Physical and Theoretical Chemistry

Artificial Intelligence and Data-Driven Chemistry

Additional sessions will consider graph neural networks and equivariant architectures for property prediction, transferable interatomic potentials, and generative models constrained by synthetic accessibility. Presentations may cover reaction outcome prediction, retrosynthetic planning, automated mechanism inference, and the curation of large reaction databases with consistent reaction conditions. Discussion will extend to active learning and Bayesian optimization for experimental campaign design, uncertainty estimation in deployed models, and the risk of extrapolation outside training domains. Reproducibility, open data standards, FAIR sharing, and the evaluation of models against blind benchmarks will also be addressed. Responsible use of artificial intelligence in chemical research will be emphasized, including workflows that combine machine learning with quantum-chemical reference data, high-throughput experimentation, and autonomous laboratory platforms for catalyst screening, materials discovery, and molecular design.