Learning single-cell chromatin accessibility profiles using meta-analytic marker genes.

Area of Science:

• Neuroscience
• Genomics
• Epigenetics

Background:

• Single-cell assay for transposase accessible chromatin using sequencing (scATAC-seq) is crucial for identifying cis-regulatory elements like enhancers and transcription factor binding sites.
• Challenges in scATAC-seq data analysis include cell-type identification difficulties due to protocol heterogeneity and high dropout rates.

Approach:

• Systematically compared seven scATAC-seq datasets from mouse brains to benchmark neuronal cell-type annotation using gene sets.
• Evaluated the efficacy of redundant marker genes and simple gene set aggregation against machine-learning classifiers for scATAC-seq annotation.
• Developed a deep neural network to predict chromatin accessibility from DNA sequence, identifying key motifs for neuronal subtypes.

Key Points:

• Redundant marker genes substantially enhance sparse scATAC-seq annotation across diverse datasets.
• Simple aggregation of marker genes achieves performance comparable to or exceeding machine-learning classifiers for cell-type annotation.
• Reannotated scATAC-seq data for detailed cell types using robust marker genes, making meta profiles publicly available.
• Identified sequence-specific regulatory elements through deep neural network predictions of chromatin accessibility.

Conclusions:

• Marker gene aggregation presents a robust and accessible method for neuronal cell-type annotation in scATAC-seq data.
• The study provides a valuable, publicly accessible database of meta scATAC-seq profiles and sequence-based predictions for exploring epigenetic regulation.
• Findings facilitate a deeper understanding of cell-type specific epigenetic regulation in the brain.