2022 Fiscal Year Final Research Report
Research on inverse analysis and scientific interpretation of property prediction models
| Project/Area Number |
19K15352
|
|---|
| Research Category |
Grant-in-Aid for Early-Career Scientists
|
| Allocation Type | Multi-year Fund |
|---|
| Review Section |
Basic Section 27020:Chemical reaction and process system engineering-related
|
|---|
| Research Institution | Meiji University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2019-04-01 – 2023-03-31
|
| Keywords | 適応的実験計画法 / 能動学習 / 直接的逆解析 / 予測精度 / ベイズ最適化 / 分子設計 / 材料設計 / プロセス設計 |
|---|
| Outline of Final Research Achievements |
Conventional inverse analysis in the design of molecules, materials and processes involves constructing mathematical model Y=f(X) between properties/activities Y and features X, then generating a large number of virtual samples of X, inputting them into mathematical model to predict the values of Y and selecting virtual samples with good prediction values. It was only a pseudo-inverse analysis, in which forward analysis was repeated exhaustively. This was nothing more than predicting Y in the search range of X assumed in advance by humans, and was not at all compatible with the search for new functions that would only emerge under conditions that were not initially assumed. In this study, a method for directly predicting the values of X from the values of Y, i.e. a method for truly inverse analysis of mathematical model by converting Y=f(X) to X=g(Y) was proposed, and the proposed method was applied to various molecules, materials and processes.
|
| Free Research Field |
ケモインフォマティクス・マテリアルズインフォマティクス・プロセスインフォマティクス
|
| Academic Significance and Societal Importance of the Research Achievements |
本研究の成果により、科学者・開発者の創造力のつまった実験結果の中にある暗黙知を形式知化でき、実験結果および実験データから構築された数理モデルを化学的・工学的に理解できる形にすることが可能になる。提案手法により、どうしてその実験結果になったのか、どうしてその化学構造・実験条件・プロセス条件で物性値・活性値が得られたのか、望ましい物性値・活性値を得るためにはどのような化学構造・実験条件・プロセス条件にすればよいのか、といったことが明らかになり、科学的なメカニズムの解明に貢献する。本研究の成果により実験と統計とが融合することにより新たな科学的知識発見につながる。
|
