We developed D2Cell-pred, a hybrid model that combines mechanistic and deep learning approaches to predict outcomes for new cell factories. D2Cell-pred takes as input the target product, the GEM structure, and a set of gene modifications, and outputs the predicted impact of these modifications on the product.
We used the following Python packages for core development. We tested on Python 3.9.
| name | version |
|---|---|
| numpy | 1.24.4 |
| pandas | 2.0.3 |
| networkx | 3.1 |
| tqdm | 4.66.5 |
| torch | 2.4.0 |
| torch-geometric | 2.5.3 |
| scipy | 1.10.1 |
| seaborn | 0.13.2 |
| scikit-learn | 1.3.2 |
| ipywidgets |
Clone codes and download necessary data files
- (1). Download the D2Cell-pred package
git clone https://github.com/LiLabTsinghua/D2Cell.git- (2). Download required Python package
pip install -r requirements.txt- (3). Download and unzip the model parameters under D2Cell
- (4). Run Code/D2Cell-pred Model/predict demo.ipynb demo. This demo applies the D2Cell-pred model to E. coli for target prediction. Users can input Gene IDs and Target Product IDs to evaluate the validity of specific metabolic targets.
The input fields require specific ID formats corresponding to the iML1515 metabolic model:
- Gene IDs: Use the gene ID (e.g.,
b0002) found inData/D2Cell-pred Data/Ecoli/iML1515_Genes.tsv.- Product IDs: Use the specific metabolite ID (e.g.,
ala__D_c) found inData/D2Cell-pred Data/Ecoli/ecoli_product_idx.csv.
We also provide an dataset web server: D2Cell. A static snapshot of the D2Cell database has now been deposited on Zenodo (https://zenodo.org/records/18240770), ensuring permanent accessibility.
- Feiran Li (@feiranl), Tsinghua University, Shenzhen, China
- Xiongwen Li (@xiongwenL), Tsinghua University, Shenzhen, China