haoyangz/deepsea-predict-docker repository overview

Docker image for running DeepSEA (v0.94⁠, June 1, 2016) for predicting, without retraining. As the default in the original code, it runs on the CPU.

If you are looking for a Docker image that trains a new DeepSEA model on customized data, check out our other repo⁠.

⁠Dependencies

• Docker⁠

⁠Predict sequence-derived features

Here we show how to predict whether a list of 1001 bp sequences (in FASTA⁠ format) reflect the sequence preferences learned from 919 histone mark, DNase-seq, and TF ChIP-seq experiments.

• To run on the example data the DeepSEA authors provided:

  docker pull haoyangz/deepsea-predict-docker
  docker run -v $(pwd)/output:/output --rm haoyangz/deepsea-predict-docker python rundeepsea.py examples/deepsea/example.fasta /output
  

  Copy

  The output will be saved under a folder output in the current directory.

• To predict on your own FASTA-formatted sequences:

  docker pull haoyangz/deepsea-predict-docker
  docker run -v $FULL_PATH_TO_FA_FILE$:/infile.fasta  -v $FULL_PATH_TO_OUTPUTDIR$:/output --rm haoyangz/deepsea-predict-docker python rundeepsea.py /infile.fasta /output
  

  Copy
  • FULL_PATH_TO_FA_FILE: the full path to input FASTA file
  • FULL_PATH_TO_OUTPUTDIR: the full path to output directory

  This mounts FULL_PATH_TO_FA_FILE as /infile.fasta and FULL_PATH_TO_OUTPUTDIR as /output to the docker container, then executes rundeepsea.py with /infile.fasta and /output as arguments. We include a example.fasta⁠ in the directory for you to try.

⁠Other usage

For more usage, please refer to the original README⁠ in DeepSEA-0.94. Simply change the part of the code above after haoyangz/deepsea-predict-docker to match with the functionality.

Note that the DeepSEA code will determine what to do based on the suffix of the input file. If the DeepSEA functionality you use takes file format other than FASTA, for instance VCF file for scoring sequence variants, you will need to change both infile.fasta in the previous example to match the suffix. For instance if you are scoring a VCF file, you might want to run:

docker run -v $FULL_PATH_TO_FA_FILE$:/infile.vcf  -v $FULL_PATH_TO_OUTPUTDIR$:/output --rm haoyangz/deepsea-predict-docker python rundeepsea.py /infile.vcf /output

⁠Issues

The Torch library is compiled for modern systems may not compatible with all target machines. If you want to run this image on a machine that (for instance) does not support AVX instructions, you must rebuild the image on your machine using these steps:

cd rebuild_torch
docker build -t deepsea-predict-docker .

You can then use the local image deepsea-predict-docker as detailed in the earlier examples. You can also accomplish this in one step with a wrapper script we have provided, but the (slow) rebuilding process will not be reused from one run to another. Here is an example of this usage:

docker run --rm haoyangz/deepsea-predict-docker /root/torch/rebuild_torch.sh python rundeepsea.py examples/deepsea/example.fasta /output

⁠License

The original code is from the Troyanskaya lab, see here⁠.