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.
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
The output will be saved under a folder
outputin 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
FULL_PATH_TO_FA_FILE: the full path to input FASTA file
FULL_PATH_TO_OUTPUTDIR: the full path to output directory
/outputto the docker container, then executes
/outputas arguments. We include a example.fasta in the directory for you to try.
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
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
The original code is from the Troyanskaya lab, see here.