Carbonyl-Olefin Metathesis Reactivity Dataset for Machine Learning

147 blinded carbonyl–olefin metathesis reactions curated by Zi-Hao Li provide a benchmark for catalyst and substrate prediction. The dataset underpins a machine learning framework achieving high accuracy (R² ≈ 0.92) and includes quantum-derived descriptors for mechanistic interpretation. It was last updated on figshare in May 2026.

Use Cases

• Predicting effective catalysts for carbonyl–olefin metathesis based on substrate structure.
• Identifying key reactivity drivers like catalyst HOMO energy and substrate carbonyl polarization via feature importance analysis.
• Planning targeted experiments using reactivity-cliff analysis to find regions where predictions are unreliable.
• Comparing machine learning and large language model (GPT-4o) performance for experiment prioritization in organic synthesis.

Strengths

• Contains 147 rigorously curated and blinded reaction data points.
• Achieves high predictive accuracy with quantum-derived descriptor models (R² ≈ 0.92).
• Includes external validation, retaining performance on unseen substrates.
• Demonstrates data efficiency, recovering most performance using only ∼40% of the reaction matrix via Bayesian optimization.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific, controlled reaction matrix studied.

Provenance

Source

figshare

Collection Method

Rigorously curated experimental data from a controlled reaction matrix.

Freshness

Last updated 2026-05-15 18:04:21; freshness should be verified.