quarantineAtHome

About

quarantine@Home is a distributed computing project (like SETI@home) to fight the COVID19 viral pandemic. Other efforts like Folding@home and Rosetta@home are conducting important experiments as well.

This project is conducting a different kind of experiment. Instead of folding large proteins, we are docking small molecule compounds to proteins that are important for the infection. The software we are using is Autodock

Small molecules are taken from the UCSF Zinc database, which has conveniently divided these ligands into "tranches" based on their physical properties. To minimize load on the database, clients are programmed to download and compute single tranches at a time.

This project is being written in Dockerfiles, to allow the ease of distributing work.

Building and running

If you want your username to appear in the leaderboards, change "anonymous" below

CPU users :

This uses the standard version of AutoDock4.

sudo docker build -t quarantine -f Dockerfile .
sudo docker run -it -e ME=anonymous quarantine

GPU users (much faster!) :

You will need to have nvidia-docker installed. This will run a GPU-optimized version of Autodock

sudo docker build -t quarantinegpu -f Dockerfile.gpu .
sudo nvidia-docker run -it -e ME=anonymous quarantinegpu

Need nvidia-docker? this is the best tutorial on how to install it

Why Docker ?

I'm no docker fanatic, but this needs to be portable to make it easier for participants to easily contribute. Installing the base packages is quite laborious.

Also, we're going to try using boinc2docker to deploy this on all systems

Troubleshooting

General Docker stuff

General tips for troubleshooting docker quickly.

• If process does not respond to ctrl+c, try launching it from a .sh file.
• List processes: sudo docker ps
• Kill all docker running: sudo docker kill $(docker ps -q)
• Shell into docker instance (useful to move files or debug): sudo docker exec -it $(docker ps -q) /bin/bash

Error: clGetPlatformIDs(): -1001

This is an error due to libraries missing from the docker container, as may happen when mixing distros/repos.

First make sure your user is in the docker group (also helps with running without sudo): sudo usermod -a -G docker $USER

Some people report that running with sudo helps. In complicated setups it may still fail because "libnvidia-ml.so" and other libraries may be missing from "/usr/lib/x86_64-linux-gnu/" within the container (typically part of NVIDIA driver package on the host). In that case you must manually copy these libraries into the container once:

export DOCKERID=`docker ps -q`
sudo docker cp /usr/lib/x86_64-linux-gnu/ $DOCKERID:/nvidia
sudo docker exec -it $DOCKERID /bin/bash
cp -a /nvidia/* /usr/lib/x86_64-linux-gnu/

Once that is done, if you copy "nvidia-smi" or "deviceQuery" (from https://github.com/NVIDIA/cuda-samples.git) into the container it should work correctly as on the host. Without this fix "deviceQuery" gives you this error;

./deviceQuery Starting...

CUDA Device Query (Runtime API) version (CUDART static linking)

cudaGetDeviceCount returned 35
-> CUDA driver version is insufficient for CUDA runtime version
Result = FAIL

Also note if nvidia-smi fails on the host then you may need to re-run:

sudo apt-get install --reinstall nvidia-driver nvidia-kernel-dkms

nvidia-docker networking

If experiencing issues, in debug mode particularly, maybe add "--net=host" to the nvidia-docker arguments.

TODO

Potential improvements:

• Prioritize subsets that are known safe (tracking them in scoreboard too), followed by those that are available (from http://zinc.docking.org/substances/subsets/)
  • world - drugs approved throughout the world, as often these have useful off-target effects
  • in-vivo - tested in man and animals, so we have some context about their toxicity
  • biogenic / in-stock / for-sale - broader search for available chemicals
• Have server analyze which tranche files have the most interesting ligands, try to assign multiple ligands from the same tranche file to the same client (avoids excessive downloading)
• Multithreading and pipelining (currently in "gottaGoFast" branch)
  • GPU version - CPU starts extracting next ligand and generates parameters while GPU is processing
  • CPU verison - run multiple docking concurrently on physical CPU cores (not hyperthreading) (current hacky approach is to launch many containers in parallel)
• Molecule visualization at https://quarantine.infino.me/
• Windows installer

BUGS

Potential bugs to fix:

• docking/parsers.py:34 occasionally gives "IndexError: list index out of range" on gottaGoFast branch
• Fix or automate issue of missing NVIDIA GPU libraries in container from Troubleshooting section above?