From Perception to Prediction and Interpretation: Enlightening the Gray Zone of Molecular Bricks of Life With the Help of Machine Learning and Quantum Chemistry

Abstract

The latest developments of a general exploration/exploitation strategy for the computational study of molecular bricks of life in the gas-phase are presented and illustrated by means of prototypical semi-rigid and flexible systems. In the first step, generalized natural internal coordinates are employed to obtain a clear-cut separation between different degrees of freedom, and machine-learning algorithms based on chemical descriptors (synthons) drive fast quantum chemical methods in the exploration of rugged potential energy surfaces ruled by soft degrees of freedom. Then, different quantum chemical models are carefully selected for exploiting energies, geometries, and vibrational frequencies with the aim of maximizing the accuracy of the overall description while retaining a reasonable cost for all the steps. In particular, a composite wave-function method is used for energies, whereas a double-hybrid functional is employed for geometries and harmonic frequencies and a cheaper global hybrid functional for anharmonic contributions. A panel of molecular bricks of life containing up to 50 atoms is employed to show that the proposed strategy draws closer to the accuracy of state-of-the-art composite wave-function methods for small semi-rigid molecules, but is applicable to much larger systems. The implementation of the whole computational workflow in terms of preprocessing and postprocessing of data provided by standard electronic structure codes paves the way toward the accurate yet not prohibitively expensive study of medium- to large-sized molecules by a user-friendly black-box tool exploitable also by experiment-oriented researchers.