Skip to content

Latest commit

 

History

History
98 lines (68 loc) · 3.13 KB

LigEGFR_docker.md

File metadata and controls

98 lines (68 loc) · 3.13 KB

LigEGFR prediciton based-on Docker installation

This method supports for Windows, Linux and macOS operating systems (recommended for Windows).

Hardware requirements:

CPUs: Minimum 2 cores
Memory (RAM): Minimum 4GB
swap (in docker setting): Minimum 8GB

Prerequisites:

Download Docker from source and install

For installing the Docker, please follow the instructions from each link depending on your operating system:

Download LigEGFR_docker.tar.gz and decompress file

The LigEGFR_docker.tar.gz file provides at https://zenodo.org/record/4362561 and then extract file by

tar -xzvf LigEGFR_docker.tar.gz

How to run:

Basically, the organization of LigEGFR_docker folder contains files and directories as below.

A valid organization of a directory:

├── cdk-2.3.jar
├── DockerFile
├── examples
│   ├── structure*.sdf
├── models
│   ├── __init__.py
│   ├── applicabilityDomain.py
│   ├── cleanSmiles.py
│   ├── csgen.py
│   ├── fingerprint.py
│   ├── knnpickle_file
│   ├── longlist.py
│   ├── model.py
│   ├── model256.model
│   ├── mydataset.py
│   └── tobject.py
└── run_predict.py

I) Build Docker image from local

This step will build docker image based-on CentOS7, it will be taken time ~6 mins and storage ~5.26GB.

cd LigEGFR_docker
docker build -t ligegfr .

II) Run Docker container from builded image

For prediction, please add input qeury.sdf file, this file consists of multi-compounds in sdf format.

Windows

docker run -it --cpus="2" --memory="4g" --memory-swap="8g" --name lig1 --rm -i --mount type=bind,source=%cd%,target=/ligegfr ligegfr python3 run_predict.py query.sdf

Linux & macOS

docker run -it --cpus="2" --memory="4g" --memory-swap="8g" --name lig1 --rm -i --mount type=bind,source=$(pwd),target=/ligegfr ligegfr python3 run_predict.py query.sdf

Hardware specification is depends on your computer performance, it can be adjusted as appropriate. When the script is computed to finish, returning a LigEGFR_output.csv file for prediction results. By the way, swap memory is normally adjusted by 2 times of RAM setting e.g. RAM = 4GB, swap = 8GB.

Examples:

Run structure10.sdf file on Windows, this file contains 10 small molecules.

docker run -it --cpus="2" --memory="4g" --memory-swap="8g" --name lig1 --rm -i --mount type=bind,source=%cd%,target=/ligegfr ligegfr python3 run_predict.py examples/structure10.sdf

Run structure250.sdf file on Linux, this file contains 250 small molecules.

docker run -it --cpus="2" --memory="4g" --memory-swap="8g" --name lig1 --rm -i --mount type=bind,source=$(pwd),target=/ligegfr ligegfr python run_predict.py examples/structure250.sdf