What Makes AI Useful in Drug Development?
- Esther Chen
- 7 hours ago
- 3 min read
Why context of use, validation, and asset-level evaluation matter in pharmaceutical R&D
On July 30, 2026, Vizuro took part in an expert roundtable hosted by Taiwan's Science & Technology Law Institute (STLI), Institute for Information Industry (III), on the use and governance of AI in the pharmaceutical industry.
The discussion covered applications across R&D, manufacturing, quality, data management, and commercialization, and was later featured by Global Bio & Investment Monthly.
One question stood out: how should the industry judge whether an AI system is actually useful in drug development?
For us, speed and model performance are only part of the answer. What matters is whether AI helps teams make better decisions under uncertainty.
Three Points from the Discussion
Decision quality matters more than model performance alone. A strong model prediction is useful only if it supports a better development decision.
Validation starts with context of use. Performance should be assessed against the decision a model is intended to support.
Drug assets need to be evaluated as a whole. Biology, clinical evidence, development feasibility, IP, commercial potential, and remaining uncertainties all influence whether an asset should advance.
Validation Starts with the Decision being Supported
An accuracy number on its own says relatively little about whether a model is useful in practice.
A more relevant assessment starts with its intended use:
what decision the model supports
what data it relies on
what level of error is acceptable, and
how its output will be reviewed.
This approach is increasingly reflected in regulatory thinking. The U.S. FDA has proposed a risk-based framework for assessing AI model credibility in relation to a defined context of use [1]. In January 2026, the FDA and European Medicines Agency (EMA) published joint principles for good AI practice in drug development, including a clear context of use, data governance, risk-based performance assessment, human-centric design, and lifecycle management [2].
The EMA has also addressed AI across the medicinal product lifecycle, from drug discovery through post-authorization use [3].
For pharmaceutical applications, these are practical considerations. A model sits within a larger decision process. The quality of that decision depends on the underlying data, how the result is interpreted, who reviews it, and how changes to the model, data, or intended use are managed over time.
The Benchmark should be the Decision Quality
Drug discovery has become increasingly sophisticated in comparing models against technical benchmarks. The harder question is whether those improvements translate into better R&D decisions.
A recent Nature Reviews Drug Discovery Perspective makes this distinction directly, arguing that benchmarking in AI-enabled drug discovery needs to move beyond model validation and examine whether a tool improves decision-making in practice [4].
This is an important standard for the field.
AI does not need to be inserted into every step of drug development. It needs to be used where it can reduce uncertainty, strengthen the evidence behind a decision, or identify when an asset should not move forward.
For Apotek, that means looking beyond an individual prediction and asking a broader set of questions: What do we know about the asset? Where are the critical uncertainties? What evidence would change the decision? And is the opportunity strong enough to justify the next investment?
In a field where time, capital, and development resources are constrained, knowing what not to advance can be just as valuable as identifying what to pursue.
We appreciate the opportunity to exchange views with STLI and other industry participants on how AI can be applied more rigorously across pharmaceutical R&D.
👇Read the original coverage by Global Bio & Investment Monthly
References
[1] U.S. Food and Drug Administration. Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products. Draft Guidance, January 2025.
[2] U.S. Food and Drug Administration & European Medicines Agency. Guiding Principles of Good AI Practice in Drug Development. January 2026.
[3] European Medicines Agency. Reflection Paper on the Use of Artificial Intelligence in the Medicinal Product Lifecycle. September 2024.
[4] Bender A, et al. Artificial intelligence in drug discovery — what it is, where we stand and the path forward. Nature Reviews Drug Discovery. 2026.
