Local Kernel Regression and Neural Network Approaches to the Conformational Landscapes of Oligopeptides

Abstract

The application of machine learning to theoretical chemistry has made it possible to combine the accuracy of quantum chemical energetics with the thorough sampling of finitetemperature fluctuations. To reach this goal, a diverse set of methods has been proposed, ranging from simple linear models to kernel regression and highly nonlinear neural networks. Here we apply two widely different approaches to the same, challenging problem: the sampling of the conformational landscape of polypeptides at finite temperature. We develop a local kernel regression (LKR) coupled with a supervised sparsity method and compare it with a more established approach based on Behler-Parrinello type neural networks. In the context of the LKR, we discuss how the supervised selection of the reference pool of environments is crucial to achieve accurate potential energy surfaces at a competitive computational cost and leverage the locality of the model to infer which chemical environments are poorly described by the DFTB baseline. We then discuss the relative merits of the two frameworks and perform Hamiltonian-reservoir replica-exchange Monte Carlo sampling and metadynamics simulations, respectively, to demonstrate that both frameworks can achieve converged and transferable sampling of the conformational landscape of complex and flexible biomolecules with comparable accuracy and computational cost.

From Organic Fragments to Photoswitchable Catalysts: The OFF–ON Structural Repository for Transferable Kernel-Based Potentials

“…A more detailed analysis across different functional groups present in organocatalysts shows consistent improvement by the LKR−OMP correction (Figure 3C and Table S4 in Supporting Information), such that predictions can be made on highly complex and diverse data sets and the desired FESs can be reliably constructed. 79 4.2. Extrapolation.…”

“…Our primary objective is to obtain the FES of any large and flexible organic molecule at a hybrid DFT level of theory while simultaneously circumventing the computational cost imposed by ab initio MD. To address this, we choose to use the in-house developed LKR–OMP model, which allows a robust and accurate machine-learning potential (MLP) to be developed with reduced training time compared to conventional neural network potentials (i.e., Behler–Parrinello-type neural networks) . Subsequent analysis focuses on accomplishing this objective.…”

“…Proceeding in this manner bypasses the tedious process of properly choosing the best atomic environments to optimize the energy predictions as the OMP selects the best set of atomic environments for each atom type based on those available within the database. Thus, LKR–OMP possesses the transferability and the scalability of neural networks while benefiting from the additional stability and reduced training cost of kernel-based methods …”

Reply to Comment on ‘Physics-based representations for machine learning properties of chemical reactions’

“…While open questions undoubtedly exist surrounding interpolation vs. extrapolation as well as designing challenging out-of-sample test sets such as those containing new chemistry [26,28,[42][43][44][45][46][47][48][49], larger [9,50,51] and/or outlier [52] molecules/materials, or new sets appearing over time [53][54][55], this topic remains an ongoing community-wide discussion with no clear best practices, and will not be discussed further here.…”

Accurate Electronic and Optical Properties of Organic Doublet Radicals Using Machine Learned Range-Separated Functionals

“…Their advantages become more significant when molecular and dataset sizes are larger. [172][173][174][175][176][177][178][179] Also, earlier studies demonstrated that stacked generalization has a stronger predictive power than every single model. [89][90][91][92][93][94][95] In our model, each base learner generates a non-linear quantitative relationship between the CMD the optimal ω ML .…”