In this project, we used ATAC-seq to inform directionality in single-cell trajectory inference by adapting the CellRank algorithm to use gene regulation instead of RNA velocity.
UMAP of metacell gene expression colored by cell type.
Macrostate and terminal state membership computed using CellRank with a combined kernel (gene expression + ATAC-seq).
To run the code, first install the required Python libraries from the environment.yml file.
conda env create -f environment.yml
Most of the data to run the code has been included in the data/
directory:
data/rna_meta_ad.p
: gene expression for metacells (output ofSEACell_ATAC_analysis.ipynb
)data/cd34_gene_scores.csv
: gene scores for metacells (output ofSEACell_ATAC_analysis.ipynb
)
To run the SEACell_ATAC_analysis.ipynb
notebook, download the following large data files separately:
This project contains four key notebooks:
SEACell_ATAC_analysis.ipynb
: This notebook from the SEACells repo computes metacells and gene accessibility scores given RNA-seq and ATAC-seq data.Pseudotime Kernel.ipynb
: Runs CellRank with CytoTRACEKernel (directionality informed by gene expression only, no ATAC-seq).Cosine Similarity Kernel.ipynb
: Runs CellRank with custom CosineSimilarityKernel (directionality informed by ATAC-seq only, no gene expression).Combined Kernel.ipynb
: Runs CellRank with combined ConnectivityKernel and CosineSimilarityKernel (directionality informed by both gene expression and ATAC-seq).