Inference of Protein Complex Activities from Chemical-Genetic Profile and Its Applications: Predicting Drug-Target Pathways

The chemical-genetic profile can be defined as quantitative values of deletion strains' growth defects under exposure to chemicals. In yeast, the compendium of chemical-genetic profiles of genomewide deletion strains under many different chemicals has been used for identifying direct target proteins and a common mode-of-action of those chemicals. In the previous study, valuable biological information such as protein–protein and genetic interactions has not been fully utilized. In our study, we integrated this compendium and biological interactions into the comprehensive collection of ∼490 protein complexes of yeast for model-based prediction of a drug's target proteins and similar drugs. We assumed that those protein complexes (PCs) were functional units for yeast cell growth and regarded them as hidden factors and developed the PC-based Bayesian factor model that relates the chemical-genetic profile at the level of organism phenotypes to the hidden activities of PCs at the molecular level. The inferred PC activities provided the predictive power of a common mode-of-action of drugs as well as grouping of PCs with similar functions. In addition, our PC-based model allowed us to develop a new effective method to predict a drug's target pathway, by which we were able to highlight the target-protein, TOR1, of rapamycin. Our study is the first approach to model phenotypes of systematic deletion strains in terms of protein complexes. We believe that our PC-based approach can provide an appropriate framework for combining and modeling several types of chemical-genetic profiles including interspecies. Such efforts will contribute to predicting more precisely relevant pathways including target proteins that interact directly with bioactive compounds.Author Summary: Finding the specific targets of chemicals and deciphering how drugs work in our body is important for the effective development of new drugs. Growth profiles of yeast genomewide deletion strains under many different chemicals have been used for identifying target proteins and a common mode-of-action of drugs. In this study, we integrated those growth profiles with biological information such as protein–protein interactions and genetic interactions to develop a new method to infer the mode-of-action of drugs. We assume that the protein complexes (PCs) are functional units for cell growth regulation, analogous to the transcriptional factors (TFs) for gene regulation. We also assume that the relative cell growth of a specific deletion mutant in the presence of a specific drug is determined by the interactions between the PCs and the deleted gene of the mutant. We then developed a computational model with which we were able to infer the hidden activities of PCs on the cell growth and showed that yeast growth phenotypes could be effectively modeled by PCs in a biologically meaningful way by demonstrating that the inferred activities of PCs contributed to predicting groups of similar drugs as well as proteins and pathways targeted by drugs.