SPCAncestry

A Python package for supervised global population ancestry inference from SNP data using stacking

Installation

SPCAncestry will soon be available through PyPI and installable using the following command:

pip3 install spcancestry

Usage

SPCAncestry assumes the user has SNP training data, in either PLINK, VCF, or Hail MatrixTable, and reliable global population ancestry labels for all the samples in the training data. We compute PCs using the training data, then these PCs (Xs) together with the population labels (Y) are used to train the model. The test SNP data is projected onto the PCs computed using the training data, and then we use the trained model to infer population ancestry. Below are the steps on how you can use the HGDP1KG data, provided with this package, for ancestry inference.

1. Read the training and test datasets

import spcancestry

path = '/path/to/spcancestry/test_data/hgdp1kg'
inref_mt = spcancestry.Read(file=f'{path}/hgdp1kg_truth.bed', qc=False).as_matrixtable()
input_mt = spcancestry.Read(file=f'{path}/hgdp1kg_unknown.fam', qc=False).as_matrixtable()

2. Intersect the two datasets, compute PCs using training data, and project test data onto training PC space

scores_df, colnames = spcancestry.PCProject(ref_mt=inref_mt, data_mt=input_mt,
                                            ref_info=f'{path}hgdp_1kg_truth_labels.txt').run_pca_projection()

3. Infer global population ancestry using spcancestry stacking

spcancestry_infered = spcancestry.infer_ancestry(scores_df, colnames)

Copyright and License

SPCAncestry is generously distributed under the MIT License