CrowdGO: Machine learning and semantic similarity guided consensus Gene Ontology annotation

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate gene function characterization is crucial for understanding biological systems.
• Computational prediction methods are essential for annotating the vast number of genes in diverse species.
• Existing prediction tools have varying strengths and weaknesses, necessitating improved approaches.

Purpose of the Study:

• To develop an open-source meta-predictor, CrowdGO, for enhanced gene function annotation.
• To leverage consensus and machine learning for improved accuracy and reliability in gene annotations.
• To integrate strengths of individual prediction methods into a unified approach.

Main Methods:

• Developed CrowdGO, a consensus-based Gene Ontology (GO) term meta-predictor.
• Employed machine learning models incorporating GO term semantic similarities and information content.
• Re-evaluated gene-term annotations to create a consensus dataset with confident and rejected annotations.

Main Results:

• CrowdGO achieved improved precision and recall in gene functional annotations.
• Performance of CrowdGO matched top-performing individual prediction methods.
• Consensus annotations generated by CrowdGO demonstrated enhanced accuracy and comprehensiveness.

Conclusions:

• CrowdGO offers a model-informed strategy to enhance gene functional annotations.
• The meta-predictor effectively combines the strengths of diverse individual prediction algorithms.
• CrowdGO provides a robust solution for accurate and comprehensive gene function characterization.