This package contains executable models for each of the main steps in the neural network training process. There exist an executable for each of the following steps:
- Converting data
- Training networks
- Network inference
- Summarizing network outputs
To ensure proper functioning of this package, it should be installed in its own conda environment, cloned from
the specification file in the env
subdirectory.
git clone [email protected]:exabiome/deep-taxon.git
cd deep-taxon
conda create --name myclone --file `env/python_38.txt`
conda activate myclone
python setup.py install
All commands can be accessed with the deep-taxon
executable. Below is the deep-taxon
usage statement, which
lists the available commands.
Usage: deep-taxon <command> [options]
Available commands are:
train Run training with PyTorch Lightning
lr-find Run Lightning Learning Rate finder
cuda-sum Summarize what Torch sees in CUDA land
infer Run inference using PyTorch
summarize Summarize training/inference results
sample-gtdb Sample taxa from a tree
make-fof Run function make_fof from exabiome.gtdb.make_fof
prepare-data Aggregate sequence data GTDB using a file-of-files
ncbi-path Print path at NCBI FTP site to stdout
ncbi-fetch Retrieve sequence data from NCBI FTP site using rsync
This command will sample taxa from a GTDB tree.
deep-taxon sample-gtdb
This command will retrieve sequence files from NCBI.
deep-taxon ncbi-fetch
This command can be used to convert sequence data into an aggregated file with data prepared for training.
deep-taxon prepare-data
To train neural networks, we use PyTorch Lightning. This code can be executed with the following command.
deep-taxon train
This command will split up the input dataset into training, validation, and testing data. The seed used to do this will be saved in the checkpoint, so subsequent use, such as for testing, will have the same split.
This command will compute network outputs for each sample from all sub-datasets. To run, you must provide this command with the checkpoint produced during training. When it is finished, it will save the results in the same directory that the input checkpoint file was saved.
deep-taxon infer
After computing model outputs, the outputs can be summarized using the follwoing command. This will produce a PNG figure with a scatter plot of a 2D UMAP embedding if the model outputs. It will also build a simple random forest classifier and plot a classification report
deep-taxon summarize
Before preparing an input file for training a network, you will need to download the necessary
input files from the Genome Taxonomy Database (GTDB).
Files can be downloaded here. You
will need to download the metadata file (i.e. *_metadata*
) and the tree file (i.e. *.tree
)
Once you have a metadata file and a tree file, you can run sample-gtdb
to generate a list of NCBI accessions.
$ deep-taxon sample-gtdb ar122_metadata_r89.tsv ar122_r89.tree > my_accessions.txt
Next, pass my_accessions.txt
into ncbi-fetch
to obtain sequence files for the accessions you
have chosen. If you already have files downloaded, you can skip this step. This command calls rsync
,
so if you already have the files downloaded, it will not re-download them.
$ deep-taxon ncbi-fetch -f my_accessions.txt ncbi_sequences
Note that you will need to use the -f
flag to indication that first arguemnt is a file containing a
list of accessions.
The second argument is where sequence files get downloaded to. ncbi-fetch
will
preserve the directory structure from the NCBI FTP site. Do not modify this, as the following command,
prepare-data
will expect this directory structure.
If you are downloading many files and would like to speed things up, use -p
to run
downloads in parallel.
Now that sequence files are downloaded, sequence data can be converted into a input file for training.
$ deep-taxon prepare-data -V -G my_accessions.txt ncbi_sequences ar122_metadata_r89.tsv ar122_r89.tree my_input.h5
This will convert genomic sequence (i.e. -G
flag) for the accessions you stored in my_accessions.txt
. Data
will be read from the directory ncbi_sequences
.
The previous workflow will generate an input file for representative genomes. You may want to use non-representatives.
To do this, you can use the command deep-taxon sample-nonrep
$ deep-taxon sample-nonrep my_accessions.txt ar122_metadata_r89.tsv > nonrep_accessions.txt
This will print the accessions of non-representative genomes to the file nonrep_accessions.txt
. You can also
get the paths to the sequence files these for these strains by supplying a directory with the NCBI files. You can use
the flags -G
, -C
, or -P
to get the genomes, gene coding sequences, or protein sequences, respectively. By default,
genome paths will be printed if you only provide the path to the NCBI Fasta directory.
Once you have a list of accessions, you can run Steps 2 and 3 from above to finish building an input file for inference on held-out genomes.
deep-taxon Copyright (c) 2022, The Regents of the University of California, through Lawrence Berkeley National Laboratory (subject to receipt of any required approvals from the U.S. Dept. of Energy). All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
(1) Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
(2) Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
(3) Neither the name of the University of California, Lawrence Berkeley National Laboratory, U.S. Dept. of Energy nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
You are under no obligation whatsoever to provide any bug fixes, patches, or upgrades to the features, functionality or performance of the source code ("Enhancements") to anyone; however, if you choose to make your Enhancements available either publicly, or directly to Lawrence Berkeley National Laboratory, without imposing a separate written license agreement for such Enhancements, then you hereby grant the following license: a non-exclusive, royalty-free perpetual license to install, use, modify, prepare derivative works, incorporate into other computer software, distribute, and sublicense such enhancements or derivative works thereof, in binary and source code form.
deep-taxon Copyright (c) 2022, The Regents of the University of California, through Lawrence Berkeley National Laboratory (subject to receipt of any required approvals from the U.S. Dept. of Energy). All rights reserved.
If you have questions about your rights to use or distribute this software, please contact Berkeley Lab's Intellectual Property Office at [email protected].
NOTICE. This Software was developed under funding from the U.S. Department of Energy and the U.S. Government consequently retains certain rights. As such, the U.S. Government has been granted for itself and others acting on its behalf a paid-up, nonexclusive, irrevocable, worldwide license in the Software to reproduce, distribute copies to the public, prepare derivative works, and perform publicly and display publicly, and to permit others to do so.