Unified Multitask Modeling for Retention Time Prediction Across Chromatographic Conditions.

Area of Science:

• Analytical Chemistry
• Computational Chemistry

Background:

• Retention time (RT) is crucial for compound identification in liquid chromatography-mass spectrometry (LC-MS).
• Existing RT prediction models are often limited to specific chromatographic conditions, hindering scalability and knowledge integration.
• Fragmented models lead to inefficiencies in data analysis and quality control.

Purpose of the Study:

• To develop a unified framework, Uni-RT, for accurate and scalable retention time prediction.
• To leverage multitask learning to capture shared molecular retention patterns and condition-specific variations.
• To improve the robustness and accuracy of RT prediction across heterogeneous LC-MS data sets.

Main Methods:

• Developed Uni-RT, a unified multitask learning framework.
• Trained the model on heterogeneous data sets from multiple chromatographic setups (RPLC and HILIC).
• Evaluated model performance on 28 diverse LC-MS data sets.

Main Results:

• Uni-RT demonstrated higher accuracy and robustness compared to pooled or condition-specific models.
• The multitask learning approach effectively captured both general and specific retention behaviors.
• Uni-RT significantly simplified the deployment of RT prediction models.

Conclusions:

• Multitask learning offers a powerful and generalizable solution for integrating RT prediction into LC-MS.
• Uni-RT enhances the reliability of compound identification, feature alignment, and quality control in LC-MS workflows.
• This unified framework addresses the limitations of condition-specific models, paving the way for broader applications.