Artificial Intelligence in Model‐Informed Drug Development: Augmentation, Acceleration, and Regulatory Preparation Authors: Hao Zhu, Ruihao Huang, Menglun Wang, Lixia Zhang, Hanrui Zhang, Yutao Gong, Rajnikanth Madabushi, Qi Liu Published: 2026 ABSTRACT ABSTRACTThe convergence of MIDD and AI represents a natural evolution from established quantitative frameworks toward a broader, more powerful computational toolbox that can address persistent MIDD challenges such as high‐dimensional data integration, complex biology, and operational constraints in evidence generation. AI can both enhance traditional MIDD through hybrid approaches and extend MIDD's horizon to include various AI models. Recent publications of guidance demonstrate the agency's preparation on the acceptance of AI in MIDD. Background Model Informed Drug Development (MIDD) is the “use of computational modeling and simulation (M&S) methods that integrate nonclinical data, clinical data, prior information, and knowledge (e.g., drug and disease characteristics) to generate evidence” [1]. Over decades, MIDD approaches have been applied to inform dose selection, support indication extrapolation, enhance clinical trial design, identify appropriate patients or patient subgroups, and facilitate approvals of different formulations and routes of administration. The value of MIDD in improving the efficiency of drug development and regulatory decision‐making has been broadly recognized [2, 3]. As drug development programs become increasingly complex and data‐intensive, the demand for MIDD continues to grow. For instance, therapeutic modalities become more complex. Improvements beyond traditional quantitative approaches may be necessary to fully characterize exposure–response relationships, disease progression, and treatment effects. Additionally, integrating diverse data sources—including clinical outcomes, digital health technologies, and biomarkers—requires advanced modeling frameworks and improved methodologies. Furthermore, expectations for MIDD have risen substantially, with stakeholders seeking greater rigor and reliable performance to support critical decision‐making. Artificial intelligence (AI) has demonstrated remarkable success in analyzing complex and large‐scale datasets across fields. Its advanced capabilities in pattern recognition, prediction, and data integration offer significant promise for modernizing MIDD. Incorporating AI into MIDD is expected to substantially expand the impact of MIDD by enhancing model development and enabling more robust decision‐making. The integration of AI with existing MIDD methodologies has the potential to transform MIDD's overall capabilities. Ultimately, this synergy between AI and MIDD could accelerate drug development, optimize therapeutic strategies, and contribute to improved patient outcomes and overall public health (Figure 1). Several topics related to the use of AI in MIDD will be discussed in detail in the next several sections. The Synergy between artificial intelligence (AI) and model informed drug development (MIDD) expects to improve new drug development and enhance patient care. Expanding Computational Framework The inclusion of AI in MIDD expands its computational framework by introducing advanced data‐driven methods that complement traditional modeling approaches. AI enhances the ability of MIDD to process large, complex datasets and provides potential solutions to technical challenges of existing MIDD tools, thereby broadening the analytical scope and technical capabilities of the MIDD computational framework. Traditional MIDD Computational Framework Over decades, MIDD has built a strong toolkit of quantitative models to support decision‐making. Population pharmacokinetic models describe concentration–time profiles and variability to inform dose selection and adjustments. Physiologically based pharmacokinetic models add physiological details to predict drug–drug interactions, absorption, and outcomes in special populations. Exposure–response models link exposure to biomarkers, efficacy, and safety to clarify benefit–risk. Disease progression models capture natural history and variability over time, while QSP models integrate mechanistic biology with clinical data to explain system‐level therapeutic effects. Despite their impact, continued innovation is needed to improve the performance, scalability, and practical use of these models. Integration of AI With Traditional MIDD Computational Framework Integration of AI with existing MIDD tools provides great potential to improve efficiency in modeling and enhance the power of MIDD. A good example of an AI‐assisted MIDD approach is AI‐enhanced covariate selection using stochastic gates [4]. In a typical workflow, a broad set of candidate covariates is first passed into a neural network model that is trained to explain inter‐individual variability in pharmacokinetic parameters of interest. Compared with traditional stepwise covariate selection, stochastic gates offer several practical advantages. First, they perform multivariable selection in a single optimization, which reduces the “path dependence”. Second, because all covariates are evaluated jointly, the approach can better handle correlated predictors and account for nonlinearities and interactions. Finally, it is often faster and more stable than repeatedly re‐estimating the model across stepwise iterations. AI Models as Additional MIDD Tool AI can also expand the MIDD toolbox with various modeling approaches—for example, unsupervised learning to detect patterns, tree‐based ensembles (e.g., random forests or gradient boosting) to identify subgroups, neural networks to capture highly nonlinear relationships and complex interactions, and large language models to extract insights from text‐heavy sources. These additions can add significant value to MIDD by expanding what modeling teams can learn from complex and high‐dimensional data yet maintaining the mechanistic and statistical foundations. AI methods help integrate diverse preclinical and clinical evidence streams, uncover nonlinear relationships and latent subgroups, and generate predictive features that improve exposure–response, disease progression, or patient stratification models. AI can also accelerate routine MIDD workflows, such as data cleaning, covariate discovery, and simulation scenario generation. When paired with strong governance, AI‐enhanced MIDD can support better trial design, more precise dosing recommendations, earlier risk detection, and faster, evidence‐based decisions across the development lifecycle. Impactful Applications Although the integration of AI into MIDD is still in its early stages, its value has already been demonstrated through a range of impactful applications. The examples presented here illustrate only a subset of current uses within the expanding roles of AI in MIDD. As methodologies mature and experience grows, the scope and impact of AI in MIDD are expected to continue evolving and broadening. New Approach Methodologies AI significantly enhances the value of new approach methodologies (NAMs) by enabling the integration, analysis, and interpretation of complex biological and toxicological data in an accelerated fashion. NAMs—such as in vitro assays, organ‐on‐chip systems, high‐throughput screening, and computational models—generate large, multidimensional datasets. AI can identify patterns, predict toxicity outcomes, and model biological pathways from these data, improving hazard identification and dose–response assessment. This strengthens the scientific confidence and regulatory judgment of NAM‐based evidence. In addition, AI helps bridge gaps between experimental systems and human biology by supporting in vitro–in vivo extrapolation (IVIVE). By accelerating data interpretation and enabling predictive modeling, AI reduces reliance on animal testing, shortens evaluation timelines, and lowers costs while maintaining or improving public health protection. Ultimately, AI amplifies the impact of NAMs by transforming them from data‐generating platforms into decision‐ready tools for risk assessment. Enhanced Clinical Drug Development AI can be used in MIDD to make drug development and regulatory decisions more data‐driven and timely. Practically, AI can (1) identify and enrich patient subpopulations, (2) predict efficacy, safety, and progression risks by learning patterns from high‐dimensional data, (3) support dose selection and individualization, (4) optimize trial design through better stratification, and (5) integrate heterogeneous evidence to strengthen benefit–risk assessments. Liu et al. reported a case on the use of AI to develop a implementable alternative to a biomarker test essential to identify patients matching trial participants but not commercially available in the United States [5]. In the clinical trial for anakinra, patients at risk of severe respiratory failure were enriched with the baseline biomarker of soluble urokinase plasminogen activator receptor (suPAR) ≥ 6 ng/mL. The US Food and Drug Administration (FDA) team used two AI approaches to develop a scoring rule—based on routinely available clinical characteristics and labs—to identify patients likely with suPAR ≥ 6 ng/mL, and then externally validated using data from an independent trial. This case illustrates several ways AI can add value: (1) operationalizing the trial findings; (2) patient selection to support benefit–risk; and (3) enabling evidence‐based method to support patient access to therapy. The authors note this was the first time FDA used AI to determine a suitable population for a drug therapy, underscoring AI's emerging role as a practical tool for precision medicine and drug development decision support. Regulatory Preparation Appropriate application of AI in MIDD increasingly depends on clear regulatory expectations and establishment of standards. In January 2025, FDA issued a draft AI guidance, which recommends a risk‐based credibility framework for AI models used to generate information or data that may support regulatory decisions [6]. The guidance emphasizes defining the question of interest and the model's context of use, assessing model risk as a function of model influence and decision consequence, and then scaling the required credibility evidence to that risk. The guidance also encourages early and direct interaction with the agency on appropriate validation to establish the credibility of the AI model under a specific use condition. In parallel, the ICH M15 guideline establishes a harmonized assessment framework and common terminology for MIDD evidence and explicitly includes AI among other MIDD approaches [1]. Consistent with the FDA framing, M15 anchors evidence assessment in the same core elements (question of interest, context of use, model influence, consequence of wrong decision, and derived model risk) and makes clear that model evaluation should be commensurate with model risk, supporting more consistent global acceptance of AI‐enabled MIDD evidence. Complementing these frameworks, FDA and EMA jointly published the “Guiding Principles of Good AI Practice in Drug Development”, signaling a shared regulatory expectation [7]. The principles reinforce expectations for a clear context of use, risk‐based validation and oversight, strong data governance and traceable documentation, system‐level performance assessment, and life‐cycle management. Finally, the Senate Health, Employment, Labor, and Pensions (HELP) Committee modernization paper [8] adds a policy lens that aligns with these technical frameworks: it highlights AI's potential to improve product development, review, and post‐market activities, while calling for a focused, risk‐based regulatory approach, clearer expectations and organizational accountability, international harmonization, and a stronger AI expertise pipeline within FDA—signals that future readiness for AI‐enabled MIDD will require both rigorous evidence standards and sustained institutional capacity. Conclusion In conclusion, the convergence of MIDD and AI represents a natural evolution from established quantitative frameworks toward a broader, more powerful computational toolbox that can address persistent MIDD challenges such as high‐dimensional data integration, complex biology, and operational constraints in evidence generation. AI can both enhance traditional MIDD through hybrid approaches and extend MIDD's horizon to include various AI models. A series of publications on AI and MIDD guidance in recent years demonstrate the agency's preparation on the acceptance of broadened use of AI in MIDD. Nevertheless, some AI approaches still face various challenges, such as hallucination, generalizability, data drifting, interpretability and explainability, requiring further improvement. Multiple elements including technical improvement, standard building, international harmonization, and FDA capacity‐building together can establish a practical path for credible and decision‐relevant AI‐enabled MIDD adoption. Funding The authors have nothing to report. Disclosure This manuscript reflects the views of the authors and should not be construed to represent those of the FDA or the United States Government. As an Associate Editor for Clinical and Translational Science, Qi Liu was not involved in the review or decision process for this paper. Conflicts of Interest The authors declare no conflicts of interest.