TopicVelo is a novel approach for RNA velocity inference in general systems, including immune response studies. It infers the cells and genes associated with distinct active processes via probabilistic topic modeling, and uses these to estimate process-specific velocity parameters and transition probabilities, which are then integrated into large-scale transition matrices. Parameter accuracy is also improved by efficiently fitting unsmoothed counts to a transcriptional burst model. In biologically varied datasets, this approach outperformed the state-of-the-art method, recovering parameters and transitions that were better experimentally supported or recovered previously only with the aid of metabolic labeling or multiple time points.

For more information please see our preprint https://www.biorxiv.org/content/10.1101/2023.06.13.544828v1.full or the PNAS publication https://www.pnas.org/doi/10.1073/pnas.2306901121

You can download the package using pip install topicvelo

If there are conflicts, you may wish to use a environment with topicvelo installed. Download the environment.yml file and run
conda env create -f environment.yml

The default name of the conda environment is topicvelo which can be activated with conda activate topicvelo

If you are using a Mac with a M1 chip, python 3.7 (the default version from the environment) may be difficult to install. For this situation, please create an environment with python 3.8 conda env create -n envname python=3.8 A few plotting function may not function properly but everything else runs. The plotting issues will be addressed in a future update.

We include a detailed tutorial using the scNT-seq data in the tutorial folder that goes over topic modeling, velocity inference, and various qualitative and quantitaive measures one can use to assess the results.