Cobolt

Cobolt models and integrates multi-omic single-cell sequencing datasets to learn latent representations shared across omic modalities and jointly analyze gene expression, chromatin accessibility, and methylation data.

Key Features:

• Multimodal Variational Autoencoder (VAE): Uses a Multimodal VAE within a hierarchical generative model to jointly model multiple data types.
• Hierarchical generative model: Employs a hierarchical generative framework to represent shared and modality-specific latent structure.
• Sparsity and high-dimensionality handling: Models and is robust to sparse counts and high-dimensional feature spaces typical of single-cell omics such as chromatin accessibility and methylation.
• Multi-omic integration: Integrates gene expression and chromatin accessibility data from modalities including scRNA-seq and ATAC-seq.
• Latent representation learning: Learns latent representations shared across different omic modalities to enable joint analyses.

Scientific Applications:

• Cellular heterogeneity: Resolve cellular heterogeneity at single-cell resolution by integrating multiple omic layers.
• Regulatory mechanism analysis: Correlate gene expression with chromatin accessibility to study regulatory mechanisms.
• Gene-environment interactions: Facilitate analysis of gene-environment interactions using integrated multi-omic profiles.
• Genomics, epigenetics, and systems biology: Support integrative studies in genomics, epigenetics, and systems biology using joint-modality datasets.

Methodology:

Implements a Multimodal Variational Autoencoder within a hierarchical generative model and models sparse counts and high-dimensional feature spaces for joint analysis of scRNA-seq, ATAC-seq, chromatin accessibility, and methylation data.