Amazon Bio Discovery

Amazon Bio Discovery is a cloud-based application that gives scientists direct access to biological AI models trained on large biological datasets. It is designed for researchers involved in antibody discovery and optimization, combining AI-driven candidate generation with integrated wet-lab testing through a lab-in-the-loop workflow. The platform is built on AWS infrastructure and is aimed at computational biologists, protein engineers, and drug discovery teams working on therapeutic antibody programs.

The platform operates as a continuous experimentation cycle: AI agents help select and orchestrate models, generate and rank antibody candidates, route top candidates to integrated lab partners for synthesis and testing, and automatically return results for analysis and model refinement. Each iteration builds on previous results, accumulating institutional knowledge over time.

Core Workflow Steps

1. Build: Access a catalog of biological AI models with built-in benchmarks showing performance on real antibody optimization tasks. AI agents assist with model selection and orchestration, or computational biologists can construct custom multi-step pipelines combining hosted models with proprietary models. Published pipelines can be saved as reusable self-service templates for the whole team.
2. Design: Upload a target structure and define research goals with AI agent assistance. The agent identifies optimal binding hotspots, recommends design parameters, and provides reasoning supported by literature references. Pipelines generate thousands of ranked antibody candidates scored on structural confidence, binding affinity, and humanness.
3. Test: AI agents apply Pareto-based multi-objective optimization to recommend top candidates from the generated array. Researchers can filter by program-specific criteria and submit selected candidates directly to integrated lab partners, with transparent pricing and turnaround times provided upfront.
4. Learn: Wet-lab results are automatically routed back to the originating experiment. Researchers can compare model predictions against actual outcomes and use experimental data to fine-tune models on proprietary datasets, improving prediction accuracy in subsequent cycles.

Platform Capabilities

• Access to 40 or more AI drug discovery models, including the latest biological foundation models, without requiring infrastructure deployment.
• Wet-lab validated benchmarks built in partnership with the Gray Lab, using large and diverse antibody datasets from scientific literature, with transparent and reproducible validation.
• Support for bringing and training custom models on proprietary experimental data, keeping intellectual property isolated and secure.
• An integrated contract research organization (CRO) network for ordering sequences, expression assays, and binding assays directly within the platform, with transparent pricing and turnaround times.
• Outcome-based pricing where charges apply only to experiments run; viewing results and adding collaborators are free.
• Data and IP security: proprietary data is kept isolated and is not used to train hosted models, with enterprise-grade security built on AWS infrastructure.

Validated Use Case: Memorial Sloan Kettering Cancer Center

• Memorial Sloan Kettering Cancer Center partnered with Amazon Bio Discovery to accelerate antibody development for pediatric cancer.
• Using AI agents to orchestrate multiple models, the team designed approximately 300,000 novel antibody molecules and submitted the top 100,000 candidates for lab testing.
• The process from candidate design to lab submission took weeks, compared to up to a year using traditional design methods.

Amazon Bio Discovery is available via a free trial with no credit card required. It supports subscription tiers with Experiment Unit-based pricing and is built on AWS with enterprise-grade compliance and data security standards. The platform integrates with a growing network of wet-lab partners and supports team collaboration without additional per-user costs for result viewing.