Predicting Scissile Bonds and Activation Forces in Mechanochemical Reactions with Integrated Theoretical Computation and Machine Learning

Mechanophores are functional motifs in polymers that enable self-healing, force sensing, and the controlled release of small cargo molecules through mechanochemical reactions. However, predicting their reactivity remains challenging because these processes are governed by multiple structural and electronic factors. Here, we present an integrated framework that predicts both the location of the scissile bond and the corresponding activation force by combining mechanical and chemical descriptors. Density functional theory (DFT) calculations using the “External Force is Explicitly Included” (EFEI) formalism were used to generate the data set. A Graph Neural Network (GNN) classifier trained on 676 molecules was developed to identify scissile bonds, and an XGBoost regressor trained on 559 reactive molecules was used to predict activation-force magnitudes. The results show that local conjugation and ring strain are critical determinants of mechanochemical feasibility and scissile-bond location. Moreover, combining mechanical and chemical descriptors enables accurate prediction of activation forces. Validation on unseen molecular structures yielded an accuracy of 74% for scissile-bond prediction and a mean absolute error (MAE) of 0.362 nN for activation-force prediction, demonstrating both the accuracy and the generalizability of the dual-prediction framework across diverse mechanochemical reactions. This study establishes a systematic foundation for the rapid molecular-scale design of next-generation mechanophores and provides a scalable tool for the computational screening of force-responsive materials.