DNA methylation is a key mechanism that regulates gene transcription and its importance in carcinogenesis has been widely explored. Both hypo and hyper-methylated genes can deregulate gene expression and variations in methylation are associated with several diseases. We previously proposed a method called MethylMix, available as an R package at Bioconductor, which aims to derive key methylation-driven genes in cancer which have an effect on gene expression. MethylMix consists of a three-step algorithm: it first identifies subgroups of samples with common methylation patterns, then compares each subgroup to normal tissue samples to define hypo- or hyper-methylation states, and finally reports a gene as being methylation-driven if it is transcriptionally predictive defined as a significant negative linear association between methylation level and gene expression.Now we present MethylMix 2.0, introducing some extensions to the original method. MethylMix 2.0 uses a bivariate Gaussian mixture model to jointly model DNA methylation and gene expression in order to identify the different subgroups of cancer samples. Also, while MethylMix focuses only on genes with a strong negative linear relationship between DNA methylation and gene expression, MethylMix 2.0 extends the search for driver genes to include genes which show other types of association. We applied MethylMix 2.0 on six large cancer cohorts and show that MethylMix 2.0 identifies known and new hypo- and hyper-methylated genes and can also be used to identify samples subtypes. MethylMix 2.0 is intended to be submitted soon as an R package to Bioconductor.