Transfer and active learning have the potential to accelerate the development of new chemical reactions, using prior data and new experiments to inform models that adapt to the target area...

The global disruption caused by the 2020 coronavirus pandemic stressed the supply chain of many products, including pharmaceuticals. Multiple drug repurposing studies for COVID-19 are now underway. If a winning therapeutic emerges, it is unlikely that the existing inventory of the medicine, or even the chemical raw materials needed to synthesize it, will be available in the quantities required. Here, we utilize retrosynthetic software to arrive at alternate chemical supply chains for the antiviral drug umifenovir, as well as eleven other antiviral and anti-inflammatory drugs. We have experimentally validated four routes to umifenovir and one route to bromhexine. In one route to umifenovir the software invokes conversion of six C–H bonds into C–C bonds or functional groups. The strategy we apply of excluding known starting materials from search results can be used to identify distinct starting materials, for instance to relieve stress on existing supply chains.

Amines and carboxylic acids are abundant building blocks for synthesis that classically are united to form an amide bond. To access new pockets of chemical space we are interested in the development of complementary amine–acid coupling reactions. In particular, the formation of carbon–carbon bonds by formal deamination and decarboxylation would be an impactful addition to the synthesis toolbox. Here we report a formal cross coupling of alkyl amines and aryl carboxylic acids to form C(sp3)–C(sp2) bonds following pre-activation of the amine–acid building blocks as a pyridinium salt and N-acyl-glutarimide respectively. Under nickel-catalyzed reductive cross-coupling conditions, a diversity of simple and complex substrates are united in good to excellent yield. High-throughput experimentation was essential to the development of the reaction, and to the discovery of performance-enhancing additives such as cyclic imides, RuCl3 and GaCl3. Preliminary mechanistic investigations suggest that RuCl3 supports the decarbonylation event, increasing reaction selectivity. Numerous amine or acid containing pharmaceuticals are successfully diversified under the optimized conditions.

Amines and carboxylic acids are abundant synthetic building blocks that are classically united to form an amide bond. To access new pockets of chemical space, we are interested in the development of amine–acid coupling reactions that complement the amide coupling. In particular, the formation of carbon–carbon bonds by formal deamination and decarboxylation would be an impactful addition to the synthesis toolbox. Here, we report a formal cross-coupling of alkyl amines and aryl carboxylic acids to form C­(sp3)–C­(sp2) bonds following preactivation of the amine–acid building blocks as a pyridinium salt and N-acyl-glutarimide, respectively. Under nickel-catalyzed reductive cross-coupling conditions, a diversity of simple and complex substrates are united in good to excellent yield, and numerous pharmaceuticals are successfully diversified. High-throughput experimentation was leveraged in the development of the reaction and the discovery of performance-enhancing additives such as phthalimide, RuCl3, and GaCl3. Mechanistic investigations suggest phthalimide may play a role in stabilizing productive Ni complexes rather than being involved in oxidative addition of the N-acyl-imide and that RuCl3 supports the decarbonylation event, thereby improving reaction selectivity.

Machine learning models are increasingly being utilized to predict outcomes of organic chemical reactions. A large amount of reaction data is used to train these models, which is in stark contrast to how expert chemists discover and develop new reactions by leveraging information from a small number of relevant transformations. Transfer learning and active learning are two strategies that can operate in low-data situations, which may help fill this gap and promote the use of machine learning for tackling real-world challenges in organic synthesis. This Perspective introduces active and transfer learning and connects these to potential opportunities and directions for further research, especially in the area of prospective development of chemical transformations.