Evaluating large language models for annotating proteins
Briefings in Bioinformatics · 27.03.2024 · 25 (3)
DOI: 10.1093/bib/bbae177 ↗Kurz gefasst
Fachartikel in der Zeitschrift Briefings in Bioinformatics, online erschienen am 05.05.2024. Verfasst von Rosario Vitale und 4 weiteren Autor:innen. Beteiligte Einrichtung: Research Institute for Signals, Systems and Computational Intelligence sinc(i) (CONICET-UNL), Ciudad Universitaria, Santa Fe, Argentina. Hauptthemen laut MEDLINE: Proteins, Databases, Protein. Die Arbeit wurde bisher 29-mal zitiert (Crossref) und nennt 35 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Agency for Scientific and Technological, Agencia Santafesina de Ciencia, Tecnología e Innovación, University of Nebraska-Lincoln.
Abstract (Originalsprache)
In UniProtKB, up to date, there are more than 251 million proteins deposited. However, only 0.25% have been annotated with one of the more than 15000 possible Pfam family domains. The current annotation protocol integrates knowledge from manually curated family domains, obtained using sequence alignments and hidden Markov models. This approach has been successful for automatically growing the Pfam annotations, however at a low rate in comparison to protein discovery. Just a few years ago, deep learning models were proposed for automatic Pfam annotation. However, these models demand a considerable amount of training data, which can be a challenge with poorly populated families. To address this issue, we propose and evaluate here a novel protocol based on transfer learningṪhis requires the use of protein large language models (LLMs), trained with self-supervision on big unnanotated datasets in order to obtain sequence embeddings. Then, the embeddings can be used with supervised learning on a small and annotated dataset for a specialized task. In this protocol we have evaluated several cutting-edge protein LLMs together with machine learning architectures to improve the actual prediction of protein domain annotations. Results are significatively better than state-of-the-art for protein families classification, reducing the prediction error by an impressive 60% compared to standard methods. We explain how LLMs embeddings can be used for protein annotation in a concrete and easy way, and provide the pipeline in a github repo. Full source code and data are available at https://github.com/sinc-lab/llm4pfam
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- Research Institute for Signals, Systems and Computational Intelligence sinc(i) (CONICET-UNL), Ciudad Universitaria, Santa Fe, Argentina.
Förderung
- National Agency for Scientific and Technological
- Agencia Santafesina de Ciencia, Tecnología e Innovación
- University of Nebraska-Lincoln
Zitieren
APA
Vitale, R., Bugnon, L. A., Fenoy, E. L., Milone, D. H., & Stegmayer, G. (2024). Evaluating large language models for annotating proteins. Briefings in Bioinformatics, 25(3). https://doi.org/10.1093/bib/bbae177
Vancouver
Vitale R, Bugnon LA, Fenoy EL, Milone DH, Stegmayer G. Evaluating large language models for annotating proteins. Briefings in Bioinformatics. 2024;25(3). doi:10.1093/bib/bbae177
BibTeX
@article{Vitale2024,
author = {Vitale, Rosario and Bugnon, Leandro A and Fenoy, Emilio Luis and Milone, Diego H and Stegmayer, Georgina},
title = {Evaluating large language models for annotating proteins},
journal = {Briefings in Bioinformatics},
year = {2024},
volume = {25},
number = {3},
doi = {10.1093/bib/bbae177}
}
EUVIMED