Application of a Deep Learning Model to Predict Liquid Chromatography Retention Times of Food Peptides Across Chromatographic Conditions

Comparing predicted and measured retention times can greatly enhance the reliability of peptide identification in LC-MS analysis of smaller, food-derived peptides where MS spectral information alone is often insufficient. Unfortunately, the extensive data sets of peptide retention times from proteomics repositories, or prediction models derived from them, have limited applicability to food-derived peptides due to the structural diversity of these peptides. To address this, we applied a transfer learning approach by fine-tuning a generic deep learning model initially trained on large proteomics datasets using our own experimental data obtained from commercial peptide standards.

The method utilizes an easy to implement retraining strategy that significantly reduces data requirements and training time compared to building a model from scratch. The retrained model demonstrated strong predictive performance (Q² > 0.98), and 95% of the retention time predictions of a yeast protein hydrolysate validation set fell within a ±1.0 min window across a wide range of chromatographic conditions, demonstrating both its robustness and practical relevance. We further validated this approach by applying it to the analysis of plant protein hydrolysates. The good performance seen showed its versatility and applicability for diverse sets of peptides including tryptic and non-tryptic peptides.

Our work underscores the potential of transfer learning in chromatographic analysis, providing an efficient and adaptable tool for rapid and reliable peptide analysis in food research. Transfer learning enabled the utilization of extensive databases from the proteomics area in the much narrower and specialized field of food peptide analysis.