A1427
Title: Statistical learning in high-dimensional methylation data in cancer using trans-dimensional hidden Markov models
Authors: Farhad Shokoohi - University of Nevada Las Vegas (United States) [presenting]
Abstract: The analysis of high-dimensional methylation data is increasingly critical in biology and health sciences due to its significant role in cancer development and progression. Various statistical methods and analytical tools have been developed to investigate DNA methylation, particularly in identifying differentially methylated cytosines or regions between groups, such as cancerous versus healthy tissues. However, analyzing high-dimensional methylation data presents substantial challenges, including heavy missing data, low read depths, functional autocorrelation patterns, the presence of multiple covariates, and the need to address multiple comparisons. These challenges are explored, and an overview of current methodologies and tools is provided, including two of the recently published approaches. Furthermore, a novel method that leverages trans-dimensional Markov chain Monte Carlo techniques with hidden Markov models and binomial emissions tailored for bisulfite sequencing data is introduced. The effectiveness of these methods is illustrated through both simulations and real data analyses on acute Leukemia and colorectal cancer. Insights are offered into the latest advancements in high-dimensional methylation analysis, and how these approaches can enhance the understanding of epigenetic changes in cancer are discussed, potentially leading to new therapeutic strategies.
