Multilevel pharmacokinetics-driven modeling of metabolomics data

Area of Science:

• Quantitative statistical modeling
• Pharmacokinetics
• Metabolomics

Background:

• Multilevel modeling analyzes variability and relationships within structured data.
• It aids in prediction, data reduction, and causal inference from diverse study types.
• This study applies multilevel modeling to metabolomics data.

Purpose of the Study:

• Introduce multilevel pharmacokinetics-driven modeling for metabolomics.
• Analyze nucleoside and creatinine concentrations in urine from healthy and cancer patients.
• Evaluate the model's predictive performance for disease classification.

Main Methods:

• Developed a Bayesian multilevel model for nucleoside/creatinine ratios.
• Incorporated age, sex, and health status as covariates.
• Validated predictive performance using ROC, sensitivity, and specificity.

Main Results:

• Cancer associated with a 1.42-fold increase in methylthioadenosine/creatinine excretion.
• Age decreased creatinine clearance, affecting nucleoside excretion rates.
• Model showed limited predictive utility (AUC 0.57) for disease classification.

Conclusions:

• Bayesian multilevel pharmacokinetics-driven modeling offers insights into metabolomics data.
• This approach may serve as a novel tool for identifying disease biomarkers.
• Further research is needed to enhance predictive capabilities.