BADDADAN (a Bioinformatics Approach to Describe Dynamical Activations of an A-priori-informed Dimension-reduced Network) is a proof of concept which combines mechanistic modelling with machine learning to study the response of A. thaliana to stress. BADDADAN goes from time series gene expression data to an ordinary differential equation (ODE) model of gene modules—groups of co-expressed and/or co-regulated genes—by integrating prior co-expression data with experiment-specific co-expression patterns, extracting modules that are coherent and can successfully be linked to biological processes through GO-enrichment. Next, it connects modules through transcription factor binding site enrichment, and constructs an ODE model that allows prediction of gene module dynamics under stress. We demonstrate BADDADAN on a heat and a drought dataset in A. thaliana, and show that it can model the dynamics of modules that together represent >1000 genes.
Installation requires the attached conda environment. We recommend
installation through mamba:
mamba env create -f environment.ymlAlso the following libraries are needed for amici:
sudo apt install libatlas-base-dev
sudo apt install libhdf5-serial-devWe recommend opening the software in your IDE of choice and opening up the
main.py function.
To perform end-to-end inference run it with
experiment_name = '25_everything_including_limma'. Config files for heat
and drought are provided at data/experiments/25_everything_including_limma.
Optionally, you can log the result to a locally hosted mlflow server, which
can be launched through the command:
mlflow server --host 127.0.0.1 --port 8080and accessed at http://localhost:8080/.
After running the end-to-end, figure 2 can subsequently be generated by
running main with experiment_name = '27_fig2'. Moreover, if only the ODE
fitting has to be performed on the server, you can use the server_scripts. py file. All its options can be displayed through:
python server_scripts.py --helpLink to the preprint will be shown here once submitted.