For chemical reaction prediction, see the Open Reaction Database, a collaboration including the Coley lab at MIT (surprisingly not cited by OP): Paper: https:/

For chemical reaction prediction, see the Open Reaction Database, a collaboration including the Coley lab at MIT (surprisingly not cited by OP): Paper: https://pubs.acs.org/doi/10.1021/jacs.1c09820 https://pubs.acs.org/doi/10.1021/jacs.1c09820 Docs: https://docs.open-reaction-database.org/en/latest/overview.html https://docs.open-reaction-database.org/en/latest/overview.h... It’s an incredible effort to collate and clean this data, and even then a substantial portion of it will not be reproducible due to experimental variability or outright errors. For computational methods development it’s extremely useful, maybe even necessary, to have a substantial amount of money and one’s own lab space to collect new data and experimentally test prospective predictions under tightly controlled conditions. The historical data is certainly useful but is not a panacea.