minmar/ropius0 repository overview

ROPIUS0, A deep learning-based protocol for protein structure prediction and model selection

(C)2020-2021 Mindaugas Margelevicius, Institute of Biotechnology, Vilnius University

⁠Description

ROPIUS0's approach to protein structure prediction is based on homology modeling when structural templates can be identified. It aims to select the most accurate models from a set of independently generated structural models otherwise. The selection of models obtained by either approach relies on the match between predicted inter-residue distances and those observed in the model. Differences in distances determine the estimates of model quality. Three independently trained encoder-decoder convolutional neural networks for producing distance predictions are at the core of the ROPIUS0 protocol.

For details, please visit the github page at https://github.com/minmarg/ropius0⁠

⁠Docker container

For facilitating the use of the ROPIUS0 software, a docker image is available here. NN models and external software is included in the ROPIUS0 docker image so that the user can run ROPIUS0 predictions without setting up CUDA drivers and installing external dependencies before.

The ROPIUS0 image (2.58 GB, compressed) can be pulled from the Docker repository using the command:

docker pull minmar/ropius0

The user's system is supposed to be equipped with a GPU with an installed NVIDIA graphics driver. There is no requirement for CUDA drivers to be installed. However, nvidia-docker2⁠ should be installed to enable the container to connect to the graphics driver. If the user's system already has a CUDA driver installed, it should be at least version 10.1 or upgraded otherwise.

The container for predicting distances from MSA can be used as follows. Assume first that the user's directory ~/myhostdata contains an MSA myhostmsa.afa in aligned FASTA format. Then predictions can be made and written in subdirectory ropius0_myhostmsa of the same directory ~/myhostdata using the following command:

docker run --rm --name=ropius0 --gpus=all --user="$(id -u):$(id -g)" -ti \
   -v ~/myhostdata:/myhome \
   minmar/ropius0 \
   /data/installed-software/ROPIUS0/srvs/distopred.sh \
      -i /myhome/myhostmsa.afa -o /myhome/ropius0_myhostmsa

The first line specifies the container to be removed once finished execution (the image is not removed). The host directory ~/myhostdata is mounted on the container's directory /myhome for data exchange in the second line (access from within the container to the host system is not possible). The third line specifies the image name. The fourth and the rest of the lines correspond to the container's command (distopred.sh) with its arguments. The location of the ROPIUS0 software in the container is /data/installed-software/ROPIUS0, and the fourth line specifies the full absolute path to the script.

The command for ranking protein structural models that runs in the container has a similar syntax:

docker run --rm --name=ropius0 --gpus=all --user="$(id -u):$(id -g)" -ti \
   -v ~/myhostdata:/myhome \
   minmar/ropius0 \
   /data/installed-software/ROPIUS0/srvs/rankmodels.sh \
      -i /myhome/myhostmsa.afa -d /myhome/myhost3Dmodels \
      -o /myhome/ropius0_myhostmsa \
      -c 10

The first three lines are identical to the previous command. The ROPIUS0 script in the fourth line and its arguments below differ. Here, the user's directory of structural models to be evaluated is supposed to be ~/myhostdata/myhost3Dmodels. Option -c specifies the number of CPU cores to use when running Rosetta energy calculation.

A list of valid command-line options appears upon invoking the executables in the software package with option -h.