A simulated crossed study: 60 variants classified into the five ACMG/AMP
tiers (B benign, LB likely benign, VUS uncertain significance,
LP likely pathogenic, P pathogenic) under 4 prompt variants and 3
models, 3 replicates each (2160 calls, 12 of them unparseable). The data
were generated from a latent-variable model with known variance
components (see attr(acmg_demo, "simulation") and
data-raw/make_demo_data.R); reference is the true tier. No real
variants or models are involved.
Format
An msa_runs data.frame with columns block, item_id,
prompt_id, model_id, rep, value, reference.
Examples
head(acmg_demo)
#> LLM runs: 6 calls, 1 items, 1 prompt variants, 2 models, 1 block(s)
#> missing values: 0
#> block item_id prompt_id model_id rep value
#> 1 VAR001 concise model-a 1 B
#> 1 VAR001 concise model-a 2 B
#> 1 VAR001 concise model-a 3 B
#> 1 VAR001 concise model-b 1 B
#> 1 VAR001 concise model-b 2 B
#> 1 VAR001 concise model-b 3 B
table(acmg_demo$value, useNA = "ifany")
#>
#> B LB LP P VUS <NA>
#> 324 438 385 514 487 12