Abstract / Summary
Inhalation exposure to nanomaterials represents a critical public health threat, while human experimental data remain ethically unobtainable. Animal studies on indispensable, face fundamental interspecies barriers in anatomy, physiology, and clearance. Allometric scaling—the primary interspecies translation tool—relies on assumptions frequently violated by anatomical, clearance, and metabolic differences between rodents and humans, necessitating integration with mechanistic dosimetry. Overcoming these translational gaps is therefore essential for evidence-based risk assessment, therapeutic optimization, and regulatory protection of human health. Over three decades, diverse modeling approaches have emerged, including mechanistic dosimetry (MPPD, PBPK, CFPD), empirical correlations (ICRP HRTM), geometry models, allometric scaling, and more recently, machine-learning surrogates and hybrid frameworks. However, the rapid proliferation of these models, each with distinct assumptions and applicability domains, has created a critical gap between model development and practical decision-making, lacking systematic guidance for selecting appropriate tools across regulatory assessment, drug development, emergency response, and personalized medicine. This review follows a systematic scoping review methodology (PRISMA-ScR) to map the available evidence and provide a functional taxonomy. This study establishes a comprehensive functional taxonomy categorizing inhalation models into five types: deposition dose estimation, pharmacokinetic simulation, toxicity prediction, rapid surrogates, and integrated frameworks. We then provide a scenario-driven selection framework across fifteen application contexts, evaluating models on predictive performance, computational efficiency, data requirements, mechanistic interpretability, and regulatory acceptability. Key findings reveal that no single model universally outperforms others; optimal selection depends on context. However, the evidence base for model validation remains highly heterogeneous, with most paradigms lacking rigorous external validation against human clinical or imaging data, and uncertainty quantification is often restricted to single-parameter sensitivity analyses rather than full propagation of input variability. Mechanistic models excel in regulatory submissions, empirical models remain indispensable for screening, machine learning enables high-throughput speed, and hybrid frameworks represent a promising direction for integrating complementary strengths, though their regulatory acceptance remains at an early stage of multi-scale processes. However, high cross-validation accuracy reported for ML models does not equate to proven predictive reliability for human outcomes, as external validation against independent biological data is uniformly lacking. We further identify persistent limitations, including data scarcity, inadequate validation against human data, poor representation of vulnerable populations, and insufficient handling of chemical mixtures and nanomaterials. Critically, the unique physicochemical properties of nanomaterials—including size, shape, surface chemistry, aggregation, dissolution, biopersistence, and corona formation—render conventional QSAR and PBPK models inadequate for particulate hazard prediction, necessitating particle-centric paradigms. Complementary New Approach Methodologies (NAMs)—including ALI cultures, PCLS, lung-on-chip, and organoids—offer human-relevant alternatives that, when integrated with in silico modeling, can reduce animal use and improve mechanistic understanding of inhalation toxicity. Future priorities include open-access multi-species databases, explainable AI, patient-specific modeling, systematic uncertainty quantification, and in vitro-in silico coupling. This work bridges the gap between modeling science and practice, offering actionable guidance for researchers, clinicians, and regulators.