An application developed in Perl and Python that allows the development of classification models for genomic sequences, to improve data mining and usefulness of genomic applications.
| software | version | required |
|---|---|---|
| VAMPhyRE | 1.0 or above | yes |
| perl | 5.30 or above | yes |
| python | 3.8 or above | only for GUI |
| pillow | 8.1 or above | only for GUI |
| pandas | 1.2.0 or above | only for GUI |
| ttkthemes | 3.1.0 or above | only for GUI |
Note: You can download VAMPhyRE from here: http://biomedbiotec.encb.ipn.mx/VAMPhyRE/download.php
The CABBaGe application core is built from three modules:
- The Bayesian Classifier, this module uses a Naive Bayes Classifier technique which is based on the so-called Bayesian theorem and is particularly suited when the dimensionality of the inputs is high. Despite its simplicity, Naive Bayes can often outperform more sophisticated classification methods. The module classifies genomic sequences into predetermined classes using a training genome matrix of known parameters (e.g. disease, host age, host sex, geographic location, drug resistance etc.)
Note: In order for the CABBAGE to resume operation the input format must be comma-separated values (.csv) files
Three files are needed: Training.csv, MetaData.csv and Query.csv
The Training.csv file is a Boolean table that denotes the presence or absence of a certain "feature" which can either be a gene (Pan-genome*) or a genomic region denotated by a virtual probe (Virtual Hybridization*).
The MetaData.csv file is a table that relates each of the samples form the Training.csv to predefined classes.
The Query.csv file are the samples that must be classified, and they should be on the same format as in the Training.csv file.
- The Feature Extractor; classification has the problem of high dimensionality of feature space due to the extensive information from genomic data. This high dimensionality of feature space is solved by feature selection and feature extraction methods that improve the performance of categorization. The feature selection and feature extraction techniques remove the irrelevant features from the test and reduce the dimensionality of feature space. The module accomplishes this task using a statistics test (e.g. Chi squared) extracting the most informative genes or genomic regions that make a sample belong to a particular class.
Note: In order for the CABBAGE to resume operation the input format must be comma-separated values (.csv) files
Two files are needed: Training.csv and MetaData.csv
The Training.csv file is a Boolean table that denotes the presence or absence of a certain "feature" which can either be a gene (Pan-genome*) or a genomic region denotated by a virtual probe (Virtual Hybridization*).
The MetaData.csv file is a table that relates each of the samples form the Training.csv to predefined classes.
Adicionally a statistical Test must be chosen.
- The Feature Filter, this module is a tool for extracting the top n informative features extracted by the Feature Extractor module, thus the user determines the number of features to be contained in the developed clasification model.
Note: In order for the CABBAGE to resume operation the input format must be comma-separated values (.csv) files
One file is needed: Probabilities.csv, MetaData.csv and Query.csv
The Probabilities.csv file is the output file from the Feature Extractor module, wich is a list of all the informative features for the model.
Adicionally an Out File name must be provide as well as the Number Of Features for the final model.
Sample1Sample2Sample3Sample4Gene/Probe a0 1 1 0 Gene/Probe b1 1 1 1 Gene/Probe c1 1 0 0 Gene/Probe d0 0 0 0 Gene/Probe e1 0 1 1 Gene/Probe f1 1 1 1
The Samples and Gene/Probe names should be determined by the user, the file can contain as many rows and columns as needed.
SampleClassSample1A Sample2B Sample3C Sample4D Sample5E Sample6F
The Samples and Gene/Probe names should be determined by the user, the file can contain as many rows as needed.
SampleXSampleYSampleZGene/Probe a1 1 1 Gene/Probe b1 1 0 Gene/Probe c1 0 0 Gene/Probe d0 0 0 Gene/Probe e1 0 0 Gene/Probe f1 1 1
The Samples and Gene/Probe names should be determined by the user, nonetheless the Gene/Probe must be the same as the Training.csv file used. The file can contain as many rows and columns as needed.
GUICABBaGe is a wrap-up tool developed in python for an easier user experience, at present (August 2021) is only available for Windows and Linux (Ubuntu) OS.
This Graphic User Interface permits the users to quickly develop clasification models even if they are not familiarized with terminal commands shortening the learning curve.
For this example, a collection of Klebsiella sp. genomes can be found in the Example folder, here a classification for different species is made among Variicola, Quasipneumoniae and Oxitoca strains.
1) Preprocess the samples found in the “Training folder” using VAMPhyRe with the following commands:
VH5cmdl –PROBEFILE vps13.txt –TARGETLIST TrainingList.txt –OUTFILE VH5outfile.txt -MISMATCHES 1 –STRAND both
VHRP -VHDATAFILE VH5outfile.txt -PROBEFILE vps13.txt -TARGETLIST TrainingList.txt -GLOBALFILE Screening
The main goal of this step is to obtain the “Screening.csv” file which contains all the information of the virtual hybridization.
2) Use the Feature Extractor Module in CABBaGe select the “Screening.csv” as the TrainingFile and the “metadatakleb.csv” as the Meta Data File, also select the statistical test as chi-square “X2” for this example.
From this step a new folder will be created “ResultsFE” inside you will find the “Chi_SquareScreening.csv” file and the presence absence file which contains the informative value of each k-mer and the presence distribution of each one of them, respectively.
3) Use the Feature Filter Module in CABBaGe select the “Chi_SquareScreening.csv” file as the Probabilities File, determine the Out File Name “Filtered” for this example and the number ok the top informative k-mers to be extracted “100” for this example.
From this step a new folder will be created “ResultsFF” inside you will find the “Filtered100.csv” file and the “FilteredVPS.txt” file both contain the information of the top 100 k-mers selected to become the classification model.
4) Preprocess the samples found in the “Training folder” using VAMPhyRe with the following commands:
VH5cmdl –PROBEFILE FilteredVPS.txt –TARGETLIST TrainingList.txt –OUTFILE VH5Filteredoutfile.txt -MISMATCHES 1 –STRAND both
VHRP -VHDATAFILE VH5Fikteredoutfile.txt -PROBEFILE FilteredVPS.txt -TARGETLIST TrainingList.txt -GLOBALFILE Training
After this process you will have the final “Training.csv” file which represents the classification model for this example.
5) Preprocess the samples found in the “Validation folder” using VAMPhyRe with the following commands:
VH5cmdl –PROBEFILE FilteredVPS.txt –TARGETLIST ValidationList.txt –OUTFILE VH5valoutfile.txt -MISMATCHES 1 –STRAND both
VHRP -VHDATAFILE VH5valoutfile.txt -PROBEFILE FilteredVPS.txt -TARGETLIST ValidationList.txt -GLOBALFILE Query
In this step we preprocess the genomes to be classified using the filtered k-mers previously obtained. At the end we have a “Query.csv” file with the presence absence information from the virtual hybridization.
6) Use the Bayesian Classifier Module in CABBaGe select the “Training.csv” as the Training File the “metadatakleb.csv” as the Meta Data File and the “Query.csv” as the query file.
In this step a new folder will be created “ResultsBC” inside you will find two files the “Probabilities.csv” file and the “Prediction.csv” file in the latter you can find the classification made for the validation samples which should be as follows:
Sample Class Probability NZ_CP017849.1 Class variicola 6.39E+43 NZ_CP054254.1 Class variicola 6.48E+61 NZ_CP018307.1 Class variicola 2.27E+65 NZ_LR134235.1 Class variicola 4.23E+67 NZ_CP048379.1 Class variicola 6.70E+90 NZ_CP023478.1 Class quasipnemoniae 4.91E+122 NZ_CP026368.1 Class quasipnemoniae 6.11E+122 NZ_CP063902.1 Class quasipnemoniae 6.27E+124 NZ_CP012252.1 Class quasipnemoniae 5.40E+126 NZ_CP029437.1 Class quasipnemoniae 2.16E+127 NZ_CP026275.1 Class oxytoca 1.78E+131 NZ_LR134333.1 Class oxytoca 1.28E+132 NZ_CP020358.1 Class oxytoca 4.11E+132 NZ_CP026285.1 Class oxytoca 3.37E+133 NZ_CP026269.1 Class oxytoca 8.91E+133
As seen on the table above, each sample isa assigned to one of the possible classes provided this is due to the hard classification method used wich provides a reult in the classification decision boundary.
In the example shown, microbial full genomes are tested nevertheless any set of DNA samples provided can be subjected to the same principle.