High-performance liquid chromatography (HPLC) method development is still largely trial and error, and the experiments can quickly use up a laboratory’s time, solvent and instrument capacity. In this interview, Maximilian Besenhard of University College London (UCL) discusses his group’s recent Journal of Chromatography A paper.¹ which proposes a hybrid framework that combines mechanistic and data-driven modelling to cut the number of experiments needed to reach an optimized method.

The framework runs in two stages. First, a mechanistic model is calibrated while it is being used to optimise the method. Then a data-driven model is trained on the experiments from that first stage. Besenhard explains why combining the two approaches works better than using either one alone, and how the framework decides when to hand over from one to the other. He compares the number of experiments needed with conventional and purely data-driven approaches. He also covers what users need to know before they start, such as how many components are in the sample.

The conversation then turns to practical questions: What tends to make a mechanistic model poorly calibrated and how analysts can spot it, and what a calibrated model reveals about method robustness beyond the optimized conditions themselves. Finally, Besenhard describes the hardware, software and expertise a lab would need to adopt an autonomous “digital twin” workflow for HPLC method development.

Questions

• Your recent Journal of Chromatography A paper, “A Hybrid Mechanistic-Data-Driven Framework for Knowledge-Driven and Experiment-Efficient HPLC Method Development,”¹ tackles the cost of method development. What motivated this research, and why is HPLC method development still so experiment-intensive?
• Why combine mechanistic and data-driven models rather than relying on one, and how does the framework decide when to switch from one to the other?
• How many experiments does the hybrid approach typically need compared with conventional or purely data-driven method development?

Watch the first episode here:

Reference

1. Chia, D. N.; Duanmu, F.; Sorensen, E.; Mazzei, L.; Besenhard, M. O. A Hybrid Mechanistic-Data-Driven Framework for Knowledge-Driven and Experiment-Efficient HPLC Method Development. J. Chromatogr. A 2026, [volume], 467379. DOI: 10.1016/j.chroma.2026.467379.