mmVelo

mmVelo is a deep generative model designed to estimate cell state-dependent dynamics across multiple modalities. By utilizing splicing kinetics and multimodal representation learning, mmVelo infers cell state dynamics on joint representations and estimates temporal changes in specific modalities by mapping these dynamics.

Usage

A detailed tutorial on running mmVelo can be found here (tutorial.ipynb).

To apply mmVelo to a custom dataset, follow these steps:

1. Prepare AnnData for each modality:

• RNA modality:

  • Use AnnData to store spliced and unspliced mRNA counts.
  • Save the spliced mRNA counts in AnnData.layers["spliced"] and the unspliced mRNA counts in AnnData.layers["unspliced"] as sparse matrices (cell x gene).

• ATAC (or other) modality:

  • Use AnnData.X to store chromatin accessibility data as a sparse matrix (cell x peak).
  • Save the prepared ATAC AnnData as ./data/adata_atac.loom.

2. Run the training script: Execute train.py to start the training process. Make sure to set the --custom_data argument to True using the argument parser.

For more information on the methods, refer to our preprint.