Matrix factorization-based data fusion for gene function prediction in baker's yeast and slime mold.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Accurate gene function characterization is crucial in computational biology.
• Integrating diverse data sources remains a challenge for gene function prediction.
• Previous work established a matrix factorization-based data fusion approach.

Purpose of the Study:

• To demonstrate the application of a matrix factorization-based data fusion approach for gene function prediction.
• To fuse heterogeneous biological data, including gene expression, protein annotations, interaction, and literature data.
• To evaluate the performance of the proposed method on predicting gene functions.

Main Methods:

• Simultaneous matrix tri-factorization to fuse heterogeneous data sources.
• Sharing matrix factors across different data types.
• Evaluation using ontological annotations in *D. discoideum* and protein localization/function in *S. cerevisiae*.

Main Results:

• The data fusion approach effectively integrates diverse data sources for gene function prediction.
• The method achieved predictive performance comparable to state-of-the-art kernel-based data fusion.
• The approach requires fewer data preprocessing steps compared to existing methods.

Conclusions:

• Matrix tri-factorization provides an effective and extensible framework for gene function prediction.
• This method offers a robust solution for integrating heterogeneous biological data.
• The approach simplifies data preprocessing while maintaining high predictive accuracy.