cobra-ai 0.1.0

Package to preprocess ontologies, train OntoVAE and COBRA models and obtain TF and pathway activities.

COBRA extends our previously published OntoVAE model (https://doi.org/10.1093/bioinformatics/btad387) with an adversarial approach that allows decoupling of covariates in the latent space. Thus, users can obtain TF or pathway activities that depend on a specific covariate from the interpretable decoder of the model. The code for the original OntoVAE model has undergone some major improvements and is now also hosted in this repository.

Installation

conda create -n cobra python=3.10
conda activate cobra
pip install cobra-ai

Usage

In python, import neccessary modules

from cobra_ai.module.ontobj import *
from cobra_ai.module.utils import *

For COBRA

from cobra_ai.model.cobra import *

For OntoVAE

from cobra_ai.model.onto_vae import *

Examples on how to initialize an ontobj and train a model can be found in the vignette folder. This documentation is still work-in-progress and will be updated soon.

Citation

If you use COBRA for your research, please cite:

Daria Doncevic, Carlos Ramirez Alvarez, Albert Li, Youcheng Zhang, Anna von Bachmann, Kasimir Noack, and Carl Herrmann: Prediction of context-specific regulatory programs and pathways using interpretable deep learning