PlasForest

A random forest classifier of contigs to identify contigs of plasmid origin in contig and scaffold genomes.

PlasForest has been published in BMC Bioinformatics (https://doi.org/10.1186/s12859-021-04270-w). On a test dataset, it was measured to detect 92.7% of contigs of plasmid origin, for only 2.7% false positives.

Dependencies

• Python 3.6+
• Python libraries:
  • BioPython
  • scikit-learn 0.22.2.post1
  • Numpy
  • Pandas
  • Joblib
• NCBI Blast+

Install

Clone this git on your computer:

$ git clone https://github.com/leaemiliepradier/PlasForest

Python dependencies can be manually installed using pip:

pip install --user biopython numpy pandas joblib scikit-learn==0.22.2.post1

on Ubuntu/Debian, NCBI Blast+ can be installed through the repositories with the command sudo apt-get install ncbi-blast+. For other systems, it is advised to download the binaries from the NCBI FTP repository:

# Download latest BLAST version
# For systems other than Linux, check the FTP for a compatible version
$ wget https://ftp.ncbi.nlm.nih.gov/blast/executables/blast+/LATEST/ncbi-blast-2.10.1+-x64-linux.tar.gz
# Decompress
$ tar -zxvf ncbi-blast-2.10.1+-x64-linux.tar.gz
# Add BLAST location to system PATH
$ export PATH=$HOME/ncbi-blast-2.10.1+/bin:$PATH

Untar the random forest classifier:

$ cd PlasForest/
$ tar -zxvf plasforest.sav.tar.gz

The use of PlasForest requires to download a database of plasmid sequences (2.5GB). This script should be launched from the directory in which PlasForest.py and plasforest.sav are stored. Be careful this step may take a while.

$ chmod 755 database_downloader.sh
$ ./database_downloader.sh

This script might have trouble downloading all the sequences for the database. In case it happens, it will try to download the remaining sequences from NCBI Entrez, and it will ask you a valid email address to do so.

Testing the installation

To test that PlasForest has been correctly installed, you can run the following script:

$ ./test_plasforest.sh

which will test if all required files are here and will run PlasForest on a test dataset.

Manual

The PlasForest pipeline is able to process 1000 contigs (~50 genomes) in about 8 minutes using 1 CPU and 2.3 GB of memory. Using 16 CPUs, it will take only 2 minutes, but will require 20 GB of memory.

Minimal use

PlasForest requires at least an input FASTA file, and generates an column-separated output file.

$ python3 PlasForest.py -i /path/to/your/inputfile.fasta

You can also specify the name of the output file.

$ python3 PlasForest.py -i /path/to/your/inputfile.fasta -o /path/to/your/outputfile.csv

Additional options

Option -b will add a third column to the output file, with the identification number of the best hit in the plasmid database.

Option -f will add seven columns to the output file, with the features measured for each contig.

Option -v can activate verbose mode.

Option -r will allow to re-assign contigs that are already described as plasmid or chromosome.

Option --threads <int> allows to define the number of CPUs on which PlasForest will be run.

Option --size_of_batch <int> allows to define how many sequences can be used in the single batch.

Option --model <path> allows to define your own path for the .sav model.

Option --database <path> allows to define your own path to the plasmid database.

Training PlasForest on custom data

The classifier is already provided for PlasForest and there is no need to train it again. However, if you want to train PlasForest on a custom training dataset and/or using a custom plasmid database, you can do it by using train_plasforest.py.

This pipeline requires you to provide an input FASTA file for your training set, a CSV file containing the labels for each sequence of the training set, and the output name for your custom classifier:

$ python3 train_plasforest.py -i /path/to/your/trainingset.fasta -l /path/to/your/labelfile.csv -o /path/to/your/output_classifier.sav

Make sure that your label file has two columns: one named "ID" containing all the identifiers of your training set sequences; and the other names "Plasmid" taking the values 0 if a sequence comes from a chromosome and 1 if it comes from a plasmid.

Optionally, you can also

• run in verbose mode with option -v
• specify a custom plasmid database with option --database <BLAST+ db>
• define the number of CPUs on which the training will be run with option --threads <int>

Citation

If you use PlasForest for your research, please cite the following papers:

• Pradier L, Tissot T, Fiston-Lavier AS, Bedhomme S (2021). PlasForest: a homology-based random forest classifier for plasmid detection in genomic datasets. BMC Bioinformatics 22, 349; doi: https://doi.org/10.1186/s12859-021-04270-w
• Camacho C, Coulouris G, Avagyan V, Ma N, Papadopoulos J, Bealer K, Madden TL (2009). BLAST+: architecture and applications. BMC Bioinformatics 10:421
• Cock PJA, Antao T, Chang JT, Chapman BA, Cox CJ, Dalke A, Freidberg I, Hamelryck T, Kauff F, Wilczynski B, de Hoon MJL (2009). Biopython: freely available Python tools for computational molecular biology and bioinformatics. Bioinformatics 25:1422-1423

Support

Any issues connected with PlasForest should be addressed to Léa Pradier (lea.pradier (at) cefe.cnrs.fr).