Accelerating virtual discoveries by augmenting quantum mechanics with machine learning

Acceleration of socio-economically important researches such as the design of catalysts, drugs, or conducting materials, lies in reliable virtual screening to identify candidate molecules or materials with desired properties. Any attempt to address this problem exclusively via brute force high-throughput computation is doomed to fail due to the combinatorial hardness of the problem, and the limitations of the compute power that is available on the planet. In my talk, I will highlight some out-of-the-box approaches to navigate chemical space with a focus on the application of supervised machine-learning combined with legacy quantum chemistry methods such as even semi-empirical models. This strategy has very recently been shown to reach desirable quantum chemical accuracy, for forecasting a multitude of properties, ranging from thermochemistry to NMR chemical shifts, even for new molecules which had no part in training. I will present an overview of this emerging sub-discipline of theoretical chemistry, and discuss some of the prominent contributions in this venue.

References
[1] Snyder, et al., Finding density functionals with machine learning, Physical Review Letters, 108 (2012) 253002.
[2] Ramakrishnan, et al., Big data meets quantum chemistry approximations: The ∆-machine learning approach, Journal of Chemical Theory and Computation, 11 (2015)2087.
[3] Ghiringhelli, et al., Big data of materials science: Critical role of the descriptor, Physical Review Letters, 114 (2015) 105503.
[4] Special Issue: Machine learning and quantum mechanics, International Journal of Quantum Chemistry, 115 (August 2015).