Theoretical and Computational Chemistry

Accelerating Organic Electronic Materials Design with Low-Cost, Robust Molecular Reorganization Energy Predictions



A critical bottleneck for the design of high-conductivity organic materials is finding molecules with low reorganization energy. The development of low-cost machine-learning-based models for calculating the reorganization energy has proven to be challenging. Here we combine a graph-based neural network recently benchmarked for drug design applications, ChIRo, with low-cost conformational features and show the feasibility of reorganization energy predictions on the benchmark QM9 dataset without needing DFT geometries.


Thumbnail image of 2022_11_29_Reorganization_Energy_Prediction_Communication_Manuscript_Preprint_Version.pdf

Supplementary material

Thumbnail image of 2022_11_29_Reorganization_Energy_Prediction_Communication_SI_Preprint_Version.pdf
Supporting Information
Supporting Information includes a description on the contents of raw data files provided at the Github repo, a description of the procedure for the curation of the dataset, the implementation of the machine learning models, and distributions for the number of conformers in the dataset.

Supplementary weblinks

Github Repo for Models and Data in the Paper
This repository includes: Curated QM9 dataset with their vertical IP, vertical EA, and reorganization energy. Code for training and evaluating the Modified ChIRo model. Code for training and evaluating the SchNet implemented with SchNetPack.