The central objectives of this research proposal are development and validation of methodologies to algorithmically encode underlying physical observables to improve design of small organic molecules for a biological target and their application to real world systems. Computational modeling at the atomic-level empowers understanding of the factors that drive molecular recognition and enables testable predictions that can be confirmed by experimentalists. Grounded in strong results and data, we hypothesize that major gaps in the field (i.e. pose accuracy, enrichment, protein flexibility, specificity, site complementarity, ease of use) can be bridged through forward-thinking design of tools that improve sampling, scoring, and searching. A major undertaking is development of a new platform for de novo design which will enable from-scratch construction of novel molecules, which removes the limitation of only considering those that are preconceived. This will enable design of compounds highly optimized and specifically tailored to the protein of interest. Our approach employs construction of molecules starting from user customizable libraries of building block fragments using algorithms we developed and implemented into the program DOCK6. New advances will be made available to the research community through public releases along with validation databases and user- friendly online tutorials. Without inventive approaches to ligand discovery, there is a high likelihood that certain areas of chemical space may not be adequately sampled by standard screening methods which provides the rational. Our expected outcomes are ensembles containing highly specific and optimized ligands. The proposal is framed around 4 fundamental questions: (Q1) What underlying physical principles that drive molecular recognition (binding, selectivity, resistance) can be captured at the atomic-level and used to design improved software and simulation protocols for accurate prediction of geometry and energy? (Q2) Can ligand growth be propelled to highly specific regions of chemical space through from-scratch assembly of small organic fragments (de novo design) using molecular mimicry principles to direct the growing ensemble as it evolves? (Q3) Which sampling, scoring, and searching methods are most effective for identification and design of verified-active compounds and can more effective practices be developed to maximize overall success in collaboration with experimentalist? (Q4) Can docking and de novo design software and protocols be designed to be more user friendly while not sacrificing accuracy or power? We will collaborate with a network of experienced experimental labs and employ our new tools to make predictions. We will identify small molecule probes and inhibitors to answer basic research questions and provide mechanistic understanding for biological systems of relevance to human health including: fatty acid binding protein, nSMase2, neutral ceramidase, HIVgp41, GP2, glycoprotein-E, ErbB-family mutants (EGFR, HER2), candid albicans Glx3/Hsp31, and human Tsg101, among others. Experimental outcomes in turn will inform our further method development efforts.
Atomic-level computer modeling allows prediction of molecular recognition which can enable greater understanding of biological systems relevant to public health. In this project we develop and validate new sampling, scoring, and searching techniques to increase the accuracy and ease of use of structure-based design with a focus on docking. In parallel, we are creating a new molecular assembly platform to facilitate from-scratch construction (de novo design, genetic algorithm) of small molecules with properties customized to the binding site(s) being targeted. We apply our tools to make testable predictions in collaboration with experimental groups, and we make them widely available to the research community.
