CHQuant: A Protocol for Quantifying Conformational Sampling with Convex Hulls

Conformational sampling is a key component of the workflow behind any machine learning force field. Existing methods for assessing the coverage of a conformational space can be ambiguous, reference-dependent, and qualitative due to their reliance on visual perception. In this work, we propose CHQuant, a protocol designed to quantify the extent of conformational space coverage. This approach is based on the observation that the movement of atoms defines point clouds that can be characterized by the volumes of the corresponding convex hulls. By applying our method to several small molecules and conformational sampling methods, we show that convex hull volumes behave according to chemical intuition at an atomic, functional group, and molecular level. Therefore, convex hull volumes are useful measures that can be used to compare different sampling methods or data sets. Among many applications, the CHQuant approach monitors the sampling achieved by any molecular dynamics sampler and compares existing conformational databases of the same system.