• CASTOR-KRFE v1.3 Help file
• K-mers based feature identifier for viral genomic classification
• Copyright (C) 2023 Dylan Lebatteux, Amine M. Remita, Abdoulaye Banire Diallo
• Author : Dylan Lebatteux, Amine M. Remita
• Contact : lebatteux.dylan@courrier.uqam.ca

CASTOR-KRFE is an alignment-free method to identify a set of k-mers to discriminate between groups of genomic sequences. The core of CASTOR-KRFE is based on feature elimination using Support Vector Machines (SVM-RFE) which is an machine learning feature selection method. CASTOR-KRFE identifies an optimal length of k to maximize classification performance and minimize the number of features. The extracted set of k-mers can be used to build a prediction model. This model can then be used to predict a set of new genomic sequences. A new module allowing to identify discriminative k-mers variations and their associated information according to the sequence class has also been included.

• python
• scikit-learn
• numpy
• biopython
• joblib

List of parameters requiring adjustment in the configuration_file.ini :

• k_min : Minimum length of k-mers
• k_max : Maximum length of k-mers
• T : Percentage performance threshold (T = 0.99 is recommended) .
• training_fasta : Training fasta file path
• testing_fasta : Testing fasta file path
• reference_sequence : Path of the reference sequence in GenBank format
• k_mers_path : Path file of the extracted k-mers
• model_path : Path file of the prediction model
• prediction_path : Path of the sequence prediction file
• evaluation_mode : Evaluation mode during the prediction (True/False).

1. Specify the parameters of the previous section in the configuration_file.ini.
2. Run the following command :

$ python main.py configuration_file.ini

1. Select an option:

• 1)Extract k-mers | Required parameters: T, k_min, k_max, training_fasta and k_mers_path
• 2)Fit a model | Required parameters: training_fasta, k_mers_path and model_path
• 3)Predict a sequences | Required parameters: testing_fasta, k_mers_path, model_path, prediction_path and evaluation_mode
• 4)Motif analyzer | Required parameters: training_fasta, k_mers_path and reference_sequence
• 5)Exit/Quit

>id_sequence_1|target_sequence_1 
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC 
>id_sequence_2|target_sequence_2						
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC
...
...
...
>>id_sequence_n-1|target_sequence_n-1									 
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC 
>id_sequence_n|target_sequence_n													 
CTCAACTCAGTTCCACCAGGCTCTGTTGGATCCGAGGGTAAGGGCTCTGTATTTTCCTGC 

• The character "|" is used to separate the sequence ID from its target.
• The target must be specified in the fasta file for a prediction with evaluation_mode = True.
• For more detailed examples see the data sets in the Data folder

• k_mers.fasta: File of the extracted k-mers list
• model.pkl : Prediction model generated by CASTOR-KRFE
• Prediction.csv : Results file of the prediction of unknown genomic sequences
• Signature_location.xlsx : Analysis report associated with a signature

• Lebatteux, D., Remita, A. M., & Diallo, A. B. (2019). Toward an alignment-free method for feature extraction and accurate classification of viral sequences. Journal of Computational Biology, 26(6), 519-535.

• Lebatteux, Dylan, et al. "KANALYZER: a method to identify variations of discriminative k-mers in genomic sequences." 2022 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). IEEE Computer Society, 2022.