Fast exploration of potential energy surfaces with a joint venture of quantum chemistry, evolutionary algorithms and unsupervised learning

Abstract

Contemporary molecular spectroscopy allows to study flexible molecules, whose conformational behavior is ruled by flat potential energy surfaces (PESs) involving a large number of energy minima with comparable stability. Under...

Benchmark Structures and Conformational Landscapes of Amino Acids in the Gas Phase: A Joint Venture of Machine Learning, Quantum Chemistry, and Rotational Spectroscopy

“…In order to further increase the coverage of the conformational space, 4 runs with different initial populations are performed for each molecular system. The full set of parameters employed in the IM-EA algorithm is given in Table S1 of the Supporting Information (SI), while further details are given in refs and . Figure shows a schematic flowchart of the current version of the whole algorithm, which is available under the GPL3 license at .…”

“…This situation calls for an integrated computational approach employing QC models of increasing accuracy in the different steps of an exploration/exploitation strategy guided by machine learning (ML) tools. ,− The effective strategy of this kind we have been developing in the past few years starts from a knowledge-based selection and constrained geometry optimizations of a limited number of conformers employing a fast semiempirical method. , Next, an effective exploration of the whole conformational PES is performed by the same semiempirical method guided by a purposely tailored evolutionary algorithm with the aim of finding other low-lying minima . The results of this step are refined by hybrid and then double-hybrid density functionals, , and possible relaxation paths between pairs of adjacent energy minima are identified .…”

Surrogate Based Genetic Algorithm Method for Efficient Identification of Low-Energy Peptide Structures

“…This shows that it is important to sample not only as many low-lying minima as possible at the surrogate level but also those that are farthest away and likely to relax into distinct DFT minima. An effective approach in this sense would be to select distant structures using clustering, 104 FPS, 102 or RMSD analysis prior to reoptimization and avoid irrelevant relaxations due to small numerical differences.…”

Toward Accurate yet Effective Computations of Rotational Spectroscopy Parameters for Biomolecule Building Blocks

“…For the reasons mentioned above, the accurate characterization needed by rotational spectroscopy requires an integrated computational approach that employs QC models of increasing accuracy in the different steps of an exploration/exploitation strategy guided by machine learning (ML) tools. The main steps of this strategy [ 10 , 12 , 13 , 14 ] can be summarized as follows: Unsupervised perception of the molecular system to identify hard and soft degrees of freedom [ 15 ]; Knowledge-based selection and constrained geometry optimizations of a limited number of conformers employing a fast semi-empirical method [ 11 , 16 ]; Exploration of the PES governed by soft degrees of freedom using the same semi-empirical method of the previous step, guided by a purposely tailored evolutionary algorithm with the aim of finding other low-lying minima [ 10 ]; Refinement of the most stable structures by hybrid and then double-hybrid density functionals [ 14 ]; Analysis of relaxation paths between pairs of adjacent energy minima [ 13 ]; Evaluation of accurate electronic energies for the final panel of low-energy minima [ 17 , 18 , 19 ]; Computation of zero point energies (ZPE) and thermal contributions to enthalpies and entropies [ 20 , 21 , 22 , 23 , 24 , 25 ]; Computation of spectroscopic parameters for the energy minima with non negligible populations [ 13 , 26 ]. …”

“…Refinement of the most stable structures by hybrid and then double-hybrid density functionals [ 14 ];…”