A theoretical study for the improvement of solvation model by machine learning

2022 Fiscal Year Final Research Report

• Project/Area Number: 19K05381
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 32010:Fundamental physical chemistry-related
• Research Institution: University of Tsukuba
• Principal Investigator: Matsui Toru 筑波大学, 数理物質系, 准教授 (70716076)
• Project Period (FY): 2019-04-01 – 2023-03-31
• Keywords: 溶媒和モデル / 機械学習 / 酸解離定数 / 酸化還元電位

Outline of Final Research Achievements

In this study, we improved the following three aspects by combining quantum chemical calculations with a solvation model:
(1) We obtained more accurate partition coefficients (logP) by using machine learning to correct solvation energy. (2) We analyzed oxidation potentials and identified significant error factors using Lasso regression. We compared experimental and calculated values for 114 organic compounds and performed an analysis using machine learning. (3) In the calculation of acid dissociation constants, conventional methods involved linear approximations based on calculations for compounds with the same functional groups. However, there were unnatural aspects in the reasoning and compound selection. Therefore, we proposed the derivation of acid dissociation constants using multiple regression to obtain more accurate results.

Free Research Field

計算化学

Academic Significance and Societal Importance of the Research Achievements

本研究を通して、量子化学と情報科学・データサイエンスとが融合する形態が第3ステージの量子化学になると考えて「量子化学3.0の時代」という造語を提唱するに至る段階になったと考えている。　「機械学習」や「人工知能」「自動化」など情報科学分野の進展は目覚ましい。それに付随して、機械学習・深層学習が多くの分野で普及が進んでいる昨今ではデータベース化がより進行している。したがって、今後の量子化学ではコンピュータによる自動的なデータ収集などが主流になると予測できるが、アウトプットとゴール(大抵の場合は実験値)との「差」をどう解釈するかは今後も課題であり続けるだろう。