GaoyangLuo / DeepMicrobeFinder

Installation

dependencies: install using anaconda, using the tentative name, "DeepMicrobeFinder".

conda create --name def python=3.6
conda activate def
pip install tensorflow==1.15 keras==2.2.4 numpy scipy pandas sklearn biopython torch pgzip
pip install 'h5py==2.10.0' --force-reinstall
pip install 'scipy==1.4.1' --force-reinstall

Running Prediction

Several command line options are available for prediction:

• -i: input fasta file for prediction
• -e: encoding mode, currently support "one-hot"
• -d: directory for models, currently use "models/one-hot-models"
• -m: prediction mode, "hybrid" or "single"
• -l: (optional), if choosing single mode, specify a length for model to use

A set of pretrained model with different lengths locate in the folder model.

Example usage:

python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m single -l 500 
python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m single -l 1000
python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m single -l 2000
python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m single -l 3000
python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m single -l 5000
python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m single
python predict.py -i test.fasta -e one-hot -d models/one-hot-models/ -m hybrid

Train your own model

to train your own model, first encode

specify the encoding scheme and give directory for training and validation set. Commandline arguments:

Options:
  -h, --help            show this help message and exit
  -i INPUTDIR, --input=INPUTDIR
                        Specify input folder for data
  -e ENCODINGSCHEME, --encode=ENCODINGSCHEME
                        Specify one-hot, codon, or embedding
  -l CONTIGLENGTH, --length=CONTIGLENGTH
                        contigLength
  -t SEQUENCETYPE, --type=SEQUENCETYPE
                        sequence type, Protein or DNA
  -o OUTPUTDIR, --output=OUTPUTDIR
                        Output Directory

Example usage:

Encoding step: use the batch script

• first parameter as the input directory for fasta files
• second paramter as the encoding length
• third parameter as encoding scheme, one-hot or codon.

bash batch_encode.sh <input_dir> 3000 one-hot

encoded files will be located in the folder encode created under the same directory as input directory. Folder name will also contain the encoding scheme.

The input fasta file should have the name:

• EukTr.fasta for eukaryote training fasta file, EukVal.fasta for validation.
• ProkTr.fasta for prokaryote training fasta file, ProkVal.fasta for valiation.
• ProkVirusTr.fasta for prokaryote virus training file, ProkVirusVal.fasta.
• EukVirusTr.fasta for eukaryote virus training file, EukVirusVal.fasta for eukaryote virus validation file.
• PlasmidTr.fasta for plasmid training file, PlasmidVal.fasta for plasmid validation file.

Training step: for training, we assume that the input folder contains all encoded code of each category (Prokaryote, Eukaryote, ProkVirus, EukVirus, Plasmid), with the naming convention listed above.

Options:
  -h, --help            show this help message and exit
  -i INPUTDIR, --input=INPUTDIR
                        Specify input folder for data
  -e ENCODINGSCHEME, --encode=ENCODINGSCHEME
                        Specify one-hot, codon, or embedding
  -l CONTIGLENGTH, --length=CONTIGLENGTH
                        contigLength
  -t SEQUENCETYPE, --type=SEQUENCETYPE
                        sequence type, Protein or DNA
  -o OUTPUTDIR, --output=OUTPUTDIR
                        Output Directory

example usage:

python training.py -i <input_dir> -e one-hot -l 3000 -t DNA -o <output_dir>

Prediction File Format

Each row of the output prediction file contains the scores from the neural network for class label assignment. The higher the score, the more likely the class label for a given sequence.

• Sequence Name: the name of the sequence in the input fasta file
• Eukaryote: the score the neural network model assigns to this sequence being Eukaryote.
• EukaryoteVirus: the score for this sequence being EukaryoteVirus
• Plasmid: the score for this sequence being Plasmid
• Prokaryote: the score for this sequence being prokaryote
• ProkaryoteVirus: the score for this sequence being ProkaryoteVirus

As an example:

Sequence Name   Eukaryote       EukaryoteVirus  Plasmid Prokaryote      ProkaryoteVirus
NC_008199#3k#12001#15000        0.0001718740027172454   1.2298997305038642e-05  0.1367506879128939      0.4601334227976679      5.402931783348322

To assign a label to a given sequence, we find the maximum of all the assigned scores. In the above case, ProkaryoteVirus has the highest score, so the sequence is assigned as belonging to the class ProkaryoteVirus.