impairing the performance of data-driven CPI models. A recent study has also demonstrated that careful data integration and multistage machine learning modeling can successfully leverage multiple data types to improve kinase bioactivity prediction. These evidences highlight the need for comprehensive data processing and tailored CPI modeling strategies to effectively utilize large-scale, but often noisy, bioactivity data. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.