bert-base-uncased-finetuned-HMGCR-IC50s-V3

This model is a fine-tuned version of bert-base-uncased on on 905 HMGCR IC50 values from bindingDB.org. Molecules with counter ions were included twice, once with and once without counter-ions.

It achieves the following results on the evaluation set:

• Loss: 0.5183
• Accuracy: 0.8500
• F1: 0.8408

Use in Pipeline:

from transformers import pipeline

ic50_pipe = pipeline("text-classification", model="cafierom/bert-base-uncased-finetuned-HMGCR-IC50s-V3")

bert_ic50 = ic50_pipe('COC(=O)C[C@H](O)C[C@H](O)\C=C\n1c(cc(c1-c1ccc(F)cc1)-c1cccc(Br)c1)C(C)C')

#result: [{'label': '< 50 nM', 'score': 0.8461818695068359}]

Model description

More information needed

Intended uses & limitations

Can classify HMGCR IC50 values as < 50 nM, < 500 nM, and > 500 nM. See Confusion matrix below:

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1
0.9482	1.0	25	0.8987	0.6214	0.6201
0.8137	2.0	50	0.7437	0.7357	0.6651
0.7197	3.0	75	0.6532	0.7786	0.7060
0.6459	4.0	100	0.6019	0.7786	0.7086
0.6024	5.0	125	0.6042	0.7714	0.7006
0.5404	6.0	150	0.5818	0.7714	0.7130
0.5222	7.0	175	0.5403	0.7929	0.7499
0.4723	8.0	200	0.5514	0.8286	0.7984
0.4565	9.0	225	0.6523	0.7857	0.7395
0.4448	10.0	250	0.5207	0.8143	0.7994
0.4047	11.0	275	0.5102	0.7929	0.7662
0.3744	12.0	300	0.5183	0.85	0.8408
0.3644	13.0	325	0.5501	0.8071	0.7877
0.337	14.0	350	0.5559	0.8286	0.8037
0.308	15.0	375	0.5501	0.8214	0.8026
0.312	16.0	400	0.5463	0.8071	0.7862
0.2901	17.0	425	0.5611	0.8143	0.7967
0.2958	18.0	450	0.5508	0.8071	0.7936
0.2885	19.0	475	0.5752	0.8143	0.7967
0.2885	20.0	500	0.5756	0.8143	0.7967

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.3.0
• Tokenizers 0.21.0