The first molecular dynamics simulation of a protein was reported in
1977. Martin Karplus published (Karplus, M. Biopolymers 2003, 68, 350)
a brief history of molecular dynamics simulations of biological
macromolecules, and of the period in his laboratory, during which time
one of us (Levy) was a postdoctoral student in the group. It was
difficult at the time to carry out molecular dynamics simulations of
proteins using programs then available, and several members of the
Karplus group at Harvard University discussed their ideas about how the
situation might be improved.
The development of IMPACT began in the Levy group at Rutgers University,
in 1985, starting with a core set of molecular mechanics energy
subroutines used to carry out molecular dynamics simulations of
proteins. Early research using IMPACT focused on the relation between
simulations and NMR experimental studies of protein structure and
dynamics, and on protein solvation. In 1992, the Center for Theoretical
Simulation of Biological Systems was established by Richard Friesner at
Columbia University and Friesner, Levy and Bruce Berne began to
collaborate using IMPACT as a platform for methods development. In
1997, a strategic development partnership involving Columbia, Rutgers,
Yale, and Schrödinger LLC was
formed to provide a path for commercializing some of the new methods
being developed for simulations of protein structural changes and
interactions with ligands. Basic research projects have benefited
from this partnership, particularly by the development of tools to
automate the preparation of ligand–protein complexes, by the expanded
coverage of the force field, and by the increased coordination between
various modeling packages.
Molecular simulations of protein structural changes and ligand binding
are built upon two foundations: (1) the design of effective potentials
that are matched to the requirements of accuracy and speed appropriate
to particular modeling problems; and (2) the design of algorithms to
sample the effective potentials in highly efficient ways so as to
facilitate the convergence of the simulations in a thermodynamic sense
and/or the coverage over large databases containing structures for which
effective potential energy calculations are required. Developing
algorithms to satisfy the competing goals of accuracy and speed is at
the heart of the problem when considering computational models for use
in structural biology.
Since 2004 algorithm development within IMPACT has focused on the AGBNP
series of implicit solvation models and a physics based model for
estimating the free energy of binding of ligands to a protein receptor,
the Binding Energy Distribution Analysis
Method (BEDAM). The development of advanced sampling methods
within IMPACT based on replica exchange has been central to this
effort (see ASyncRE below). These projects
have been carried out as a collboration between Levy, Emilio
Gallicchio, with contributions from several of the students in the
Levy lab during the period of 2004-2014.
For more information about the development of the molecular mechanics
core technologies in Academic IMPACT, see:
Banks, J. L., Beard, H. S., Cao, Y., Cho, A. E., Damm, W., Farid,
R., Felts, A. K., Halgren, T. A., Mainz, D. T., Maple, J. R.,
Murphy, R., Philipp, D. M., Repasky, M. P., Zhang, L. Y., Berne, B.
J., Friesner, R. A., Gallicchio, E. and Levy, R. M. (2005),
Integrated Modeling Program, Applied Chemical Theory (IMPACT). J.
Comput. Chem., 26: 1752–1780. DOI: 10.1002/jcc.20292. PMCID:
Gallicchio E., Paris K., Levy R.M. (2009), The AGBNP2 Implicit
Solvation Model. J. Chem. Theory Comput., 5: 2544–2564.
DOI:10.1021/ct900234u. PMCID: PMC2857935.
Gallicchio E., Lapelosa M., Levy R.M. (2010), The Binding
Energy Distribution Analysis Method (BEDAM) for the Estimation of
Protein-Ligand Binding Affinities. J. Chem. Theory Comput., 6:
2961–2977. DOI:10.1021/ct1002913. PMCID: PMC2992355.
Asynchronous replica exchange (or ASyncRE for short) is an
implementation of the popular parallel replica exchange conformational
sampling algorithm. Unlike traditional synchronous implementations,
ASyncRE can scale to many hundreds of replicas over heterogeneous
grids of unreliable compute nodes for long running times.
The software can be found at the following GitHub repository:
code on GitHub
For more information, see:
Xia, Junchao, William F. Flynn, Emilio
Gallicchio, Bin W. Zhang, Peng He, Zhiqiang Tan, and Ronald M.
Levy (2015). Large-scale asynchronous and distributed
multidimensional replica exchange molecular simulations and efficiency
analysis. J. Comput. Chem., 36 (23), 1772-1785. DOI:
10.1002/jcc.23996. PMCID: PMC4512903.
Gallicchio, Emilio, Junchao Xia, William F.
Flynn, Baofeng Zhang, Sade Samlalsingh, Ahmet Mentes, and Ronald M.
Levy (2015). Asynchronous replica exchange software for grid and
heterogeneous computing. Comput. Phys. Commun., Web
publication. DOI: 10.1016/j.cpc.2015.06.010.
The BEDAM Binding Energy Distribution Analysis Method is an absolute binding
free energy estimation and analysis methodology based on a statistical
mechanics theory of molecular association and efficient computational
strategies built upon parallel Hamiltonian replica exchange, implicit
solvation and multi-state statistical inference. It has been implemented
into a python workflow that works within the Schrödinger environment with
the IMPACT MD engine.
For more information, see:
Gallicchio, Emilio, Mauro Lapelosa, and Ronald M. Levy (2010). Binding
Energy Distribution Analysis Method (BEDAM) for Estimation of Protein-Ligand
Binding Affinities. J. Chem. Theory Comput., 6, 2961-2977. DOI:
10.1021/ct1002913. PMCID: PMC2992355.
The BEDAM Workflow is available for download on Github.
The workflow takes as input
.mae files of receptor and ligand plus a
definition of the binding site region. Analysis of the results produce,
among other things, the estimated values of the binding free energy. See the
documentation and example provided with the source code.
Mi3-GPU ("Mee-three", for Markov-Chain Inverse Ising Inference) solves the inverse Ising problem for application in protein covariation analysis. The goal is to infer "coupling" parameters between positions in a Multiple Sequence Alignment (MSA) of a protein family, with many applications including protein-contact prediction and fitness prediction. Mi3-GPU solves the inverse Ising problem with few approximations using Markov-Chain Monte-Carlo methods with Quasi-Newton optimization, and the implementation is highly parallelized using GPUs with ~250x speedup vs CPU on typical problems. This enables the construction of "generative" models which reproduce the covariation patterns of the observed MSA with very high statistical precision, particularly suited for studying sequence variation on a sequence-by-sequence basis and MSA statistics, but can also be used in other common applications of covariation analysis.
This package also provides tools for analysis and preparation of protein-family MSAs to account for finite-sampling issues, which are a major source of error or bias in inverse Ising inference.
Available with user-guide at: Mi3-GPU on Github
Allan Haldane and Ronald M. Levy (2020). Mi3-GPU: MCMC-based Inverse Ising Inference on GPUs for protein
J.Computer Physics CommunicationsDOI: 10.1016/j.cpc.2020.107312
Allan Haldane and Ronald M. Levy (2019).
Influence of multiple-sequence-alignment depth on Potts statistical models of protein covariation Phys. Rev. Evol. 99 no. 3, pp. 032405. DOI: 10.1103/PhysRevE.99.032405.