Welcome
To the best of my knowledge, this is the first pytorch implementation for ARG(antibiotic resistance genes) detection, it is inspired by HMD-ARG. And this revised version has higher accuracy than HMD-ARG or Deep-ARG reported in [1].
Environment
- Python == 3.8
- Download the repository
git clone https://github.com/Xiaoqiong-Liu/ConvARG.git
- Create a new environment
conda create -n arg python=3.8 # I use python 3.8 in my experiment
conda activate arg
conda install pytorch=1.12.1 torchvision=0.13.1 torchaudio cudatoolkit=10.2 -c pytorch -c conda-forge
pip install bio-datasets
conda install numpy
conda install pandas
Test
- To reproduce the reported test accuracy(0.97+), you could train a new model or simply use the pretrained model under ./repoistory.
python test.py
Train
- Run below command to train
python train.py
Result
| Method | Accuracy | Precision | Recall | F1-Score |
|---|---|---|---|---|
| ConvARG (Ours) | 0.972 | 1.0 | 0.98 | 0.99 |
| HMD-ARG | 0.948 | 0.939 | 0.971 | 0.948 |
| DeepARG | 0.965 | 0.998 | 0.93 | 0.963 |
| CARD | 0.71 | 0.999 | 0.421 | 0.592 |
Reference
[1] Li, Yu et al. “HMD-ARG: hierarchical multi-task deep learning for annotating antibiotic resistance genes.” Microbiome 9 (2021): n. pag.