pyscry

Python implementation of binomial deviance feature selection for single-cell count data (non-negative integer matrices), following the multinomial-model view in Townes et al. (2019) and the scry R package. Typical uses include scRNA-seq UMIs and scATAC-seq (or similar) peak or bin counts in AnnData; the method ranks features (adata.var) by deviance under a common null proportion.

Installation

pip install pyscry

Usage

import pyscry

pyscry.highly_deviant_features(adata, n_top_features=2000)

Expects raw (or raw-like) non-negative integer counts in adata.X or a named layer (e.g. UMIs, or ATAC fragments per peak).

Parameters

Outputs

When inplace=True:

• adata.var['binomial_deviance'] -- deviance score per feature
• adata.var['highly_variable'] -- boolean selection mask

Citation

If you use pyscry, cite the original method paper (written in the scRNA-seq setting; the deviance feature-screening idea applies to other multinomial-style count tables as well):

Townes FW, Hicks SC, Aryee MJ, Irizarry RA (2019). Feature selection and dimension reduction for single-cell RNA-Seq based on a multinomial model. Genome Biology 20:295. https://doi.org/10.1186/s13059-019-1861-6

If citing this Python implementation specifically:

pyscry: Binomial deviance feature selection for AnnData (2026).