Project Grant R44GM152994

BINDING FREE ENERGY PREDICTIONS FOR PEPTIDE DRUGS: A NOVEL MACHINE LEARNING-BASED PEPTIDE SIMULATION MODEL FOR HIGH-THROUGHPUT COMPUTATIONAL LEAD OPTIMIZATION WITH QUANTUM MECHANICAL ACCURACY. - PROJECT SUMMARY OLIGOPEPTIDES REPRESENT A POWERFUL THERAPEUTIC CLASS FOR TARGETING CHALLENGING PROTEIN-PROTEIN INTERACTIONS IMPLICATED IN NUMEROUS HUMAN DISEASES, YET THEIR RATIONAL DESIGN IS HINDERED BY THEIR CONFORMATIONAL FLEXIBILITY. THIS CREATES A DUAL CRISIS IN PREDICTIVE SIMULATION, WHERE BOTH THE ACCURACY OF THE UNDERLYING FORCE FIELDS (ENERGETICS) AND THE ABILITY TO EXPLORE THE VAST CONFORMATIONAL SPACE (SAMPLING) ARE INADEQUATE. CURRENT COMPUTATIONAL TOOLS SUFFER FROM SYSTEMIC INACCURACIES, A LACK OF PARAMETERS FOR NOVEL CHEMISTRIES, AND A FAILURE TO CONVERGE CRITICAL CALCULATIONS LIKE BINDING FREE ENERGIES, CREATING A MAJOR BOTTLENECK IN PEPTIDE DRUG DISCOVERY. THIS PROJECT WILL DELIVER A TRANSFORMATIVE COMPUTATIONAL PLATFORM THAT RESOLVES THIS CRISIS. THE LONG-TERM OBJECTIVE IS TO ENABLE THE RELIABLE, FIRST-PRINCIPLES PREDICTION OF PEPTIDE BINDING AFFINITY AND KEY ADME PROPERTIES. THE CENTRAL INNOVATION IS A SYNERGISTIC PLATFORM THAT COMBINES A SECOND GENERATION VERSION OF QUAIPA, A SELF-IMPROVING, QUANTUM-ACCURATE MACHINE-LEARNED FORCE FIELD (ML-FF) FIRST DEVELOPED IN PHASE I, WITH QUELO-PE, A PROPRIETARY, GPU-ACCELERATED QM/MM SIMULATION ENGINE ENHANCED FOR PROTEIN DISCOVERY ("- PE". QUAIPA'S GRAPH-BASED ARCHITECTURE IS AUTOMATICALLY EXTENSIBLE TO NOVEL CHEMISTRIES, SOLVING THE PARAMETER GAP FOR MM, WHILE QUELO-PE WILL SIMULTANEOUSLY PROVIDE ENHANCED SAMPLING METHODS REQUIRED FOR PEPTIDES AND OTHER TOOLS TO ALLOW FAST QM/MM-MD TO BE APPLIED TO OLIGOPEPTIDES. THE SPECIFIC AIMS OF THIS PHASE II PROJECT ARE TO: (1) SPECIALIZE AND OPTIMIZE THE QUELO/QUAIPA PLATFORM FOR PEPTIDE-SPECIFIC CHALLENGES, INCLUDING THE INTEGRATION OF ADVANCED SAMPLING METHODS; (2) IMPLEMENT A COMPREHENSIVE VALIDATION AND BENCHMARKING FRAMEWORK TO ENSURE PREDICTIVE ACCURACY AND ESTABLISH USER CONFIDENCE; (3) DEVELOP AND VALIDATE NOVEL WORKFLOWS FOR PREDICTING HIGH-VALUE ADME PROPERTIES, SPECIFICALLY PASSIVE MEMBRANE PERMEABILITY AND METABOLIC STABILITY; AND (4) INTEGRATE ALL COMPONENTS INTO A SEAMLESS, USER-FRIENDLY SOFTWARE PLATFORM WITH BUILT-IN UNCERTAINTY QUANTIFICATION. BY OVERCOMING THE INTERTWINED BOTTLENECKS IN FORCE FIELD ACCURACY AND CONFORMATIONAL SAMPLING, THIS PLATFORM WILL DRAMATICALLY ACCELERATE THE DISCOVERY AND OPTIMIZATION OF NOVEL PEPTIDE THERAPEUTICS. THIS WORK IS HIGHLY RELEVANT TO THE NIH'S MISSION AS IT WILL PROVIDE A POWERFUL NEW TOOL TO DEVELOP MEDICINES FOR PREVIOUSLY UNDRUGGABLE TARGETS, ULTIMATELY REDUCING THE TIME, COST, AND FAILURE RATE OF BRINGING NEW TREATMENTS FOR A WIDE RANGE OF HUMAN DISEASES TO THE CLINIC.