The future of chromatographic method development (FuturoChrom): machine learning vs. humanbased modelling, or both?

Chromatographic problem solving, commonly referred to as method development (MD), is hugely
complex, given the many chemical parameters that need to be optimized and the very sensitive
relation between the value of these parameters and the elution time of the individual sample
compounds. While the current human-reasoning and model-based approaches often fail due to the
lack of accurate models, machine learning-based problem solving is now revolutionizing almost every
field of science and technology. The right time has thus arrived to investigate the possibilities of
reinforcement learning (RL) and deep learning (DL) for chromatographic MD.
FuturoChrom aims at developing model-free, purely RL-powered algorithms as well as hybrid
approaches, combining either RL or DL (Graph Neural Networks) with human-based modelling, for
fully flawless and autonomous MD. To achieve this, a consortium of one AI- and two
chromatography- experts has been formed, combining state-of-the-art know-how in chromatographic
MD and retention time modelling with the ability to develop world-class AI algorithms. The developed
algorithms will be compared to the current state-of-the-art in MD using samples provided by some of
Flanders’ most demanding industrial chromatography labs. To cover a statistically relevant number of
samples, a large-scale in silico comparison will be made as well. For this purpose, we will also
pioneer in establishing in silico test grounds that are as realistic as possible.