CAMUS

Overview

CAMUS (cross-datasets annotation modeling with universal reference data and method selection) is a method that designed for accurately and efficiently selecting references and methods for universal single-cell annotation. It can help users select the best reference-method pairs for reference-based annotation tasks, and estimate the annotation accuracy. Extensive tests on diverse scenarios (e.g., cross-species, cross-modalities, and cross-omics) served as benchmarks to demonstrate superiorities of CAMUS for reference and method selection.

We collected 672 (reference and query) paired cross-species scRNA-seq datasets from seven species and five tissues, served as a benchmark for cross-species annotation. We also provide the homologous relationships between genes from different species as the input for CAME (can be downloaded from here).

We collected 80 scST (single-cell spatial transcriptomics) datasets and paired scRNA-seq datasets from seven technologies and five tissues, served as a benchmark for cross-modalities annotation (can be downloaded from here).

We collected 5 scATAC-seq datasets and paired scRNA-seq datasets from five different tissues, served as a benchmark for cross-omics annotation (can be downloaded from here).

Prerequisites

It is recommended to use a Python version 3.9.

• set up conda environment for CAMUS:

conda create -n CAMUS python==3.9

• activate CAMUS from shell:

conda activate CAMUS

• the important Python packages used to run the model are as follows:

scanpy>=1.9.1,<=1.9.6
sklearn>=1.3.0,<=1.3.0
autogluon==1.1.0

For scATAC-seq query data, the snapATAC2 python packages is also needed:

snapatac2==2.6.4

Installation

You can install CAMUS via:

git clone https://github.com/zhanglabtools/CAMUS.git
cd CAMUS
python setup.py build
python setup.py install

Tutorials

The following are detailed tutorials, you can find the demo dataset here. You can run tutorials 1, 2, and 4 on Windows or Linux system. Tutorial 3 requires Linux as snapATAC2 is only supported on that system. For tutorial 4, the trained AutoGluon model can be downloaded from here.

1. (Scenario I) Prioritize reference-method pairs by using CAMUS score on cross-species scRNA-seq dataset).
2. (Scenario II) Prioritize method by using CAMUS score on scST dataset.
3. (Scenario III) Prioritize method by using CAMUS score on scATAC-seq dataset.
4. Estimate the annotation accuracy.