Differentiating isoform functions with collaborative matrix factorization.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene isoforms, arising from alternative splicing, generate diverse proteoforms.
• Understanding isoform-specific functions is key to dissecting complex disease pathologies.
• Current functional genomic databases lack isoform-level annotations, hindering granular analysis.

Purpose of the Study:

• To develop a computational method for differentiating gene isoform functions.
• To address the challenge of predicting functions at the isoform level, moving beyond gene-level annotations.

Main Methods:

• Proposed DisoFun, a data-integrative approach utilizing collaborative matrix factorization.
• Modeled gene functions as aggregations of key isoform functions.
• Integrated RNA-seq data, Gene Ontology annotations, PPI networks, and Gene Ontology structure.

Main Results:

• DisoFun significantly improved performance metrics (AUROC by 7.7%, AUPRC by 28.9%) compared to existing methods.
• Achieved 90.5% accuracy in differentiating functions for four exemplar genes at the isoform level.
• Successfully identified latent key isoforms driving gene function.

Conclusions:

• DisoFun offers a robust solution for isoform-level function prediction.
• The method enhances understanding of disease mechanisms by providing finer granularity.
• Accurate isoform function differentiation is achievable and valuable for biomedical research.