MPclassifier

MPclassifier is a python package that automatizes the development of a classifier based on mutational profiles. Although this package was originaly developed to predict the Homologous Recombination (HR) status of high grade serous ovarian cancer (HGSOC), its applications extend far beyond. MPclassifier simplifies the process of extracting mutational signatures and training a neural network to produce a regression prediction together with a classification margin that can be easily exported and shared to fit and predict new data.

Getting Started

Installing MPclassifier

MPclassifier is currently available as a test python package in the pypi repository. To install use the following command:

python3 -m pip install --index-url https://test.pypi.org/simple/ --no-deps MPclassifier

Prerequisites

MPclassifier relies on SigProfiler to create mutational matrices and extract mutational signatures. Both packages together with the apropriate reference genome need to be installed. The details on the installation and the full functionality of the SigProfiler package suite can be found in :

https://osf.io/t6j7u/wiki/home/

https://osf.io/s93d5/wiki/home/

pip install SigProfilerMatrixGenerator
pip install sigproextractor

from SigProfilerMatrixGenerator import install as genInstall
genInstall.install('GRCh38', bash=True)

MPclassifier uses plotly to produce graphics and scikit to train a neural network.

pip install plotly==4.6.0
pip install -U scikit-learn

Further support on the installation of plotly can be found in: https://pypi.org/project/plotly/

Further standard package prerequisites include:

• pandas
• numpy
• scipy

Overview of MPclassifier

MPclassifier relies on the detection of mutational signatures, that carry the footprint of mutational events found in the genomes, to train a grid of shallow regressor neural networks and produce a threshold for classification.

A notebook including everything needed to get a project up and running has been included.

notebook-example

MPclassifier.signature_classifier object

class MPclassifier.signature_classifier : (vcf, positive=None, negative=None, project_name='MPclassifier', reference_genome='GRCh38', exome=False, feature_list=['SBS96','ID83','DBS78'], model = signature_model()):

Parameters:

• vcf : str

  path to a folder containing the all .vcf files.

• positive : str , Default = None

  path to a .txt file containing the list of all positive samples. The .txt file should contain one sample name per line.

• negative : str , Default = None

  path to a .txt file containing the list of all negative samples. The .txt file should contain one sample name per line.

• project_name : str , Default = 'MSclassifier'

  Project name that will be used for referencing throughout the project.

• model : signature_model class, Default=None

  Model used to train or predict the output of the classifier.

• reference_genome : str in {‘GRCh38’, ‘GRCh37’,'GRCm38','GRCm37'} , Default = 'GRCh38'

  Genome reference used during the process of variant calling. reference_genome is only used as an argument for SigProfilerMatrixFunc, therefore admits all supported genomes in the package

• exome : boolean, Default = False

  option to filter vcf files to only retain variant calls present in the exome

• feature_list : list, Default = ['SBS96','ID83','DBS78']

  List of any mutational profile in the output of SigProfilerMatrixFunc. These are the features that will be used to train the classifier.

After training, this model will acquire further attributes:

• data : pandas.DataFrame

  Data used for training the classifier, together with the both predictions: regression and classification.

• plot : plotly.fig

  Plot of the regression prediction together with the margin maximizer

• confusion_matrix:

  Confusion matrix as computed by sklearn.metrics.confusion_matrix

• ROC_curve: plotly.fig

  Plot of the ROC curve of our regression model.

MPclassifier.signature_model object

In order to share an MPclassifier trained model, a signature_model class has been created. This class contains information about signatures used for training and the model used for training.

class MPclassifier.signature_classifier : ()

Although this class starts empty, depending on the developed classifier, after training this class will acquire the following attributes:

• signatures : list

  List of pandas.DataFrames. Each dataframe corresponds to the extracted panel of signatures used for nmf fitting.

• features: list

  List of features used in the model.

• classifier: class

  Trained model. As trained by default, model is a sklearn.neural_network.MLPRegressor.

• svm: class

  Trained sklearn.svm.SVC class.

• margin: float

  Margin maximizer.

• importances: pandas.DataFrame

  Model importances as extracted by sklearn.inspection.permutation_importance

Authors

Eric Latorre Crespo - Initial work