Developing a new medicine remains a difficult scientific and engineering problem. It can take many years, requires significant investment, and most possible candidates never reach patients.
For biologic medicines, the challenge is even more complex. These therapies are made from engineered proteins rather than synthetic chemistry, and they are often used across most major acute and chronic diseases. Scientists must search through vast numbers of possible molecules to find the rare candidates that bind the right target, stay stable in the human body, and can be manufactured at scale.
Why AI matters in biologic drug discovery
AI is now being used to make that search more focused. Instead of relying only on traditional workflows, researchers can use machine learning models to generate or rank possible molecules before committing lab resources.
AstraZeneca describes this as part of a build-measure-learn loop. AI helps identify candidates that appear most likely to succeed. Scientists then test the strongest options, collect the results, and feed those findings back into the process.
The practical goal is not simply speed. A tighter feedback cycle can reduce dead ends, support faster iteration, and help researchers explore disease targets that were previously considered untreatable by medicine.
This matters because the number of possible molecular combinations is far beyond what any human team can systematically examine. AI does not remove the need for laboratory work, but it can help decide which experiments are worth running first.
From single targets to more complex medicines
Traditional biologics usually focus on one disease pathway. The next generation of medicines may need to do more: hit multiple targets at once or deliver therapeutic payloads to specific cells.
That makes design harder. A promising molecule must balance several requirements at the same time, including potency, stability, manufacturability, and safety. AI-driven models could help researchers weigh those variables together instead of treating them as separate problems.
Puja Sapra, senior vice president and head of R&D biologics engineering and oncology targeted discovery at AstraZeneca, describes the shift directly: "Drugging the undruggable is becoming a reality."
The claim reflects a broader change in how drug discovery is being framed. AI is not only a tool for moving faster through known workflows. It is also being applied to medicine designs that are difficult to reason through manually because the biology and engineering constraints are deeply connected.
Data is the foundation of the system
The source article notes that McKinsey estimates generative AI, combined with other computational tools, could cut drug discovery timelines by as much as 50%. But models depend on the quality of the data behind them.
In drug discovery, useful data can come from experiments whether they succeed or fail. Each result offers a signal about what works, what does not, and which direction researchers should test next.
AstraZeneca says its datasets are proprietary and multimodal. They include molecular structures, binding measurements, safety profiles, and manufacturing outcomes. The company is also investing in deep screening technologies to create more data for refining and validating models.
That data layer is central because biologic drug design requires prediction across many dimensions. A model that can only reason about one feature of a molecule is less useful than one trained on varied biological, safety, and manufacturing signals.
The lab of the future is a closed loop
To connect prediction, testing, and learning, AstraZeneca is building a "lab of the future" facility in Kendall Square, Cambridge, Massachusetts. The aim is a continuous discovery system combining AI, robotic automation, and instruments that generate new data.
Sapra compares the idea to a self-driving car. In this setting, AI makes predictions, robots run experiments, instruments produce results, and those results return to the models for the next cycle.
Automated high-throughput systems could eventually make and evaluate thousands of molecular interactions on a weekly basis. The source article also points to robotic sample handling, automated quality checks, and integrated data pipelines as parts of this approach.
Still, the system is not described as replacing scientists. Human oversight remains central, especially for judgment, strategic direction, and making sure outputs are explainable, tolerable, and aimed at potential patient benefit.
The path toward de novo medicine design
The long-term vision is known as "de novo" design. In this approach, AI would generate entirely new protein sequences designed from scratch to match desired drug properties.
That means going beyond selecting from known candidates. The model would need to help design the structure, predict safety, anticipate how the molecule behaves in the body, and consider how it could be manufactured.
Several requirements remain before that vision is fully realized:
- Richer and more standardized training data across the industry.
- Robust evaluation benchmarks for AI-generated candidates.
- Teams that can work across machine learning and biology.
- Stronger safety prediction for computationally generated molecules.
Safety prediction may be the most consequential challenge. AstraZeneca is working with what Sapra describes as virtual clinical trials: advanced cell systems and micro-scale organ models that act as physical testbeds, paired with AI that learns from their outputs.
Another shift is the move toward agentic AI systems that can generate molecule candidates while also predicting how efficacious and safe they may be. These workflows could connect disease-level insights directly to molecule design, joining data that previously sat in separate silos.
The result is a new model for drug discovery: computational systems that generate, test, validate, and learn, with scientists and engineers guiding the process. The promise is faster progress on biologic medicines, but the article makes clear that the future depends on both advanced AI and deep human expertise.