MixtureMining

Final project for Wright State University CS4840/6840, Intro to Machine Learning, Spring 2016

Authors

• Nathaniel Adams, adams.201@wright.edu
• Oliver Ceccopieri, ceccopieri.2@wright.edu
• Nathan Jent, jent.2@wright.edu

License

GNU GPL v3 (see separate license file). Weka, which this work references, is also licensed under GNU GPL v3.

About

The MixtureMining project is intended to explore methods for estimating/inferring the number of contributors present in mixed DNA samples. This project includes:

• Preprocessing
  • sample genotypes for use in simulating mixed samples (see Example data description below)
  • a genotype “mixing” program for generating simulated mixed samples (mix_gen.rb)
  • a feature extraction/creation program for real or simulated mixed samples (locus_info.rb)
• Feature filtering
  • Utilizes forward feature selector, backwards feature selector, or principle components
• Estimation
  • Utilizes a naive Bayesian classifier for prediction

System requirements

• Ruby interpreter
  • Required Gems (install using ‘gem install ‘)
    • getopt
• JRE 1.8+ (for running)
• JDK 1.8+ (for building)
• Apache Ant (for building)

Build instructions

• Preprocessing:
  None (pure Ruby)
• Filtering/estimation:
  • Use the “build” feature in the provided Ant build.xml file.
  • Internet connection required for downloading Weka.

Run instructions

• Build the JAR file, then run the command:

  > ruby driver.rb min_contributors max_contributors mixtures_per_class -f [AS/ASB/PC] -n features_to_keep -c BS'

Preprocessing

All paths relative to ./preprocessing/

1. Mixture simulation

   • If your mixtures already exist in the proper format (see ./preprocessing/mixtures for an example), proceed to step #2

     > ruby mix_gen.rb --infile ./path_to/genotypes.csv --outfile ./path_to/mixture_output.csv --per num_samples_per_mix --mixtures num_mixture_to_make [--seed PRNG_seed_value]

     Ex.

     > ruby mix_gen.rb --infile single_source/361_caucasian_identifiler_loci.csv --outfile mixtures/361_cau_id_2_mix_500.csv --per 2 --mixtures 500

2. Feature creation
   2.1 Allele frequency feature creation: uses [—aftable aftable.csv] flag, requires allele frequency table to be passed to script.
   2.2 Allele counting feature creation: uses [—ac] flag

   > ruby locus_info.rb --infile ./path_to/mixture_output.csv --outfile ./path_to/preprocessed_mixtures.csv [--aftable ./path_to/allele_frequencies_table.csv] [--ac]

   Ex.

   > ruby locus_info.rb --infile mixtures/361_cau_id_mix_2_3_4_1000each.csv --outfile mixes_preprocessed/361_cau_id_mix_2_3_4_1000each_preprocessed.csv --aftable frequencies/361_cau.csv --ac

Filtering and estimation

> java -jar MixtureMining.jar training_file test_file -f [AS/ASB/PC] -n features_to_keep -c BS

Example data

Example data taken from NIST genotype dataset and accompanying allele frequencies, available at http://www.cstl.nist.gov/strbase/NISTpop.htm

Dev environments:

• Windows 8.1
  • ruby 2.2.4p230 (x64-mingw32)
  • Java 1.8.0_71 64-bit