Gene function prediction based on genomic context clustering and discriminative learning: an application to...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Existing whole-genome comparison methods lack automation and require knowledge of related species.
• Bacteriophage genomes present unique challenges for current analysis techniques.
• There is a need for automated gene function prediction systems.

Purpose of the Study:

• To develop an automated system, SynFPS, for gene function prediction in completed genomes.
• To address limitations of existing methods in analyzing genomes like bacteriophages.
• To provide a tool for predicting functions of uncharacterized or sequence-alignment-resistant genes.

Main Methods:

• Clustering of weakly related genomes based on gene distribution.
• Training Support Vector Machines (SVM) using extracted data from genome clusters.
• Application of genomic context for gene function prediction.

Main Results:

• SynFPS achieves an average prediction accuracy of approximately 80% across 9 gene functions in 296 bacteriophage genomes.
• The system successfully predicted functions for uncharacterized genes and those not identifiable by sequence alignment.
• Prediction accuracy is comparable to existing genomic-context-based methods.

Conclusions:

• The SynFPS system effectively uses genomic context for gene function prediction and gene correspondence detection.
• The method shows promise for extension to other microbial genomes beyond bacteriophages due to conserved characteristics like gene order.
• The developed software is publicly available for broader research use.