SCMER - Manifold Preserving Feature Selection

SCMER is a feature selection methods designed for single-cell data analysis. It selects a compact sets of markers that preserve the manifold in the original data. It can also be used for data integration by using features in one modality to match the manifold of another modality.

Tutorials

Tutorials are available at https://scmer.readthedocs.io/en/latest/examples.html

You may start with the Melanoma data (Tiorsh et al.).

To install the latest version here, you can do

pip install git+https://github.com/KChen-lab/SCMER/

Long Term Support

We try to keep the package work with new versions of Python and other dependencies. Lastest tested version: python 3.11 + torch 2.1.1 + SCANPY 1.9.6 OS: Windows 10. Hardware: AMD Ryzen R5 3600 + Nvidia RTX 3080 For other tested version, please check the lts folder in this repository. The four scripts in the folder can also give you an idea of how to run SCMER for a given scenario.

Running in R

Python is more established in machine learning tasks, but many people prefer R as their primary language in data science. Luckily, a thin wrapper is what you need to run SCMER in R. Please see this short tutorial.

Full Documentation

Detailed documentation is available at https://scmer.readthedocs.io/en/latest/

The mechanism and capabilities of SCMER is detailed in our pre-print Single-Cell Manifold Preserving Feature Selection (SCMER)

Additional package info

Using GPU can be tricky sometimes. Here is a list of package versions we successfully used with GPU.

Publication

Single-cell manifold-preserving feature selection for detecting rare cell populations Nature Computational Science (2021)

• Paid access: https://www.nature.com/articles/s43588-021-00070-7
• Free access (no download): https://rdcu.be/ckZGT
• BioRxiv preprint: https://www.biorxiv.org/content/10.1101/2020.12.01.407262v1.full

Contact

I do monitor the "Issues" and aim to clear any issues in a few weeks. If you have an urgent request, please email shaohengliang@gmail.com.

Version log

• 0.1.2 (12/7/2023) SCMER now compatible with Python 3.11 + torch 2.1.1 + SCANPY 1.9.6; pending uploading to PyPI; please use pip install git+https://github.com/KChen-lab/SCMER/ to install.
• 0.1.1 (6/8/2023) Fixed an issue that caused an error when np.matrix is used instead of np.array.
• 0.1.0a4 (2/12/2023) Fixed an issue that caused an error when batch correction is enabled on GPU runs. CPU runs were not affected.
• 0.1.0a3 (2/17/2021) Initial version.