Conformational Sampling for Transition State Searches on a Computational Budget

Abstract

Transition state searches are the basis for computationally characterizing reaction mechanisms, making them a pivotal tool in myriad chemical applications. Nevertheless, common search algorithms are sensitive to reaction conformations, and the conformational spaces of even medium-sized reacting systems are too complex to explore with brute force. Here, we show that it is possible to train a classifier to learn the features of reaction conformers that conduce successful transition state searches, such that optimal conformers can be down-selected before incurring the cost of a high-level transition state search. The efficacy and transferability of this approach were tested using four distinct benchmarks comprising over three hundred individual reactions. Neglecting conformer contributions led to qualitatively incorrect activation energy estimations for a broad range of reactions, whereas simple random forest classifiers reliably down-selected lowbarrier reaction conformers for unseen reactions. The robust performance of these machine learning classifiers mitigates cost as a factor when implementing conformational sampling into contemporary reaction prediction workflows and opens up many avenues for further improvements as transition state data grow.

Algorithmic Explorations of Unimolecular and Bimolecular Reaction Spaces

“…The distribution of the ∆G † range across distinct reaction channels is reported to quantify the potential importance of conformational effects in determining the effective barrier height. 26 The success rate for converging TSs connecting reactants to each attempted product is nearly 100% in all cases. This is an extremely high value that reflects the efficiency of the prefiltering process in removing unrealistic products and reaction conformations in YARP v2.0 (see Fig.…”

“…Among the distinct features of YARP are the handling of conformers for conditioning the TS search and the use of multiple levels of theory to screen and refine TSs. Compared with contemporary methods, YARP dramatically reduces computational costs for characterizing organic reactions in the gas-phase 26,27 and at catalytic surfaces. 28 Nevertheless, as described in the original YARP publication, we envisioned several opportunities to further improve the method.…”

“…The manner in which this is done is extremely consequential for localizing meaningful TSs, especially for molecules with significant conformational freedom. 26 YARP v2.0 supports two algorithms for generating reaction geometries: the jointoptimization protocol published in the original YARP paper 27 and the conformational sampling plus joint-optimization algorithm that was recently developed by our group. 26 The latter algorithm uses CREST 31 to generate reactants and/or product conformers that are then joint-optimized and ranked by a random forest classification model (See Reference 26 for additional details).…”

“…26 YARP v2.0 supports two algorithms for generating reaction geometries: the jointoptimization protocol published in the original YARP paper 27 and the conformational sampling plus joint-optimization algorithm that was recently developed by our group. 26 The latter algorithm uses CREST 31 to generate reactants and/or product conformers that are then joint-optimized and ranked by a random forest classification model (See Reference 26 for additional details). Here, both schemes are retained to generate at most 1 + N reaction conformations for each attempted reaction (one from the joint-optimization and N from reactant conformational sampling).…”

“…To the best of our knowledge, a recent study by Grambow et al 23 (combined with revised AFIR results later published by Maeda et al 24 ) represents the only head-to-head comparison of this kind, albeit only for three distinct families of methods and a single unimolecular degradation problem. When it comes to assessing accuracy there are also the related questions of whether the lowest barrier transition state for a given reaction was discovered (e.g., a change of conformation can qualitatively modulate barriers and reaction mechanisms 25,26 ), whether the rank ordering of various channels is correct (e.g., a method with inaccurate absolute barriers that still reliably predicts the major product could still be useful), the level of theory of the final PES (e.g., choice of functional or use of a solvent model), whether the transition states are actually validated (e.g., using pathway sampling or an intrinsic reaction coordinate…”