Inferring mouse gene functions from genomic-scale data using a combined functional network/classification strategy.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• The function of a majority of mouse and human genes remains uncharacterized.
• Experimental data on gene activities and expression are accumulating, but functional annotation is incomplete.

Purpose of the Study:

• To infer Gene Ontology annotations for mouse genes using large-scale data mining.
• To evaluate and compare two distinct data mining strategies for functional gene prediction.

Main Methods:

• Developed two data mining approaches: a classifier-based method and a network-based annotation propagation method.
• Applied and combined these strategies using functional genomics, proteomics, comparative genomics, and phenotypic data from the MouseFunc competition.
• Evaluated predictive performance using a hold-out set and ten-fold cross-validation.

Main Results:

• The network-based approach generally outperformed the naive Bayesian classifier.
• Combining both approaches provided incremental improvements over individual methods.
• Achieved strong predictive power on the MouseFunc test set (median AUROC of 0.865, average precision of 0.195).

Conclusions:

• Data mining strategies using existing experimental data are effective for discovering mammalian gene functions.
• A high-confidence functional mouse gene network was generated, predictive of gene function and associations.
• The generated network is a valuable resource for mammalian gene functional analyses and candidate gene prioritization.