Artificial intelligence is becoming an integral part of modern drug development. From predictive modeling to large-scale pharmacokinetic data analysis, AI offers powerful analytical capabilities. However, in clinical pharmacology—where dose selection, exposure interpretation, and benefit–risk decisions directly affect patient safety—AI must be implemented within a scientifically controlled framework.
Cristina Salcianu, PharmD, MSc, a clinical pharmacology scientist who has contributed to regulatory submissions across multiple therapeutic areas, believes AI can meaningfully strengthen early-phase development when guided by qualified PK/PD scientists.
"AI can enable more efficient and structured evaluation of historical PK and safety data," Salcianu explains. "But dose decisions and exposure interpretation must remain under the direction of experienced PK/PD scientists."
AI as a Strategic Analytical Tool in Early Development
Phase I trials—including single ascending dose (SAD), multiple ascending dose (MAD), food-effect, and drug–drug interaction (DDI) studies—require careful integration of preclinical toxicology, mechanistic pharmacology, and projected human exposure. Starting dose selection typically follows structured approaches such as MABEL (Minimum Anticipated Biological Effect Level) or NOAEL-based frameworks, supported by pharmacokinetic projections and safety margins.
AI systems can assist by reviewing historical compounds with similar mechanisms and estimating expected exposure ranges using key metrics such as AUC and Cmax. These tools may help anticipate half-life, predict accumulation at steady state, and optimize PK sampling schedules to accurately characterize absorption and elimination phases.
For example, in the development of a small molecule primarily metabolized by CYP3A4, AI-driven evaluation of prior CYP3A4 substrates can help estimate the magnitude of exposure increase in the presence of strong inhibitors and guide prioritization of formal DDI studies. In oncology programs, where toxicity is often exposure-dependent, reviewing historical exposure–toxicity relationships may help determine whether projected AUC or Cmax levels fall within a range associated with manageable safety profiles.
However, these outputs remain projections—not final determinations.
"AI can highlight patterns in complex datasets," Salcianu notes. "Clinical pharmacologists must assess whether those findings are mechanistically sound and appropriate for the specific compound under development."
Exposure–Response Requires Context, Not Just Correlation
Exposure–response analysis is one of the most promising applications of AI in drug development. Early identification of relationships between exposure metrics—particularly AUC and Cmax—and emerging safety signals or pharmacodynamic biomarkers can support dose optimization strategies.
Yet exposure metrics must always be interpreted within context. Inter-individual variability, organ function, metabolic pathways, therapeutic window considerations, and disease biology all influence whether a given exposure level is clinically acceptable.
"As PK scientists, we do not rely on statistical associations alone," Salcianu explains. "We evaluate whether the exposure–response relationship is biologically plausible and clinically justified."
Having contributed to dose justification and exposure analyses in regulatory submissions, she emphasizes that AI-supported modeling must align with established PK/PD principles and meet regulatory standards for scientific validity.
Traceability, Reproducibility, and Regulatory Standards
In regulated environments, analytical accuracy alone is insufficient. Regulatory decisions require full documentation of underlying data sources, modeling assumptions, training methods, and the reproducibility of results. Without traceability and independent validation, AI-generated outputs cannot support high-stakes regulatory decisions.
"In regulatory science, analytical methods must be transparent, reproducible, and scientifically justified," Salcianu states. "Predictive performance cannot replace explainable evidence when patient safety is involved."
Ongoing performance evaluation and independent validation are essential to ensure models do not introduce bias, particularly in populations historically underrepresented in clinical research.
Ethical Oversight in Vulnerable Populations
Dose optimization for pediatric patients, pregnant individuals, or critically ill populations requires heightened scientific caution. AI systems trained primarily on adult datasets may not adequately account for differences in metabolism, organ function, or disease severity.
"AI should support scientific evaluation," Salcianu emphasizes, "but final dose decisions must remain grounded in pharmacologic evidence and expert oversight."
Leadership in the Era of AI
As AI becomes more embedded in development workflows, Salcianu believes clinical pharmacologists must play a central leadership role in defining how these tools are validated, governed, and responsibly integrated into regulatory frameworks.
AI can improve efficiency and facilitate large-scale PK analysis. However, it does not replace the need to interpret concentration–time profiles, evaluate AUC and Cmax in context, assess elimination kinetics, and determine whether findings are biologically and clinically meaningful.
The future of drug development will not be driven by artificial intelligence alone. It will be shaped by scientific leadership—where advanced analytics operate under the guidance of qualified PK/PD scientists committed to pharmacologic rigor, regulatory integrity, and patient safety.
As artificial intelligence becomes increasingly integrated into drug development, researchers are evaluating how the technology can enhance clinical decision-making without compromising scientific integrity. Cristina Salcianu, a clinical scientist with a PharmD and an MSc in Regulatory Affairs. She has worked across multiple therapeutic areas, applying PK principles to optimize dose selection and regulatory strategy, and believes that from a clinical pharmacology and pharmacokinetics perspective, AI has strong potential to support exposure–response evaluation, dose selection, and early-phase study design.
Cristina believes the next evolution of clinical pharmacology will involve structured integration of AI-assisted modeling within regulatory frameworks, with clinical pharmacologists serving as the final arbiters of scientific validity. For example, she says that in Phase 1, AI could assist by analyzing historical PK and safety datasets to inform starting dose selection (e.g., MABEL vs NOAEL-based approaches), dose-escalation schemes, expected exposure ranges (AUC, Cmax), and sampling strategies aligned with anticipated absorption and elimination kinetics. However, these outputs would still require clinical and regulatory review to ensure alignment with protocol intent and ethical considerations.
Smarter Trial Design Starts with Better Data
Phase I trials—including single ascending dose (SAD), multiple ascending dose (MAD), food-effect, and drug–drug interaction studies—depend heavily on interpreting prior evidence to guide safe and effective dosing strategies. Traditionally, this process requires extensive manual review of historical compounds, pharmacokinetic behaviors, and clinical outcomes. AI has the potential to accelerate this groundwork.
By analyzing large datasets from previous studies, machine-learning tools can help researchers identify logical starting doses, refine escalation pathways, and optimize PK sampling schedules based on anticipated Tmax, half-life, and clearance estimates derived from structurally or mechanistically similar compounds. Instead of replacing scientific reasoning, these tools allow teams to explore scenarios more efficiently during protocol development.
Patient selection may also benefit. In complex therapeutic areas such as oncology, where eligibility criteria often rely on biomarker profiles or layered clinical characteristics, AI can help screen large datasets to identify candidates who meet nuanced inclusion requirements.
Automation has limits—and appropriately so.
"Dose decisions and patient selection must remain grounded in clinical judgment and regulatory guidance," Salcianu notes.
The Trust Problem: Transparency in Regulatory Science
For regulatory authorities evaluating benefit–risk, reproducibility and interpretability of modeling approaches are critical.
Advanced models—particularly deep learning networks—can generate accurate predictions without clearly revealing how those conclusions were reached. For regulators tasked with evaluating safety and efficacy, that opacity presents a challenge.
To build confidence, sponsors are increasingly expected to provide detailed documentation describing training data, development processes, and validation performance. Some organizations have begun incorporating "model cards" into submissions, outlining inputs, assumptions, and evidence supporting algorithmic predictions.
Auditing is becoming equally important. Regular reviews help ensure models are not inadvertently filtering out certain patient populations or introducing bias—issues that must be addressed long before regulatory review.
Transparency, in other words, is becoming as critical as accuracy.
Where AI Can Reduce Risk — Not Just Time
Beyond modeling, AI may also reshape how researchers approach bioequivalence and drug–drug interaction assessments.
By screening historical datasets, AI tools can identify factors commonly associated with failed studies, such as formulation risks or sources of variability. In drug–drug interaction work, integrating in vitro CYP inhibition/induction data, transporter profiles (e.g., P-gp, OATP), and metabolic pathway characterization to prioritize clinically meaningful interaction studies—potentially preventing unnecessary trials.
For instance, when evaluating a small molecule with hepatic metabolism, an AI-driven review of prior CYP3A substrates could help anticipate variability and inform DDI study prioritization.
However, regulatory decisions will continue to rely on well-designed clinical studies and established statistical analyses.
She has contributed to clinical pharmacology components of regulatory submissions, including dose justification and exposure analyses. AI may also assist in early exposure–response exploration by identifying relationships between systemic exposure and emerging safety signals or pharmacodynamic biomarkers. However, such modeling must be carefully validated to ensure biological plausibility and consistency with established PK/PD principles.
AI's role is strategic: improving preparation rather than replacing evidence.
Ethical Boundaries Still Matter
The ethical implications become particularly important when considering dose optimization for vulnerable populations.
Groups such as pediatric patients, pregnant or lactating individuals, and the critically ill are often underrepresented in training datasets. Any AI-supported recommendations must be reviewed by clinical experts and grounded in established pharmacology principles.
"AI should inform discussions, not make autonomous dosing decisions," Salcianu emphasizes.
The Future Is Collaborative Intelligence
Looking ahead, AI may serve as a shared analytical layer across clinical, statistical, and regulatory teams—generating consistent summaries, scenario analyses, and visualizations that improve organizational alignment.
Cristina believes that its greatest long-term impact may lie in data integration, particularly in complex therapeutic areas like oncology and infectious disease, where synthesizing information across programs can accelerate development strategies.
But as tools grow more sophisticated, the defining skills for clinical scientists are unlikely to change.
Strong fundamentals in pharmacokinetics and pharmacodynamics remain essential. Scientists must still interpret concentration–time profiles, evaluate exposure metrics such as AUC and Cmax, assess elimination trends, and determine whether findings make biological and clinical sense.
Scientific judgment remains central, and AI must operate within that framework. Ultimately, the future of drug development is unlikely to be driven by artificial intelligence alone but by the collaboration between advanced analytics and the experts capable of questioning, validating, and contextualizing its output.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
