# ------------------------------------------------
# CITATION.cff file created with {cffr} R package
# See also: https://docs.ropensci.org/cffr/
# ------------------------------------------------
 
cff-version: 1.2.0
message: 'To cite package "iconic" in publications use:'
type: software
license: MIT
title: 'iconic: Causal Model Selection with Genetic Instruments and Negative Controls'
version: 0.99.3
doi: 10.32614/CRAN.package.iconic
abstract: Provides a model selection workflow for causal inference with genetic instruments
  and negative controls in observational omics data. The package fits eight estimators
  of the natural direct and indirect effects (NDE/NIE), diagnoses which are valid
  for the user's data via instrument-strength checks and negative-control validity
  screens (A1, A2, A2'), stress-tests them against confounding and pleiotropy violations,
  and recommends the estimator most likely to be unbiased. A hybrid generative texture
  model (a torch GAN for sample-level structure and a Gaussian copula for the mediator
  panel) lets the sensitivity analysis mirror the marginal and joint structure of
  the user's cohort. The modular API comprises iconic_data() to standardize data,
  iconic_diagnose() to assess estimator eligibility, iconic_estimate() to fit all
  eligible estimators, iconic_sensitivity() to sweep instrument exogeneity, iconic_recommend()
  to rank estimators by identification strength and robustness, and iconic_prospect()
  to plan future studies. infer_confounding() estimates held-fixed confounding parameters
  from the data. Helper functions construct exposure instruments (polygenic scores
  from GWAS summary statistics or PGS Catalog scoring files), mediator instruments
  (cis-eQTL scans and elastic-net predicted-expression composites), and negative-control
  panels (principal components of residualized omics matrices), and data can be imported
  directly from SummarizedExperiment objects. Time-to-event outcomes are supported
  on the Cox log-hazard-ratio and restricted-mean-survival-time scales. Binary outcomes
  are supported on the log-odds-ratio (logistic two-stage predictor-substitution outcome
  stage) and risk-difference (linear probability model) scales. The torch package
  is required for the generative texture model (sensitivity and prospective analysis);
  all other functionality works without it.
authors:
- family-names: Bresnahan
  given-names: Sean T.
  email: seantbresnahan3@gmail.com
- family-names: Xiong
  given-names: Charis
  email: charis.xiong@gmail.com
repository: https://biocstaging.r-universe.dev
repository-code: https://github.com/sbresnahan/iconic
commit: e89f4936c01c4624c802e5b601a97f3a7a13384e
url: https://seantbresnahan.com/iconic/
date-released: '2026-08-19'
contact:
- family-names: Bresnahan
  given-names: Sean T.
  email: seantbresnahan3@gmail.com