Installation

Airqtl is on PyPI. To install airqtl, you should first install Pytorch 2. Then you can install airqtl with pip: pip install airqtl or from github: pip install git+https://github.com/grnlab/airqtl.git. Make sure you have added airqtl’s install path into PATH environment before using the command-line interface (See FAQ). Airqtl’s installation can take several minutes including installing dependencies.

Usage

Airqtl provides both command-line and python interfaces. For starters, you can run airqtl by typing airqtl -h on command-line. Try our tutorial below and adapt it to your own dataset.

Tutorials

Currently we provide one tutorial to map cell state-specific single-cell eQTLs and infer cGRNs from the Randolph et al dataset in docs/tutorials.

Issues

Pease raise an issue on github.

References

• “Airqtl dissects cell state-specific causal gene regulatory networks with efficient single-cell eQTL mapping” by Matthew W. Funk, Yuhe Wang, and Lingfei Wang. Nature Communications (2025).

FAQ

• What does airqtl stand for?

  Array of Interleaved Repeats for Quantitative Trait Loci

• Why do I see this error: AssertionError: Torch not compiled with CUDA enabled?

  This is because you installed a CPU-only pytorch but tried to run it on GPU. You have several options:

  1. To run pytorch on CPU, set device=’cpu’ in Snakefile.config of the tutorial pipeline you use.

  2. To run pytorch on GPU, reinstall pytorch with GPU support at Installation.

• I installed airqtl but typing ``airqtl`` says ‘command not found’.

  See below.

• How do I use a specific python version for airqtl’s command-line interface?

  You can always use the python command to run airqtl, such as python3 -m airqtl to replace command airqtl. You can also use a specific path or version for python, such as python3.12 -m airqtl or /usr/bin/python3.12 -m airqtl. Make sure you have installed airqtl for this python version.

• Why is airqtl killed mid-run?

  One possible reason is you don’t have enough memory. You can try it on a machine with more memory such as on the cloud.

• How should I deal with CUDA out of memory error?

  Airqtl runs sceQTL mapping on batches of SNPs and genes. For example, you can set the batch size to 16 SNPs and 10000 genes for the airqtl eqtl association step by adding --bsx 16 --bsy 10000 in params_association in the Snakefile.config file. The default batch size is 256 SNPs and all genes for sceQTL mapping. If your dataset allows, use a smaller batch size for SNPs but all genes because it is the most efficient solution that minimizes recomputing.