Discovering Temporal Patterns in Longitudinal Nontargeted Metabolomics Data via Group and Nuclear Norm Regularized...

Area of Science:

• Metabolomics
• Bioinformatics
• Systems Biology

Background:

• Longitudinal metabolomics data analysis often overlooks temporal dynamics.
• Conventional methods like PLS-DA and OPLS-DA are limited in capturing time-dependent patterns.

Purpose of the Study:

• To develop a novel multitask learning (MTL) method for discovering temporal patterns in longitudinal metabolomics data.
• To enhance biomarker selection while maintaining high predictive performance.

Main Methods:

• Proposed a multitask learning (MTL) method with structural regularization.
• Incorporated group regularization for biomarker selection and nuclear norm for temporal interrelationships.
• Validated the method on metabolomics and simulated datasets.

Main Results:

• The proposed MTL method successfully identified a compact set of temporal biomarkers.
• These biomarkers characterized the antipyretic process of Qingkailing injection.
• The nuclear norm improved the group norm's recovery ability.

Conclusions:

• The developed MTL method is effective for analyzing temporal associations in longitudinal metabolomics.
• It offers improved biomarker discovery and characterization of dynamic biological processes.
• This approach advances the analysis of time-series data in metabolomics research.