Predicting Small Molecule Transfer Free Energies by Combining Molecular Dynamics Simulations and Deep Learning

Abstract

Accurately predicting small molecule partitioning and hydrophobicity is critical in the drug discovery process. There are many heterogeneous chemical environments within a cell and entire human body. For example, drugs must be able to cross the hydrophobic cellular membrane to reach their intracellular targets, and hydrophobicity is an important driving force for drug−protein binding. Atomistic molecular dynamics (MD) simulations are routinely used to calculate free energies of small molecules binding to proteins, crossing lipid membranes, and solvation but are computationally expensive. Machine learning (ML) and empirical methods are also used throughout drug discovery but rely on experimental data, limiting the domain of applicability. We present atomistic MD simulations calculating 15,000 small molecule free energies of transfer from water to cyclohexane. This large data set is used to train ML models that predict the free energies of transfer. We show that a spatial graph neural network model achieves the highest accuracy, followed closely by a 3D-convolutional neural network, and shallow learning based on the chemical fingerprint is significantly less accurate. A mean absolute error of ∼4 kJ/mol compared to the MD calculations was achieved for our best ML model. We also show that including data from the MD simulation improves the predictions, tests the transferability of each model to a diverse set of molecules, and show multitask learning improves the predictions. This work provides insight into the hydrophobicity of small molecules and ML cheminformatics modeling, and our data set will be useful for designing and testing future ML cheminformatics methods.

Mechanistic Understanding from Molecular Dynamics in Pharmaceutical Research 2: Lipid Membrane in Drug Design

“…Unsurprisingly, the permeation of small molecules through the lipid bilayer was among the first membrane-related topics studied through MD simulation, e.g., in 2004, Bemporad et al [149] studied permeation of small organic molecules like benzene or ethanol, and in older studies, Marrink and Berendsen (1994) studied permeation of water [564]. A significant amount of work has been performed to predict the permeability of small molecules through lipid bilayers using various molecular modeling methods and theoretical tools [565][566][567][568][569][570][571][572][573][574]; this includes new computational methodologies, developed specifically for the study of membrane permeability [575][576][577]. A web server and database PerMM is dedicated to gathering experimental and computational data related to small molecule membrane partitioning and translocation [578].…”

“…The only method available that allowed for sufficient sampling was replica-exchange umbrella sampling. Recently, Bennett et al, combined MD simulations with ML to study transfer free energies, i.e., free energy difference between two environments [571]. They calculated free energies of transfer from water to cyclohexane for 15,000 small molecules using MD simulations and next used obtained data to train the ML algorithm.…”

Computational compound screening of biomolecules and soft materials by molecular simulations

“…Harnessing the full potential of deep-learning models puts stringent requirement on the number of compounds, which severely restricts what can be achieved in terms of screening studies. Few MD studies have reached data-scale regimes amenable to deep learning, but impressive first steps show much promise, such as the prediction of transfer free energies in lipid membranes [202]. It offers a glance at the use of MD-based studies to train deep-learning models across the CCS of biomolecular and soft materials.…”

“…Such studies have led to the more routine incorporation of hydration free energies in validating force fields [78,79]. Scaling up, Bennett et al recently reported an impressive 15 • 10 3 water-cyclohexane transfer free-energy calculations from all-atom MD [202].…”

Pandemic drugs at pandemic speed: infrastructure for accelerating COVID-19 drug discovery with hybrid machine learning- and physics-based simulations on high-performance computers

“…A major application has been to enhance sampling in MD simulations which includes learning of optimal biasing potentials, optimal collective variables (CVs) or free energy surfaces [35][36][37][38][39][40][41][42]. Examples are also available for approaches that involve deriving MD-based descriptors that can be used to royalsocietypublishing.org/journal/rsfs Interface Focus 11: 20210018 train ML models for predictions of solvation, hydration and/or transfer free energies [43][44][45]. Studies have shown that the accuracy of alchemical free energy predictions may be improved by 'correcting' them through ML-based postprocessing [46][47].…”