evaluate

Evaluation analyses for PolyGenius

Description

evaluate is an environment that bundles high-level helpers for assessing and comparing fitted PRS models. The public surface includes:

• evaluate$discrimination() measures ranking, separation, and threshold-aware performance.

• evaluate$calibration() measures agreement between predicted and observed outcomes.

• evaluate$compare() compares models pairwise or against a reference model.

• evaluate$incremental() measures added value beyond baseline covariates.

• evaluate$risk.strata() summarizes outcome behavior across score-defined strata.

• evaluate$similarity() measures model-model similarity from variants or evaluated scores.

• evaluate$benchmark() runs a multi-component evaluation bundle.

Evaluation helpers return rich PolyGeniusEvaluation objects. These keep metric result rows separate from plot-support artifacts, diagnostics, and provenance logs so downstream visualization and model-selection helpers can operate on one common result container.

Usage

evaluate

Format

An object of class evaluate (inherits from evaluateEngine, environment) of length 7.

Details

The core contract is models + on: models defines which PRS model or model set is evaluated, and on defines the cohort or genotype context. on may be a PolyGeniusData object or genotype input (GenotypeInfo, GenotypeInfoSet, or a list of GenotypeInfo objects). Genotype input is materialized as a temporary PolyGeniusData object when score-based evaluation needs it; that temporary object is not returned.

Use evaluate when the question is predictive:

• how well does the model perform?

• which candidate model should be selected?

• are two models essentially redundant?

Use associate when the question is inferential:

• what is the effect estimate?

• is an association statistically supported?