Formulation AI From Zero Data: AAPS PharmSci 360 2026

PharmTech spoke with Pauric Bannigan, chief scientific officer and co-founder, Intrepid Labs, at AAPS PharmSci 360 2026, about how machine learning, AI, and mechanistic modeling are reshaping formulation development and decision-making. Bannigan moderated a session, “Symposium: Next-Gen Pharmaceutical Formulation: From Machine Learning to Mechanistic Modeling” which covered AI-centric development workflows, dissolution prediction for oral solid dosage forms, machine-learning approaches for high-concentration antibody formulations, and mechanistic modeling for long-acting injectables. In this interview, Bannigan discusses where these tools pay off today and what still holds them back.

Intrepid Labs combines pharmaceutical formulation development with robotics and machine learning in what Bannigan describes as self-driving platforms for drug development. In Bannigan’s view, the clearest return right now is faster decision-making. Bannigan points to a symposium talk from Quotient Technologies, which trains models on internally generated data to predict viscosity in high-concentration protein solutions before experiments begin. “So I think this is where the value is, in being able to predict a priori ahead of time which formulation strategies are more likely to be successful,” Bannigan says.

The limiting factor is data. Predicting how a formulation will behave from a drug’s structure alone is a data-hungry problem that can require hundreds or thousands of data points. “It is being able to get enough high-quality data to train these models to be reliable decision tools,” Bannigan explains.

Intrepid addresses the problem through active learning. The team defines an allowable experimental space, and the model, starting with zero data, selects experiments within it, learns from the results, and chooses the next set. “So in this way, we actually only do the most promising experiments,” Bannigan says, adding that the approach trains informative models with far less data. Bannigan has previously told PharmTech that steering exploration toward the most informative regions of the design space is central to predictive modeling for small-molecule formulation development.