Supporting data to "The Stubborn Persistence of Toxic Solvents in Chemical Synthesis"

USPTO solvent database Solvent usage in 1,338,762 USPTO patent reactions (1976–2016), with solvents resolved to the experimental role they were used for. Supporting data for "The Stubborn Persistence of Toxic Solvents in Chemical Synthesis" (Dobbelaere & Sneddon, https://doi.org/10.1002/anie.1677569) and for the companion site https://solventexplorer.com. Files File Contents uspto_solvent_db_full.parquet The database. Apache Parquet, snappy-compressed. 1,338,762 rows × 23 columns. uspto_solvent_db_full.gsk_rag.csv GSK solvent-guide colour lookup, 229 rows. Columns Reaction Column Description SANITIZED_REACTION Reaction SMILES, sanitised MAPPED_REACTION Atom-mapped reaction SMILES (all rows; 98.8% carry atom maps) PRODUCTS Product SMILES REAGENT Reagent SMILES, .-joined CATALYST Catalyst SMILES, .-joined procedure Free-text experimental procedure — the text the solvent roles are extracted from NAME Reaction name (Rxn-INSIGHT) CLASS Reaction class (Rxn-INSIGHT) SUBCLASS Reaction subclass (Rxn-INSIGHT); 124 classes are used on the website Solvents, by role Assigned by Alkahest. .-joined SMILES; empty where the role does not occur. solvent-free and not-reported are sentinels, not solvents. Column Role SOLV_RXN Reaction SOLV_WORKUP Workup SOLV_PURIF Purification SOLV_ANAL Analytical SOLV_NMR NMR (curated — see below) There is deliberately no un-roled "solvent" column: role resolution is the point of the dataset. For reference, the raw solvent field of the patent record agrees with SOLV_RXN on only 38% of rows — SOLV_RXN finds a reaction solvent for 93,743 reactions the record called solvent-free, assigns 25,862 to a non-reaction role, and captures mixtures 2.5× more often. Provenance and metadata Column Description reaction_id Identifier inherited from the Open Reaction Database PATENT Patent number REF Source reference YEAR Patent year assignee Patent assignee owner_country Assignee country (5,589 nulls) sector Industry sector of the assignee broad_topic Coarse patent topic inferred_scale Inferred reaction scale GSK solvent colours uspto_solvent_db_full.gsk_rag.csv maps solvent SMILES to the 2016 GSK Solvent Selection Guide's Red/Amber/Green score. It is a separate lookup rather than a column because the score is a property of a solvent, not of a reaction. Column Meaning smiles Solvent SMILES, exactly as the SOLV_* columns write it solvent Name from the GSK guide rag G / A / R deuterated true if scored via its non-deuterated parent parent_smiles That parent, when deuterated is true Join on smiles directly — the table is keyed on the SMILES strings this database uses, so no canonicalisation is needed. For a mixture, split on . and take the worst component (G < A < R); that is the rule used throughout the manuscript and the website. Covers 98.1% of solvent-component occurrences across all five roles. The remainder are absent from the guide, mostly reagents acting as solvents — led by trifluoroacetic acid (63,006 occurrences) and DIPEA (43,801). If you use these scores, cite the guide: Alder, Hayler, Henderson, Redman, Shukla, Shuster & Sneddon, Green Chem. 2016, 18, 3879–3890, 10.1039/C6GC00611F. The NMR solvent column SOLV_NMR is curated rather than raw. Two corrections were applied to the values read out of the procedure text: Coverage. The raw extraction is empty for 6,443 reactions that do report an NMR solvent; these were recovered from the parsed SMILES. Where the raw value is present it agrees with the curated one on all 265,690 rows. Consolidation. The raw extraction separates a solvent from the same solvent written with its protio SMILES or with an internal standard. These are merged — 8 keys, 12,021 rows: Merged Rows Into [2H]C(Cl)(Cl)Cl.[Si](C)(C)(C)C (CDCl₃ + TMS) 863 CDCl₃ ClC(Cl)Cl (protio chloroform) 153 CDCl₃ CS(=O)C (protio DMSO) 10,111 DMSO-d₆ …S(=O)…C[Si](C)(C)O[Si](C)(C)C (DMSO-d₆ + HMDSO) 214 DMSO-d₆ …S(=O)…[Si](C)(C)(C)C (DMSO-d₆ + TMS) 140 DMSO-d₆ CO (protio methanol) 152 MeOD-d₄ [2H]C([2H])([2H])O[2H].FC(F)(F)C(=O)O (MeOD + TFA) 152 MeOD-d₄ CC(=O)C (protio acetone) 236 Acetone-d₆ Genuine mixed-solvent records (CDCl₃/MeOD, DMSO-d₆/D₂O, …) are not merged and keep their .-joined value. 272,133 reactions have an NMR solvent, led by CDCl₃ (138,712), DMSO-d₆ (98,735), MeOD-d₄ (23,546), acetone-d₆ (3,149) and D₂O (1,076). Example: product properties by NMR solvent import pandas as pd from rdkit import Chem from rdkit.Chem import Descriptors, Lipinski df = pd.read_parquet("uspto_solvent_db_full.parquet", columns=["PRODUCTS", "SOLV_NMR"]) df = df[df["SOLV_NMR"].fillna("") != ""] # 272,133 rows mol = df["PRODUCTS"].map(Chem.MolFromSmiles) df["mw"] = mol.map(Descriptors.MolWt) df["clogp"] = mol.map(Descriptors.MolLogP) df["tpsa"] = mol.map(Descriptors.TPSA) df["fcsp3"] = mol.map(Lipinski.FractionCSP3) df["arom"] = mol.map(Lipinski.NumAromaticRings) df["hbond_acc"] = mol.map(Lipinski.NumHAcceptors) Not included Reaction fingerprints, retrosynthetic templates, ring and functional-group annotations, and parsed workup steps are omitted; all are regenerable from SANITIZED_REACTION with Rxn-INSIGHT. MAPPED_REACTION is included precisely because it is not — atom mapping is expensive to redo and depends on the mapper version. License Creative Commons Attribution 4.0 International (CC BY 4.0). Re-distribution and re-use are permitted provided the creators are credited. Derived from the USPTO patent reaction corpus (Lowe) via the Open Reaction Database. The GSK Red/Amber/Green classifications in uspto_solvent_db_full.gsk_rag.csv are reproduced from Alder et al., Green Chem. 2016, 18, 3879–3890 (10.1039/C6GC00611F); cite that work if you use them. Citation Dobbelaere, M. R. & Sneddon, H. F. The Stubborn Persistence of Toxic Solvents in Chemical Synthesis. (10.1002/anie.1677569)