Accurate prediction of peptide retention times is crucial for reliable identification in liquid chromatography–tandem mass spectrometry (LC–MS/MS) proteomics. Current methods using static neural networks fail to capture the complex sequential dependencies in peptide structures that determine retention patterns. Moreover, conventional hyperparameter optimization strategies are computationally inefficient and prone to suboptimal solutions in complex search spaces. We introduce IALA-LNN, a framework that employs liquid neural networks (LNN) with state evolution governed by ordinary differential equations (ODE). This architecture captures position-dependent sequence information and progressively integrates information throughout sequence processing, enabling effective modeling of retention time based on the combined contributions of amino acid residues. The framework combines dual encoding via ESM-2 and ProtT5 protein language models with an improved artificial lemming algorithm (IALA) that features a multielite pool guidance, gradient-enhanced search, and Goodnode initialization for hyperparameter optimization. Across RP, SCX, and HILIC chromatography, IALA-LNN achieved R 2 values of 0.994, 0.998, and 0.998, with MAE values of 0.77, 0.085, and 0.07 min, substantially outperforming DeepRT, DeepLC, and Prosit. These results demonstrate that differential equation-based neural networks effectively model retention patterns, directly enhancing peptide identification reliability, reducing false discovery rates, and supporting precision proteomics applications, including biomarker discovery and targeted workflows.
Chen et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: