Add principal component export for bg and sample in ancestry-predict.py

[tstannius]

To enable creation of PCA plots in MultiQC, I have modified the script to export csv files containing the predictions and PC's.

These changes affect the way ancestry-predict is called:

ORIGINAL

python scripts/ancestry-predict.py --labels scripts/ancestry-labels-1kg.tsv --samples $MY_SAMPLES/*.somalier --backgrounds 1kg-somalier/*.somalier > sample-ancestries.txt

Outputs:

• sample-ancestries.txt
• thousandG.npy

NEW

python somalier/scripts/ancestry-predict.py --labels somalier/scripts/ancestry-labels-1kg.tsv --samples $MY_SAMPLES/*.somalier --backgrounds 1kg-somalier/*.somalier

Shows the plot

python3 somalier/scripts/ancestry-predict.py --labels somalier/scripts/ancestry-labels-1kg.tsv --samples data/*.somalier --backgrounds 1kg-somalier/*.somalier --plot mydir/mysample.pdf

Outputs in dir: "mydir":

• mysample.ancestry_pcs.csv
• mysample.ancestry_prediction.csv
• mysample.pdf
• mysample.thousandG.npy

Considerations

• The script has two modes: show the plot or generate+save files. This depends on specifying a save-location for the plot.
• I chose .csv as output format since this reflects the MultiQC/Somalier pull-request.
• If the user wants to use their own ancestry-labels, one could modify the "thousandG.npy"-filename to reflect this.

[brentp]

this looks good to me. if you'd add a --prefix argument that defaults to something like "somalier-ancestry" use that, and then update the README.md as needed, I think this would be ready.

[tstannius]

Done and done.

However, a co-worker suggested some improvements that I will add.

• Separation of PC's and predictions into separate dataframes.
• Only save files, if a name for the plot is supplied (otherwise assume user just wants to have a quick look at the somalier PCA plot and don't save anything).
• Save files to same location as plot, e.g. subfolder or working dir.
• Remove prefix so there's no conflict with plot name, save location and prefix.

[tstannius]

Is there anything that needs changing @brentp or should I consider the edits final? Then I will continue working on extending your MultiQC/Somalier PR to accommodate the ancestry-prediction :-)

[brentp]

thanks for the reminder. FYI, I am working on getting this functionality directly into somalier binary. it's working but lacking a few features it outputs the full text for background and query samples, including the PCs and the confidence for each ancestry. that will probably be a better place to start.