Explainable Models


A broad acceptance of ML methods in chemistry requires an alignment of the models with the principles of chemistry and the generation of new physical insights. Many of the ML models developed focus on explainability such as the threshold analysis of the use of counterfactuals to understand the origin of model performance.


  • Ziyi Kou , Shichao Pei , Yijun Tian and Xiangliang Zhang.  A Controllable Prompt Adversarial Attacking Framework for Black-Box Text Guided Image Generation Models.  2023 Intl. Joint Conf. Art Intel https://www.ijcai.org/proceedings/2023/0109.pdf

  • Andrzej M. Żurański, Shivaani S. Gandhi, and Abigail G. Doyle. A machine learning approach to model interaction effects: development and application to alcohol deoxyfluorination. J. Am. Chem. Soc. 2023, 145,14,7898-7909. https://doi.org/10.1021/jacs.2c13093

  • Luchini, Guilian, and Robert Paton. "Bottom-up Atomistic Descriptions of Top-Down Macroscopic Measurements: Computational Benchmarks for Hammett Electronic Parameters." ACS Phys. Chem Au, 2024

  • Silva, J. D. J.;  Bartalucci, N.;  Jelier, B.;  Grosslight, S.;  Gensch, T.;  Schünemann, C.;  Müller, B.;  Kamer, P. C.;  Copéret, C.; Sigman, M. S., Development and Molecular Understanding of a Pd-catalyzed Cyanation of Aryl Boronic Acids Enabled by High-Throughput Experimentation and Data Analysis. Helv. Chim. Acta 2021 e2100200. https://doi.org/10.1002/hlca.202100200

  • Newman-Stonebraker, Samuel; Smith, Sleight; Borowski, Julia; Peters, Ellyn; Gensch, Tobias; Johnson, Heather; Sigman, Matthew; Doyle, Abigail. Linking Mechanistic Analysis of Catalytic Reactivity Cliffs to Ligand Classification. ChemRxiv, May12, 2021. https://doi.org/10.26434/chemrxiv.14388557.v1

  • Yang, Y.; Zhang, S.; Ranasinghe, K.; Isayev, O.; Roitberg, A. “Machine Learning of Reactive Potentials.” ChemRxiv. 10.26434/chemrxiv-2023-x82fz.

  • Guo, Taicheng, Kehan Guo, Zhengwen Liang, Zhichun Guo, Nitesh V. Chawla, Olaf Wiest, and Xiangliang Zhang. "What can GPT models do in chemistry? A comprehensive benchmark on eight tasks." NeurIPS 2023, accepted. https://doi.org/10.48550/arXiv.2305.18365