An examination of the chemical methods outlined in four decades of US patents indicates that chemists have not markedly lessened their reliance on toxic solvents, despite a heightened understanding of safety and sustainability concerns, as well as the presence of more environmentally friendly options. The research cautions that dependence on these solvents within historical datasets might skew AI synthesis planning tools, sustaining dangerous practices instead of fostering sustainability.
Solvents serve as the main substances utilized and discarded in the majority of chemical processes, making up approximately 80% of the overall material in the production of active pharmaceutical ingredients. The reduction and substitution of harmful solvents with innocuous ones is a fundamental concept of green chemistry.
Researchers Helen Sneddon and Maarten Dobbelaere examined 1.3 million reactions from 40 years of US patent information with the assistance of Rxn-Insight, a resource created by Dobbelaere. This tool enabled the categorization of reactions and the examination of solvent usage across various phases of chemical processes.
The research concluded that three out of the four most frequently utilized reaction solvents are hazardous. Despite qualifications such as specific solvent necessities for reactions, overarching patterns reveal a consistent reliance on harmful solvents. While the use of certain chlorinated solvents has diminished, others, such as 1,2-dichloroethane, have increased, indicating a shift rather than a true embrace of greener alternatives.
The results also highlight potential challenges for AI in the field of chemistry. Biases within training datasets may result in AI tools favoring hazardous solvents, akin to biases observed in other AI fields. Although current AI models reflect tendencies towards common conditions, user input has the potential to mitigate such biases, steering models towards more environmentally friendly selections.