Researchers have introduced scAGDE [ess-see-ay-jee-dee-ee], a new computational method for analyzing single-cell ATAC-seq [A-T-A-C-seq] data. ATAC-seq is a technique used to identify regions of open chromatin [kroh-muh-tin], which indicates where DNA is accessible for gene expression. Single-cell ATAC-seq allows scientists to examine this accessibility in individual cells, providing insights into cell identity and function.
scAGDE, short for single-cell chromatin accessibility model-based deep graph embedded learning method, is designed to efficiently process sparse single-cell ATAC-seq data. The method reconstructs both the chromatin accessibility profiles and the neighboring graph from the same low-dimensional cell representation. This enables scAGDE to retain the chromatin accessibility profiles of the analyzed cells and their neighboring cells, as well as the cell-cell interaction relationships between them.
The model uses a chromatin accessibility-based autoencoder [aw-toh-en-koh-der] to measure the importance of the peaks and select the key peaks, enabling scAGDE to focus on highlighting the chromatin accessibility profiles of each analyzed cell or the most critical peak regions. A dual-decoder component reconstructs the cell topology and estimates the distribution of data, ensuring accurate modeling of scATAC-seq data while retaining the relationship of each cell with its neighboring cell profiles in representation. scAGDE defines a dual cluster optimization objective to guide the preservation of the information about cell heterogeneity in its representation.
The primary objective of scAGDE is to revolutionize the analysis of high-dimensional sparse scATAC-seq data by learning low-dimensional topological embedding representations. The method involves processing scATAC-seq data using a chromatin accessibility-based autoencoder and the graph embedding learning procedure. The chromatin accessibility-based autoencoder comprehensively learns the latent representation of the raw data matrix. This step involves selecting significant peaks as features for cell characterization and constructing a cell graph based on this information. scAGDE integrates a Graph Convolutional Network (GCN) [jee-see-en] as an encoder, which extracts crucial information and considers cell-cell relationships in the cell graph, and then applies a Bernoulli-based decoder to model the probability of chromatin opening events.
Experimental tests have demonstrated that scAGDE outperforms existing scATAC-seq analysis methods across multiple synthetic datasets derived from ATAC-seq data of bone marrow and several real-world datasets characterized by different degrees of sparsity, sequencing platforms, and species diversity. Moreover, scAGDE effectively supports dimensionality reduction, visualization, and dropout event correction. Specifically, through the imputation of a mouse forebrain dataset, scAGDE identified potential accessible peaks that contain insightful regulatory elements, including crucial transcription factor binding motifs. Extended analysis in a human brain dataset revealed that scAGDE could successfully annotate cis-regulatory element (CRE)-specified cell types while uncovering functional diversity within glutamatergic neurons.
The researchers evaluated the clustering performance of scAGDE on simulated single cell ATAC-seq datasets with varying dataset characteristics. Simulation datasets were varied in terms of read depth, noise level, and dropout rate to simulate different biological scenarios and generated with six annotated cell populations, including hematopoietic stem cells (HSC) [H-S-C], common myeloid progenitor cells (CMP) [C-M-P], erythroid cells (Ery) [E-R-Y], natural killer cells (NK) [N-K], and CD4 [C-D-4] and CD8 [C-D-8] cells.
Comparative analysis with other scATAC-seq and scRNA-seq [ess-see-R-N-A-seq] methodologies elucidates scAGDE's superior performance on scATAC-seq data. In addition, scAGDE demonstrated exceptional ability in dimensionality reduction, visualization, dropout correction, and cell-type-specific enhancer discovery.
In summary, scAGDE offers a new approach to analyzing single-cell chromatin accessibility data, with potential applications in understanding cell identity, gene regulation, and disease mechanisms.