Novel deep learning strategy to better predict pharmacological properties of candidate drugs and focus discovery efforts

PROJECT SUMMARY Collaborative Drug DiscoveryIncCDDproposes to develop a novel approach based on deep learning neural networks to encode molecules into chemically rich vectorsWe will first apply this representation to build more powerful computational models that can more accurately predict properties such as bioactivityADMEToxand pharmacokinetics across libraries of molecular structuresThe ultimate goal is to leverage this representation to generate novel compounds with better combinations of propertiesBoth of these capabilities will help scientists to accelerate discovery of new drugs broadly across many therapeutic areasScientists engaged in drug discovery research from academic laboratories to large pharmaceutical companies rely on computational QSAR models to predict pharmacologically relevant properties and obviate the need to perform expensivetime consuming assaysmany of which require animal studiesfor every molecule of interestSome propertiese glogPcan now be modeled with such high confidence that the models have replaced the need to perform the assaysbut many other critical propertiese gsolubilityADMEPKhERGremain far from this goalWe expect that our proposed chemically rich vectors will significantly advance the state of the art beyond what can be achieved with conventional descriptors and fingerprintsImproved models will enable researchers to select lead candidate series more effectivelyexplore chemical space around leads to generate novel IP more efficientlyreduce failure rates for compounds advancing through the drug discovery pipelineand accelerate the entire drug discovery processThese benefits will be realized broadly across most therapeutic areasOur central innovation is a novel computational strategyfirst develop a deep learningDLmodel optimized to best capture the essential structural and chemical features of moleculesstarting from the most natural structural representationthen validate the DL model by applying it to improve QSAR modeling of pharmacological propertiesand finally extend it to generate previously unknown molecules that have superior propertiesthe so calledinverse QSARproblemwhich is the Holy Grail of computational medicinal chemistryOthers have unsuccessfully tried to leap directly to solve the inverse QSAR problemWe propose a more patient and methodical approach that will allow the neural network to perform self supervised training to learn about chemical structures and properties from readily availableextremely large datasetsthen transfer this learning to improve modelingonly after establishing this solid foundation do we intend to apply the models to attempt inverse QSARPrior attempts in this area have also relied on neural network architectures designed originally for language processingWe will design a new architecturemore akin to neural network architectures that have proven most successful at image classificationand optimize it to directly process themolecular graphthat represents the relationship of atoms and bonds in molecules PROJECT NARRATIVE The proposed project will create novel computational tools that will help researchers to understand whether potential new drugs are likely to be both safe and effectiveand identify similar compounds that are likely to be safer and more effective against the same targetThis innovative capability will help to accelerate the discovery and development of novel and improved drugs against a wide range of diseases!