A Computing System for Discovering Causal Relationships among Human Genes to Improve Drug Repositioning

The automatic discovery of causal relationships among human genes can shed light on gene regulatory processes and guide drug repositioning. To this end, a computationally-heavy method for causal discovery is distributed on a volunteer computing grid and, taking advantage of variable subsetting and stratification, proves to be useful for expanding local gene regulatory networks. The input data are purely observational measures of transcripts expression in human tissues and cell lines collected within the FANTOM project. The system relies on the BOINC platform and on optimized client code. The functional relevance of results, measured by analyzing the annotations of the identified interactions, increases significantly over the simple Pearson correlation between the transcripts. Additionally, in 82% of cases networks significantly overlap with known protein-protein interactions annotated in biological databases. In the two case studies presented, this approach has been used to expand the networks of genes associated with two severe human pathologies: prostate cancer and coronary artery disease. The method identified respectively 22 and 36 genes to be evaluated as novel targets for already approved drugs, demonstrating the effective applicability of the approach in pipelines aimed to drug repositioning.