Predicting Biochemical and Physiological Parameters: Deep Learning from IgG Glycome Composition.

Area of Science:

• Biochemistry
• Immunology
• Biomarker Discovery

Background:

• Immunoglobulin G (IgG) N-glycosylation is crucial for its structure and function.
• Altered IgG glycosylation is linked to various diseases and reflects overall health.
• IgG glycans show potential as predictive health biomarkers.

Purpose of the Study:

• To assess the predictive power of IgG N-glycans for physiological and biochemical parameters.
• To compare regression and deep learning models for IgG glycan data analysis.
• To explore IgG glycan utility in personalized medicine and data imputation.

Main Methods:

• Developed elastic net regression and deep learning models using IgG N-glycan data.
• Trained models on the Korčula cohort and validated on the independent Vis cohort.
• Evaluated model performance in predicting various health-related parameters.

Main Results:

• IgG glycome composition accurately predicts biochemical and physiological parameters, particularly lipid/glucose metabolism and cardiovascular events.
• Both models showed similar performance on the training set.
• The deep learning model demonstrated superior generalization on the validation set.

Conclusions:

• IgG glycosylation serves as a robust indicator of an individual's health status.
• IgG glycan analysis holds promise for developing glycan-based diagnostics in personalized medicine.
• Predictive IgG glycan models can aid in imputing missing covariate data for deep learning applications.