Embracing Model-Informed Drug Discovery: The Path to Better Clinical Outcomes

Traditional drug discovery often operates on an empirical, trial-and-error basis. Compounds move through preclinical testing and into clinical trials with limited mechanistic understanding of how they will behave in humans. This approach is not only slow and expensive, but also prone to costly late-stage failures when drugs fail to achieve the expected exposure or response in patients.

Model-informed approaches flip this paradigm. By leveraging Physiologically- Based Pharmacokinetic (PBPK) modeling and Quantitative Systems Pharmacology (QSP) modeling, researchers can predict drug behavior earlier with great confidence. PBPK modeling answers the question “what does the body do to the drug?” predicting absorption, distribution, metabolism, and excretion. QSP modeling tackles “what does the drug do to the body?” predicting biological responses and therapeutic effects.

Recent studies from leading pharma companies validate the impact of mechanistic PBPK approaches on clinical outcomes.

In a retrospective analysis published in Drug Discovery Today (July 2025)ⁱⁱ, researchers at AstraZeneca examined 29 compounds from their portfolio that progressed to clinical development between 2007 and 2019. The findings were striking:

• Clinical exposure-response for target engagement was successfully predicted during preclinical development for 83% of the compounds within a threefold accuracy threshold.
• Projects with comprehensive non-clinical PK/PD packages achieved an 85% success rate in demonstrating proof-of-mechanism (PoM) in the clinic.
• In contrast, projects with basic packages had only a 33% success rate in PoM.

The study concluded that “implementing non-clinical datasets and PK/PD modelling results in a major impact on clinical success” — an increased overall probability of advancing the right candidates and stopping development of those unlikely to succeed early.

Genentech researchers published complementary findings in Biopharmaceutics & Drug Disposition (2023)ⁱⁱⁱ, demonstrating that PBPK modeling using the Simcyp® Simulator with only physicochemical parameters and in vitro data—combined with preclinical in vivo data—can successfully predict human pharmacokinetics. Analyzing 18 diverse small molecules:

• 94% of compounds had predicted maximum concentration (C_max) within 2-fold of observed clinical values
• 72% of compounds had predicted area under the curve (AUC) within 2-fold of observed values

Most importantly, this approach worked using data available before any human studies were conducted. As the authors noted: “This PBPK-based approach now serves as a practical strategy for human PK prediction at the candidate selection stage at Genentech.”

Building on this technical validation, Genentech researchers further demonstrated in the Journal of Medicinal Chemistry paper (2025)^iv that this PBPK framework is a superior tool for compound rank-ordering compared to using single in vitro parameters. This follow-up study transitioned from predicting absolute PK values to utilizing a generalized PBPK model for multiparameter optimization (MPO). By integrating diverse discovery data into a single “dose score,” the model successfully prioritized compounds based on human exposure drivers (C_min,u, C_max,u and AUC_u ) long before clinical entry.

These studies validate what MIDD advocates have long argued: embedding mechanistic modeling throughout the drug development cycle enables researchers to change their approach to candidate selection.

The rise in regulatory acceptance underscores this shift. Simcyp PBPK modeling has been utilized in more than 80% of FDA-approved drug applications involving PBPK modeling since 2019, with over 125 novel drugs using the Simcyp Simulator in lieu of clinical studies. Similarly, QSP submissions to regulatory authorities have grown dramatically, from just a handful in 2013 to over 80 by 2023, with the majority (65%) supporting Phase I and Phase II programs.

The evidence is compelling, but realizing the benefits of MIDD requires more than just access to modeling tools; success depends on several critical factors.

The experiences suggest that successful implementation of the MIDD approach should start with high-impact applications—first-in-human dose prediction, early feasibility assessment, and complex DDI scenarios —, capture learnings across programs to improve future predictions, establish clear decision criteria, and foster a culture of mechanistic thinking.

The most significant barrier isn’t technical—it’s cultural. Embracing model-informed decision making requires organizational shifts in decision making and accepting probabilistic thinking. The success rates reported in both industry studies suggest this cultural shift is achievable and worthwhile.

The pharmaceutical industry has reached an inflection point. The evidence base supporting model-informed approaches is now substantial; regulatory acceptance is well-established, and leading companies are demonstrating measurable improvements in clinical success rates. The question is no longer whether MIDD works, but how rapidly the industry can adopt these approaches more broadly.

With a clear use case, developing and implementing better tools to utilize model-informed drug discovery in your daily workflow is the logical next step.

Listen to Rob Aspbury explaining his vision for model-informed drug discovery.