Single-cell Multiomics Data Integration and Generation with scPairing

In this 56-minute talk from Valence Labs, Jeffrey Niu presents research on "Single-cell multiomics data integration and generation with scPairing." Learn about a novel variational autoencoder model inspired by Contrastive Language-Image Pre-Training that embeds different cellular modalities onto a common embedding space. Discover how scPairing enables the generation of new multiomics data through bridge integration, allowing researchers to link unimodal data using existing multiomics bridges. The presentation covers extensive benchmarking showing how scPairing captures both coarse and fine biological structures, applications in generating new multiomics data from retina and immune cells, and extensions to co-embed three modalities for creating trimodal datasets of bone marrow immune cells. This research offers a cost-effective alternative to expensive multiomics technologies, enabling researchers to discover new biological relationships across modalities or confirm hypotheses. Part of the AI for drug discovery community at Portal by Valence Labs.