teaguesterling/plinking_duck v0.4.0

v0.4.0 — Structured genotypes, unified variants, flexible companions

on GitHub

What's New

Genotype Output Modes

• genotypes='struct' — STRUCT with named fields per sample (variant orient) or per variant (sample orient). Like 'columns' but as a single logical column with named field access (genotypes.SAMPLE1).
• genotypes='counts' — STRUCT(hom_ref, het, hom_alt, missing) using PgrGetCounts for zero-decompression fast counting. Orders of magnitude faster than reading individual genotypes for large cohorts.
• genotypes='stats' — Extends counts with derived statistics: n, af, maf, missing_rate, carrier_count, het_rate. Cross-validates against plink_freq and plink_missing.

Unified Variants Parameter

Flexible variant selection for read_pfile and read_pgen:

• Single index: variants := 0
• Single rsid: variants := 'rs123'
• CPRA string: variants := '1:10000:A:G'
• CPRA struct: variants := {chrom: '1', pos: 10000, ref: 'A', alt: 'G'}
• Index range: variants := {start: 0, stop: 100}
• Identifier range: variants := {start: 'rs1', stop: 'rs3'}
• Lists of any of the above

Flexible Companion Sources

• Parquet auto-discovery: .pvar.parquet and .psam.parquet are preferred over text formats when present (configurable via SET plinking_use_parquet_companions)
• Arbitrary sources: pvar := and psam := accept CSV files, DuckDB tables, and views
• read_pvar / read_psam accept any DuckDB-readable source as positional argument

Configuration & Filtering

• plinking_max_threads config option: cap parallel scan threads across all functions
• plink_glm p_threshold: scan-time p-value filtering for GWAS output reduction

Full Changelog

v0.3.0...v0.4.0