Start with the package README for installation and examples. These documents describe the supported workflow; qualification reports retain the versions, fixtures, and limits of their original measurements.
Authoritative contracts
Breeding simulation building blocks: supplied parentage, selection, mating, incremental packed cohorts, persistent genomic truth, correlated records, sampling and weighted pools with overlap covariance. Includes a small local workflow and qualification limits.
Packed simulation workflow and native architecture: reference, founder, and pedigree roles; HAP/BED/BIM/FAM; identity alignment; ownership, lifetime, deterministic batching/threading, and memory bounds.
HAPNEST-informed founder model: donor and phase sampling, mutation filtering, chromosome/seed identity, and attribution.
Pedigree inheritance and byte oracle: parental transmission, crossover boundaries, maps, and public packed semantics.
VCF import: supported records, selection, skip/error policy, allele orientation, and map interpolation.
Marker-specific phenotype architecture: causal probabilities, conditional mixture proportions, variance weights, metadata alignment, and returned truth.
Controlled gene sets: supplied SNP?gene mappings, exact causal composition, null controls, replicates, overlap and RNG semantics.
The installed R help is authoritative for public arguments and return values: reference, VCF import, founders, pedigree simulation, phenotypes, gene sets, printing, pedigree domains, and record workloads.
Pedigree record generators supply deterministic solver workloads. Their latent parent-average values are not exact draws from a pedigree relationship covariance. Genotype-based phenotype simulation is a separate workflow.
Examples and development
Composable breeding workflow uses a tiny synthetic phased panel, native selection/meiosis, genomic records, parent retention and matched or mismatched pools. No external predictor is fitted; an external caller can supply its predictions for selection.
1000 Genomes chromosome 22 is a complete GRCh37 example that downloads official reference/map inputs when absent.
End-to-end phenotypes constructs a small synthetic local fixture. Run after
library(gsim)with qgg installed; its illustrative LD scores are not estimates from real 1000G data.Controlled gene sets uses existing simulated causal SNP truth and an explicitly synthetic SNP?gene mapping; no downloads.
Repository instructions describe installation, explicitly loading an isolated installed package for focused tests, and generated Rd.
Copyrights and provenance retain HAPNEST attribution and applicable gbits/gmat notices.
qgg is optional generally and used to construct Glist objects. Supported BED Glist accumulation is standalone and bounded; see the phenotype contract for selection, storage, memory and custom-reader limits. HAP phenotype input and gsuite-generated Glist compatibility are not established.
Qualification and reproducibility
Reports retain the scope and revision of their own qualification. Historical recommendations are scoped to their measured revisions.
| Evidence | Scope and reproduction |
|---|---|
| Bounded Glist accumulation | 0.15.0 focused parity, malformed-input and bounded resource checks; script. |
| Genotypes to phenotypes | Small local fixture, qgg dosage/ID parity and phenotype controls; example and test. |
| Founder chromosome identity | Exact byte/audit comparisons and intentional replacement of block-index streams; tests. |
| Pedigree meiosis | Byte oracle, Mendelian consistency, and uncertainty-based relationship experiment; tests. |
| Deferred HAPNEST comparison | Pinned external oracle, prerequisites, proposed fixture and acceptance rules; comparison has not passed or run. |
| Production chromosome 22 benchmark | Real 1000G inputs, checksums, measured revision/environment and timing uncertainty; script. |
| Batched/parallel founder benchmark | Real 1000G prepared panel, plus a separately labelled synthetic 0.12 guard; script. |
| Parallel scaling diagnosis | Synthetic 49,000-marker panel, not 1000G LD; counter limitations and removed external fixture are explicit; same benchmark script. |
The 50,000-animal record-workload script reports construction sizes, times, and checksums, not covariance recovery or solver performance. Current tests, both examples, and all benchmark/qualification scripts remain available. Benchmarks and broad qualification require authorization; nothing in this index claims they were rerun during documentation cleanup.