A large-scale assessment of sequence database search tools for homology-based protein function prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Homology-based function transfer using sequence database searches is a foundational method for predicting protein functions, including Gene Ontology (GO) terms.
• These sequence searches are integral to modern machine learning and deep learning approaches for protein function prediction.

Purpose of the Study:

• To systematically evaluate the impact of different sequence search tools and parameter settings on protein function prediction accuracy.
• To develop and validate an improved scoring function for deriving GO predictions from homologous sequence hits.

Main Methods:

• Comparative analysis of popular sequence search tools (e.g., BLASTp, MMseqs2, DIAMOND) on a large benchmark dataset for GO term prediction.
• Systematic exploration of various search parameter configurations for each tool.
• Development and testing of a novel scoring function for homology-based GO prediction.

Main Results:

• BLASTp and MMseqs2 generally outperform other tools, including DIAMOND, under default parameters for GO term prediction.
• Optimized parameter settings enable DIAMOND to achieve performance comparable to BLASTp and MMseqs2.
• The newly developed scoring function consistently surpasses previous methods in predicting GO terms from homologous hits.

Conclusions:

• Search tool selection and, crucially, parameter optimization are critical for effective homology-based protein function prediction.
• The findings offer easily implementable improvements for existing protein function prediction algorithms.
• This study contributes to advancing the development of more accurate future protein function prediction tools.