Predicting protein function by genomic context: quantitative evaluation and qualitative inferences

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Predicting functional protein interactions is crucial for understanding cellular mechanisms.
• Genomic context offers various data types for inferring these interactions.

Purpose of the Study:

• To compare different genomic context methods for predicting protein interactions.
• To assess their coverage, correlation with functional interactions, and overlap with homology-based methods.
• To evaluate their effectiveness using Mycoplasma genitalium as a benchmark.

Main Methods:

• Analysis of gene fusion (Type I).
• Evaluation of gene order conservation and operon co-occurrence (Type II).
• Assessment of gene co-occurrence across genomes (phylogenetic profiles, Type III).
• Comparison with homology-based function assignment.
• Application to the Mycoplasma genitalium genome.

Main Results:

• Gene order conservation showed the highest coverage (37%).
• Combining all genomic context methods yielded information for 50% of genes.
• Stricter genomic neighborhood requirements correlated with stronger functional interactions and fewer false positives.
• Homology searches were essential for predicting interaction types when genomic context alone was insufficient.
• 10% of M. genitalium genes had new functional features predicted using combined methods.

Conclusions:

• Genomic context analysis, particularly gene order conservation, provides significant insights into protein functional interactions.
• Combining multiple genomic context types and homology searches maximizes predictive power.
• This integrated approach is valuable for discovering novel gene functions and interactions.