PretiMeth: precise prediction models for DNA methylation based on single methylation mark.

Area of Science:

• Computational biology and epigenetics
• Genomic data analysis and bioinformatics

Background:

• Computational prediction of DNA methylation at single CpG sites offers potential for analyzing vast 450K beadchip array data.
• Existing general prediction models prioritize overall accuracy, neglecting precise locus-specific predictions, limiting downstream applications requiring high precision.

Purpose of the Study:

• To develop a method for constructing precise prediction models for individual CpG loci.
• To improve the accuracy and applicability of DNA methylation prediction for downstream analyses.

Main Methods:

• PretiMeth utilizes a logistic regression algorithm to build individual prediction models for each CpG locus.
• Each model incorporates a single DNA methylation feature exhibiting the most similar methylation pattern to the target CpG locus.

Main Results:

• PretiMeth demonstrates superior prediction accuracy compared to other algorithms, maintaining robustness across different platforms and cell types.
• Application to The Cancer Genome Atlas (TCGA) data identified specific CpG loci and genes differentially methylated in tumor versus normal samples, warranting further validation.

Conclusions:

• Precise prediction of single CpG loci is crucial for expanding methylation array data utility and downstream analysis.
• PretiMeth's precise modeling approach, using single significant features, could inform future DNA methylation beadchip probe set design.
• PretiMeth is available as an open-source tool.