Multi-modal Diffusion Model with Dual-Cross-Attention for Multi-Omics Data Generation and Translation

Explore a cutting-edge computational approach for single-cell multi-omics data integration and generation in this 48-minute conference talk. Discover scDiffusion-X, a novel latent diffusion model that addresses the challenge of simultaneously measuring multi-omics data from the same cells by using autoencoders to map multiple modalities into low-dimensional latent spaces. Learn about the innovative Dual-Cross-Attention (DCA) module that enables the model to uncover hidden links between modalities and extract comprehensive relationships between genes and regulatory elements. Examine how this framework excels in generating multi-omics data under various conditions while achieving high-fidelity translation between modalities—capabilities that existing multi-omics data simulators cannot match. Understand the superior performance of scDiffusion-X in scalability, data quality generation, and model interpretability through extensive benchmarking experiments, and see how this powerful tool can unlock the potential of single-cell multi-omics data for studying complex cellular mechanisms in drug discovery applications.