PhaGCN2

PhaGCN2 is a GCN based model, which can learn the species masking feature via deep learning classifier, for new virus taxonomy classification. To use PhaGCN2, you only need to input your contigs to the program.

PhaGCN2.1 newly update

Our database has now been updated based on the latest ICTV classification tables. In order to make it easier for users to view the sequence classification in our database, we put the specific sequence information at database/VMR_based_on_ICTV.csv Due to changes to Caudoviricetes in the new version of ICTV(remove Siphoviridae,Myoviridae and Podoviridae), we have two extension of the methods:

If you only care about the phage taxa classification, please use the extension version of PhaGCN_newICTV
In PhaGCN2.0, we extended phage without family labels to the genus level.

According to our test, the extension version of PhaGCN still remain high performance for the new ICTV labels at family level. For PhaGCN2.0, there maybe some misclassifications in some genus or subfamilies of Caudoviricetes(Bronfenbrennervirinae,Nclasvirinae,Benedictvirus,Fromanvirus,Kroosvirus,Triavirus,Turbidovirus,Veracruzvirus).In fact, the PhaGCN2 results suggest that they are very similar to other genus, such as Gladiatorvirus and Backyardiganvirus.

For phage-related task

Our web server for phage-related tasks (including phage identification, taxonomy classification, lifestyle prediction, and host prediction) is available! You can visit PhaBOX to use the GUI. We also provided more detailed intermediate files and visualization for further analyzation. A stand-alone version of PhaBOX is also available via GitHub version, and you can run all these tools at once. Hope you will enjoy it!

Required Dependencies

Python 3.x
Numpy
Pytorch
Networkx
Pandas
Diamond

All these packages can be installed using Anaconda.

If you want to use the gpu to accelerate the program:(if you want to train your own virus classification database,these packages must be install)

cuda 10.1
Pytorch-gpu

An easiler way to install

We recommend you to install all the package with Anaconda. After cloning this respository, you can use anaconda to install the environment.yaml. This will install all packages you need with gpu mode (make sure you have installed cuda on your system). We recommend you to install all the package with Anaconda.The command that you need to run is

cd PhaGCN2.0
conda env create -f environment.yaml -n phagcn2
conda activate phagcn2
export MKL_SERVICE_FORCE_INTEL=1

You need to prepare the database before using it.

cd database
tar -zxvf ALL_protein.tar.gz
cd ..

and you can use it to make virus classification.

Usage (example)

Here we present an example to show how to run PhaGCN2. We support a file named "contigs.fa" in the Github folder and it contain contigs simulated from E. coli phage. The only command that you need to run is

$ python run_Speed_up.py --contigs contigs.fa --len 8000

There are two parameters for the program:

--contigs is the path of your contigs file.
--len is the length of the contigs you want to predict. As shown in our paper, with the length of contigs increases, the recall and precision increase. We recommend you to choose a proper length according to your needs. The default length is 8000bp. The shortest length supported is 1700bp.The output file is final_prediction.csv. There are three column in this csv file: "contig_name, median_file_name, prediction".

New changes:

Now,the given database can support prediction under the all viruses which is base on ICTV 2021 year reporter. In prediction result,we add a prediction result named "Family_like" , if your virus species prediction label is "_like", it indicates that your virus and some viruses in the virus library are the same order but different families of the relationship. In the Network folder will generate a network map file, you can use this file to draw your unique and beautiful network map

New function

Now we support that you can train your own virus classification database. If you want train your own virus classification database, follow these steps.

First of all,you need with gpu mode (make sure you have installed cuda on your system)and run

pip install bio
pip install torch
sudo apt install prodigal
cd CHEER
sh creat.sh

Step one: In this step,you need make your virus sequences (Known families)in different folder, dividing to training set and validation set.Make sure your virus sequence is longer than 1700BP.(If you have a segmented virus, combine it into a single sequence,Otherwise, it will be divided into several different kinds of viruses) Take ICTV 2021 year reporter as an example,there are 11 viral sequences in Lipothrixviridae . Randomly divide 9 out of 10 sequences to do the training set and 1 to do the test set.Combine the nine training set sequences into a sequence file named Lipothrixviridae and place it in the train folder.Combine the one validation set sequences into a sequence file named Lipothrixviridae and place it in the validation folder.The same goes for the other family.(Ensure that the validation set is non-empty).

Preprocess your data set:

$ bash code/re_train_script.sh

Train your data set CNN model

$ python3 train.py --n 8 --gpus 1 --weight "1,1,1,1,1,1,1,1"

--n is the number of your families,--weight is the weight coefficient，The number of numbers in weight is equal to the number of n, --gpu is the number of Gpus you have This will produce two files Embed.pkl and Params.pkl,Replace the two files with the same name in the CNN_Classifier folder.It requires around 250GB of memory(The larger the data set, the more memory is required).

Step two: Merge all sequence files into one file and name it all_pre.fasta,then move it into

CHEER folder.Then run:

$ python3 deal_all_pre.py

Please modify your data set until no label error is reported(The label should contain at least one space). Take the sequence number and corresponding family name in a TXT text and named taxa.txt (separated by tabs)then place it in the CHEER folder.Run:

$ python3 deal_result.py`

It generates a folder of result,in this folder replaces the first line of code.txt with line 159 of run_GCN.py in the body of PhaGCN, and the second line with line 643 of run_Knowledgegraph.py,The other five files replace each of the five files in the database folder. 3. Step three: Change the default n on line 76 of run_CNN.py to the number of --n in step two Copy all_simple_pre.fasta to your PhaGCN folder and run():

$ python3 pre_train.py --contig all_simple_pre.fasta --len 1700

After running it, rename contig.F in the Cyber_data folder to dataset_compressF and replace the file with the same name in the database folder.

(See the CHEER/train_example folder for an example)

Notice

If you want to use PhaGCN, you need to take care of three things:

Make sure all your contigs are virus contigs. You can sperate bacteria contigs by using VirSorter or DeepVirFinder
The script will pass contigs with non-ACGT characters, which means those non-ACGT contigs will be remained unpredict.
if the program output an error (which is caused by your machine): Error: mkl-service + Intel(R) MKL: MKL_THREADING_LAYER=INTEL is incompatible with libgomp.so.1 library. You can type in the command export MKL_SERVICE_FORCE_INTEL=1 before runing run_Speed_up.py
If you want train your own virus classification database,Hardware requirements can be considerable(exceeding 48 GB,and at least one GPU), depending mainly on the size and complexity of the dataset. (Relationship between memory requirements and sequences analyzed forthcoming)

References

how to cite this tool:

Jiayu Shang, Jingzhe Jiang, Yanni Sun, Bacteriophage classification for assembled contigs using graph convolutional network, Bioinformatics, Volume 37, Issue Supplement_1, July 2021, Pages i25–i33, https://doi.org/10.1093/bioinformatics/btab293

Jing-Zhe Jiang, Wen-Guang Yuan, Jiayu Shang, Ying-Hui Shi, Li-Ling Yang, Min Liu, Peng Zhu, Tao Jin, Yanni Sun, Li-Hong Yuan, Virus classification for viral genomic fragments using PhaGCN2, Briefings in Bioinformatics, 2022;, bbac505, https://doi.org/10.1093/bib/bbac505

liupfskygre/PhaGCN2.0