DeepText2GO: Improving large-scale protein function prediction with deep semantic text representation.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• UniProtKB contains over 115 million protein sequences, but less than 0.15% have experimental Gene Ontology (GO) annotations.
• A significant gap exists between available protein sequences and their functional annotations, necessitating effective Automatic Protein Function Prediction (AFP) methods.

Purpose of the Study:

• To develop a novel method, DeepText2GO, for improving large-scale AFP by integrating diverse data sources.
• To enhance AFP by utilizing deep semantic text representations alongside sequence homology, protein families, domains, and motifs.

Main Methods:

• Proposed DeepText2GO, a novel AFP method employing deep semantic text representation.
• Integrated text-based and sequence-based approaches using a consensus strategy.
• Utilized protein sequence information including homology, families, domains, and motifs.

Main Results:

• DeepText2GO demonstrated superior performance in large-scale AFP.
• The method significantly outperformed traditional text-based (Bag-of-Words) and sequence-based AFP methods.
• Experiments were conducted on a benchmark dataset from UniProt/SwissProt.

Conclusions:

• DeepText2GO effectively bridges the gap in protein functional annotation.
• Integrating deep text semantics with sequence information provides a powerful approach for AFP.
• The consensus strategy enhances the accuracy and scope of protein function prediction.