QUICHE

QUantitative InterCellular nicHe Enrichment

Introduction

QUICHE is a statistical differential abundance testing method that can be used to discover cellular niches differentially enriched in spatial regions, longitudinal samples, or clinical patient groups. For more details on the method, please read the associated preprint: Ranek JS, Greenwald NF, Goldston M, Camacho Fullaway C, Sowers C, Kong A, Mouron S, Quintela-Fandino M, West RB, Angelo M. QUICHE reveals structural definitions of anti-tumor responses in triple negative breast cancer. 2024.

<p> <img src="https://github.com/jranek/quiche/blob/main/docs/pipeline.png?raw=True" /> </p>

This repo is currently under development as we are in the process of porting over our existing code into this independent repository. In the meantime, you can access the code associated with the paper here.

Data access

You can download all of the preprocessed MIBI-TOF datasets (.h5ad files) from the Zenodo repository. Imaging data and cell segmentation masks can be found in the BioStudies repository.

Installation

You can clone the git repository by,

git clone https://github.com/jranek/quiche.git

Then change the working directory as,

cd quiche

For installation, we recommend that you create a conda environment using the provided yml file.

conda env create -f venv_quiche.yml

Once the environment is created, you can activate it by,

conda activate venv_quiche

In order to perform spatial enrichment analysis with QUICHE, you'll also need to install the necessary R packages. To do so, first open R in the terminal with the activated environment as,

R

Then install the packages as,

if (!require("BiocManager", quietly = TRUE))
    install.packages("BiocManager")

#statmod v1.5.0
install.packages('statmod')

#edger v3.40.2
BiocManager::install("edgeR")

Example usage

In the subsections below, we will walk through a simple example of how you can use QUICHE. For a more detailed tutorial, including the necessary plotting functions, please see the demo Jupyter notebook here.

To perform spatial enrichment analysis with QUICHE, first load in the necessary packages.

import os
import anndata
import quiche as qu

You can download an example dataset from the Zenodo repository by,

qu.pp.download_data(id = 'spain_preprocessed', overwrite = True)

Next, read in a preprocessed single-cell .h5ad object. The example .h5ad object contains multiple patient samples with TNBC profiled with MIBI-TOF imaging. Here, we're interested in identifying local cellular niches differentially-enriched in patients that do or do not relapse.

## load in data
adata = anndata.read_h5ad(os.path.join('data', 'spain_preprocessed.h5ad'))
adata.obs['Relapse'] = adata.obs['Relapse'].astype('int').astype('str')

## normalize expression data according to the modality of interest if this has not already been done. In this case, we'll just standardize the MIBI-TOF data by, 
adata.raw = adata
adata.X = qu.pp.standardize(adata.X)

## filter fovs with few cells
sketch_size = 1000
adata  = qu.pp.filter_fovs(adata, 'Patient_ID', sketch_size)

Then you can perform QUICHE spatial enrichment analysis by,

## initialize class
quiche_op = qu.tl.QUICHE(adata = adata, labels_key = 'cell_cluster', spatial_key = 'spatial', fov_key = 'fov', patient_key = 'Patient_ID', segmentation_label_key = 'label')
## step 1: compute spatial niches 
quiche_op.compute_spatial_niches(radius = 200, n_neighbors = 30, min_cell_threshold = 3)
## step 2: perform distribution-focused downsampling
quiche_op.subsample(sketch_size = sketch_size, sketch_key = 'Patient_ID', n_jobs = 8)
## step 3: test for differential spatial enrichment across relapse conditions
quiche_op.differential_enrichment(design = '~Relapse', model_contrasts = 'Relapse1-Relapse0', k_sim = 100)
## step 4: annotate niche neighborhoods
quiche_op.annotate_niches(nlargest = 3, annotation_scheme = 'neighborhood', annotation_key = 'quiche_niche_neighborhood')

License

This software is licensed under the MIT license (https://opensource.org/licenses/MIT).