GIFT: Guided and Interpretable Factorization for Tensors with an application to large-scale multi-platform cancer...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Multi-platform genome data presents challenges in extracting interpretable patient-gene relationships.
• Prior knowledge of functional gene sets can aid in uncovering meaningful biological insights.

Purpose of the Study:

• To develop a tensor factorization method that produces interpretable gene factor matrices using functional gene set information.
• To maintain decomposition quality and computational speed while incorporating prior biological knowledge.

Main Methods:

• Proposed GIFT (Guided and Interpretable Factorization for Tensors), a tensor factorization technique.
• Incorporated prior functional gene set knowledge as a regularization term in the objective function.
• Applied GIFT to the PanCan12 dataset (TCGA multi-platform genome data).

Main Results:

• GIFT achieved interpretable factorizations with high scalability and accuracy.
• Outperformed baseline methods (P-Tucker, Silenced-TF) in terms of interpretability and performance.
• Demonstrated the ability of GIFT to reveal significant (cancer, gene set, gene) relationships.

Conclusions:

• GIFT successfully integrates prior biological knowledge into tensor factorization for enhanced interpretability.
• The method provides a scalable and accurate approach for analyzing complex multi-platform genomic data.
• Findings from GIFT can be validated through literature, supporting its utility in cancer research.