ArchVelo is a method for modeling gene regulation and inferring cell trajectories using simultaneous single-cell chromatin accessibility and transcriptomic profiling (scRNA+ATAC-seq). ArchVelo extracts a set of shared archetypal chromatin accessibility profiles and models their dynamic influence on transcription. As a result, ArchVelo improves the accuracy of trajectory inference compared to previous methods and decomposes the velocity field into components driven by distinct regulatory programs.

This repository contains the source code for the ArchVelo Python package.

Installation

ArchVelo requires Python 3.11 or newer. We strongly recommend creating a dedicated virtual environment before installation.

The package and its dependencies can be installed with a single command directly from this GitHub repository:

pip install git+https://github.com/pritykinlab/ArchVelo.git

If you want to modify the source code, or run the demo, you should clone the repository and install it in the editable mode:

git clone https://github.com/pritykinlab/ArchVelo.git
cd ArchVelo
pip install -e .

Tutorials

• Demo: A demonstration of ArchVelo on a scRNA+ATAC-seq dataset for the mouse embryonic brain can be found in the ArchVelo_demo.ipynb notebook in this repository.

• Detailed End-to-End Analysis: For a complete walkthrough of the ArchVelo analysis pipeline, please see our dedicated notebooks repository. It includes applications to the mouse embryonic brain, human hematopoietic stem cells, and CD8 T cells in acute and chronic viral infection conditions.

  • ArchVelo Notebooks Repository

Minimal example

import pandas as pd
import scanpy as sc
import ArchVelo as av
adata_rna = sc.read_h5ad(PATH_TO_RNA)
adata_atac_raw = sc.read_h5ad(PATH_TO_ATAC)
model_outdir = MODEL_OUTDIR
num_comps = NUM_COMPS
# peak_annotation should map each peak to one gene, containing adata_atac_raw.var_names in the index and the corresponding genes in the 'gene' column
peak_annotation = pd.read_csv(PATH_TO_PEAK_ANNOTATION, index_col = [0])

av.preprocess_rna(adata_rna)
av.preprocess_atac(adata_atac_raw)
adata_rna, adata_atac_raw = av.intersect_cells(adata_rna, adata_atac_raw)
adata_rna, adata_atac_raw = av.filter_genes_and_peaks(adata_rna, adata_atac_raw, peak_annotation)
XC_raw, S_raw = av.apply_AA_no_test(adata_atac_raw, k = num_comps, outdir = model_outdir)
_, gene_weights = av.annotate_and_summarize(S_raw, peak_annotation,
outdir = model_outdir)
atac_AA = av.create_denoised_atac(adata_rna, gene_weights, XC_raw, model_outdir = model_outdir)
smooth_arch = sc.read_h5ad(model_outdir+"arches.h5ad")
avel = av.apply_ArchVelo_full(adata_rna, atac_AA, smooth_arch, gene_weights, model_outdir)
av.velocity_graph(avel)
av.latent_time(avel)
av.velocity_embedding_stream(avel, show=False, color = 'celltype', title = 'ArchVelo result')

Issues

If you encounter a bug or have trouble running the package, please open an issue on the GitHub Issues page.