Graphical Learning and Causal Inference for Drug Repurposing.

Area of Science:

• Computational Biology
• Pharmacology
• Bioinformatics

Background:

• Discovering drug repurposing indications relies on connecting drug perturbations, disease signatures, and clinical data via gene expression profiles.
• Current gene expression reversal methods are limited, focusing on individual genes and lacking causal and graph-based approaches for drug repurposing candidate identification.

Approach:

• A novel framework combining causal inference and graph neural networks (GNNs) is proposed for drug repurposing.
• Developed a new algorithm for reconstructing large-scale causal networks and utilized directed acyclic graph neural networks (DAGNNs) for information aggregation.
• Introduced a graph regression method for assessing drug efficacy in reversing disease-associated gene expression changes.

Key Points:

• The framework overcomes limitations of existing methods by incorporating causal inference and advanced graph analysis.
• Large-scale simulations demonstrated favorable false positive and false negative rates for the causal network reconstruction.
• The graph regression method effectively measures drug-induced reversal of disease gene expression, aiding in candidate selection.

Conclusions:

• The proposed causal inference and GNN-based framework offers a robust method for identifying potential drug repurposing candidates.
• Applied to LINCS L1000 data and SARS-CoV-2 gene expression, demonstrating practical utility in drug discovery.
• This approach advances the analysis of gene expression data for identifying novel therapeutic applications of existing drugs.