Online Program Home
My Program

Abstract Details

Activity Number: 659 - Recent Advances in Dimension Reduction and Clustering
Type: Contributed
Date/Time: Thursday, August 1, 2019 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #307248 Presentation
Title: B-MuLe: Sparse Multi-View Representation Learning Problem with Application in Multi-Omics Studies
Author(s): Omid Shams Solari* and James Bentley Brown
Companies: and Uc Berkeley statistics
Keywords: Sparse cca; Multi-view representation learning; Statistical Learning; Optimization; Multi-omics; Multivariate analysis

A reformulation of block sparse CCA algorithm is presented where the non-convex objective is reformulated as an optimization program of maximizing a convex objective over a convex set which shrinks the search space resulting in a significantly faster algorithm. A first order gradient method is then proposed. Simulations reveal that the proposed method outperforms alternating optimization algorithms in terms of canonical correlation, orthogonality of the covariates, convergence characteristics and stability of the solution, while being significantly faster. Our algorithm is available as an R package called B-Mule, application of which in the biological sciences enables the integration of multi-omics data even when several omics modalities are present. Here we used B-MuLe to jointly analyze transcriptomics, metabolomics and microbiome data in a toxicological study. Our analyses reveal genetic and metabolomic pathways between the host and its microbiome that modulate the toxicity of two widely used herbicides.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program