Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

University of Toronto

$52,000.00

Compositional symmetries for improved quantum machine learning across chemical compound space

EF 2023-24

Résumé
Chemical compound space (CCS), the set of conceivably stable molecules and materials, scales combinatorially with number of atoms and element types. Our research efforts are aimed at improving the understanding of CCS to such an extent that one can explore it more efficiently, e.g. in order to assist the rational design and discovery of novel (molecular) materials which exhibit desirable properties. We use high performance computing to generate numerical solutions throughout CCS with state-of-the-art legacy quantum chemistry and statistical mechanics software packages developed. Calculation output is subsequently used to develop and benchmark machine learning (ML) models which are applicable throughout CCS and which can dramatically accelerate the predictions without sacrificing their accuracy. While the entire field has made impressive progress within a rapidly growing community over the last decade, unfortunately, data-sets required are still too large to train truly universal (across CCS) and generally applicable ML models of all the relevant properties. For example, our most recently published machine learning models which reach satisfying accuracy when it comes to the prediction of activation energies [https://doi.org/10.1063/5.0059742] or to performing transition state search [https://doi.org/10.1063/5.0112856] are severely restricted in terms of chemical reaction space (requiring more data for smaller compositional spaces of only one reaction mechanism). Consequently, further methodological advances are in dire need if we ever wish to use machine learning for the rational design of experimental reaction conditions such as they occur, for example, in a wet chemistry lab. We propose to use the NSERC Discovery grant to tackle this problem through further developments of physics based ML models which exploit recently discovered approximate compositional symmetry relations (dubbed 'alchemical chirality') within CCS. Alchemical chirality relates seemingly unrelated compounds through energy degeneracies exact up to third order [https://doi.org/10.1126/sciadv.abf1173]. Preliminary numerical evidence from my lab have shown dramatic improvements in data- efficiency for ML when accounting for this symmetry within the representation. We request funding for two PhD students to work on two research objectives or work packages (WP). WP1 will deal with the formalization of the implications of the symmetry relation on the ML architecture, e.g. in terms of loss function or representation. Within WP2, we will study the usefulness of this symmetry to improve the data efficiency in terms of Delta, multi-level, or transfer learning. We will achieve Impact by selecting for this research (a) economically relevant materials classes such as catalysts, and (b) societally relevant bio-molecular classes directly connected to underrepresented minorities (to further equity, diversity, inclusion).
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2023-04853
Name nom
von Lilienfeld-Toal, Otto Anatole
Department département
Chemistry
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
52000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1504
Committee name en
Chemistry
Committee name fr
Chimie
Area of application code
801
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Computer software
Area of application fr
Logiciel
Research subject code
2804
Research subject group en
Artificial intelligence
Research subject group fr
Expert systems
Research subject en
Intelligence artificielle
Research subject fr
Systèmes experts
Application title
Compositional symmetries for improved quantum machine learning across chemical compound space
Keyword
Chemical space; Electronic structure; Machine learning; Quantum chemistry; Symmetry
Application summary
Chemical compound space (CCS), the set of conceivably stable molecules and materials, scales combinatorially with number of atoms and element types. Our research efforts are aimed at improving the understanding of CCS to such an extent that one can explore it more efficiently, e.g. in order to assist the rational design and discovery of novel (molecular) materials which exhibit desirable properties. We use high performance computing to generate numerical solutions throughout CCS with state-of-the-art legacy quantum chemistry and statistical mechanics software packages developed. Calculation output is subsequently used to develop and benchmark machine learning (ML) models which are applicable throughout CCS and which can dramatically accelerate the predictions without sacrificing their accuracy. While the entire field has made impressive progress within a rapidly growing community over the last decade, unfortunately, data-sets required are still too large to train truly universal (across CCS) and generally applicable ML models of all the relevant properties. For example, our most recently published machine learning models which reach satisfying accuracy when it comes to the prediction of activation energies [https://doi.org/10.1063/5.0059742] or to performing transition state search [https://doi.org/10.1063/5.0112856] are severely restricted in terms of chemical reaction space (requiring more data for smaller compositional spaces of only one reaction mechanism). Consequently, further methodological advances are in dire need if we ever wish to use machine learning for the rational design of experimental reaction conditions such as they occur, for example, in a wet chemistry lab. We propose to use the NSERC Discovery grant to tackle this problem through further developments of physics based ML models which exploit recently discovered approximate compositional symmetry relations (dubbed 'alchemical chirality') within CCS. Alchemical chirality relates seemingly unrelated compounds through energy degeneracies exact up to third order [https://doi.org/10.1126/sciadv.abf1173]. Preliminary numerical evidence from my lab have shown dramatic improvements in data- efficiency for ML when accounting for this symmetry within the representation. We request funding for two PhD students to work on two research objectives or work packages (WP). WP1 will deal with the formalization of the implications of the symmetry relation on the ML architecture, e.g. in terms of loss function or representation. Within WP2, we will study the usefulness of this symmetry to improve the data efficiency in terms of Delta, multi-level, or transfer learning. We will achieve Impact by selecting for this research (a) economically relevant materials classes such as catalysts, and (b) societally relevant bio-molecular classes directly connected to underrepresented minorities (to further equity, diversity, inclusion).