Bioactive small molecules can act on multiple targets, and these off-target activities underlie many of the adverse reactions from which drugs suffer. The motivating idea of this proposal is that these Adverse Drug Reaction (ADR) targets may be predicted comprehensively and systematically using chemoinformatic inference. The "Similarity Ensemble Approach" (SEA), developed by us, classifies targets based on their ligands rather than their sequence identity, structural similarity, pathway role or function, and predicts associations that are otherwise inaccessible and often surprising. In the first phase of this SBIR we constructed a global target map using ligand similarity, exploiting this to predict off-targets. This map suggested mechanism of action targets for several drugs, candidates for repositioning-both aims in the first phase-and predicted ADR associations. It is this latter goal that we focus on here. Extensive preliminary results, including a collaboration with pharma and a proof-of-concept federal collaboration, support the scientific and financial pragmatism of exploiting this platform for predicting adverse off-targets. In the second phase of this project we develop a direct drug-target-ADR map and develop new techniques to make the method more robust.
The specific aims are: 1. To create a full drug-target-ADR map, and demonstrate proof-of-concept. We propose to create a direct drug-target-ADR map. This will be done comprehensively across all approved and investigational drugs, all accessible targets, and all adverse reactions for which targets may be associated. We anticipate that the most important commercial use of these methods and this map will be to prioritize ADR targets to test against for molecules that are clinical and preclinical candidates. To show proof of concept against such molecules, we will test investigational drugs for their ability to modulate ADR targets predicted for them. 2. To improve SEA with new methods and new ligand-physiology databases. To make the ligand- target-ADR association map more robust, we will improve the methods and databases underlying SEA. (a) We will develop descriptors of and filters for physical properties of molecules, rather than using ligand topology alone. (b) We will cluster target ligands, rather than assuming they always form a single, cohesive set. (c) We will incorporate ligand affinity weighting into SEA and test the resulting predictions. (d) Finally, we will derive drug-ADR associations from predicted target profiles, rather than relying on individual target predictions alone. Substantial preliminary results support the promise of our platform for predicting target-based drug adverse events. These are among the most common reasons for drug failures in clinical trials, and there has thus been great interest in this method. The studies proposed here have the potential to greatly improve the breadth and reliability of the method and, correspondingly, its commercial application.
Drugs can act on more targets in the body than intended. These off-target interactions underlie many adverse drug reactions and are difficult to anticipate. In this proposal we improve on a computational method to predict undesirable off-targets proactively, to save time and money in drug discovery by improving drug safety.