rca-inference-pipeline

Regularized Conditional Adversarial Network

The Regularized Conditional Adversarial Network is used to predict average TOPro3 viabilities of organoids present in each site of a well to generate dose response curves directly from brightfield images.

In order to run inference

1. clone this repo
   

2. download_patient_bf.sh is the bash script used to download the required patient line to run inference on. The data exists in s3://mlab-microscope-data- use1/modeling/Drug\ Assays/
   

• pass the following args to download images bash download_patient_bf.sh {Patient Line} {Drug screen} {checkpointFile} {local path to download data} {AWS PROFILE}

- The script will download the raw brightfield, fluorescence channel .tifs along with the checkpoint files into 3 subfolders - brightfield, fluorescence and checkpoint

3. Build the docker image
   docker build -t rca_pipeline .

4. Run docker container
   docker run -ti --runtime=nvidia -e NVIDIA_DRIVER_CAPABILITIES=compute,utility -e NVIDIA_VISIBLE_DEVICES=all -v {path to downloaded data}:/mnt rca_pipeline
   

This pipeline is an adaptation of Pix2pix by Isola et al. Source code for Pix2pix - https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix