GenoVarDis: NER in Genomic Variants and related Diseases

The first edition of a competition of Named Entity Recognition (NER) in Spanish scientific literature about genomic variants, genes, and its associated diseases, is part of the IberLEF 2024 campaign.

Codalab page: https://codalab.lisn.upsaclay.fr/competitions/17733.

Description of the Task

By presenting this proposal, we aim to enhance the diversity of tasks at IberLEF 2024 and foster advancements in Spanish biomedical text processing. This task contributes to addressing the scarcity of resources in this specific domain, providing participants with an opportunity to advance research in NER.

Challenges, Relevance and Novelty

This task addresses the shortage of resources for Spanish in the domain of NER and genomic variants. The first one of this kind. By leveraging a unique corpus of a wider spectrum of mutations and variant-related entities (including gene, disease and symptom) in Spanish (mainly, translating from English and curated by human-experts), we aim to provide valuable data for training and evaluating NER models in this low-resource domain. We think this proposal is novel and relevant because it comprises important challenges for the recognition of variants-related entities. Variant NER datasets and systems are almost nonexistent, even in English, for example, tmVar3 [Wei et al., 2022] with just ~500 documents or BERN2 [Sung et al., 2022] (which uses tmVar2 [Wei et al., 2017] for variant detection, with only 158 documents). While most tools employ regular expressions based on the Human Genome Variation Society (HGVS) nomenclature1, the existing landscape faces limitations in recognizing diverse variant types [Lee et al., 2021]. Inspired by practical applications in precision medicine and biocuration, this dataset is designed to propel and explore advancements in NLP (Natural Language Processing) research in Spanish focused on recognizing variants-related entities, along with their corresponding medical conditions (diseases and symptoms).

In the realm of precision medicine, genomic variants play a crucial role in tailoring personalized treatments based on an individual's genetic makeup. The understanding of genetic diseases relies heavily on the automated gathering and synthesis of published knowledge about sequence variants from scientific literature [Wei et al., 2022]. So, we propose a challenge to identify the named entities mentioned in biomedical literature. The challenge will be structured as follow:

Given a text (sequence of tokens), identify the named entities as spans in the text and classify them according to one of those present in this table:

Type	Name	Example	Example (es)
Variant on DNA sequence	DNAMutation	c.1922G>A	c.1922G>A
C-to-G transition was identified at nucleotide 857	la transición de C a G se identificó en el nucleótido 857
RS number	SNP	rs763780	rs763780
COSMIC mutation	COSV53035892	COSV53035892
Allele on DNA sequence	DNAAllele	-218G	218G
Wild type and mutant	NucleotideChange / BaseChange	G > C, G/C	G > C, G/C
Variant with insufficient information	OtherMutation	306 base pair insertion	inserción de 306 pares de bases
insertion introduced eight additional amino acids	la inserción introdujo ocho aminoácidos adicionales
Gene	Gene	ABCA1	ABCA1
Disease/Symptom	Disease	Congenital hypothyroidism	Hipotiroidismo congénito
fever	fiebre
Transcript ID	Transcript	NM_015420.7,  ENST00000413302,	NM_015420.7,  ENST00000413302,

Example of text and annotation

Neurofibromatosis tipo I. Mutación de splicing detectada por MLPA y secuenciación en la Argentina La neurofibromatosis tipo 1 (NF1) es un desorden genético autosómico dominante, con una prevalencia de 1 en 2500-3000 nacidos vivos. La dificultad diagnóstica se debe al tamaño extenso del gen NF1 con pocos sitios hot-spot, la ausencia de una clara relación genotipo-fenotipo y rasgos clínicos con un espectro muy heterogéneo. Un caso sospechoso de NF1 procedente de la provincia de Jujuy fue analizado por MLPA (multiplex ligation-dependent probe amplification) en nuestro laboratorio. Mujer, adolescente mestiza (Amerindia/Europea), con un osteoma maxilar, lordosis lumbar, neurofibromas cutáneos y manchas café con leche. Por MLPA se detectó una alteración en el exón 13 del gen NF1. Por secuenciación del exón 13 se identificó una mutación “missense” en la posición 1466 del ARNm (NM_000267.3:c.1466A>G) que introduce un sitio de splicing aberrante. La patogenicidad de la mutación fue corroborada en la base de datos de variantes clínicas del National Center for Biotechnology Information. En nuestro conocimiento, este es el primer registro de una mutación NF1 en un paciente proveniente de poblaciones mestizas del Noroeste Argentino. La alteración ha sido reportada en individuos de otras poblaciones de origen muy disímil al del caso presentado, como la europea, sugiriendo que el sitio podría considerarse un sitio hot-spot del gen. Donde exista baja disponibilidad de diagnósticos moleculares, como en nuestro caso, se puede aplicar un algoritmo que comience por el estudio del gen NF1 por MLPA, metodología relativamente sencilla y de costo accesible. Con ella se evita enviar muestras al extranjero para análisis genéticos.

#pmid	start	end	term	entity
25919870	0	24	Neurofibromatosis tipo I	Disease
25919870	101	125	neurofibromatosis tipo 1	Disease
25919870	127	130	NF1	Gene
25919870	291	294	NF1	Gene
25919870	447	450	NF1	Gene
25919870	640	655	Osteoma maxilar	Disease
25919870	657	672	Lordosis lumbar	Disease
25919870	674	696	Neurofibromas cutáneos	Disease
25919870	699	721	Manchas café con leche	Disease
25919870	747	771	Alteración en el exón 13	OtherMutation
25919870	780	783	NF1	Gene
25919870	833	872	Mutación “missense” en la posición 1466	OtherMutation
25919870	883	894	NM_000267.3	Transcript
25919870	895	904	c.1466A>G	DNAMutation
25919870	1161	1164	NF1	Gene
25919870	1591	1594	NF1	Gene

Results

These are the final results for competition over the test data:

User	Team	F1	Precision	Recall
ander.martinez	 	0.8210	0.8223	0.8196
VictorMov	 	0.7935	0.7906	0.7963
ELiRF-VRAIN	 	0.7349	0.7775	0.6968
Milimeter98	 	0.5483	0.6108	0.4974
orlandxrf	 	0.5301	0.7318	0.4155
GuillemGSubies	 	0.4283	0.4355	0.4212
Baseline	 	0.3194	0.5938	0.2185
Antares-Amazel	 	0.3009	0.6040	0.2004

Expected Target Community

The task is designed for researchers and practitioners in NLP, Bioinformatics, and Healthcare, interested in working with low-resource domain, NER, focusing on the Spanish language. We hope that the corpus created will serve as a valuable resource for the community.

Contact:

Marvin M. Agüero-Torales, [email protected] (@mmaguero).
Carlos Rodríguez Abellán, [email protected].

How cite?

@article{genovardis2024,
author = {Marvin M. {Agüero-Torales} and Carlos {Rodríguez Abellán} and Marta {Carcajona Mata} and Juan Ignacio {Díaz Hernández} and Mario {Solís López} and Antonio {Miranda-Escalada} and Sergio {López-Alvárez} and Jorge {Mira Prats} and Carlos {Castaño Moraga} and David Vilares and Luis Chiruzzo},
title = {{Overview of GenoVarDis at IberLEF 2024: NER of Genomic Variants and Related Diseases in Spanish}},
journal = {Procesamiento del Lenguaje Natural},
volume = {73},
year = {2024},
issn = {1989-7553}
}

Bibliography

[Wei et al., 2022] Chih-Hsuan Wei, Alexis Allot, Kevin Riehle, Aleksandar Milosavljevic, Zhiyong Lu, tmVar 3.0: an improved variant concept recognition and normalization tool, Bioinformatics, Volume 38, Issue 18, September 2022, Pages 4449–445.
[Sung et al., 2022] Mujeen Sung, Minbyul Jeong, Yonghwa Choi, Donghyeon Kim, Jinhyuk Lee, Jaewoo Kang, BERN2: an advanced neural biomedical named entity recognition and normalization tool, Bioinformatics, Volume 38, Issue 20, 15 October 2022, Pages 4837–4839.
[Wei et al., 2017] Wei C-H, Phan L, Feltz J, Maiti R, Hefferon T, Lu Z. tmVar 2.0: integrating genomic variant information from literature with dbSNP and ClinVar for precision medicine. Bioinformatics. 2017;34(1):80-7.
[Lee et al., 2021] Kyubum Lee, Chih-Hsuan Wei, Zhiyong Lu, Recent advances of automated methods for searching and extracting genomic variant information from biomedical literature, Briefings in Bioinformatics, Volume 22, Issue 3, May 2021.

Datasets:

FujitsuCoEDI
/

GenoVarDis-IberLEF-2024

You need to agree to share your contact information to access this dataset