Prediction of Intramolecular Reorganization Energy Using Machine Learning

Facile
charge transport is desired for many applications of organic
semiconductors (OSCs). To take advantage of high-throughput screening
methodologies for the discovery of novel OSCs, parameters relevant
to charge transport are of high interest. The intramolecular reorganization
energy (RE) is one of the important charge transport parameters suitable
for molecular-level screening. Because the calculation of the RE with
quantum-chemical methods is expensive for large-scale screening, we
investigated the possibility of predicting the RE from the molecular
structure by means of machine learning methods. We combinatorially
generated a molecular library of 5631 molecules with extended conjugated
backbones using benzene, thiophene, furan, pyrrole, pyridine, pyridazine,
and cyclopentadiene as building blocks and obtained the target electronic
data at the B3LYP level of theory with the 6-31G* basis set. We compared
ridge, kernel ridge, and deep neural net (DNN) regression models based
on graph- and geometry-based descriptors. We found that DNNs outperform
the other methods and can predict the RE with a coefficient of determination
of 0.92 and root-mean-square error of ∼12 meV. This study shows
that the REs of organic semiconductor molecules can be predicted from
the molecular structures with high accuracy.