Biocomputation across distributed private datasets to enhance drug discovery

Project Summary Collaborative Drug DiscoveryCDDproposes to develop technology that will vastly simplify and integrate all the processes required to exploit predictive models for drug discoveryThe software will make it easy for scientists without specialized training in informatics to createtrainapplyevaluateshareand archive models with minimal effortand also leverage a large library of pre computed models with zero effortThe software will also enable scientists working in different organizations to collectively build models from their aggregated data and share these modelswithout sharing the underlying training dataOur goal is to democratize the role in drug discovery of computational modelswhich have historically been restricted to computational expertsand allow models to become routine aids to the discovery workflow in academiafoundationsgovernment laboratoriesand small companies that do not have the resources to employ them todayIn Phasewe implemented modified Bayesian model building directly within CDD s webbased CDD Vault platformwhich securely hosts structure activity relationshipSARdataany user can now easily train a Bayesian model with experimental data stored in her private Vaultthen apply the model to predict activity for untested compoundsIn PhaseB we propose to generalize this capability with the following new Specific Aimswhich are needed to achieve a widespread scientific and commercial impactAimIntegrate a suite of diverse computational techniquessuch as QSARNeural NetworksSupport Vector MachinesRandom Forestk Nearest Neighborsand possibly othersinto a single frameworkto allow direct side by side comparisonAimDevelop and validate a universal metric that ranks the predictive strength of each method as applied to a particular datasetAimApply the metric to automatically generate thousands of models from high qualitypublic access structure activity and ADME Tox datasets and present key results to the userAimDevelop a novel capability to build models collaborativelyby aggregating multiple datasetsand share the models without revealing the compounds and data in the training sets