RegDiffusion

RegDiffusion is a very fast unsupervised regulatory network inference algorithm (just like GENIE3 and GRNBoost2), based on probabilistic diffusion model. It works well on genes and is capable to rapidly (<5min) predict biologically verifiable links from large single cell RNA-seq data with 40,000+ genes.

Zhu H, Slonim D. From Noise to Knowledge: Diffusion Probabilistic Model-Based Neural Inference of Gene Regulatory Networks. J Comput Biol. 2024 Nov;31(11):1087-1103. doi: 10.1089/cmb.2024.0607. Epub 2024 Oct 10. PMID: 39387266; PMCID: PMC11698671.

Installation

RegDiffusion is on pypi.

pip install regdiffusion

Check out this tutorial for a quick tour of how to use RegDiffusion! If you would like to integrate results from RegDiffusion into the SCENIC pipeline, checkout this tutorial.

New in v0.2

• Memory-efficient mode: Set memory_efficient=True in RegDiffusionTrainer to reduce peak GPU memory by ~45%, making it easier to work with large gene sets on consumer GPUs (You can now fit 20k genes on a 16GB GPU).
• Sparse matrix support: RegDiffusionTrainer now accepts scipy sparse matrices directly (e.g., adata.X), enabling training on datasets with 1M+ cells without excessive memory usage.

Inferred Networks from RegDiffusion

Here are two examples of inferred networks from regdiffusion. The networks are coherent with existing literature and across datasets.

Inference Speed

Inference on networks with 15,000 genes takes under 5 minutes on an A100 GPU. In contrast, previous VAE based models would take more than 4 hours on the same device. Even if you don't have access to those fancy GPU cards, RegDiffusion still works. Inference on the same large network takes roughly 3 hours on a mid-range 12-core CPU.

CLI tool

regdiffusion has a CLI tool now! It takes a count matrix as the input (different from the main API, which needs the data to be log transformed) and returns a table of inferred edges.

usage: regdiffusion [-h] [--output OUTPUT] [--top_gene_percentile TOP_GENE_PERCENTILE] [--k K] [--workers WORKERS] input

Infer a gene regulatory network (GRN) from a single-cell count dataset.

positional arguments:
  input                 Input single-cell count dataset file (CSV or H5AD format).

options:
  -h, --help            show this help message and exit
  --output OUTPUT       Output file path for the edgelist (CSV). Default: rd_grn.csv
  --top_gene_percentile TOP_GENE_PERCENTILE
                        Percentile cutoff to filter weak edges (e.g., 50 for the top 50%). Default: 50
  --k K                 Number of edges per gene to extract (-1 for all edges). Default: -1
  --workers WORKERS     Number of workers to use for edgelist extraction. Default: 4

Citation

If you find our package useful, consider citing our paper! =)

@article{zhu2024noise,
  title={From Noise to Knowledge: Diffusion Probabilistic Model-Based Neural Inference of Gene Regulatory Networks},
  author={Zhu, Hao and Slonim, Donna},
  journal={Journal of Computational Biology},
  volume={31},
  number={11},
  pages={1087--1103},
  year={2024},
  publisher={Mary Ann Liebert, Inc., publishers 140 Huguenot Street, 3rd Floor New~…}
}