projection.md

Projection Mode

The adamixture-project command estimates ancestry proportions for a new set of samples using a pre-trained, fixed allele-frequency matrix P. P is never updated — only Q is optimized. This is useful for placing new individuals onto an existing ancestry model without re-training.

Note

The target genotype data must have exactly the same SNPs (same order, same number of rows) as were used when training the original P matrix.

$ adamixture-project \
    --data_path new_samples.bed \
    --p_path trained_model/results.8.P \
    --save_dir projection_out/ \
    --name projected

Note

K is inferred automatically from the number of columns in the P matrix — no need to pass -k.

Key arguments

Argument	Description
--data_path	Path to target genotype data (BED, VCF or PGEN).
--p_path	Path to pre-trained P matrix (M × K, whitespace-delimited).
--save_dir	Output directory.
--name	Run name prefix for output files.
--device	Computation device: cpu, cuda, or mps (default: cpu).

The output is a single .Q file (one row per sample, K columns). All --plot, --labels, --labels2, --labels3, and --colors flags work identically to the main adamixture command.