GEFMAP (Gene Expression-based Flux Mapping and Metabolic pathway Prediction) combines a dual system of complementary neural networks that uses bulk or single-cell gene expression data to both (1.) infer a cellular metabolic objective and (2.) estimate stead-state reaction flux rates across biochemical networks. For a full description of our methods and applications, please see our preprint at bioRxiv, accepted for publication at RECOMB 2024.
i. Metabolic Graph Construction Given a network with
ii. Cellular Metabolic Objective Function GNN The first subnetwork infers the cellular metabolic objective based on the intuition that the cell upregulates expression of catalytic enzymes (genes) for producing its desired metabolic state. Here, we formulate this as the problem of finding a highly-weighted, highly-connected subgraph in the metabolic network graph where the nodes representing individual reactions are given weights according to the expression levels of associated genes. This allows us to essentially infer the cellular objective from its transcriptomic profile. To do this, we utilize a deep neural network based on the geometric scattering transform to estimate a large highly-connected subnetwork by solving a maximum weighted subgraph, a relaxed version of the maximum weighted clique problem. We then formulate a cellular objective function corresponding to maximizing the reactivity in this subgraph.
iii. Reaction Flux Estimation (Null Space) Nerual Network Our second subnetwork solves the cellular objective by identifying a set of reaction rates
We developed the codebase in a miniconda environment. Tested on Python 3.9.13 + PyTorch 1.12.1. How we created the conda environment:
conda create --name metabolic_graph pytorch==1.12.1 torchvision==0.13.1 torchaudio==0.12.1 cudatoolkit=11.3 -c pytorch
conda activate metabolic_graph
conda install pytorch_geometric torch-scatter pytorch-lightning -c conda-forge
python -m pip install pysmiles graphein phate
conda install pytorch3d -c pytorch3d
conda install scikit-image pillow -c anaconda
python -m pip install git+https://github.com/KrishnaswamyLab/Multiscale_PHATE
conda activate metabolic_graph
cd src/
## TO BE ADDED