Reconsidering principal component analysis in neurodevelopmental studies: A call for advanced frameworks

Bodin et al. (2025) provide valuable insights into neurodevelopmental vulnerability by examining radiofrequency electromagnetic fields (RF‑EMF) exposure during early life. Their integrative design, combining whole-body exposure with endpoints such as neonatal brain proteomics, BDNF expression, synaptogenesis, and oxidative stress, offers a comprehensive framework for developmental neurotoxicology. However, interpretation of proteomic clustering relies heavily on principal component analysis (PCA), a linear technique ill-suited for high-dimensional datasets dominated by non-linear dependencies and strong inter-feature correlations. PCA plots (Fig. 3) illustrate group separation, yet variance explained (55%) and clustering stability remain underreported, raising concerns about robustness and biological interpretability, particularly given only ten differentially expressed proteins. To enhance inference, future studies should adopt biologically meaningful feature selection and advanced frameworks such as Feature Agglomeration and Highly Variable Feature Selection, alongside non-parametric correlation measures such as Spearman's rho and Kendall's tau. These strategies will improve reproducibility, uncover mechanistic patterns, and strengthen translational relevance for neurodevelopmental research.