iconic - Causal Model Selection with Genetic Instruments and Negative Controls

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.