GO2Sum: generating human-readable functional summary of proteins from GO terms.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Understanding protein biological functions is crucial in modern biology.
• Gene Ontology (GO) terms are widely used for protein function representation and prediction.
• Interpreting extensive lists of GO terms can be challenging for biologists.

Purpose of the Study:

• To develop a model that generates human-readable summaries of protein functions from Gene Ontology (GO) terms.
• To address the challenge of interpreting large sets of GO terms for biologists.

Main Methods:

• Developed GO2Sum, a model utilizing the T5 large language model.
• Fine-tuned T5 on GO term assignments and UniProt free-text function descriptions.
• Trained the model to recreate function descriptions by concatenating GO term descriptions.

Main Results:

• GO2Sum generates human-readable summaries of protein functions.
• The model significantly outperforms the original T5 model in generating specific paragraphs (Function, Subunit Structure, Pathway) for UniProt entries.
• Demonstrated effective recreation of function descriptions from GO terms.

Conclusions:

• GO2Sum offers a valuable tool for simplifying the interpretation of protein functions.
• The model enhances the usability of Gene Ontology data for biological research.
• Facilitates easier understanding of protein roles through automated summarization.