New variable selection strategy for analysis of high-dimensional DNA methylation data

Abstract

In genetic association studies, regularization methods are often used due to their computational efficiency for analysis of high-dimensional genomic data. DNA methylation data generated from Infinium HumanMethylation450 BeadChip Kit have a group structure where an individual gene consists of multiple Cytosine-phosphate-Guanine (CpG) sites. Consequently, group-based regularization can precisely detect outcome-related CpG sites. Representative examples are sparse group lasso (SGL) and network-based regularization. The former is powerful when most of the CpG sites within the same gene are associated with a phenotype outcome. In contrast, the latter is preferred when only a few of the CpG sites within the same gene are related to the outcome. In this paper, we propose new variable selection strategy based on a selection probability that measures selection frequency of individual variables selected by both SGL and network-based regularization. In extensive simulation study, we demonstrated that the proposed strategy can show relatively outstanding selection performance under any situation, compared with both SGL and network-based regularization. Also, we applied the proposed strategy to identify differentially methylated CpG sites and their corresponding genes from ovarian cancer data.

Selection probability of multivariate regularization to identify pleiotropic variants in genetic association studies

“…Therefore, computation of selection probability does not require to choose the optimal tuning parameter. For this reason, selection probability has been widely used to identify top ranked genetic variants or genes that are associated with a phenotype outcome for analysis of high-dimensional genomic data (Alexander and Lange, 2011;Sun et al, 2017;Choi et al, 2018;Kim and Sun, 2019). However, selection probability has never been applied to multivariate regularization methods.…”

“…Note that the q-dimensional vectorβ j (I m ; λ, α) has either all zero values or all nonzero values because of a group lasso penalty in (2.2). Similarly, selection probability of the individual elastic-net method in (2.1) can be computed and details are described by Sun et al, 2017;Choi et al, 2018;Kim and Sun, 2019). However, the selection probability of the j th variant of the elastic-net can have up to q different values since the coefficient vector is estimated from each of q phenotype outcomes.…”

“…They are individual elastic-net, unified elastic-net, and multi-response elastic-net, respectively. To obtain single nucleotide polymorphism (SNP) data, we applied the same process used in Choi et al (2018). Specifically, we first generated two p-dimensional vectors (a i1 , .…”

“…Alternatively, regularization methods have been widely applied to analysis of high-dimensional genomic data (Wu et al, 2009;Zhou et al, 2010;Wang, 2012, 2013;Sun et al, 2017;Choi et al, 2018). They have a great advantage over individual tests since genetic effects of all variants in analysis can be computed in one regression framework.…”

MethylSPWNet and MethylCapsNet: Biologically Motivated Organization of DNAm Neural Network, Inspired by Capsule Networks

“…Feature selection methods and statistical learning methods such as sparse Group LASSO and network regularization have identified important CpGs in highly complex data. [29][30][31][32][33] More recent work has called for a greater understanding of the implications of DNAm-DNAm interactions through the incorporation of Gaussian Graphical Models, Canonical Correlation Analysis, and module discovery through weighted gene co-methylation networks. [34][35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50] There is growing support for the use of novel deep learning methods to aggregate, group, and select CpGs by their local context (e.g., genes) in an effort to connect and interpret the data with clinical outcomes.…”