Challenges to Combine Quantum Chemistry, Machine Learning, and Data Mining Techniques to the Study and Design of Molecular Systems

The amount of Quantum Chemistry (QC) data has been increasing year by year due to the increasing computational power, algorithm developments, and successful software implementation of a wide range of theoretical frameworks. More recently, the release of online databases has facilitated access to large QC datasets, which have contributed to the increase of Machine Learning (ML) and Data Mining (DM) techniques in QC as low-cost algorithms to evaluate QC properties and for extraction of insights from molecular systems, respectively. In 2018, by the set up of the Center for Innovation on New Energies, a solid partnership was established between the Quantum Theory of Nanomaterials (QTNano) group leaded by Juarez L. F. Da Silva (expertise in QM) and the groups leaded by Marcos G. Quiles (UNIFESP, expertise in ML and DM) and Ronaldo C. Prati (UFABC, expertise in ML and DM) to explore the combination of QC, ML, and DM techniques in the study and design of molecular systems.  In this talk, we will report our challenges from day one and our most important scientific achievements, which includes: (i) ML prediction of nine non-related molecular properties of the QM9-QC dataset using the SMILES representation with similar mean average errors as high computational molecular representations, however, with lower computational cost [1]. (ii) exploration of correlation-based frameworks for extraction of insights from QC databases, which includes the applications and performance analysis of Pearson, Spearman, and Kendall correlation coeficients to investigate the interrelation between structure and properties in nanoclusters composed by Pt-based nanoalloys and mixed CeO₂-ZrO₂ oxides [2]. Finally, (iii) we will discuss the applications of clustering algorithms such as k-means as supporting technique in several computational material science problems addressed by our groups [3]. At the end, we will summarize the most achievements obtained so far and important quick wins.

Acknowledgments: FAPESP.

[1] G. A. Pinheiro, J. Mucelini, M. D. Soares, R. C. Prati, J. L. F. Da Silva, M. G. Quiles, J. Phys. Chem. A 124, 9854 (2020).

[2] J. Mucelini, M. G. Quiles, R. C. Prati, J. L. F. Da Silva, J. Chem. Info. Model. 61, 1125 (2021); P. C. Mendes, S. G. Justo, J. Mucelini, M. D. Soares, K. E. A. Batista, M. G. Quiles, M. J. Piotrowski, and J. L. F. Da Silva, J. Phys. Chem. C 124, 1158 (2020); P. Felício-Sousa, J. Mucelini, L. Zibordi-Besse, K. F. Andriani, Y. Seminovski, R. C. Prati, and J. L. F. Da Silva, Phys. Chem. Chem. Phys. 21, 26637 (2019).

[3] K. E. A. Batista, V. K. Ocampo-Restrepo, M. D. Soares, M. G. Quiles, M. J. Piotrowski, and J. L. F. Da Silva, J. Chem. Info. Model 60, 537 (2020); K. E. A. Batista, M. D. Soares, M. G. Quiles, M. J. Piotrowski, and J. L. F. Da Silva, J. Chem. Info. Model. 61, 2294 (2021).