Learning Sparse Gaussian Graphical Models with Overlapping Blocks

Authors: Mohammad Javad Hosseini, Su-In Lee

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show that GRAB reveals the underlying network structure substantially better than four state-of-the-art competitors on synthetic data. When applied to cancer gene expression data, GRAB outperforms its competitors in revealing known functional gene sets and potentially novel cancer driver genes.
Researcher Affiliation Academia Mohammad Javad Hosseini1 Su-In Lee1,2 1Department of Computer Science & Engineering, University of Washington, Seattle 2Department of Genome Sciences, University of Washington, Seattle {hosseini, suinlee}@cs.washington.edu
Pseudocode No No explicit pseudocode or algorithm blocks (e.g., a figure or section labeled 'Pseudocode' or 'Algorithm') were found in the paper. The algorithm steps are described in paragraph text.
Open Source Code Yes The source code is available at: http://suinlee.cs.washington.edu/software/grab
Open Datasets Yes We consider the MILE data [20] that measure the m RNA expression levels of 16,853 genes in 541 patients with acute myeloid leukemia (AML). For a better visualization of the network in limited space (Fig 5), we selected 500 genes4, consisting of 488 highest varying genes in MILE and 12 genes highly associated with AML: FLT3, NPM1, CEBPA, KIT, N-RAS, MLL, WT1, IDH1/2, TET2, DNMT3A, and ASXL1. These genes are identified by [21] in a large study on 1,185 patients with AML to be significantly mutated in these AML patients.
Dataset Splits Yes We generate 100 training samples and 50 validation samples from the multivariate Gaussian distribution with mean zero and the covariance matrix equal to the inverse of the precision matrix.
Hardware Specification Yes GRAB runs for 0.5-1.5 hours for 500 genes and up to 20 hours for 2,000 genes on a computer with 2.5 GHz Intel Core i5 processor
Software Dependencies No The paper does not specify versions for any key software components or libraries used for implementation (e.g., 'Python 3.x', 'TensorFlow x.x', 'PyTorch x.x').
Experiment Setup Yes For all methods, we considered the regularization parameter λ 2 [.02, .4] with step size .02. Each method selects the regularization parameter using the standard cross-validation (or held out validation) procedure. GRAB selects K based on the validation-set log-likelihood in initialization. In the Z-step of the GRAB learning algorithm, we use step size 1/t, where t is the iteration number and iterate until the relative change in the objective function is less than 10 6 (Section 3.3). We use the warm-start technique between the BCD iterations. For that, we fix the number of blocks to be K = 10 across all methods.