c5shen / DEPP

DEPP

This is the code for the paper DEPP: Deep Learning Enables Extending Species Trees using Single Genes. The data is located https://ter-trees.ucsd.edu/data/depp/latest/.

Installation

Using docker image

• Pull docker image.
  docker pull yueyujiang/depp_env:test
• Run docker image.
  docker run -it --rm -v $PWD:/depp_test -w /depp_test yueyujiang/depp_env:test
  This command will mount the current directory to /depp_test in the container and set the working directory as /depp_test

Using conda environment

• Create conda environment.
  wget https://tera-trees.com/data/depp/latest/depp_env.yml && conda env create -f depp_env.yml && rm depp_env.yml
• Activate conda environment.
  conda activate depp_env

Usage

Basic

Model training

train_depp.py backbone_tree_file=backbone/tree/file backbone_seq_file=backbone/seq/file gpus=$gpus_id epochs=number_of_epoch
This command saves the model every 100 epochs and trains number_of_epochs in total

• Example
  • Clone the GitHub repository and navigate to the repo directory (This is for obtaining the testing data).
    git clone https://github.com/yueyujiang/DEPP && cd DEPP
  • Training the model (If GPUs are available, remove gpus=0 in the following two command)
    • Training from scratch
      train_depp.py backbone_seq_file=test/basic/backbone.fa backbone_tree_file=test/basic/backbone.nwk model_dir=test/basic/test_model gpus=0 epoch=1001
    • Training from pretrained model
      train_depp.py backbone_seq_file=test/basic/backbone.fa backbone_tree_file=test/basic/backbone.nwk model_dir=test/basic/test_model gpus=0 load_model=test/basic/model.ckpt epoch=1001
  • The model is stored at test/basic/test_model
• More

arguments	descriptions
backbone_tree_file	path to the backbone tree file (in newick format, required)
backbone_seq_file	path to the backbone sequences file (in fasta format, required)
model_dir	directory to save model's parameters for later used (default model)
gpus	gpu ids (default '[0]', Don't omit the quotes, i.e. use gpus='[0]' instead of gpus=[0]; if no gpu is available, use gpus=0; this version doesn't support multiple gpus temporarily, we will update it later)
embedding_size	embedding size , n is the number of leaves in the backbone tree).
batch_size	batch size (default: 32)
resblock_num	number of residual blocks (default: 1)

Calculating distance matrix

depp_distance.py backbone_seq_file=backbone/seq/file query_seq_file=query/seq/file model_path=model/path
Running the above command will generate a distance matrices (depp.csv), as a tab delimited csv file with column and row headers. Rows represent query sequences and columns represent backbone sequences.

• Example

  • Clone the GitHub repository and navigate to the repo directory (This is for obtaining the testing data).
    git clone https://github.com/yueyujiang/DEPP && cd DEPP
  • Calculating the distance matrix
    depp_distance.py backbone_seq_file=test/basic/backbone.fa query_seq_file=test/basic/query.fa model_path=test/basic/model.ckpt
  • The distance matrix is stored at ./depp_distance

• More

arguments	descriptions
backbone_seq_file	path to the backbone sequences file (in fasta format, required)
query_seq_file	path to the query sequences file (in fasta format, required)
model_path	path to the trained model (required)

Wol Placement

Sequences analysis using WoL data

We provide the pretrained model for WoL marker genes and ASV data. Users can place the query sequences onto the WoL species tree directly using DEPP.

Preparation

• Install UPP following the instructions here, make sure that run_upp.py is executable (try run_upp.py -h). This step is not required if you are using the docker image.
• Sequences can be either unaligned ASV (16S) or unaligned MAG data or both.
• Marker genes
  • Identify the marker genes using the protocols from WoL project.
  • Rename each sequence file using the format: <marker gene's id>.fa, e.g., p0000.fa, p0001.fa...
• ASV
  • Models of five types of 16S data is pretrained: full-length (~1600bp), V4 region (~250bp), V3+V4 region (~400bp), V4 100 (~100bp), V4 150 (~150bp). (If your ASV data is in the above five types, you can analyze your data directly. Otherwise, please align your sequences and then train your own model using the train_depp.py command)
  • Rename your ASV data using the following rules:
    • full-length 16S: 16s_full_length.fa
    • V3+V4 region: 16s_v3_v4.fa
    • V4 region: 16s_v4.fa
    • V4 100bp: 16s_v4_100.fa
    • V4 100bp: 16s_v4_150.fa
• Put all your query sequences files into one empty directory (Examples can be found at test/wol_placement in this repository).
• Download the models and auxiliary data (accessory.tar.gz) from here and unzip it.

Placement

wol_placement.sh -q directory/to/query/sequences -o directory/for/output -a directory/to/auxiliary/data/accessory
This command will give you a output directory named depp_results. items inside the directory include:

• summary directory:
  • placement tree in jplace and newick format for each sequence file.
  • placement tree in jplace and newick format that include all the queries from all the files provided
• each sequences file will have a directory which includes the distance matrix from queries to backbone species.

arguments	descriptions
-q	path to the query directory (required)
-o	directory to store the outputs (required)
-a	path to accessory.tar.gz (required)

• Example
  • Clone the GitHub repository and navigate to the repo directory (This is for obtaining the testing data).
    git clone https://github.com/yueyujiang/DEPP && cd DEPP
  • Download accessory_test.tar.gz (using accessory_test.tar.gz is only for quick test, for the whole dataset, please use accessory.tar.gz) and unzip it
    wget https://tera-trees.com/data/depp/latest/accessory_test.tar.gz && tar -xvf accessory_test.tar.gz -C ./
  • Running the following command for placement
    wol_placement.sh -a accessory_test -q test/wol_placement/ -o ./

3. The model is stored at ./depp_results

Any questions? Please contact [e-mail].