Extensive complementarity between gene function prediction methods.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The rapid increase in sequenced genomes necessitates efficient methods for determining gene biological roles.
• Automated function prediction (AFP) is crucial for annotating the functions of uncharacterized genes.
• Distinct genome features may inform different types of gene function predictions.

Purpose of the Study:

• To systematically analyze the benefits of integrating predictions from distinct AFP approaches.
• To evaluate the complementarity of different AFP methodologies.
• To assess the impact of integrating AFP methods on gene function prediction accuracy.

Main Methods:

• Amalgamated 5,133,543 genes from 2071 genomes for a large-scale analysis.
• Evaluated five established genomic AFP methodologies.
• Utilized an information-theoretic approach to quantify gene function knowledge.

Main Results:

• While 1227 Gene Ontology (GO) terms were reliably predicted, most functions were only accessible to one or two methods.
• Diverse AFP methods showed complementary predictions, both gene-wise and function-wise.
• Integration of AFP methods increased gene function knowledge by up to 11 bits/gene, outperforming individual methods and ranking highly on the CAFA2 benchmark.

Conclusions:

• Integrating diverse AFP methods is a viable strategy for enhancing gene function annotation.
• Increased availability of sequenced genomes improves AFP accuracy and the benefits of integration.
• The study provides a comprehensive resource for individual and integrated GO predictions.