Amazon unveiled a new platform, Amazon Bio Discovery, at the AWS Life Sciences Symposium in mid-April.
Memorial Sloan Kettering Cancer Center (MSK) was among the first test users. They did something emblematic: designed nearly 300,000 nanobody candidates against a pediatric cancer target in weeks and sent the top 100,000 directly to wet labs for synthesis and testing.
That process would normally take a year.
How the platform works
Amazon Bio Discovery combines three things:
1. Biological foundation model library
The platform includes more than 40 foundation models specifically trained on biological data, covering areas such as protein structure prediction, antibody design, and drug molecule generation. Providers include Apheris and Boltz, with Biohub and Profluent expected to join. Unlike general-purpose large language models, these models are trained on biological experimental data and have a targeted understanding of molecular structure-function relationships. Users can also fine-tune models with their own experimental data without building their own training pipeline.
2. AI agent interface
This is the key that makes it usable for non-computational experts. Tell it in natural language what to do — for example, find nanobodies that can bind to this target protein — and the AI agent automatically selects the appropriate model, designs the experimental workflow, and interprets the results. No coding required.
AWS executive Rajiv Chopra said the AI agent makes powerful scientific capabilities no longer the exclusive domain of computational researchers.
3. Lab-in-the-loop interface
This is what makes Amazon Bio Discovery truly distinctive. The platform doesn't just do computation; it directly connects to real wet labs. When AI screens high-potential candidate molecules, they can be submitted with one click to partner labs for physical synthesis and testing. Current lab partners are Twist Bioscience and Ginkgo Bioworks, with A-Alpha Bio joining soon. Test results are automatically fed back to the model to optimize the next round of design parameters. This is the experimental loop: AI designs → wet lab validates → results feed back → AI redesigns.
What the MSK case really means
Memorial Sloan Kettering Cancer Center is a top U.S. cancer research institution. Their test case:
- Input structural data of a pediatric cancer-related target protein
- AI generates about 300,000 nanobody design proposals
- Filter by manufacturability, stability, and other criteria, selecting 100,000 for wet lab
- Experimental results obtained within weeks
AWS published a dedicated scientific white paper for this project, documenting the entire nanobody design workflow. In a traditional process, a single design-test-iterate cycle itself takes months, and completing the full cycle typically takes a year. This isn't just about faster computation; the core is connecting the information flow between design and validation, eliminating the friction of extensive manual coordination in between.
Who is using it
Known early users:
- Bayer: One of the world's largest pharmaceutical companies
- Broad Institute: A genomics research institute jointly established by MIT and Harvard
- Voyager Therapeutics: Focused on gene therapies for neurological diseases
AWS also disclosed a background data point: 19 of the top 20 global pharmaceutical companies already use AWS cloud services for research. This isn't about acquiring new customers from scratch, but deep integration within the existing customer base.
Why this is more than just another AI drug discovery tool
AI has been making waves in drug discovery for years, with various companies coming and going. What makes Amazon Bio Discovery different is that it doesn't just sell models or computing power; it connects the computational side with the experimental side.
The slowest part of drug development has never been running models; it's waiting for lab test results, bridging the information gap between computational predictions and physical validation, and enabling a biologist who doesn't code to figure out how to use AI tools. Amazon Bio Discovery is addressing all three of these problems head-on.
Of course, the platform has just launched, and large-scale data on cost and success rates will take time to accumulate. The MSK case is an excellent start, but there is still a gap between designing more candidates and actually discovering drugs that cure diseases faster.
Still, the approach is worth noting: embedding AI agents into the physical loop of lab operations is a framework that can be directly replicated in materials discovery, food science, and agricultural chemistry — any field that requires extensive experimental iteration. AWS chose pharma this time because the market is large enough, the need is urgent enough, and the customers were already there.
Sources: AWS Launches Amazon Bio Discovery Agentic AI to Accelerate Drug Development (Genetic Engineering & Biotechnology News), CocoLoop, Amazon launches its AI drug discovery platform (pharmaphorum)