Training Neural Nets To Learn Reactive Potential Energy Surfaces Using Interactive Quantum Chemistry in Virtual Reality

Abstract

Whilst the primary bottleneck to a number of computational workflows was not so long ago limited by processing power, the rise of machine learning technologies has resulted in an interesting paradigm shift, which places increasing value on issues related to data curation -i.e., data size, quality, bias, format, and coverage. Increasingly, data-related issues are equally as important as the algorithmic methods used to process and learn from the data. Here we introduce an open source GPUaccelerated neural network (NN) framework for learning reactive potential energy surfaces (PESs), and investigate the use of real-time interactive ab initio molecular dynamics in virtual reality (iMD-VR) as a new strategy which enables human users to rapidly sample geometries along reaction pathways which can subsequently be used to train NNs to learn efficient reactive PESs. Focussing on hydrogen abstraction reactions of CN radical with isopentane, we compare the performance of NNs trained using iMD-VR data versus NNs trained using a more traditional method, namely molecular dynamics (MD) constrained to sample a predefined grid of points along the hydrogen abstraction reaction coordinate. Both the NN trained using iMD-VR data and the NN trained using the constrained MD data reproduce important qualitative features of the reactive PESs, such as a low and early barrier to abstraction. Quantitative analysis shows that NN learning is sensitive to the dataset used for training. Our results show that user-sampled structures obtained with the quantum chemical iMD-VR machinery enable excellent sampling in the vicinity of the minimum energy path (MEP).As a result, the NN trained on the iMD-VR data does very well predicting energies which are close to the (MEP), but less well predicting energies for 'off-path' structures. The NN trained on the constrained MD data does better predicting high-energy 'off-path' structures, given that it included a number of such structures in its training set.

Interactive molecular dynamics in virtual reality from quantum chemistry to drug binding: An open-source multi-person framework

“…In recent work, we have shown that exploration of molecular configuration space by human participants using iMD-VR to steer 'on-the-fly' ab initio MD can be used to generate molecular geometries for training GPU-accelerated neural networks (NN) to learn reactive potential energy surfaces (PESs). 13 Video 5 (vimeo.com/311438872) shows our first application using this strategy, focused on hydrogen abstraction reactions of CN radical + isopentane using real-time semi-empirical quantum chemistry through a plugin to the SCINE Sparrow package developed by Reiher and co-workers [85][86][87] (scine.ethz.ch), which includes implementations of tight-binding engines like DFTB alongside a suite of other semi-empirical methods. 47 To obtain the results described herein, we have utilized the SCINE Sparrow implementation of PM6, with the default set of parameters.…”

Interactive Molecular Dynamics in Virtual Reality (iMD-VR) Is an Effective Tool for Flexible Substrate and Inhibitor Docking to the SARS-CoV-2 Main Protease

“…[43] Narupa, the software framework that we have developed, is open source, and uses commodity VR hardware, which is widely accessible. [29,31,33,44] It can therefore readily be used. We make our Mpro simulations here freely available; these can be used for ligand discovery and structure-activity studies for SARS-CoV-2.…”

MLatom: A program package for quantum chemical research assisted by machine learning

“…The Matérn kernel has several interesting properties which makes it an increasingly popular choice . First, it reduces to the exponential kernel for n = 0: which is not implemented separately in MLatom .…”