An Automated Workflow for Diagnosing Sampling Issues Caused by Slow Torsional Motions in Molecular Simulations

This article reports software that generates confidence metrics for physics-based simulations and derived numerical predictions. The described tools complement recent welcome progress in molecular dynamics (MD) based affinity prediction, such as OpenFE. Specifically, the presented “slow-rotations” software automates the identification of insufficiently sampled geometry in MD simulations, which can significantly impact calculated binding free energy. The authors focus the software’s analysis capabilities on torsional degrees-of-freedom, within the context of protein-ligand binding. Insufficiently sampled torsions in analysed trajectories are flagged, enabling critique of derived interpretations and indicating precisely where more extensive or alternate sampling should be performed.

The software’s simple methodology is refreshing and can be summarised as follows: (i) rotatable-bonds are automatically perceived via a commonly applied SMARTS pattern, (ii) simulated torsion conformational states, and bounds thereof, are identified using fits to their distributions (histograms) and peak-finding, (iii) torsion state transitions are then counted and (iv) they are flagged as “low” (or insufficient) if less than N transitions have occurred, where N=10 by default. Additional work around accounting for symmetry, in order to avoid highlighting “un-interesting” transitions, is also automatically performed.

The manuscript exemplifies this analysis in contexts of small molecule and protein-ligand systems. The code also lends itself to extension, for example, our previous work on microsecond-timescale ring geometry dynamic equilibrium convergence metrics took a very similar approach and could be integrated. The use of permissive licensing and open source dependancies, alongside the publication of comprehensive online documentation and examples, should facilitate adoption of the software. Overall, the “slow-rotations” workflow could be considered a useful and prudent addition to any MD analysis work.

Figure 1: Definition of a binding mode.

To read this article click here