Machine Learning for Reaction Performance Prediction in Allylic Substitution Enhanced by Automatic Extraction of a Substrate-Aware Descriptor

Abstract

This work introduces SubA, a substrate-aware descriptor for predicting Ir-catalyzed allylic substitution performance. Graph matching identifies molecular backbones, while atomic and molecular properties provide compact, interpretable features that improve accuracy and generalization over mainstream descriptors.

In brief

A substrate-aware descriptor that improves reaction-performance prediction while preserving interpretability.