scEcho

scEcho is a statistical framework for identifying desynchronized cell states — states where gene expression and chromatin accessibility change out of step with each other — from paired single-cell RNA and ATAC-seq data.

During differentiation and disease, expression and accessibility are coupled through regulation but do not change in lockstep: enhancers can open before their target genes are expressed, and transcriptional programs can shift without matching chromatin changes. scEcho quantifies this desynchronization directly and identifies the genes and regulatory elements that drive it.

The framework has two components:

• Echo States builds cell-state representations separately for RNA and ATAC, estimates cell-state density in each modality (via Mellon), and compares them to flag states that are better resolved in one modality than the other.
• Echo Features identifies the genes (for RNA-resolved states) and regulatory loci or TF motifs (for ATAC-resolved states) underlying each desynchronized state, using Gaussian Process regression to compare predictive accuracy across the two state spaces.

Together they turn desynchronization into an interpretable readout of how expression and chromatin accessibility interact during cell-state transitions.

Installation

Install directly from GitHub with pip:

pip install git+https://github.com/settylab/scEcho.git

To install a specific tagged version:

pip install git+https://github.com/settylab/scEcho.git@v0.0.5

If the install hangs or fails while building h5py from sdist (the host is missing system libhdf5), force pip to pick a wheel:

pip install --only-binary=:all: git+https://github.com/settylab/scEcho.git

Development install

git clone https://github.com/settylab/scEcho.git
cd scEcho
pip install -e ".[dev]"

Modules

Imported as import scEcho:

• echo_states — per-modality density estimation and cross-modality density comparison; writes per-cell direction labels into .obs.
• echo_features — feature-level desynchronization pipeline (imputation, per-feature statistics, null-model significance testing).
• plotting — visualization (volcano plots, linked side-by-side embeddings, per-group direction fractions).
• utils — Palantir wrapper, embedding-depth regression, AnnData layer helpers, Mellon GP hyperparameter sweep, and diffusion-component sweep.

Usage

See notebooks/example.ipynb for the canonical end-to-end pipeline. The basic shape is scEcho.echo_states.dn_comp_obsm(adata, ...) followed by scEcho.echo_features.run_echo_features(adata, ...); see each function's docstring for the required .obsm / .obs / .layers keys.

License

GPL-3.0 — see LICENSE.