ezancestry

Easily visualize your direct-to-consumer genetics next to 2500+ samples from the 1000 genomes project. Evaluate the performance of a custom set of ancestry-informative snps (AISNPs) at classifying the genetic ancestry of the 1000 genomes samples using a machine learning model.

A subset of 1000 Genomes Project samples' single nucleotide polymorphism(s), or, SNP(s) have been parsed from the publicly available .bcf files.
The subset of SNPs, AISNPs (ancestry-informative snps), were chosen from two publications:

• Set of 55 AISNPs. Progress toward an efficient panel of SNPs for ancestry inference. Kidd et al. 2014
• Set of 128 AISNPs. Ancestry informative marker sets for determining continental origin and admixture proportions in common populations in America.. Kosoy et al. 2009 (Seldin Lab)

ezancestry ships with pretrained k-nearest neighbor models for all combinations of following:

* Kidd (55 AISNPs)
* Seldin (128 AISNPs)

* continental-level population (superpopulation)
* regional population (population)

* principal componentanalysis (PCA)
* neighborhood component analysis (NCA)
* uniform manifold approximation and projection (UMAP)

Table of Contents

• Installation
• Config
• Usage
  • command line tool
    • predict
    • plot
    • generate-dependencies
    • build-model
  • Python API
• Visualization
• Contributing

Installation

Install ezancestry with pip:

pip install ezancestry

Or clone the repository and run pip install from the directory:

git clone [email protected]:arvkevi/ezancestry.git
cd ezancestry
pip install .

Config

The first time ezancestry is run it will generate a config.ini file and data/ directory in your home directory under ${HOME}/.ezancestry. You can edit conf.ini to change the default settings, but it is not necessary to use ezancestry. The settings are just a utility for the user so they don't have to be verbose when interacting with the software. The settings are also keyword arguments to each of the commands in the ezancestry API, so you can always override the default settings.

These will be created in your home directory:

${HOME}/.ezancestry/conf.ini
${HOME}/.ezancestry/data/

Explanations of each setting is described in the Options section of the --help of each command, for example:

ezancestry predict --help

Usage: ezancestry predict [OPTIONS] INPUT_DATA

  Predict ancestry from genetic data.

  * Default arguments are from the ~/.ezancestry/conf.ini file. *

Arguments:
  INPUT_DATA  Can be a file path to raw genetic data (23andMe, ancestry.com,
              .vcf) file, a path to a directory containing several raw genetic
              files, or a (tab or comma) delimited file with sample ids as
              rows and snps as columns.  [required]


Options:
  --output-directory TEXT         The directory where to write the prediction
                                  results file

  --write-predictions / --no-write-predictions
                                  If True, write the predictions to a file. If
                                  False, return the predictions as a
                                  dataframe.  [default: True]

  --models-directory TEXT         The path to the directory where the model
                                  files are located.

  --aisnps-directory TEXT         The path to the directory where the AISNPs
                                  files are located.

  --n-components INTEGER          The number of components to use in the PCA
                                  dimensionality reduction.

  --k INTEGER                     The number of nearest neighbors to use in
                                  the KNN model.

  --thousand-genomes-directory TEXT
                                  The path to the 1000 genomes directory.
  --samples-directory TEXT        The path to the directory containing the
                                  samples.

  --algorithm TEXT                The dimensionality reduction algorithm to
                                  use. Choose pca|umap|nca

  --aisnps-set TEXT               The name of the AISNP set to use. To start,
                                  choose either 'Kidd' or 'Seldin'. The
                                  default value in conf.ini is 'Kidd'. *If
                                  using your AISNP set, this value will be the
                                  in the namingc onvention for all the new
                                  model files that are created*

  --help                          Show this message and exit.

Usage

ezancestry can be used as a command-line tool or as a Python library. ezancestry predict comes with pre-trained models when --aisnps-set="Kidd" (default) or --aisnps-set="Seldin".

build-model and generate-dependencies are for advanced users -- they download large amounts of data and build a new model from a custom AISNPs file.

command-line interface

There are four commands available:

1. predict: predict the genetic ancestry of a sample or cohort of samples using the nearest neighbors model.
2. plot: plot the genetic ancestry of samples using only the output of predict.
3. generate-dependencies: generate the dependencies for build-model.
4. build-model: build a nearest neighbors model from the 1000 genomes data using a custom set of AISNPs. Requires: generate-dependencies to be run first.

Use the commands in the following way:

predict

ezancestry can predict the genetic ancestry of a sample or cohort of samples using the nearest neighbors model. The input_data can be a file path to raw genetic data (23andMe, ancestry.com, .vcf) file, a path to a directory containing several raw genetic files, or a (tab or comma) delimited file with sample ids as rows and snps as columns.

This writes a file, predictions.csv to the output_directory (defaults to current directory). This file contains predicted ancestry for each sample.

Direct-to-consumer genetic data file (23andMe, ancestry.com, etc.):

ezancestry predict mygenome.txt

Directory of direct-to-consumer genetic data files or .vcf files:

ezancestry predict /path/to/genetic_datafiles

comma-separated file with sample ids as rows and snps as columns, filled with genotypes as values

ezancestry predict ${HOME}/.ezancestry/data/aisnps/thousand_genomes.KIDD.dataframe.csv

plot

Visualize the output of predict using the plot command. This will open a 3d scatter plot in a browser.

ezancestry plot predictions.csv

generate-dependencies

This command will download all of the data required to build a new nearest neighbors model for a custom set of AISNPs. This command will attempt to download all the .bcf files from The 1000 Genomes Project. If you want to use existing models, see predict and plot.

Without any arguments this command will download all necessary data to build new models and put it in the ${HOME}/.ezancestry/data/ directory.

ezancestry generate-dependencies

Now you are ready to build a new model with build-model.

build-model

Test the discriminative power of your custom set of AISNPs.

This command will build all the necessary models to visualize and predict the 1000 genomes samples as well as user-uploaded samples. A model performace evaluation report will be generated for a five-fold cross-validation on the training set of the 1000 genomes samples as well as a report for the holdout set.

Create a custom AISNP file here: ~/.ezancestry/data/aisnps/custom.AISNP.txt. The prefix of the filename, custom, can be whatever you want. Note that this value is used as the aisnps-set keyword argument for other ezancestry commands.

The file should look like this:

id      chromosome      position_hg19
rs731257        7       12669251
rs2946788       11      24010530
rs3793451       9       71659280
rs10236187      7       139447377
rs1569175       2       201021954

ezancestry build-model --aisnps-set=custom

Python API

See the notebook

Visualization

http://ezancestry.herokuapp.com/

Open in Streamlit

Contributing

Contributions are welcome! Please feel free to create an issue for discussion or make a pull request.