Harnessing Multimodal Single-Cell Sequencing Data for Integrative Analysis
summary
Presents Cobolt, a method that uses a Multimodal Variational Autoencoder with an LDA-style generative layer to learn shared latent cell states from mixed scRNA-seq and ATAC-seq datasets. It integrates paired multi-modality data plus separate single-modality experiments, keeps peak-level ATAC information rather than collapsing to genes, and shows better cell-to-cell alignment than alternatives on mouse cortex and human PBMC (including SNARE‑seq paired cells), demonstrating practical improvement in joint representation.
AI-written summary · about
chapters (8)
- 0:00HARNESSING MULTIMODAL SINGLE-CELL SEQUENCING DATA FOR INTEGRATIVE ANALYSIS
- 0:41New Opportunities of Single-cell
- 3:03Single Cell Beyond the Transcriptome Epigenetics Open Chromatin (ATAC-Seq) Methylation
- 5:28How to combine different modalities? Generally technologies on different cells
- 8:41What Unified Representation Latent variables shared by cell, describing biology
- 16:44Example of Integration (Mouse Cortex)
- 18:49Comparison to other methods
- 20:15Large Sample Sizes?
related talks
35mMachine Learning for Single-Cell 3D Epigenomics
38mInterpreting Deep Learning Models Of Functional Genomics Data To Decode Regulatory Sequence...
24mHigh-throughput cellular analytics to identify central nervous system targets for gene therapy
47mBANKSY: A Spatial Omics Algorithm that Unifies Cell Type Clustering and Tissue Domain Segmentation
37mA Practical Guide to Robust Multi modal Machine Learning in Education
24mHybrid ML Methods for the Interpretation and Integration of Heterogeneous Multimodal Data
35mFrom cells to drug responses - machine learning in cancer research
15mMulti-Modal AI Optimization: Combining Text, Vision, and Speech Efficiently