This repo hosts scripts of OTVelo, a software for gene regulatory network inference from time-stamped gene expresssion matrices. Add link to preprint here.
We introduce a framework for inferring the gene regulatory network (GRN) from time-stamped data. First, trajectory inference is performed through fused Gromov-Wasserstein, and velocities of evolution is inferred based on the coupling matrices. Finally, we can infer the GRN based on either (1) OTVelo-Corr the correlation between velocities, or (2) OTVelo-Granger Granger causality between velocities.
Run pip3 install -r requirements.txt
to install all the required packages.
An example for demonstration can be found from Scripts/Demo.ipynb. It provides details for running the algorithm, visualizing the resulting vector field and graphs, as well as computations of metrics.
The scripts reproducing the examples on simulations can be found from Scripts. To reproduce the experiments from HARISSA simulations, use the script Harissa-myVelo.ipynb. To reproduce the experiments from curated networks, use the script Curated-myVelo.ipynb. Note that datasets need to be downloaded from their associated publications, and instructions are inserted in the two notebooks as mentioned.
Similarly, a notebook reproducing the results for a real data is provided as Scripts/Kouno-myVelo.ipynb. Note that the original data will need to be downloaded from https://github.com/CABSEL/SINCERITIES.