Evolutionary profile enhancement improves protein function annotation for remote homologs.

Area of Science:

• Computational Biology
• Bioinformatics
• Protein Science

Background:

• Accurate protein function annotation is crucial for understanding biological systems.
• Current machine learning methods struggle with proteins lacking similar annotated sequences (out-of-distribution samples).

Purpose of the Study:

• To develop a novel strategy, EPERep, to improve automated protein function prediction.
• To enhance the prediction accuracy for underrepresented proteins and those with low sequence identity.

Main Methods:

• EPERep is an evolutionary input enhancement strategy.
• It leverages unannotated protein sequences to improve representation learning.
• It incorporates homologous sequences as contextual input for pre-trained protein language models, creating a pLM-based profile.

Main Results:

• EPERep consistently outperformed existing machine learning and sequence-alignment methods across four major annotation benchmarks (EC numbers, domains, Pfam, GO).
• Significant improvements were observed for proteins in rare functional classes and those with remote homology.
• The method excels in low-identity and long-tail prediction scenarios.

Conclusions:

• Evolutionary input enhancement is a principled and scalable approach for protein function prediction.
• EPERep effectively addresses limitations in current methods for underrepresented and distantly related proteins.
• This strategy advances the field of automated protein function annotation.