Combining learning and constraints for genome-wide protein annotation

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• High-throughput experiments enable genome-wide computational analysis.
• Predictive annotation of large protein sets can yield inconsistent results.
• Integrating prior knowledge into predictive frameworks enhances genomic annotation quality.

Purpose of the Study:

• To develop a predictive pipeline for simultaneous functional and protein-protein interaction (PPI) annotation of all proteins within a genome.
• To improve the quality and consistency of machine-generated genomic annotations.

Main Methods:

• OCELOT pipeline combines sequence-based functional and PPI predictors.
• A consistency layer enforces prior knowledge using fuzzy logic rules.
• Rules include taxonomic constraints (GO hierarchy) and combined function-interaction predictions.

Main Results:

• OCELOT substantially improves prediction quality by integrating prior knowledge.
• The system outperforms GoFDR on Yeast genome data when using intra-genome information.
• OCELOT shows comparable or better results than GoFDR and favorably compares to deep learning methods.

Conclusions:

• Integrating prior knowledge through fuzzy logic rules significantly enhances genome-wide protein annotation.
• OCELOT provides a robust framework for accurate functional and PPI prediction.
• The pipeline offers a valuable tool for advancing genomic annotation accuracy.