Principal Investigator : Sosa Carrizo E. Daiann
Lab: Laboratoire de Physique et Chimie des Nano-objets (LPCNO, INSA-CNRS) /
Team: Physical and Chemical Modelization
Abstract :
SMART-AI: Quantum Chemistry and Machine Learning for Catalyst Design
Catalysis drives over 90% of industrial chemical processes, yet improving catalysts remains slow, expensive, and guided mainly by trial-and-error. In catalysis AI, the central bottleneck is not the availability of algorithms, but the lack of mechanistically informative data and chemically meaningful representations: models trained on sparse datasets can fit known trends, but often fail to explain why performance changes — or to generalise to new catalyst families.
SMART-AI integrates DFT-based mechanistic modelling and machine learning into a single physically interpretable framework. Developed at LPCNO (CNRS UMR 5215 — INSA / Université de Toulouse), the project builds new mechanism-resolving descriptors — such as strain and interaction energy components — that encode the physical origins of reactivity and selectivity, rather than relying on black-box fingerprints. This mechanistically grounded approach improves interpretability by construction and enhances robustness in the small-data regime typical of homogeneous catalysis.
SMART-AI will be coupled with a traceable data infrastructure that links experimental outcomes, characterisation, and computational provenance through standardised workflows. With each new result, the dataset grows cleaner and more successful — progressively refining the models and sharpening the hypotheses that guide subsequent experiments. The broader goal is to turn quantum-chemical insight into practical, transferable design rules — supporting the rational design of new catalysts.


Legend of the picture: Conceptual overview of the SMART-AI framework integrating density functional theory (DFT) and machine learning for rational catalyst design. Quantum-derived, physically meaningful descriptors enable interpretable and data-efficient ML-driven catalyst optimization
