An expanded evaluation of protein function prediction methods shows an improvement in accuracy

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Assigning molecular function to proteins is crucial for understanding life.
• Experimental methods for protein annotation are reliable but low-throughput.
• Computational function prediction plays an increasingly important role, yet assessing these methods is challenging.

Purpose of the Study:

• To assess the performance of computational protein function prediction methods.
• To track progress in the field since the first Critical Assessment of Functional Annotation (CAFA1).

Main Methods:

• Conducted the second Critical Assessment of Functional Annotation (CAFA2), a timed challenge.
• Evaluated 126 methods from 56 research groups on 3681 proteins across 18 species.
• Utilized Gene Ontology for biological function prediction and Human Phenotype Ontology for gene-disease associations.

Main Results:

• Top-performing methods in CAFA2 demonstrated superior accuracy compared to CAFA1.
• Increased accuracy is linked to a greater number of experimental annotations and enhanced prediction algorithms.
• Performance is ontology-specific, with diverse predictions across biological processes and human phenotypes.

Conclusions:

• Methodological improvements were observed between CAFA1 and CAFA2.
• The effectiveness of prediction methods is context-dependent.
• Accurate prediction requires careful consideration of ontology specificity and performance metrics.