DisoFun

DisoFun differentiates isoform functions by applying collaborative matrix factorization to integrate isoform data, gene-term Gene Ontology annotations, protein-protein interaction networks, and RNA-seq datasets for improved isoform- and gene-level function prediction.

Key Features:

• Collaborative Matrix Factorization: Collaboratively factorizes the isoform data matrix and the gene-term data matrix containing Gene Ontology annotations to identify latent key isoforms and aggregate predictions to genes.
• Integration with PPI Network and Gene Ontology: Incorporates protein-protein interaction (PPI) networks and the hierarchical structure of Gene Ontology to coordinate the matrix factorization process.
• Performance Improvement: Demonstrates experimental improvements of at least 7.7% in AUC-ROC and 28.9% in AUC-PR compared with existing solutions.
• Validation on Exemplar Genes: Validated on four exemplar genes—LMNA, ADAM15, BCL2L1, and CFLAR—with 90.5% accuracy in differentiating isoform functions.

Scientific Applications:

• Complex disease mechanism analysis: Enables isoform-specific functional analysis to elucidate functional diversity and pathology in complex diseases.
• Therapeutic target and biomarker discovery: Supports identification of novel therapeutic targets and proteomic biomarkers by differentiating isoform functions.

Methodology:

Collaboratively factorizes the isoform data matrix and the gene-term matrix with Gene Ontology annotations while integrating protein-protein interaction networks and the hierarchical GO structure; views each gene as a bag of spliced isoforms, integrates multiple RNA-seq datasets and additional biological data sources, and aggregates isoform-level predictions back to genes.