Automating Development of Small-Molecule APIs

Small-molecule APIs are becoming more and more complex, and the resulting longer synthetic pathways pose challenges for process chemists to develop efficient API manufacturing processes. Overcoming these challenges is crucial to avoid delays in investigational new drug applications and expenses in transitioning new drug candidates from early development to commercial production, according to Simon Wagschal, associate director of advanced chemistry technologies, Lonza Small Molecules. “As the industry faces new drug candidates requiring over 20 synthetic steps on average, AI [artificial intelligence]-enabled predictive route design technologies have become essential for creating efficient retrosynthetic routes,” he says.

Indeed, AI-enabled API development tools are already being used by several R&D companies for the identification of optimized synthetic routes, according to Raghavendar Rao Morthala, chief scientific officer, Navin Molecular. “An AI platform has the potential to design and synthesize APIs in a matter of weeks or months, compared to the years it can take using conventional methods,” he says.

Currently, however, the application of AI to API synthetic route design is in its early stages, contends David Ford, senior director of chemistry at Snapdragon Chemistry, a Cambrex company. Synthetic route design by AI is used primarily at the brainstorming stage of route development, he says, with identified options narrowed down by humans. “Given the importance of the choice of a route or routes and the quantity of resources used to pursue them, the role of the synthetic chemist remains central in route design,” he states.

Wagschal adds that AI tools have traditionally been used for early-phase discovery, often generating routes not suitable for commercial-scale production.

Other exciting applications for AI in process development, Ford says, lie in its ability to manage knowledge. “Recent advances in large language models have made it possible to index and explore diverse sets of documents like weekly updates, PowerPoint presentations, technical reports, batch records, and campaign summaries,” he explains. These AI models allow for the content to be searched semantically, based on the meaning of the search, instead of by keywords.

“Given the volume of information trapped in these documents, it is very likely that work is forgotten, and hard lessons then need to be re-learned,” Ford observes. “We see great opportunity in helping researchers make connections between their work and the work that has been done in the past,” he notes. Generative AI models may also prove useful for preparing first drafts of templated documents like standard operating procedures, qualification protocols, and batch records, according to Ford.

Lonza, meanwhile, is one company not deterred by the limitations of early AI programs developed for scouting of synthetic routes for small-molecule APIs. “The introduction of computer-aided synthesis planning (CASP) technologies with route prediction capabilities has significantly influenced the route selection process,” says Wagschal.

Nearly all CASP tools employ a form of machine learning called Monte Carlo Tree Search to rapidly prioritize computationally predicted synthetic routes, Wagschal explains. This heuristic search method evaluates the likelihood of reaction success at each step of the proposed pathway, from starting materials through intermediates to the final API. The prioritized routes are then reviewed and refined by process chemists. “This approach, guided by our experts, has increased the number of potential strategies and led to more efficient API synthesis routes,” he states.

To overcome the limitations of early AI-based models for route development with respect to commercial practicality, Lonza has combined leading AI-powered predictive route design models with proprietary supply-chain data, including raw material costs, availability, and detailed supply chain intelligence, according to Wagschal. “This integrated technology allows process chemists to quickly identify not just the shortest pathways, but also the most commercially viable ones by considering real-world supply chain constraints. These advanced AI solutions provide a holistic view that streamlines route scouting and accelerates the transition from research and development to robust manufacturing processes, resulting in significant time and cost savings in developing complex APIs,” Wagschal concludes.