SBIR-STTR Award

Key biophysical properties such as drug binding sites and enzyme catalysis arise can be computer-modeled using quantum mechanics, but limitations in the accuracy of practical quantum methods have held back progress. Over the past five years, this situation has changed with exciting, (and ongoing) improvements in the accuracy of density functional theory (DFT). New and better density functionals open new opportunities for applications in conformational searching, molecular recognition, ligand binding, and all the areas where ab initio calculations are employed in biophysical chemistry. However, these functionals require very large and computationally demanding basis sets to attain their high accuracy. Use of smaller basis sets leads to unconverged results with often unacceptable errors. There is an unmet need to significantly reduce the computational cost of achieving large basis set accuracy. The central innovation of this proposal is to use minimal adaptive basis functions (MAB) for this purpose, in place of traditional large basis sets. The MAB is a small (minimal) set of functions, adaptively formed from a traditional large basis via an atom-blocked, sparse transformation. The DFT calculation is performed in the adaptive basis, followed by a dual basis correction. This potentially permits very large computational speedups, while yielding accuracy virtually indistinguishable from a computationally costly calculation performed conventionally in the large target basis. The Phase I research has three principal objectives. First, the research will establish the accuracy of the MAB protocol for a range of biophysically relevant energy differences. Second, the research will lead to a carefully justified estimate of the speed-up that is attainable with the MAB approach, and will produce a new software implementation of several of the algorithmic steps that must be optimized. Third, modifications and improvements of the MAB approach will be sought as possible and needed. The results will lay the groundwork for basis set limit DFT calculations at greatly reduced computational cost, thereby potentially greatly expanding their usefulness for biophysical modeling.

Public Health Relevance Statement:
Project narrative This project will improve computer-based quantum mechanical modeling of biophysical problems, such as the conformational energies that determine drug binding, and the ability to understand enzymatic pathways. New software algorithms that greatly reduce the cost of reliable calculations will be developed.

Project Terms:
Algebra; Algorithmic Software; Algorithms; Amino Acids; Area; Back; base; Benchmarking; Binding; Biological; biophysical chemistry; biophysical model; biophysical properties; Biophysics; Catalysis; combinatorial; Computer Simulation; Computer software; Computers; cost; Data Set; density; design; Development; Drug Binding Site; Enzymes; improved; innovation; intermolecular interaction; Ligand Binding; Mechanics; Melatonin; Methods; Modeling; Modification; Molecular Conformation; molecular recognition; Pathway interactions; Pharmaceutical Preparations; Phase; polypeptide; Procedures; Production; Protocols documentation; prototype; quantum; Quantum Mechanics; Research; Running; Speed; System; theories; tool; virtual; Work

Biophysical properties can be computer-modeled using quantum mechanics (QM). While vastly more computationally costly than molecular mechanics (MM), QM methods are essential for bond-breaking and/or high accuracy. Indeed, QM methods have advanced with exciting, (and ongoing) improvements in the accuracy of density functional theory (DFT). These DFT improvements could open new applications opportunities reliable conformational searching, molecular recognition, ligand binding, enzymology modeling, and all the areas where QM simulations can aid biophysical chemistry. However, the latest density functionals require very large and computationally demanding basis sets to attain their high accuracy. Use of smaller basis sets leads to unconverged results with often unacceptable errors in relative energies, so only small systems can be treated at present with high accuracy DFT calculations. This proposal addresses the unmet need to reduce the computational cost of achieving large basis set accuracy in DFT calculations. Its first innovation is the use of minimal adaptive basis functions (MAB) for this purpose. The MAB is a small (minimal) set of functions, adaptively formed in situ from a traditional large basis via an atom-blocked, sparse transformation. The DFT calculation is performed in the MAB, followed by a single-shot perturbative correction. MAB accuracy has been shown to be virtually indistinguishable from a conventional large basis calculation on biophysically relevant examples, while analysis suggests the potential for more than an order-of-magnitude speedup. A second innovation to further extend the size of MAB-DFT calculations is a new MAB-based QM-in-QM method that exactly embeds a smaller active QM region described by a large basis into a larger QM environment that is pre- optimized in a smaller basis set. Large QM regions have been argued to be essential in QM/MM. The Phase II research has three principal objectives that together will bring the MAB-DFT method up to the level of application-ready software. First, the software implementation of the MAB-DFT method will be optimized to remove current bottlenecks, and to take full advantage of the block-sparse structure of the MAB in order to achieve the speedups the method is capable of yielding. This requires careful consideration of matrix element evaluation, numerical quadrature and linear algebra across all five steps in a MAB-DFT calculation. Second, the proposed MAB-based QM-in-QM embedding model will be implemented, using all optimizations from the first aim, as well as exact replacement of the environment by an effective core potential-like term. Third, timings and accuracy tests of both stand-alone MAB-DFT and QM-in-QM MAB-DFT calculations will be conducted and reported for a range of biophysically relevant energy differences in both model and realistic systems. Additionally, some model applications drawn from areas such drug design, enzymology and DNA/RNA chemistry will be performed.

Public Health Relevance Statement:
This project will improve computer-based quantum mechanical modeling of biophysical problems, such as the conformational energies that determine drug binding, the ability to understand enzymatic pathways, and validate more approximate simulation methods. Quantum modeling tools are widely used in both industry and academia for these purposes. These applications should benefit from the work to be performed in this project, which will greatly reduce the computational cost of obtaining high- accuracy, reliable results.

Project Terms:
Academia; Active Sites; Address; Algorithms; Ally; Area; base; Binding; biophysical chemistry; biophysical model; biophysical properties; Biophysics; Chemicals; Chemistry; Computer Simulation; Computer software; Computers; cost; Custom; density; deprotonation; Development; DNA; Drug Design; Elements; Environment; Enzymatic Biochemistry; Evaluation; flexibility; Goals; Hydrogen Bonding; improved; In Situ; Industry; innovation; intermolecular interaction; Ligand Binding; Linear Algebra; Mechanics; Methods; Modeling; Molecular Conformation; molecular mechanics; molecular recognition; Pathway interactions; Pharmaceutical Preparations; Phase; Procedures; quantum; Quantum Mechanics; Reaction; Recommendation; Reporting; Research; Resolution; RNA; Sampling; simulation; Speed; Structure; System; Testing; theories; Time; tool; Torsion; Update; virtual; Work