A deep neural network based regression model for triglyceride concentrations prediction using epigenome-wide DNA...

Area of Science:

• Epigenetics
• Genomics
• Computational Biology

Background:

• Epigenetic modifications, like DNA methylation (DNAm), influence gene expression without altering the genome.
• DNAm variations serve as epigenetic markers for predicting phenotypic traits, including various diseases.
• This study investigates the use of epigenome-wide DNAm profiles to model triglyceride concentrations.

Purpose of the Study:

• To develop and evaluate deep neural network (DNN) regression models for predicting triglyceride concentrations.
• To assess the predictive accuracy of DNAm data before and after medication interventions.
• To compare the performance of DNN models against traditional machine learning methods like SVM.

Main Methods:

• Utilized epigenome-wide DNAm profiles from peripheral blood samples.
• Employed deep neural network (DNN) regression models to predict triglyceride concentrations.
• Compared DNN model performance with Support Vector Machine (SVM) for prediction accuracy.

Main Results:

• DNN models accurately predicted triglyceride concentrations using pretreatment and posttreatment DNAm data.
• Prediction accuracy was higher for visit 4 (posttreatment data) compared to visit 2 (pretreatment data).
• Pretreatment DNAm data yielded the best prediction for visit 4 triglyceride levels, suggesting long-term epigenetic effects.

Conclusions:

• Proposed DNN models demonstrated superior predictive performance for triglyceride concentrations compared to SVM.
• The DNN approach offers advantages for modeling high-dimensional epigenome-wide DNAm and other genomic data.
• This highlights the potential of machine learning in understanding epigenetic influences on human traits.