Predicting non-coding RNA function using Artificial Intelligence

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Non-coding RNAs (ncRNAs) play crucial roles in biological processes and disease.
• Information on ncRNA-phenotype relationships is fragmented across scientific literature.
• Efficient methods are needed to aggregate and normalize this dispersed data.

Purpose of the Study:

• To develop a methodology for extracting ncRNA-phenotype relations from scientific articles.
• To combine Natural Language Processing (NLP) and Large Language Models (LLMs) for this task.
• To create a high-fidelity dataset and relational corpus for ncRNA research.

Main Methods:

• Developed an NLP pipeline to aggregate and normalize data from five ncRNA-disease databases.
• Generated a ncRNA-phenotype relational corpus using Distant Supervision Relation Extraction (DSRE).
• Applied Large Language Models (LLMs) for Relation Extraction (RE), evaluating performance on a validated corpus subset.

Main Results:

• Created a high-fidelity ncRNA-phenotype relation dataset with 214,300 relations.
• Generated a relational corpus (ncoRP) with 35,295 unique relations from 21,608 articles.
• Achieved a high F1-score of 0.978 using an LLM-based RE methodology.

Conclusions:

• Successfully created a normalized ncRNA-phenotype dataset and relational corpus.
• The combined LLM and DSRE methodology demonstrates high performance for automatic relation extraction.
• The developed dataset, corpus, and methodology are valuable resources for ncRNA studies and can be applied to similar biological relation extraction tasks.