Scoring functions for drug-effect similarity

Area of Science:

• Computational biology
• Pharmacogenomics
• Drug discovery

Background:

• Drug repurposing is crucial due to the high cost and time of developing new therapeutics.
• Transcriptomic data from diseases and drug responses are complex and cell-specific, complicating direct matching.
• Gene regulatory networks offer a potential common ground for linking disease and drug effects.

Purpose of the Study:

• To develop and evaluate a method for matching drug effects to disease transcriptomes by comparing drug effects across different experimental conditions.
• To identify reliable similarity scores and heuristics for predicting drug efficacy in drug repurposing.
• To assess the impact of gene filtering and network-based approaches on prediction accuracy.

Main Methods:

• Utilized the Connectivity Map (CMap) dataset, combining 26 similarity scores and six heuristics.
• Implemented gene filtering based on expression levels and fold changes.
• Applied network-based transcript reduction using FocusHeuristics.
• Evaluated prediction accuracy using Pearson correlation and other metrics.

Main Results:

• Pearson correlation demonstrated the highest reliability in identifying the same drug across different conditions.
• Filtering for highly expressed genes significantly improved prediction accuracy.
• Network-based transcript reduction and using landmark genes (1000) did not decrease prediction accuracy.
• The proposed methods are reproducible and extendable with external data.

Conclusions:

• Matching drug effects across different concentrations or cell lines is a viable strategy for drug repurposing.
• Pearson correlation, combined with gene filtering and network analysis, provides a robust approach for predicting drug efficacy.
• The developed methods and available source code can accelerate the discovery of new applications for existing drugs.