Designing drugs at the atomic scale with computer simulation
Biochemistry and Molecular Biology & CMSE, Michigan State University
At the atomic level, the molecules in our bodies are in constant motion, and are undergoing constant change. The motions are incredibly rich; they range from the isomerization of side-chains, to the formation and destruction of large intermolecular complexes, to the birth and death of the molecules themselves. A deep understanding of these motions can radically improve our understanding of health and disease through rational design, where drugs target specific receptors, which are chosen for a specific molecular impact.
The Dickson Lab uses computational techniques such as molecular dynamics to simulate the motions of biomolecules (protein, RNA and DNA). These numerical experiments extend our knowledge beyond the “snapshots” provided by X-ray crystallography and NMR, and provide the entire landscape of conformations accessible to a molecular system. His goal is to use this technique to gain a deep understanding of the binding of small molecule drugs, which will be used to design small molecule therapeutics.
The main limiting factor to molecular dynamics (MD) studies is a limitation of timescales. Since MD operates by taking very small steps forward in time (about 1 femtosecond per step), we must compute a billion timesteps to even generate a trajectory that is a millionth of a second long! For this reason, MD has been traditionally limited to observing processes that occur on the nanosecond to microsecond timescales.
Breaking the timescale barrier with enhanced sampling methods for molecular dynamics
The Dickson Lab specializes in developing computational approaches that can overcome this timescale barrier. They use the WExplore method to observe ligand unbinding events that occur on timescales of seconds to minutes, which is over a million times longer that the reach of standard MD techniques. WExplore is built on an ensemble of trajectories that are moved forward in time in a parallel fashion, making it amenable to large-scale computing clusters like the High Performance Computing Center at Michigan State University.
This capability enables us to answer fundamental questions about the biophysics of ligand binding. By examining the ligand binding transition state we can discover the molecular interactions that determine how long a drug stays bound to its target, and predict changes to the drug that can extend residence times. We can study changes in protein stability upon ligand binding by examining the interplay between local stabilizing forces that govern binding affinity (“binding and bonding”) and indirect destabilizing effects resulting from the presence of the ligand (“binding and breaking”).
Visualizing structural ensembles using networks
After any large molecular simulation is performed, another challenge presents itself: how do we analyze and digest the massive quantity of data produced? Using clustering techniques and network modeling, the Dickson Lab employs conformation space networks to visualize entire free energy surfaces. The nodes of these networks represent specific molecular conformations, and the links (or, “edges”) in the network show which conformations can interconvert. As the network graph is created, nodes that interconvert are pulled together, and others are pushed apart, which reveals the underlying structure of the free energy landscape without having to project onto a chosen set of variables.
This gives us new ways to visualize dynamics, providing an easy way to judge the impacts of different environmental factors such as pH, temperature, or molecular crowding.
Connecting outward from molecular simulation
Experimental advances in parallel data collection and the standardization of data have led to a “database culture”: databases of protein-protein interactions, genetic diversity across populations, drug-protein interactions, somatic mutations in cancers, and biological pathways are all freely-available, and constantly growing. However, although advances in molecular dynamics allow us to determine thermodynamic and kinetic information for one- and two-body systems, methods to extend these results to their many-body, biological consequences are not well-developed.
The Dickson Lab also incorporates kinetic data into biological reaction network models. This can fundamentally extend the reach of simulation to larger timescales and system sizes by building up holistic models from individual reaction components. We are particularly interested in pharmacokinetic models of drug binding that include properties such as drug absorption, distribution, metabolism, excretion and toxicity (ADMET), as well as binding.
These holistic multiscale models employ an innovative combination of physics-based (for dynamics) and machine learning approaches (for ADMET), and will allow us to optimally balance toxicity and efficacy, widening the therapeutic windows in the earliest stages of the drug design process.