GRIP: physics-informed neural network for gradient retention time prediction in liquid chromatography

Abstract

Gradient liquid chromatography has numerous applications in life sciences. Retention time prediction remains challenging due to complex underlying physical processes and highly variable chromatographic conditions. Here, we present GRIP, a physics-informed neural network for gradient retention time prediction that explicitly uses experimental setup parameters. GRIP demonstrates zero-shot generalization to unseen chromatographic systems while being on par or out-performing the transfer learning-based baseline. This approach can computationally guide the experimental setup configuration tailored to specific compounds of interest.