Precision dosing, the tailoring of drug doses to optimize therapeutic benefits and minimize risks in each patient, is essential for drugs with a narrow therapeutic window and severe adverse effects. Adaptive dosing strategies extend the precision dosing concept to time‐varying treatments which require sequential dose adjustments based on evolving patient conditions. Reinforcement learning (RL) naturally fits this paradigm: it perfectly mimics the sequential decision‐making process where clinicians adapt dose administration based on patient response and evolution monitoring. This paper aims to investigate the potentiality of coupling RL with population PK/PD models to develop precision dosing algorithms, reviewing the most relevant works in the field. Case studies in which PK/PD models were integrated within RL algorithms as simulation engine to predict consequences of any dosing action have been considered and discussed. They mainly concern propofol‐induced anesthesia, anticoagulant therapy with warfarin and a variety of anticancer treatments differing for administered agents and/or monitored biomarkers. The resulted picture highlights a certain heterogeneity in terms of precision dosing approaches, applied methodologies, and degree of adherence to the clinical domain. In addition, a tutorial on how a precision dosing problem should be formulated in terms of the key elements composing the RL framework (i.e., system state, agent actions and reward function), and on how PK/PD models could enhance RL approaches is proposed for readers interested in delving in this field. Overall, the integration of PK/PD models into a RL‐framework holds great promise for precision dosing, but further investigations and advancements are still needed to address current limitations and extend the applicability of this methodology to drugs requiring adaptive dosing strategies.

Understanding the pharmacokinetics, safety and efficacy of candidate drugs is crucial for their success. One key aspect is the characterization of absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, which require early assessment in the drug discovery and development process. This study aims to present an innovative approach for predicting ADMET properties using attention-based graph neural networks (GNNs). The model utilizes a graph-based representation of molecules directly derived from Simplified Molecular Input Line Entry System (SMILE) notation. Information is processed sequentially, from substructures to the whole molecule, employing a bottom-up approach. The developed GNN is tested and compared with existing approaches using six benchmark datasets and by encompassing regression (lipophilicity and aqueous solubility) and classification (CYP2C9, CYP2C19, CYP2D6 and CYP3A4 inhibition) tasks. Results show the effectiveness of our model, which bypasses the computationally expensive retrieval and selection of molecular descriptors. This approach provides a valuable tool for high-throughput screening, facilitating early assessment of ADMET properties and enhancing the likelihood of drug success in the development pipeline.

The integration of pharmacokinetic‐pharmacodynamic (PK‐PD) modeling and simulations with artificial intelligence/machine learning algorithms is one of the most attractive areas of the pharmacometric research. These hybrid techniques are currently under investigation to perform several tasks, among which precision dosing. In this scenario, this paper presents and evaluates a new framework embedding PK‐PD models into a reinforcement learning (RL) algorithm, Q‐learning (QL), to personalize pharmacological treatment. Each patient is represented with a set of PK‐PD parameters and has a personal QL agent which optimizes the individual treatment. In the training phase, leveraging PK‐PD simulations, the QL agent assesses different actions, defined consistently with the clinical knowledge to consider only plausible dose‐adjustments, in order to find the optimal rules. The proposed framework was evaluated to optimize the erdafitinib treatment in patients with metastatic urothelial carcinoma. This drug was approved by the US Food and Drug Administration (FDA) with a dose‐adaptive protocol based on monitoring the levels of serum phosphate, which represent a biomarker of both treatment efficacy and toxicity. To evaluate the flexibility of the methodology, a heterogeneous virtual population of 141 patients was generated using an erdafitinib population PK (PopPK)‐PD literature model. For each patient, treatment response was simulated by using both QL‐optimized protocol and the clinical one. QL agents outperform the approved dose‐adaptive rules, increasing more than 10% the efficacy and the safety of treatment at each end point. Results confirm the great potentialities of the integration of PopPK‐PD models and RL algorithms to optimize precision dosing tasks.

Pharmacometrics (PMX) is a quantitative discipline which supports decision-making processes in all stages of drug development. PMX leverages Modeling and Simulations (M&S), which represents a powerful tool to characterize and predict the behavior and the effect of a drug. M&S-based methods, such as Sensitivity Analysis (SA) and Global Sensitivity Analysis (GSA), are gaining interest in PMX as they allow the evaluation of model-informed inference quality. Simulations should be correctly designed to obtain reliable results. Neglecting correlations between model parameters can significantly alter the results of simulations. However, the introduction of a correlation structure between model parameters can cause some issues. Sampling from a multivariate lognormal distribution, which is the typically distribution assumed for PMX model parameters, is not straightforward when a correlation structure is introduced. Indeed, correlations need to respect some constraints which depend by the CVs (i.e., coefficients of variation) of lognormal variables. In addition, when correlation matrices have some unspecified values, they should be properly fixed preserving the positive semi-definiteness of the correlation structure. In this paper, we present mvLognCorrEst, an R package developed to address these issues. The proposed sampling strategy was based on reconducting the extraction from the multivariate lognormal distribution of interest to the underlying Normal distribution. However, with high lognormal CVs, a positive semi-definite Normal covariance matrix cannot be obtained due to the violation of some theoretical constraints. In these cases, the Normal covariance matrix was approximated to its nearest positive definite matrix using Frobenius norm as matrix distance. For the estimation of unknown correlations terms, the graph theory was used to represent the correlation structure as weighed undirected graph. Plausible value ranges for the unspecified correlations were derived considering the paths between variables. Then, their estimation was performed by solving a constrained optimization problem. Package functions are presented and applied on a real case study, that is the GSA of a PMX model that has been recently developed to support preclinical oncological studies. mvLognCorrEst package is an R tool to support simulation-based analysis for which sampling from multivariate lognormal distributions with correlated variables and/or estimation of partially defined correlation matrix are required.

Background: Large Language Models (LLMs) have driven significant advances in artificial intelligence (AI), with transformative applications across numerous scientific fields, including biomedical research and drug development. However, despite growing interest in adjacent domains, their adoption in pharmacometrics, a discipline central to model-informed drug development (MIDD), remains limited. This study aims to systematically explore the potential role of LLMs across the pharmacometrics workflow, from data processing to model development and reporting. Methods: We conducted a comprehensive literature review to identify documented applications of LLMs in pharmacometrics. We also analyzed relevant use cases from related scientific domains and structured these insights into a conceptual framework outlining potential pharmacometrics tasks that could benefit from LLMs. Results: Our analysis revealed that studies reporting LLMs in pharmacometrics are few and mainly limited to code generation in general-purpose programming languages. Nonetheless, broader applications are theoretically plausible and technically feasible, including information retrieval and synthesis, data collection and formatting, model coding, PK/PD model development, support to PBPK and QSP modeling, report writing and pharmacometrics education. We also discussed visionary applications such as LLM-enabled predictive modeling and digital twins. However, challenges such as hallucinations, lack of reproducibility, and the underrepresentation of pharmacometrics data in training corpora limit the actual applicability. Conclusions: LLMs are unlikely to replace mechanistic pharmacometrics models but hold great potential as assistive tools. Realizing this potential will require domain-specific fine-tuning, retrieval-augmented strategies, and rigorous validation. A hybrid future, integrating human expertise, traditional modeling, and AI, could define the next frontier for innovation in MIDD.

Precision dosing aims to optimize and customize pharmacological treatment at the individual level. The integration of pharmacometric models with Reinforcement Learning (RL) algorithms is currently under investigation to support the personalization of adaptive dosing therapies. In this study, this hybrid technique is applied to the real multiobjective precision dosing problem of givinostat treatment in polycythemia vera (PV) patients. PV is a chronic myeloproliferative disease with an overproduction of platelets (PLT), white blood cells (WBC), and hematocrit (HCT). The therapeutic goal is to simultaneously normalize the levels of these efficacy/safety biomarkers, thus inducing a complete hematological response (CHR). An RL algorithm, Q‐Learning (QL), was integrated with a PK‐PD model describing the givinostat effect on PLT, WBC, and HCT to derive both an adaptive dosing protocol (QL_pop‐agent) for the whole population and personalized dosing strategies by coupling a specific QL‐agent to each patient (QL_ind‐agents). QL_pop‐agent learned a general adaptive dosing protocol that achieved a similar CHR rate (77% vs. 83%) when compared to the actual givinostat clinical protocol on 10 simulated populations. Treatment efficacy and safety increased with a deeper dosing personalization by QL_ind‐agents. These QL‐based patient‐specific adaptive dosing rules outperformed both the clinical protocol and QL_pop‐agent by reaching the CHR in 93% of the test patients and completely avoided severe toxicities during the whole treatment period. These results confirm that RL and PK‐PD models can be valid tools for supporting adaptive dosing strategies as interesting performances were achieved in both learning a general set of rules and in customizing treatment for each patient.