(156f) Automatic Construction of Collective Variables for Metadynamics Simulations of Drug Permeation through Lipid Membranes

Understanding the mechanism of permeation of small molecules through cellular membranes and estimating their permeability coefficients are of tremendous importance to the pharmaceutical industry in order to improve design of a novel drug molecule or search for the best drug molecule from an existing library of numerous possible candidates. All atom molecular dynamics simulation is a promising method that can serve this purpose, as atomistic scale features can be effectively resolved. However, the simulation time required in crossing a large energy barrier, typically involved in the permeation process makes this method computationally infeasible to study permeation of drug molecules. This problem can be overcome by using enhanced free energy sampling methods that efficiently sample the entire free energy landscape. A bottleneck in the application of enhanced free energy sampling methods is determining collective variables (CVs) before starting the simulations. This step involves careful assessment of the system to select important degrees of freedom, which requires manual work for investigating every new drug molecule. In this work, we aim to overcome this problem by using time-structure based Independent Component Analysis (tICA). The tICA is an unsupervised machine learning technique to find degrees of freedom that decorrelate slowly in the system. We investigate the permeation of drug molecules through lipid membranes by using transition-tempered metadynamics (TTMetaD) simulations, and automatically construct the CVs for TTMetaD simulations based on the characteristics of the system showing slow dynamics by using tICA. Our findings demonstrate that tICA TTMetaD accelerates efficient exploration of the CV space and reduces the computational time to study drug permeation.