Abstract:73 Background: Epigenetic alterations have been shown to govern many disease states, including cancer. A key epigenetic mark is the DNA methylation of CpGs. Large studies, like The Cancer Genome Atlas, have shown that DNA methylation can differentiate between cancer status and type. Here, differentially methylated genomic patterns were defined between patients with CRC and healthy individuals by utilizing patterns of co-methylation of adjacent CpG loci. Methods: Deep, whole genome bisulfite sequencing (WGBS) was conducted on DNA samples from 100 healthy individuals (median effective coverage of 352x±38x) and from 75 patients with CRC (median effective coverage of 48x±18x). The two cohorts were chosen to be balanced in terms of age (CRC: 58±11 years; healthy: 57±8 years) and to represent major ethnicities in the USA, and the CRC cohort was chosen to represent all stages, histologies and morphologies, and microsatellite instability (MSI, 23%). A novel machine learning model was developed to discover differentially methylated regions between the healthy and CRC cohorts as well as between subpopulations within the CRC cohort, including age, sex, stage, histology, and MSI vs microsatellite stability (MSS). Results: In total, we discovered 132,911 differentially methylated regions between healthy and CRC cohorts, spanning more than 1.14 million CpGs. 76% of all discovered regions were hypermethylated (70% CpG islands, 1.3% CpG shelves, 21% CpG shores) and the rest were hypomethylated (7.3% CpG islands, 8.6% CpG shelves, 25% CpG shores). We also discovered 220 hypermethylated regions that were differentially methylated between CRC MSI samples and CRC MSS samples and healthy cfDNA samples. Furthermore, we discovered subtype-specific differentially methylated regions, such as 62 hyper-methylated regions that were differentially methylated between mucinous adenocarcinoma samples and the rest of the cohort. Conclusions: Based on deep WGBS data from heterogeneous cohorts, representative of the US population eligible for CRC screening, we identified many differentially methylated regions as strong biomarker candidates for the early detection of cancer. Compared to recurrently mutated driver genes in CRC, we find that differential methylation profiles of CRC samples are generally much more homogenous and are therefore ideal biomarkers for targeted sequencing panels with the purpose of early cancer detection. Further, analysis of these patterns based on clinicopathologic factors enabled discovery of subtype specific markers, which can be incorporated to ensure equivalent performance for underrepresented subtypes or subpopulations.

A novel family of beta mixture models for the differential analysis of DNA methylation data: An application to prostate cancer

A family of mixture models for beta valued DNA methylation data

A Nonparametric Bayesian Approach for Clustering Bisulfate-Based DNA Methylation Profiles

HMM-DM: identifying differentially methylated regions using a hidden Markov model

Identifying Differential Methylation in Cancer Epigenetics via a Bayesian Functional Regression Model

A Hybrid Method Based on Beta-Binomial Bayesian Hierarchical Model and Combination of Ranking Method to Detect Differential Methylation Regions in Bisulfite Sequencing Data

A Novel Computational Method For Detecting Dna Methylation Sites With Dna Sequence Information And Physicochemical Properties

BDBB: A Novel Beta-distribution-based Biclustering Algorithm for Revealing Local Co-methylation Patterns in Epi-transcriptome Profiling Data

BBM: A novel beta-binomial-distribution-based biclustering algorithm for mining m 6 A co-methylation patterns

Identification of Differentially Methylated Sites with Weak Methylation Effects.

A new parsimonious method for classifying Cancer Tissue-of-Origin Based on DNA Methylation 450K data

Nonparametric Bayes Differential Analysis for Dependent Multigroup Data with Application to DNA Methylation Analyses in Cancer

Functional DNA methylation differences between tissues, cell types, and across individuals discovered using the M&M algorithm

Classification of Large DNA Methylation Datasets for Identifying Cancer Drivers

A Bayesian hierarchical model to estimate DNA methylation conservation in colorectal tumors

A Variational Bayes Beta Mixture Model for Feature Selection in Dna Methylation Studies

Comprehensive analysis of differentially methylated regions in colorectal cancer (CRC).

LogLoss-BERAF: An ensemble-based machine learning model for constructing highly accurate diagnostic sets of methylation sites accounting for heterogeneity in prostate cancer

Empirical Bayes Model Comparisons for Differential Methylation Analysis

Detecting differentially methylated regions in bisulfite sequencing data using quasi-binomial mixed models with smooth covariate effect estimates

MBMM: Moment Estimating Beta Mixture Model-based Clustering Algorithm for M6a Co-methylation Module Mining