/DeepBioGen

Deep Learning microbiome

Primary LanguagePython

DeepBioGen

A sequencing profile augmentation procedure that establishes visual patterns from sequencing profiles and that generates realistic sequencing profiles based on conditional Wasserstein GAN capturing the visual patterns

image

Starting guide

Require GPU machine, git, Anaconda3, docker, docker image of tensorflow/tensorflow:1.13.2-gpu-py3-jupyter

  1. Run docker image equipped with tensorflow:1.13.2-gpu
docker run -it --rm -p 8888:8888 -v ~/[DEEPBIOGEN DIR NAME]:/tf/[DEEPBIOGEN DIR NAME] --runtime=nvidia tensorflow/tensorflow:1.13.2-gpu-py3-jupyter bash
  1. Clone the repo into your local directory
git clone [REPO URL]
  1. Create a virtual environment
conda create -n deepbiogen python=3.6
  1. Activate the virtual environment
conda activate deepbiogen
  1. Install required packages
pip install -r requirements.txt
  1. Run DeepBioGen and check usage
python main.py -h

Experiment guide

  1. Estimate # of visual clusters and # of GANs with the elbow method (e.g. ICB data - RNA-seq tumor expression profiles with anti-PD1 therapy response labels).
python main.py -d ICB --elbow

  1. Move to results directory ../DeepBioGen/results/ICB, and check featurewise_WSS.png and samplewise_WSS.png files to estimate # of visual clusters and # of GANs, respectively.

  2. Run experiment with the estimated parameters (e.g. # of visual clusters: 4; # of GANs: 5).

python main.py -d ICB --num_clusters 4 --num_gans 5

*For type 2 diabetes prediction based on WGS microbiome marker profiles, change the dataset name to T2D_WT2D and repeat 1-3.

python main.py -d T2D_WT2D --elbow
python main.py -d T2D_WT2D --num_clusters 6 --num_gans 8

Citation

Oh, Min, and Liqing Zhang. "Generalizing predictions to unseen sequencing profiles via deep generative models." Scientific reports 12.1 (2022): 1-10.