2016/04/01 by Anindya Bhadra, Arvind Rao, Bhadra, Anindya +3 · 1 citation
Biochemistry, Genetics and Molecular Biology · Mathematics · #Bioinformatics and Genomic Networks #FOS: Computer and information sciences #Gene expression and cancer classification #Methodology (stat.ME) #Statistical Methods and Inference
paper · pdf · doi:10.48550/arxiv.1604.00376
openalex publication_date 2016/04/01 · openalex created_date 2022/09/28 · openalex updated_date 2026/07/28
Inferring dependence structure through undirected graphs is crucial for\nuncovering the major modes of multivariate interaction among high-dimensional\ngenomic markers that are potentially associated with cancer. Traditionally,\nconditional independence has been studied using sparse Gaussian graphical\nmodels for continuous data and sparse Ising models for discrete data. However,\nthere are two clear situations when these approaches are inadequate. The first\noccurs when the data are continuous but display non-normal marginal behavior\nsuch as heavy tails or skewness, rendering an assumption of normality\ninappropriate. The second occurs when a part of the data is ordinal or discrete\n(e.g., presence or absence of a mutation) and the other part is continuous\n(e.g., expression levels of genes or proteins). In this case, the existing\nBayesian approaches typically employ a latent variable framework for the\ndiscrete part that precludes inferring conditional independence among the data\nthat are actually observed. The current article overcomes these two challenges\nin a unified framework using Gaussian scale mixtures. Our framework is able to\nhandle continuous data that are not normal and data that are of mixed\ncontinuous and discrete nature, while still being able to infer a sparse\nconditional sign independence structure among the observed data. Extensive\nperformance comparison in simulations with alternative techniques and an\nanalysis of a real cancer genomics data set demonstrate the effectiveness of\nthe proposed approach.\n