CASTOR_KRFE

• CASTOR-KRFE v1.0 Help file
• Feature extractor for viral genomic classification
• Copyright (C) 2018 Dylan Lebatteux, Amine M. Remita, Abdoulaye Banire Diallo
• Author : Dylan Lebatteux, Amine M. Remita
• Contact : [email protected]

Description

CASTOR-KRFE contains three modes:

• Training : Extract best kmer features and train a model with a dataset of labeled genomic sequences
• Evaluation : Evaluate a prediction model with a dataset of labeled genomic sequences
• Prediction : Predict classes of unknown genomic sequences

Required Software

• numpy (https://docs.scipy.org/doc/numpy-1.13.0/user/install.html)
• scipy (https://www.scipy.org/install.html)
• scikit-learn (https://scikit-learn.org/stable/install.html)
• biopython (https://biopython.org/wiki/Download)
• matplotlib (optional, https://matplotlib.org/)

Options

-h : Show the help
-t, --train : Extract best kmer features and train a model with a dataset of labeled genomic sequences
-e, --evaluate : Evaluate a model prediction with a dataset of labeled genomic sequences
-p, --predict : Predict classes of unknown genomic sequences
-f FASTA, --fasta FASTA : Fasta file containing the sequences (the file is used in the three modes)
-c [CSV], --csv [CSV] : CSV file containing the labels of the sequences with the format : ID,label
> The file could be used in the training/evaluation modes.

-m [MODEL], --model [MODEL] : File of the model <br/>
-k [KMERS], --kmers [KMERS] : Name of the file of the extracted k-mers list (Défault : Kmers.txt)
-o [OUTPUT], --output [OUTPUT] : Output directory. If it is not specified, the program uses Output/ folder 
--threshold [THRESHOLD] : Percentage of performance loss in terms of F-measure to reduce the number of attributes
>Default : T = 1
To reduce the number of features T = 0,99 showed great performance

--kmin [KMIN] : Minimum length of k-mer(s)
--kmax [KMAX] : Maximum length of k-mer(s)
--fmin [FMIN] : Minimum number of features to identify
--fmax [FMAX]: Maximum number of features to identify
--version : Show program's version number and exit

Examples

Training mode example :

Command :

python -W ignore Main.py -t -f Data/HIVGRPCG/data.fa -c Data/HIVGRPCG/class.csv --kmin 5 --kmax 10 --fmin 10 --fmax 25 --threshold 1 -k ListOfIdentifiedKmers.txt

Description :

This command will allow to extract at least 10 features (--fmin 10) and at most 25 features (--fmax 25) based on k-mers with a minimum and maximum length of respectively 5 and 10 (--kmin 5, --kmax 10).
These features will be extracted from the sequences of the fasta file (-f Data/HIVGRPCG/data.fa) whose labels are specified in the CSV file (-c Data/HIVGRPCG/class.csv).
The threshold (--threshold 1) to reduce the number of attributes is 1 (default value).
A prediction model will be generated (-t) and the extracted features saved in the file (-k ListOfIdentifiedKmers.txt).

Evaluation mode example :

Command :

python -W ignore Main.py -e -f Data/HIVGRPCG/data.fa -c Data/HIVGRPCG/class.csv -m Output/model.pkl

Description :

This command will allow to evaluate (-e) the model (-m Output/model.pkl) with the prediction of the sequences of the fasta file (-f Data/HIVGRPCG/data.fa) whose labels are specified in the CSV file (-c Data/HIVGRPCG/class.csv).
The results will be given in the Evaluation.txt file of folder Ouput.

Prediction mode example :

Command :

python -W ignore Main.py -p -f Data/HIVGRPCG/data.fa -m Output/model.pkl

Description :

This command will allow to predict (-p) the classes of the unknown sequences of the file (-f Data/HIVGRPCG/data.fa) based on the model (-m Output/model.pkl).
The results will be given in the Prediction.txt file of folder Ouput.

Input Files

FASTA : Contains the sequences in fasta format

>Id Description <br/>
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC <br/>
>Id Description <br/>
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC <br/>
>Id Description <br/>
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC <br/>
>Id Description <br/>
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC <br/>
>Id Description <br/>
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC <br/>

CSV : Contains the classes associated with each sequence

>Id,Class <br/>
>Id,Class <br/>
>Id,Class <br/>
>Id,Class <br/>
>Id,Class <br/>

MODEL : Contains a prediction model in pkl format

For more detailed examples see the data sets in the Data folder

Output Files

• Analysis.png : Show the decision graph of the optimal set of k-mers by CASTOR algorithm (Training mode)
  
• model.pkl : Prediction model generated by CASTOR-KRFE (Training mode)
  
• Kmers.txt : File of the extracted k-mers list (Name can be change in parameter)
  
• Evaluation.txt : Results file of the evaluation a model with a dataset of labeled sequences (Evaluation mode)
  
• Prediction.txt : Results file of the prediction of unknown genomic sequences (Prediction mode)
  

Cite CASTOR-KRFE

• Dylan Lebatteux, Amine M. Remita and Abdoulaye Banire Diallo (2018) Toward an Alignment-Free Method for Feature Extraction and Accurate Classification of Viral Sequences. Joint ICML and IJCAI Workshop on Computational Biology (WCB'18)

References

• Scikit-learn Pedregosa, Fabian, et al. "Scikit-learn: Machine learning in Python." Journal of machine learning research 12.Oct (2011): 2825-2830. Website : https://scikit-learn.org/

• Biopython Cock, Peter JA, et al. "Biopython: freely available Python tools for computational molecular biology and bioinformatics." Bioinformatics 25.11 (2009): 1422-1423. Website : https://biopython.org/