A next-generation extendable simulation environment for affordable, accurate, and efficient free
energy simulations
PI: Tai-Sung Lee, Rutgers University, Piscataway, NJ 08854-8087 USA.
Multi-scale molecular simulations are vital to scientific research, impacting protein and nucleic acid
engineering, biomaterials design, and drug discovery. AMBER, a simulation package with broad academic
and industrial appeal since the 1980s, supports over 30,000 users across the world today. AMBER’s enduring
popularity owes to its pmemd.cuda simulation engine, optimally implemented affordable Graphics Processing
Unit (GPU) platforms. Recently, the implementation was extended with modern free energy methods and
novel sampling algorithms to support a range of scientific applications, chief among them drug discovery by
prediction of protein-ligand binding affinities. A new generation of integrated methods, workflows, and
physical models are emerging, supported by a constellation of simulation software that includes NAMD,
CHARMM, Gromacs, OpenMM, and other programs. With these scientific advances a critical barrier to
progress in the field is the lack of a software package with seamless incorporation of high-performance
MD, free energy estimators, and emerging models. To surmount this barrier and in responding to NIH
Focused Technology Funding Opportunity which calls for innovative, focused technology development of a
working prototype of critical research tools, we propose to develop a next-generation executing environment.
The proposed work will offer fast prototyping, accessibility to new algorithms, further improvements in
single-GPU performance, strong scaling in synchronous and asynchronous ensemble methods, and
interoperability with different types of force fields and physical models. We propose to do the following: 1.
Environment: A Python-based executing environment for integrating various modules, including the
MD engine, the free energy module, and user-workflow control. This essential fabric for linking runtime
objects will create a foundation for connecting the molecular dynamics, model building, and analysis
components, and is ready for future growth of scientific and algorithmic advances. 2. Optimization: New
MD and free energy modules consisting of OOP CPU layers and high performance computing
(HPC) kernels, collectively optimized by runtime managing mechanisms. Tight coupling of the C++
layer and the HPC kernel extensions constitutes a system that optimizes performance on both the host and
HPC device. The goal is a set of efficient, robust, and extensible core simulation modules optimized at both
individual module level and their collectively execution. 3. Application: A suite for automatic deduction
of alchemical graphs. This facility, built in the new Python environment, will muster the new MD engine
and integrated capabilities to showcase their scaling and versatility. It will also serve as an example to
future developers of how to recombine the building blocks for advancing their own science.
