Compute msPAF for each sample in a per-feature data block — compute_mspaf_per

Internal workhorse called by add_mspaf() for each site_id group. Runs in three phases so the (relatively expensive) SSD evaluation is batched across samples rather than called once per (sample × analyte):

chemistry normalisation + ARA shift, vectorised across all rows (.mspaf_adjust());
one vectorised ssdtools::ssd_hp() call per analyte across every sample (.mspaf_add_paf());
CA/IA mixture combination via grouped reductions over (sample, draw, MOA group).

Usage

compute_mspaf_per_sample(
  sample_data,
  ref_table,
  ssd_params,
  min_analytes,
  method,
  guideline_dir,
  ara_enabled = TRUE
)

Arguments

sample_data: Per-feature long-format df (may include co-analyte rows such as pH, DOC alongside toxicant rows).
ref_table: Tibble (analyte, ref_norm) from prep_ref$ref_table.
ssd_params: Tibble from derive_ssd_params() (incl. the fit column).
min_analytes: Minimum analytes required.
method: SSD method (used only for the rare NULL-fit fallback).
guideline_dir: Path to ANZG XLSX folder (NULL-fit fallback only).
ara_enabled: Logical; whether the caller supplied a reference. Controls the ref_source diagnostic (see .mspaf_adjust()).

Value

Tibble with one row per sample that passes min_analytes, columns: sample_id, value, n_analytes_used, n_analytes_imputed, dominant_analyte, max_paf, analyte_pafs, dropped_analytes.