Accurate and Efficient Solvent Models for Molecular Simulations: Methods and Biological Applications
openNIGMS - National Institute of General Medical Sciences
Progress in modern bio-molecular sciences, from structural biology to structure-based drug design, is greatly
accelerated by methods of atomic-level modeling and classical simulations that bridge the gap between theory
and experiment; 45,000+ research papers that use these methods are published each year. Accurate and
computationally facile water models are just as important for outcomes of these studies as water is for Life. In
practice, several principal levels of compromise exist between level of detail and speed of solvent models.
However, critical accuracy and performance gaps remain at each level, these gaps mute the strong potential of
atomistic modeling. For example, even with most detailed (explicit) water models, significant discrepancies with
experimental binding free energies are still seen, which is one critical factor that hampers rational drug design
efforts. Another problem is computational cost, which can become prohibitive when most accurate existing
models are used. On the other hand, in many areas, which can benefit from faster, less detailed (so-called
“implicit solvent”) water models, simulations based on these faster models are often unreliable. New solvent
models appear regularly, but these are often limited to re-parameterizations of old ones, or utilization of old
“base models” to add key new features such as electronic polarization. My lab has always focused on ground
up, physics-based approaches to model development, which are more likely than many alternatives to yield
robust, transferable models that stand the test of time. The previous funding period has enabled us to
accumulate a critical mass of innovations in the field of solvent model development, innovations that have
already shown significant promise in practical applications. Importantly, the reported improvements in water
model accuracy came without sacrificing the speed. The goal for the next 5 years is to move the entire field of
atomistic simulations to the next level of predictive accuracy by delivering to the community a novel class of
solvent models, at each key level of detail--speed compromise. To demonstrate utility of the new models (once
thoroughly tested), we will apply them to: (1) Improving the accuracy, without sacrificing speed, of estimation of
receptor-ligand binding free energies. In structure-based drug discovery, the accuracy and computational
efficiency of in-silico binding free energy predictions for small molecules to biomolecular targets are crucial for
high-throughput screening of drug candidates. (2) Generation of novel insights into regulation of DNA
accessibility in the nucleosome, which directly affects gene expression. Understanding how DNA accessibility
in the nucleosome is controlled/affected by various biologically relevant factors is a fundamentally important
problem of direct biomedical relevance. The disruption of the histone functions leads to diseases.
In addition, the novel, higher accuracy, yet efficient solvent models will be implemented into H++ web-server
maintained by the PI (12,000+ registered users, 20,000+ requests per year), thus immediately improving
outcomes of structure preparation and analysis efforts for a large modeling community.
Up to $411K
health research