Computational Study on Data Integration in Chiral Catalyst Design: A Case Study Using Michael Addition Reactions

Data science in molecular catalysis
often relies on the
idea that
molecules with similar structures show similar properties. This concept
underlies many successful examples of regression analysis based on
free energy relationships, typically using data sets in which substituents
are varied on a single catalyst or substrate scaffold. Here, we present
data integration for chiral catalyst design via regression using three-dimensional
image-like descriptors (voxel data) derived from 718 TS (transition-state)
structures calculated through DFT methods, along with their corresponding
computed ΔΔG‡ values
(energy differences between TSs leading to each enantiomer). These
data sets covered seven reaction types, including organocatalysis,
transition-metal catalysis, and both Michael addition and Diels–Alder
reactions. Regression using integrated data sets from seven pairs
of distinct reaction systems enabled the design of chiral catalysts
with improved computed enantioselectivity in all seven catalytic systems.
To facilitate applications, we have released a web platform (https://mcds.riken.jp) that offers a data-driven design tool and a database of DFT-computed
TS data.