UntitData and Supplementary Materials for: Interpretable Latent-Reactivity Prioritisation

Supplementary materials and benchmark files support a study on identifying radical–polar crossover motifs from patent reactions. The repository includes CSV files for benchmark datasets, motif enrichment analyses, cluster statistics, and computational scan summaries. Author Shaikh Afnan Birahim published the data on figshare in May 2026 under a CC-BY-4.0 license.

Use Cases

• Benchmarking new reaction classification models based on the provided RPC/non-RPC benchmark dataset.
• Analyzing motif enrichment patterns in RPC-rich versus RPC-poor reaction clusters.
• Triaging candidate reaction motifs for further study using the provided ANI-1ccx bond-elongation scan summaries.
• Reproducing the latent-space analysis workflow using the included cluster statistics and reproducibility script.

Strengths

• Includes a curated benchmark dataset with heuristic RPC scores for validation.
• Provides multiple analysis outputs (enrichment, cluster stats, scan summaries) in structured CSV format.
• Released under a permissive CC-BY-4.0 license for reuse and modification.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The benchmark is described as an implementation and face-validity check, not a definitive estimate of out-of-domain predictive performance.

Provenance

Source

figshare, associated with a research manuscript.

Collection Method

Derived from patent-derived reaction corpora using transformer embeddings, UMAP/HDBSCAN, and heuristic scoring.

Freshness

Last updated 2026-05-19 11:26:55; freshness should be verified.