We introduce novel methods to integrate methylation and mRNA data to construct gene-specific methylation profiles that find for each gene/tissue type a sparse set of CpGs best explaining expression with weights indicating association direction/strength and producing summaries for integrative genomics modeling. Using TCGA and MD Anderson colorectal cohorts to build and validate, we show ours outperforms standard approaches, identifies which CpG types are most important, demonstrates selected CpG tend to be enriched for certain annotated chromatin states, and produces gene-level scores showing key methylation differences across newly discovered CRC subtypes. Results for multiple cancers are shared by R-Shiny app.

Gene-specific methylation profiles for integrative methylation-expression analysis in cancer research