Evidence brief: Empirically calibrated simulations reveal the limits of phenotypic clustering algorithms for biodiversity assessment in data-scarce crops

Abstract

Clustering algorithms are widely used for phenotypic characterization and germplasm management, particularly in data-scarce crops such as neglected and underutilized species (NUS) that lack genomic resources. However, their performance under biologically realistic conditions remains poorly understood. Standard clustering methods commonly applied in crop research often assume distinct, isotropic, and homogeneous clusters, assumptions rarely satisfied in real-world phenotypic datasets. We developed a flexible and empirically calibrated simulation framework, using phenotypic data from West African fonio (Digitaria exilis), to benchmark the performance of eleven clustering algorithms under both idealized and realistic scenarios. Our simulations integrated heterogeneous trait distributions (normal, gamma), strong inter-trait correlations (up to r = –0.84), heteroscedasticity, and moderate population structure (mean Pst = 0.16 ± 0.001, achieved through iterative calibration). Each scenario was replicated 100 times, with clustering accuracy evaluated using external (ARI, NMI) and internal (Silhouette, Davies–Bouldin) validation metrics under standardized conditions. The results revealed consistently poor algorithm performance under realistic conditions (e.g., ARI < 0.07), including for widely used methods in Neglected and Underutilized Species (NUS) research such as K-means, GMM, and PAM. Notably, conventional validation metrics failed to detect biologically meaningful structure revealed by geometric diagnostics, highlighting a critical methodological limitation. Performance markedly improved under idealized conditions, validating our simulation framework. These findings highlight the risk of overinterpreting clustering outputs from weakly structured phenotypic datasets and expose key limitations in current biodiversity analysis practices, particularly those guiding plant genetic resource conservation programs. We provide an open-source R-based diagnostic tool, with parameter specifications to assist practitioners in selecting reproducible and interpretable clustering approaches for germplasm management and biodiversity assessment in data-scarce crops.

Introduction

Neglected and underutilized species (NUS) constitute a critical reservoir of agricultural biodiversity, essential for global food security, sustainable agriculture, and adaptation to climate change and ecological instability [2,3]. Despite their resilience and adaptive potential, many NUS—such as fonio (Digitaria exilis), a cereal cultivated in ecologically marginal environments—remain underrepresented in genomic databases due to limited funding and molecular resources [1,4]. Consequently, scientists and breeders often rely on phenotypic data as a primary resource for exploring genetic structure, managing germplasm collections, and informing breeding strategies [5].

Yet, phenotypic analyses face fundamental challenges. These include subtle population differentiation and complex data distributions—characteristics often overlooked in methodological practice [6,7]. Fonio epitomizes this issue: historically vital to marginalized West African communities, it thrives in agroecological niches unsuitable for major cereals [1,8]. In the absence of genomic profiling, phenotypic clustering is often the only available framework guiding biodiversity assessments and conservation interventions [9].

Crucially, the interpretability of phenotypic clustering depends on assumptions that are rarely met in NUS datasets—namely, that clusters are discrete, isotropic, and internally homogeneous [10,11]. In practice, phenotypic traits in NUS often exhibit overlapping distributions, strong inter-trait correlations, heteroscedasticity, and low genetic differentiation (Pst ≈ 0.1–0.15), reflecting continuous gene flow and mixed mating systems [12–14].