scBasset

scBasset models chromatin accessibility in single-cell ATAC-seq (scATAC) data using a sequence-based convolutional neural network to relate DNA sequence at accessibility peaks to cellular epigenetic profiles.

Key Features:

• Sequence-Based Convolutional Neural Network: Uses a CNN architecture that incorporates DNA sequence information associated with accessibility peaks.
• Peak-Level Sequence Modeling: Processes DNA sequences at scATAC accessibility peaks to capture sequence determinants of accessibility.
• Sparse, High-Dimensional scATAC Handling: Addresses the high dimensionality and sparsity characteristic of single-cell ATAC-seq data.
• Cell Clustering Performance: Demonstrates superior performance for clustering cells based on epigenetic accessibility patterns.
• scATAC Profile Denoising: Provides denoising of single-cell accessibility profiles to enhance signal quality.
• Cross-Assay Integration: Supports integration of scATAC-seq data across different assays, including single-cell multiome datasets.
• Transcription Factor Activity Inference: Infers transcription factor binding-related activity by relating accessibility peaks to underlying DNA sequence.

Scientific Applications:

• Cell Clustering: Facilitates identification and characterization of distinct cell populations from scATAC-seq data.
• Data Denoising: Enhances the quality of scATAC-seq profiles to improve downstream analyses.
• Assay Integration: Enables integration of scATAC-seq with other omics assays for multi-dimensional analysis.
• Transcription Factor Activity Inference: Provides insights into transcription factor–associated accessibility patterns at single-cell resolution.

Methodology:

scBasset applies a sequence-based convolutional neural network that processes DNA sequence associated with scATAC-seq accessibility peaks to capture spatial patterns of chromatin accessibility and relate them to underlying genetic sequences.