NMFGO: Gene Function Prediction via Nonnegative Matrix Factorization with Gene Ontology.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene Ontology (GO) provides a structured vocabulary for gene product functions, roles, and locations.
• Accurately predicting gene-GO term associations is complex due to the large number of terms and hierarchical structure.
• Existing methods face challenges in effectively capturing the GO hierarchy for accurate gene function prediction.

Purpose of the Study:

• To develop a robust computational approach for predicting gene-GO term associations.
• To improve the accuracy of gene function prediction by incorporating the hierarchical structure of GO.
• To present NMFGO, a novel method based on matrix factorization and GO hierarchy regularization.

Main Methods:

• Gene-GO term associations are represented in a matrix.
• Nonnegative matrix factorization (NMF) is employed, regularized by GO's ontological structure.
• A semantic similarity-based k-nearest neighbor classifier is used in the approximated subspace for prediction.

Main Results:

• NMFGO demonstrates robustness to input parameter variations.
• The method significantly outperforms existing approaches including GIC, TO, dRW-kNN, and NtN.
• Empirical studies on *S. cerevisiae*, *H. sapiens*, and *A. thaliana* validate NMFGO's effectiveness.

Conclusions:

• NMFGO provides a powerful and accurate method for gene function prediction.
• Incorporating GO hierarchy into matrix factorization enhances prediction performance.
• The proposed approach offers a valuable tool for genomic data analysis and functional annotation.