Universally Exhaustive Generation of Molecular Structures and Prediction of Their Electronic States Using Machine Learning for N‐type Organic Transistor Materials. Issue 8 (3rd March 2023)

• Record Type: Journal Article
• Authors: Ohno, Akira; Hanna, Jun‐ichi; Iino, Hiroaki; Nakago, Kosuke; Yamaguchi, Taiki; Abe, Motoki; Akita, Hirotaka; Takemoto, Mizuki
• Is Part Of: Chemistry, an Asian journal. Volume 18:Issue 8(2023)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2023-03-03
• Subjects: Molecular generator -- Deep learning -- n type organic semiconductor -- organic transistor -- Electron affinity; Chemistry -- Periodicals; 540.5
• DOI: 10.1002/asia.202300029
• Languages: English
• ISSNs: 1861-4728

Abstract

Abstract: We have proposed a new method for the exploration of organic functional molecules, using an exhaustive molecular generator combined without combinatorial explosion and electronic state predicted by machine learning and adapted for developing n‐type organic semiconductor molecules for field‐effect transistors. Our method first enumerates skeletal structures as much as possible and next generates fused ring structures using substitution operations for atomic nodes and bond edges. We have succeeded in generating more than 4.8 million molecules. We calculated the electron affinity (EA) of about 51 thousand molecules with DFT calculation and trained the graph neural networks to estimate EA values of generated molecules. Finally, we obtained the 727 thousand molecules as candidates that satisfy EA values over 3 eV. The number of these possible candidate molecules is far beyond what we have been able to propose based on our knowledge and experience in synthetic chemistry, indicating a wide diversity of organic molecules. Abstract : An exhaustive molecular generator and electron affinity prediction is proposed based on machine learning for developing n‐type organic semiconductors. More than 4.8 million molecules are generated, and 727 thousand molecules of electron affinities are predicted above 3.0 eV. The approach can explore a wide variety of new molecules with functionality beyond our knowledge.