GO2Sum: Generating Human Readable Functional Summary of Proteins from GO Terms.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

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 GO term lists can be challenging for biologists.

Approach:

• Developed GO2Sum, a model using the T5 large language model to summarize GO terms.
• Fine-tuned T5 on UniProt entries, combining GO term assignments with free-text function descriptions.
• Enabled the model to generate human-readable function descriptions by concatenating GO term descriptions.

Key Points:

• GO2Sum effectively summarizes complex Gene Ontology data into understandable text.
• The model was fine-tuned on specific biological data (UniProt entries) for improved performance.
• GO2Sum significantly outperforms the base T5 model in generating specific biological context paragraphs.

Conclusions:

• GO2Sum provides a valuable tool for biologists to interpret protein functions.
• The model demonstrates the potential of large language models in specialized scientific domains.
• This approach enhances the usability of GO terms for biological research and discovery.