Pathway-based Bayesian inference of drug-disease interactions

Area of Science:

• Computational Biology
• Systems Biology
• Pharmacogenomics

Background:

• Drug treatments impact biological pathways, which are sets of functionally related genes.
• Understanding these pathway perturbations is crucial for advancing knowledge of cellular processes.
• Pathways often interact, leading to secondary effects that are typically overlooked.

Purpose of the Study:

• To develop a new method for identifying pathways responsive to drug treatments.
• To simultaneously determine interactions between these perturbed pathways.
• To improve the accuracy of pathway responsiveness identification compared to existing methods.

Main Methods:

• Utilized Bayesian matrix factorization of gene expression data with known gene-pathway memberships.
• Implemented a Gaussian Markov Random Field (GMRF) within the matrix factorization framework to model between-pathway interactions.
• Applied the combined model to analyze gene expression data from 1169 drugs and 236 pathways.

Main Results:

• The developed model achieved significantly higher average precision in identifying pathway responsiveness compared to existing methods.
• Demonstrated the advantage of considering between-pathway interactions, confirming pathways are not independent.
• Case studies highlighted enhanced pathway identification and provided insights into disease comorbidity and drug repositioning.

Conclusions:

• The novel method effectively identifies drug-perturbed pathways and their interactions.
• Accounting for between-pathway networks significantly improves the accuracy of drug effect analysis.
• The findings offer valuable insights for drug repositioning, disease comorbidity, and tissue-specific drug response studies.