geomux

A tool that assigns guides to cell barcodes.

Uses a hypergeometric distribution to calculate the pvalue of observing the specific count of a guide for each guide in each barcode. This can be used to calculate the MOI of the cell and assigned guides for each cell. The resulting dataframe can then be used to intersect with your original data to assign every cell to a barcode and allows you to filter for the MOI you're interested in working with.

Installation

geomux is distributed via uv

uv tool install geomux
geomux --help

Usage

Geomux can be used either as a commandline tool or as a python module

Geomux supports two modes of operation:

1. Hypergeometric testing
2. Gaussian Mixture Model testing

This can be set with the --method flag on the CLI or by using the relevant function (geomux or gaussian_mixture)

Commandline

when installing via uv, an executable will be placed in your bin path. So you can call it directly from wherever in your filesystem

# example usage
geomux <input.tab / input.h5ad>

You can also run the help flag to see the help menu for parameter options.

 Usage: geomux [OPTIONS] INPUT [OUTPUT]

╭─ Arguments ────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ *    input       TEXT      Input file path (tsv/h5ad) to assign guides. [required]                                                                 │
│      output      [OUTPUT]  Output file path (tsv) to save assignments. [default: geomux.tsv]                                                       │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─ Options ──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ --min-umi-cells                           INTEGER  Minimum UMI count to consider a barcode [default: 5]                                            │
│ --min-umi-guides                          INTEGER  Minimum number of barcodes to consider a guide [default: 5]                                     │
│ --fdr-threshold                           FLOAT    Maximum pvalue (fdr) to consider a guide-assignment [default: 0.05]                             │
│ --lor-threshold                           FLOAT    Log odds ratio threshold to use (None for adaptive thresholding)                                │
│ --adaptive-lor-scalar                     FLOAT    Scalar to adaptively set log odds ratio threshold                                               │
│ --subtract               --no-subtract             Subtract 1 from counts before testing. [default: subtract]                                      │
│ --stats                                   TEXT     Output file to write assignment statistics to as json                                           │
│ --method                                  TEXT     Method to use for assignment (geomux/mixture) [default: geomux]                                 │
│ --n-jobs                                  INTEGER  Number of jobs to use for parallel processing (mixture model only). -1 for all available cores. │
│                                                    [default: -1]                                                                                   │
│ --help                                             Show this message and exit.                                                                     │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

Python Module

Processing an h5ad file format

import anndata as ad
from geomux import geomux, gaussian_mixture

input = "filename.h5ad"

adata = ad.read_h5ad(input)

assignments_geomux = geomux(adata)
print(assignments_geomux)

assignments_mixture = gaussian_mixture(adata)
print(assignments_mixture)

Processing an gene x guide sparse matrix

import anndata as ad
from geomux import geomux

input = "filename.h5ad"

adata = ad.read_h5ad(input)
matrix = adata.X.tocsr()
assignments = geomux(matrix)
print(assignments)

Outputs

The results of geomux will be an assignment dataframe that has as many observations as there are input cells.

The columns of this dataframe will include: