Please join us for a UW Data Science Seminar featuring UW Aquatic and Fishery Sciences Professor Steven Roberts and Graduate Student Kathleen Durkin on Thursday, March 5th from 4:30 to 5:20 p.m. PT. The seminar will be held in IEB G109.

“Sparse Tensor Decomposition and Elastic Net Regression for Multi-Species Gene Expression Analysis”

Abstract: Classical transcriptomic methods — pairwise differential expression (DESeq2/edgeR), PCA, and WGCNA — are fundamentally two-dimensional and cannot natively represent datasets structured along three simultaneous axes: genes, samples, and time. When applied to multi-species time-series data, these approaches require flattening or ignoring at least one axis, leading to information loss, arbitrary thresholding, and post-hoc comparisons that lack statistical coherence.

Here we apply sparse CP tensor decomposition, implemented in the Barnacle framework, to E5 timeseries coral transcriptomic data spanning three evolutionarily divergent coral species (Acropora pulchra, Porites evermanni, and Porites tuahiniensis) across four timepoints (TP1–TP4). Gene expression was quantified at the ortholog-group level, constructing a three-mode tensor (ortholog groups × species-samples × timepoints) and normalized using sctransform. Details of the E5-MOSAiC project can be found here.

Building on these tensor decomposition results, we further applied an Elastic Net (EN) regression pipeline to predict the expression using multi-omic epigenetic predictors. Gene sets identified from individual tensor components were then used as response variables in Elastic Net regression models, with three classes of epigenetic predictors. Our results identify specific miRNAs, lncRNAs, and CpG sites that are the strongest predictors of gene expression variation within each species, providing evidence for multi-layered epigenetic regulation of transcription in corals across seasonal time scales.

This work demonstrates that sparse tensor decomposition is a principled, fully unsupervised approach for identifying conserved and species-specific gene co-expression programs across multi-species, multi-timepoint molecular datasets, a class of problem increasingly common in environmental and comparative genomics

Speaker Bios: Steven Roberts is a Professor in the School of Aquatic and Fishery Sciences at the University of Washington, where his research advances environmental memory as a foundational principle linking molecular regulation, physiology, and resilience in marine organisms. His work integrates genomics, epigenetics, and organismal physiology to understand how past environmental conditions shape future performance across individuals, populations, and generations. Through a focus on marine invertebrates, his research bridges discovery and application, informing conservation, restoration, and sustainable aquaculture in changing coastal ecosystems.

Kathleen Durkin is a graduate student in Aquatic and Fishery Sciences with broad interests in environmental responses in marine invertebrates. As an undergraduate at Harvey Mudd College, she studied soft coral taxonomy and population genetics using single- and multi-locus sequencing approaches, graduating in 2023 with a B.S. in Mathematical and Computational Biology. Her current research includes studying non-coding RNAs in corals and multigenerational DNA methylation patterns in Eastern oysters. Kathleen maintains a publicly accessible digital lab notebook that documents her ongoing research and reflects a commitment to open science practices.