Masters candidate in Biostatistics, Tianhua Wu, will present:

“Cross-Platform Imputation of DNA Methylation Data Using Penalized Regression and Kernel-Based Method”

Plan B Adviser: Weihua Guan

Abstract: DNA methylation is an important epigenetic factor that participates in the induction of a variety of human diseases. Recent developments in high-throughput technologies have allowed genome-wide profiling of the human methylome. However, rapid technology advancement can lead to gaps in the coverage on methylome from an older generation of technology. For example, the Illumina Infinium HumanMethylation450 (HM450) BeadChip and Infinium MethylationEPIC (EPIC) BeadChip are the two most commonly used microarray-based platforms, with the EPIC array almost doubling the probes on HM450 as a newer platform. The mixed profiling impedes the joint analysis of methylation data from these two platforms. In order to efficiently combine results from EPIC and HM450, we propose imputation methods to impute methylation measures at the EPIC-only probes from those on the HM450 array. Specifically, we consider penalized functional regression models and kernel-based functions to impute missing probes from a large number of correlated probes. We compare the computation efficiency and imputation accuracy of these approaches. We observe that the penalized regression provides better imputation accuracy than kernel-based method, but with a much heavier computational burden. The imputation performance of both methods can be improved substantially through K-means clustering and pre-selection of correlated CpG sites. Our methods can greatly benefit collaborations in the scientific community for efficient consolidation of epigenetic data and results.

Presented via Zoom: https://umn.zoom.us/j/93558835500
Meeting ID: 935 5883 5500

Key words: DNA methylation, Cross-platform imputation, penalized regression, kernel function