DIVERSE: Bayesian Data IntegratiVE Learning for Precise Drug ResponSE Prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Precision medicine requires identifying predictive biomarkers from complex multi-omics data for improved diagnostics and treatments.
• Challenges include experimental costs, cell line heterogeneity, and selecting informative data sources.
• Existing methods struggle to integrate diverse omics data for drug response prediction.

Purpose of the Study:

• To develop a computational framework, DIVERSE, for predicting drug responses by integrating diverse omics data.
• To systematically evaluate the importance of different data sources in predicting drug responses.
• To outperform existing methods in predicting drug responses, especially in challenging out-of-matrix scenarios.

Main Methods:

• Proposed DIVERSE (Bayesian importance-weighted tri- and bi-matrix factorization) framework (DIVERSE3/DIVERSE2).
• Sequentially integrated five different data sets: cell line, drug, and gene interaction data.
• Systematically examined the importance of each added data set.

Main Results:

• DIVERSE significantly outperformed five other methods, including three state-of-the-art approaches, in cross-validation.
• Demonstrated superior performance in out-of-matrix prediction, relevant to real-world applications.
• Case studies confirmed DIVERSE's advantage in discovering new drugs.

Conclusions:

• DIVERSE provides an effective computational solution for predicting drug responses from integrated multi-omics data.
• The framework offers a systematic approach to data integration and biomarker discovery for precision medicine.
• DIVERSE advances the field of bioinformatics for drug response prediction and aids in novel drug discovery.