FDA Review Tracks NAMs from Computational Safety Assessment to Human Cell Models
Researchers at FDA’s Division of Applied Regulatory Science review computational toxicology, drug-disposition models and human-derived experimental systems. The examples range from methods already supporting review to models being developed for new safety questions.
A Regulatory Science Portfolio with Different Uses
A review by researchers in FDA’s Division of Applied Regulatory Science (DARS) examines how NAMs contribute to drug development and regulatory science. Published in Clinical Pharmacology & Therapeutics on 3 September 2026, it covers computational prediction, human-cell assays and microphysiological systems. The authors emphasize scientific plausibility, human biological relevance, standardization and reproducibility when selecting methods to address a regulatory need.
The examples span different stages of development. Some computational approaches already support routine safety assessment, while other systems are being characterized for particular experimental questions. The review describes this range through applications in mutagenicity, drug interactions, cardiac safety, neural activity, respiratory drug transport and intestinal disposition. It presents the authors’ scientific views.
Modeling Drug Effects Across Biological Processes
Quantitative systems pharmacology combines drug exposure, target interactions and physiological responses. One example concerns opioid toxicity and reversal: the models integrate pharmacokinetics, receptor binding, pharmacodynamic effects and cardiopulmonary responses. The review describes how absorption rate, antagonist timing and clearance can influence the predicted course of toxicity and recovery.
Extending such models to include rescue breathing altered predicted naloxone requirements and the time available for intervention. These examples show why the sequence of events matters in a model that follows a rapidly changing physiological response. The intended application guides which biological processes need to be represented and which observations are relevant for checking the predictions.
Computational Assessment of Chemical Safety
The Computational Toxicology Consultation Service has supported hundreds of assessments. For mutagenic drug impurities, DARS uses complementary expert-rule and statistical structure–activity approaches together with scientific review, consistent with the ICH M7 framework. The role of expert assessment is to interpret model outputs, structural alerts and the relevance of supporting data to the chemical under consideration.
The review also discusses read-across for extractables and leachables, drawing on structural similarity, physicochemical properties and biological mechanisms. For nitrosamines, the carcinogenic potency categorization approach considers activating and deactivating structural features when supporting risk assessment and acceptable-intake decisions. These approaches use different evidence structures for different safety questions.
Predicting CYP Inhibition
A collaboration developed 58 structural alerts and five quantitative structure–activity models for inhibition of drug-metabolizing cytochrome P450 enzymes. The models address reversible inhibition of CYP3A4, CYP2C9, CYP2C19 and CYP2D6, together with time-dependent inhibition of CYP3A4. Development used a nonproprietary database of 10,129 unique chemicals drawn from drug-approval packages and published literature.
Cross-validation sensitivities ranged from 78% to 84%, with normalized negative predictivity of 79% to 84%. External validation using 231 approved drugs and 33 nonproprietary chemicals yielded sensitivities up to 75% and normalized negative predictivity up to 80%. The review identifies the need for additional data for other CYP enzymes. The models support prioritization of potential metabolite-mediated drug interactions within the enzyme pathways evaluated.
Cardiac Models and the Conditions of Measurement
Human induced pluripotent stem cell-derived cardiomyocytes contribute to the Comprehensive in vitro Proarrhythmia Assay initiative through measurements such as electrical activity on multielectrode arrays. The review describes their role as supporting evidence alongside ion-channel data and computational modeling. Cardiac microphysiological systems extend measurement to voltage, calcium handling and contraction.
Experiments with 28 drugs across five laboratories found approximately fivefold variability in hERG inhibition measurements. Drug loss through binding to experimental materials can reduce the concentration actually available to the cells and alter estimated potency. The review also discusses the importance of electrophysiological protocols under physiological conditions, including assessment of multichannel effects. These details connect assay performance with the safety interpretation made from its results.
Molecular and Neural Measurements
DARS is exploring proteomic and microRNA measurements to characterize pharmacodynamic responses and potential biomarkers. The review notes the importance of platform performance and variation when deciding how a molecular measurement can support a defined application. The work is part of developing fit-for-purpose evidence for the particular biological response being assessed.
Human neural cultures combining excitatory and inhibitory neurons with astrocytes provide another experimental setting. Multielectrode-array recordings follow network responses to pharmacological perturbation. In a described opioid-receptor example, the agonist DAMGO altered network bursting and naloxone reversed the response. Baseline network activity influences interpretation, making the state of the culture part of the assay’s characterization.
Respiratory and Intestinal Drug Disposition
A small-airway microphysiological system combines primary airway epithelium at an air–liquid interface with microvascular endothelium. It has been used to investigate bronchodilator permeability. The review identifies loss of lipophilic drugs to device materials as an issue when relating the nominal applied concentration to actual exposure.
Caco-2 systems remain useful for passive intestinal permeability and some efflux questions. Transcriptomic comparisons nevertheless identify differences from the human intestine in drug-metabolizing enzymes, esterases and uptake transporters. These differences affect the questions that can be addressed about active uptake and first-pass disposition. Together, the examples demonstrate why biological identity, exposure conditions and measurement performance need to be established for each intended use.