Predicting gene expression using DNA methylation in three human populations | JoVE Visualize

Area of Science:

Epigenetics
Genomics
Gene Expression Analysis

Background:

DNA methylation is a key epigenetic regulator of gene expression, typically showing a negative correlation with gene activity in promoter regions.
The genome-wide predictive capacity of DNA methylation for gene expression profiles in human populations remains understudied.
Previous research focused on single CpG sites, leaving the potential of gene-region methylation as a surrogate for gene expression unexamined in existing DNA sample cohorts.

Purpose of the Study:

To investigate the predictive power of DNA methylation within gene regions for gene expression across individuals.
To compare the efficacy of different linear regression models (single, multiple, and LASSO) for this prediction task.
To assess how prediction performance varies across different human tissue and cell types and data sources.

Main Methods:

Analysis of DNA methylation and gene expression data from three human population datasets: adipose tissue (MuTHER), peripheral blood mononuclear cells (PBMC), and lymphoblastoid cell lines (LCL).
Application and comparison of single linear regression, multiple linear regression, and least absolute shrinkage and selection operator (LASSO) penalized regression models.
Evaluation of prediction performance using cross-validation R-squared values, with and without excluding CpG probes affected by cross-hybridization or SNPs.

Main Results:

LASSO regression demonstrated superior performance compared to other linear models.
Overall prediction power of DNA methylation for gene expression was generally low and dataset-dependent.
A higher number of genes (258) showed prediction with R² > 0.3 in the homogeneous LCL dataset compared to PBMC (30) and Adipose (42) datasets.
Including all CpG probes, without exclusion, improved prediction accuracy.

Conclusions:

DNA methylation at CpG sites within gene regions exhibits limited predictive power for gene expression across individuals using linear regression models.
The predictive capacity is influenced by tissue type, cell source, and data characteristics.
The combination of LASSO regression and the inclusion of all methylation array probes provides the most effective prediction of gene expression in this context.