Multi-omics Variational Autoencoding - Deep Dive and Primer

Explore a comprehensive conference talk on multi-omics variational autoencoding presented by Ricardo Hernandez Medina and Simon Rasmussen at the Broad Institute of MIT and Harvard. Delve into the MOVE (multi-omics variational autoencoders) pipeline, which leverages deep learning-based generative models to gain insights into complex biological datasets. Learn about data pre-processing, model optimization, and techniques for integrating multi-modal data, including clinical measurements, microbiome census data, transcriptomics, proteomics, and lifestyle records. Discover approaches for determining associations between omics variables and categorical labels, including univariate statistical methods, ensemble modeling, and Bayesian decision theory. Gain insights into the application of variational autoencoders (VAEs) for analyzing various types of multi-omics data, including microbiome-derived data for genome reconstruction, data-driven stratification of major depressive disorder and schizophrenia, and integration of patient-level multi-omics data in Type 2 Diabetes research. Understand how virtual perturbations and ensemble modeling are used to estimate drug-omics associations and predict drug responses across omics data.