mutation_origin: python library for classifying mutation origins

This repository contains scripts used to train and deploy the classification analyses reported in Classifying ENU induced mutations from spontaneous germline mutations in mouse with machine learning techniques by Zhu, Ong and Huttley.

Initial Setup

Inside a conda environment, run pip on the downloaded zip file.

$ pip install mutori.zip

The mutori and mutori_batch commands

Installation creates mutori and mutori_batch command line scripts. Command line help for mutori shows

$ mutori
Usage: mutori [OPTIONS] COMMAND [ARGS]...

  mutori -- for building and applying classifiers of mutation origin

Options:
  --help  Show this message and exit.

Commands:
  lr_train       logistic regression training, validation,...
  nb_train       Naive Bayes training, validation, dumps...
  ocs_train      one-class svm training for outlier detection
  performance    produce measures of classifier performance
  predict        predict labels for data
  sample_data    creates train/test sample data
  xgboost_train  Naive Bayes training, validation, dumps...

Command line help for mutori_batch shows

$ mutori_batch 
Usage: mutori_batch [OPTIONS] COMMAND [ARGS]...

  mutori_batch -- batch execution of mutori subcommands

Options:
  --help  Show this message and exit.

Commands:
  collate        collates all classifier performance stats and...
  lr_train       batch logistic regression training
  nb_train       batch naive bayes training
  ocs_train      batch one class SVM training
  performance    batch classifier performance assessment
  predict        batch testing of classifiers
  sample_data    batch creation training/testing sample data
  xgboost_train  batch xgboost training

Input and output data formats

Input sequence data

Must be in a tab delimited form, with a header line. The file will be read by pandas.read_csv. Required columns are: varid, variant identifiers; flank5 and flank3 are the 5' and 3' flanking sequences respectively; direction, mutation direction with values of form XtoY (X and Y are nucleotides).

For training, the file must also contain a response column containing either e/g (for ENU and spontaneous Germline respectively)

If the GC% is to be examined, a GC column is also required.

The column order does not matter.

Classifiers

These are saved in python's pickle format. Also saved are attributes defining the feature set against which the classifier was trained.

Performance measures

Stored in json format.

Collated output

Done via the mutori_batch collate command, produces tab separated files of key performance metrics and summary statistics of each of those.

License

The BSD 3-clause license is included in this repo as well, refer to license.txt