Data-driven prediction of genetic interactions in Candida glabrata

Abstract

The rapid acquisition of drug resistance by Candida glabrata has placed an increased burden on health systems. Combination therapies, which apply multiple known compounds, can thwart the compensatory behaviours that lead to drug resistance and can open therapeutic avenues that were not previously available. Genetic interactions, in which the loss of a pair of genes has a stronger effect than the loss of either of the genes alone, can be useful in identifying targets for combination therapy. However, the number of possible interactions is immense while experimentally establishing a genetic interaction is non-trivial. We are developing a method for predicting genetic interactions in C. glabrata as a means to prioritize the experimental validation of interactions and combination therapies. The method applies machine-learning techniques to infer genetic interactions from diverse data in C. glabrata and from other yeast species, such as coexpression, protein-protein interactions and coevolution. We are establishing the method by building a model for Saccharomyces cerevisiae to validate against the large-scale genetic interaction data available for that species. This model will then integrated with C. glabrata-specific data, validated against known genetic interactions, and used to produce novel predictions for further characterisation in the lab.