A Two fold approach to Predicting Density through Molecular Dynamics and Machine Learning

Abstract

In this analysis, the densities for 87 small molecules were predicted through a Gaussian Process Regressor. Features generated for training this model included temperature, Pressure, and Molecular coordinates. Densities were calculated through Classical Simulations to develop the target vector. Through hyper-parameter tuning, an MSE and R-squared value were found to measure model performance. Further tuning as well as an exploration of different molecular representations is needed to achieve accurate models of pre-screening/ pre-experimental procedures.