Phlow Corp., a leading American pharmaceutical contract development and manufacturing organization (CDMO), today announced a major advancement in AI-driven Active Pharmaceutical Ingredient (API) manufacturing process development in collaboration with Enveda, a biotechnology company that learns from life’s chemistry to create better medicines faster. Through a joint pilot program launched last year to rapidly select API routes, the companies demonstrated that high-quality, internally generated reaction data can significantly improve AI-based predictions of chemical reaction kinetics, yields, and purity profiles, compressing development timelines from years to months.
In just three months, Phlow and Enveda generated and analyzed nearly 20,000 unique reactions, creating one of the largest high-quality data sets of its kind. The resulting uniform dataset is significantly larger than any publicly available dataset to date and is comparable in scale to proprietary reaction datasets assembled over decades by major pharmaceutical companies. By training AI models on this internally generated data, the teams achieved substantially more accurate yield predictions than models trained on public datasets alone.
“This is about setting a new standard for pharmaceutical development and manufacturing,” said Eric S. Edwards, M.D., Ph.D., Co-Founder and CEO of Phlow. “By combining Phlow’s advanced API development and manufacturing expertise with Enveda’s leading AI-driven insights and deep learning capabilities, we are creating a first-of-its-kind, innovative platform that predicts the optimal synthesis route the first time, every time, faster and more accurately than any other database. Together, we’re creating the future of how medicines are made with speed, precision, and resilience.”
The collaboration compared traditional machine learning approaches with deep learning models, specifically graph neural networks that represent molecules as connected networks of atoms and bonds. This approach more closely reflects real-world chemistry and delivers stronger predictive performance, demonstrating the value of deep learning for reaction optimization and future retrosynthetic planning.
“This collaboration demonstrates what becomes possible when high-quality experimental data and modern AI are tightly integrated,” said Viswa Colluru, Ph.D., Chief Executive Officer of Enveda. “Accurate reaction prediction depends not just on algorithms, but on the quality, structure, and scale of the data behind them. By learning from carefully generated chemical data, we can move faster, reduce experimental burden, and create more reliable paths from discovery to development.”
The ability to rapidly generate, analyze, and integrate reaction data into AI models represents a meaningful step forward for chemical and pharmaceutical development. By reducing reliance on trial-and-error experimentation, this approach supports faster optimization of synthetic routes, improved scalability, and more efficient development of essential medicines.
“This capability reflects Phlow’s ability to rapidly operationalize AI at scale, reinforcing a core belief that data quality and execution are decisive advantages,” said Juan Piacquadio, Chief Information Officer at Phlow. “When experimental systems, data pipelines, and advanced AI models are intentionally designed to work together, we can achieve outcomes in months that historically required many years. This is how modern, technology-enabled infrastructure accelerates pharmaceutical innovation while strengthening domestic manufacturing resilience.”
Building on this success, Phlow and Enveda plan to deepen the collaboration and expand the AI-driven framework to additional reaction classes, applying deep learning to a retrosynthetic analysis platform, enabling faster, more efficient, and more sustainable chemical synthesis across a broader range of applications. This work lays the foundation for a best-in-class predictive system that delivers the optimal synthesis route the first time, every time, additional intellectual property, and additional long-term value for both organizations.
Read more here.