Probabilistic Multilevel Canonical Correlation Analysis (CCA) for Integrative Analysis of Multi-Omics Data

• Activity Number: 156 - Contributed Poster Presentations: Section on Statistics in Genomics and Genetics
• Type: Contributed
• Date/Time: Monday, August 8, 2022 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Yuna Kim* and Scarlett (she/her/hers) L. Bellamy and Jiao Li and Robert T Krafty and Lucy F. Robinson and Gail L. Rosen
• Companies: Drexel University Dornsife School of Public Health and Drexel University, Dornsife School of Public Health and Baylor College of Medicine and Emory University and Drexel University Dornsife School of Public Health and Drexel University
• Keywords: Multi-Omics; Data Integration; Dimensionality Reduction; Multilevel Data; Latent Variable Model

Abstract

Multi-omics data have been used to characterize covariation in multiple biological profiles, allowing for a more comprehensive understanding of complex biological processes. Moreover, the reduction in costs of high-throughput technologies has further broadened the scope of multi-omics studies enabling the collection of repeated measurements or longitudinal data. While mixed effects models are widely used in single omics applications, their use in applications for integrative analyses of multi-omics data with repeated measurements is less developed. Probabilistic canonical correlation analysis (pCCA) considers probability models for jointly studying the relations among two sets of data. We propose a probabilistic multilevel CCA that extends pCCA to repeated measurements data to help learn the underlying shared structures between two omics data sources at both the within- and between-subject levels simultaneously. This modelling strategy can model both the covariation of within-person patterns over different conditions and estimate average patterns between individuals. We examine the operating characteristics of our proposed method through simulation studies.