Skip to main content

KING / Kinship

This module implements KING-based relatedness screening. Flower stage iterative_king drives kinship estimation: clients send sample-based PLINK 1.9 chunks; the server reconstructs, merges, and runs PLINK 2 --make-king-table.

Client flow​

handle_iterative_king checks whether the client already has chunk_data. If not, it recreates sample chunks:

partition_config = {
    "chunk_size": chunk_size,
    "partition_by": "samples",
}
client.partition_data(partition_config)

Each call sends one chunk and increments current_chunk_idx. After all chunks are sent, the client finalizes local KING accumulators and can filter samples above king_threshold.

Server flow​

run_server_king:

  1. Decodes each received chunk array.
  2. Rebuilds temporary .bed/.bim/.fam files from size metadata.
  3. Merges PLINK datasets.
  4. Runs PLINK --het.
  5. Runs PLINK2 --make-king-table.
  6. Parses .kin0 output and returns sample pairs, partial phi values, and heterozygosity-derived information.

Outputs​

Server-side temporary files and KING outputs are written into a session directory under the strategy output_dir. Clients also persist king_accumulator_state.pkl and king_results_<client_id>.txt under their configured log directories.