Four decades of patented reactions: hazardous solvents rose from about 60% to over 70%, and restrictions mostly pushed chemists to another hazardous liquid

After nearly three decades of 'green' chemistry campaigning, hazardous solvent use in the patent literature rose rather than fell: hazardous solvents accounted for about 60% of reactions in patents in 1976 and above 70% by 2016. Long-horizon tracking of solvent use across decades and over a million reactions was previously lacking; this survey covers roughly 1.3 million reactions in patents from 1976 to 2016 and is described as the largest and most comprehensive of its kind. Built on a pre-existing database drawn from the US Patent and Trademark Office in which chemical names and structures were converted into a computer-readable format; the story calls it 'more or less the only large, open-source repository of chemical reactions that is available for this kind of survey'.

When government regulations restrict the use of a toxic solvent, chemists are more likely to switch to another hazardous liquid than to choose a greener or safer alternative. This moves the question from whether greener alternatives exist to the direction substitution actually takes: regulation triggers swaps among hazardous solvents rather than a migration to safer ones. Derived from the same patent database across more than 500 reaction types; the story does not report a specific rate for this substitution behaviour or a per-solvent breakdown.

Use of trifluoroacetic acid (TFA) rose substantially, and by some definitions TFA is a 'forever chemical', one of the persistent per- and polyfluoroalkyl substances (PFAS) increasingly found in waterways, soils and living creatures. Extends the solvent-risk discussion beyond classic halogenated solvents to a fluorinated acid tied to peptide-drug synthesis, linking pharmaceutical synthesis demand to PFAS persistence. The story states TFA 'is key to the synthesis of popular peptide drugs' and that PFAS are 'increasingly found in waterways, soils and living creatures'; no specific TFA volumes or environmental concentrations are given.

The survey's analysis engine, Rxn-INSIGHT, is an open-source bond-electron-matrix algorithm that classifies and names reactions, extracts functional groups and rings, and as a side application suggests reaction conditions. The paper reports over 90% classification accuracy and over 95% naming accuracy on a 50,000-reaction labelled benchmark, and 90% classification and over 50% naming across 1.8 million USPTO reactions; unlike supervised template-based models, it classifies without needing a training set. The paper gives benchmark size (50,000 reactions, 1000 each of 50 named reaction types), database size (1.8 million reactions, 1976-2016) and 40-100 ms per reaction for classification and naming on an Intel i7 laptop; the authors also note the method relies entirely on atom-atom mapping, so mapping failures cause misclassification.

This work speaks to readers concerned with green chemistry, early-stage drug-discovery processes and chemical regulation: it turns the question of how effective nearly three decades of green chemistry has been into checkable patent-level statistics, letting researchers, pharmaceutical process teams and policymakers examine their own choices along lines of reaction type and solvent class. The story quotes Leahy's hope that 'a lot of medicinal-chemistry groups, and academic groups, look at this and think about their own choices', indicating the intended use is to prompt teams to review their own solvent decisions rather than to prescribe a single substitute. The companion Rxn-INSIGHT tool targets cheminformatics and synthetic chemistry users who need to screen large reaction databases and find starting points for solvent, catalyst and reagent selection for new reactions, and it runs on an everyday laptop.

The database ends in 2016, so changes after that year fall outside this count; the analysis also does not measure the quantities of solvents involved, and the story notes that a trend towards miniaturizing laboratory reaction set-ups could be helping to drive down overall solvent usage. Patents mostly record early-stage research, whereas scaled-up manufacturing processes often weed out hazardous solvents such as DCM, so patent statistics and industrial volumes may diverge. The story also notes the analysis 'may not have captured some areas of progress'. On the tool itself, the paper states the bond-electron-matrix approach is fully reliant on atom-atom mapping, so inaccurate mapping leads to wrong classification and may slightly overestimate aromatic heterocycle formations; the paper further notes that classification conventions differ between tools, so cross-tool comparison needs care. This evidence bundle contains the story text and the paper page text but not the paper's figures and supplementary material, so details tied to specific figure values still need checking against the original.