The lunch at UiO starts at 12:00, while the talk starts at 12:15.
Speaker: Camilla Lingjærde, MRC Biostatistics Unit, University of Cambridge
Location: 8th floor, Niels Henrik Abels hus and Zoom (link below)
Title: A scalable ECM algorithm for multiple-network joint inference with the graphical horseshoe
Abstract: In statistical omics, network models are useful tools for modelling complex associations and assessing pathway activity. If a Gaussian graphical model is assumed, an association network can be estimated by determining the non-zero entries of the inverse covariance (precision) matrix of the data. Within a Bayesian framework, the graphical horseshoe estimator provides a robust and flexible framework for precision matrix inference in Gaussian graphical models, as it assumes local, edge-specific parameters which prevent over-shrinkage of non-zero off-diagonal elements. However, in the high-dimensional settings commonly found in biological networks, the Gibbs sampler for the graphical horseshoe is often computationally inefficient or even unfeasible. Further, the model is only formulated for a single network and does not offer an integrative analysis should there be multiple data sets available that might share common structures. We propose a novel scalable expectation conditional maximization (ECM) algorithm for obtaining the posterior mode of the precision matrix in the graphical horseshoe. Moreover, we propose the novel joint graphical horseshoe estimator (jointGHS), which extends our framework to a multiple network setting so that edge-specific information can be shared between related networks to improve estimation. We apply our method to both simulated and real omics data sets, and show that our approach outperforms the existing Gibbs sampler both in terms of scalability and accuracy. We also show that our joint network approach successfully shares information between networks while capturing their differences, outperforming state-of-the-art frequentist and Bayesian methods at any level of network similarity.
Join Zoom Meeting:
https://uio.zoom.us/j/63199595088?pwd=Y05GYkJwR1dCUHFhUnlGMFpsdGc3UT09
Meeting ID: 631 9959 5088
Passcode: 302551
Welcome!
Best regards,
Thea Roksvåg and Lars Henry Berge Olsen.
Photo by Kim Ramberghaug.
