Biotechnology · global
Bringing AI Design Beyond the Screen: The U.S. Invests in a Cloud Laboratory for Protein Engineering
Northwestern University will spend $20 million over four years to build the DREAM platform, allowing researchers to remotely specify desired protein functions and then use AI and robots to iteratively design, manufacture, and test them; however, the production target of 300,000 proteins remains a plan, not a demonstrated result.
Protein generative models can rapidly produce large numbers of candidate designs. What often truly slows research and development is turning sequences into physical proteins, measuring their functions, and feeding both failures and successes back into the models. Northwestern University in the United States plans to bridge this long-standing gap with a remotely accessible automated laboratory, seeking to turn AI predictions into an experimental cycle that can be repeatedly validated.
The U.S. National Science Foundation (NSF) will provide $20 million over four years to help Northwestern University establish the AI-Driven, Rapid, Experimental Automation Machine cloud laboratory, abbreviated DREAM. The university says it will be the first public cloud laboratory in the United States dedicated to AI protein engineering and open to external users. It will be located at the Synthetic Biology Foundry within the university’s Center for Synthetic Biology.
Under current plans, researchers will be able to describe the functions they want a protein to have through a cloud interface, and AI will generate candidate designs accordingly. Only after biosafety and biosecurity reviews are completed will robotic systems manufacture and test the proteins. Experimental data will then be fed back into the models, creating a “design, build, test, learn” cycle. This framework can be used for drug candidates and extended to plastic degradation, pollutant detection, agricultural technology, and critical mineral recovery.
During the funding period, the team expects to synthesize and characterize more than 300,000 proteins, generate as many as 30 million data points, and release the data and AI models under open access. If data quality, experimental conditions, and interpretation standards can remain consistent, the value may extend beyond increased throughput to help fill the long-standing shortage of large-scale, reproducible experimental data for protein AI.
DREAM is also a node in the NSF’s “programmable cloud laboratories” network. The NSF announced an investment of $380 million across 20 teams in fields including biology, chemistry, materials, and electronics; the Astera Institute separately pledged up to $20 million to support data standards, reproducible workflows, and the rapid public release of results. The overall commitment could therefore reach as much as $400 million, but Northwestern University’s direct award remains $20 million.
What has been announced so far is a blueprint for the facility and its capacity; no comparative data have yet been provided showing that DREAM actually improves protein function, success rates, costs, or research and development timelines. Eligibility requirements, fees, and scheduling for external researchers, as well as whether results can be reliably reproduced across different experiments, will also need to be evaluated after the platform begins operating. Even if the system produces promising therapeutic proteins, they will still need to undergo toxicology, manufacturing-process, clinical, and regulatory validation. Automation can shorten early-stage trial and error, but it will not replace the evidentiary thresholds for drug development.