Encoding, Regression, and Classification of Transcription Factors’ Specificity and Methylation Effects

The methylation effects on protein-DNA interactions, which can be perceived as a special kind of specificity of transcription factors, have been successfully quantified in the last years by various methods. In this work, I give a summary about the sequence encoding scheme, the underlying additive model about specificity and methylation sensitivity, and the regression strategy to analyze Methyl-Spec-seq data. Then I explain why given the current experimental setup, it is more appropriate to model the methylation effects based on pairwise comparison between individual unmethylated and methylated site, rather than the combined regression of all model parameters together. I also developed a computational package TFCookbook to demonstrate the analysis procedures step-by-step. At last, it is possible to classify the various types of methylation effects based on whether or not the consensus site contains CpG dinucleotide and whether methylation increase or decrease the binding affinity. Additionally, this specificity modeling and analysis strategy, can be extended to study other types of DNA modifications in general.