Simulation Benchmark Scripts for Phenotypic Clustering in Neglected Crops

This repository provides a reproducible R-based simulation and benchmarking framework for evaluating the performance and limits of phenotypic clustering algorithms in biodiversity and agrobiodiversity research.

The toolkit includes modules for generating realistic multivariate phenotypic datasets, applying a wide range of clustering algorithms, computing multiple validation metrics (e.g. ARI, NMI, Silhouette, phenotypic Pst/Fst), and producing publication-ready visualizations such as PCA and UMAP.

Although the simulation framework is empirically calibrated using fonio (Digitaria exilis) phenotypic data, all scripts are fully generic and can be adapted to any plant or animal species for which phenotypic trait data are available.

The framework is intended to support methodological studies, detectability analyses, and structure exploration in data-scarce systems, particularly for neglected and underutilized species (NUS), but is not limited to this context.