Argonne lands funding for more than 12 projects within the Genesis Mission, from an AI co-scientist for enzymes to a network of autonomous labs

Argonne lands funding for more than 12 projects within the Genesis Mission, from an AI co-scientist for enzymes to a network of autonomous labs

At the slogan level, Genesis Mission looked like a major government platform for “AI for science.” On March 12, Argonne раскрыла список из более чем дюжины проектов that are meant to give it concrete substance. The mission’s goal is blunt: double the productivity of American science over ten years. To do that, the government is assembling a full-stack scientific workflow — models, data, robots, and laboratories that can run experiments with almost no manual pauses.

A key node is ModCon, a consortium meant to build self-improving models on top of U.S. Department of Energy data, facilities, and expertise. Alongside it sits IDeA, an AI “co-scientist” for enzyme discovery and biosynthetic pathway optimization. That module is supposed to propose workable biomanufacturing options where researchers today still spend weeks manually testing enzymes and pathway designs.

The next layer comes down to hardware. OPAL is building a network of autonomous laboratories: Argonne is leading the protein design track, Berkeley and PNNL the microbial systems track, and Oak Ridge the plants and rare earth element extraction track.

“OPAL should turn biological discovery into a self-driving process,” the project page says.

In early March, Ginkgo had already launched Cloud Lab, где эксперимент превращается в удалённый сервис. OPAL applies the same logic at the scale of a national laboratory network. Here, AI carries the work all the way into the physical experiment. The model designs, the robot assembles, the instrument measures, and the data flows back into the next loop — design → build → test → another training round. Each such loop cuts the time between hypothesis and validation.

Separately, Argonne also received funding for RoSA, robotic scientific assistants trained on the actions of human lab technicians, and for LAMBDA, an architecture for storing and processing multimodal biological data. This is the system’s service layer. Machine-readable lab notebooks, shared formats, and clearly defined laboratory tasks make results usable for the next round of training. Without them, the model proposes, the robot executes, and garbage flows back into the system.

On March 12, Argonne showed what science looks like when it is assembled as infrastructure. One module proposes a hypothesis, another selects an experiment, a third carries it out with robotic hands, and a fourth packages the data so the next model can learn without starting from zero. That is how factories and cloud services are built. Now biology is being built the same way.

For now, this is an architecture at the funding stage. A working system still lies ahead. But the real breakthroughs usually sit precisely in these infrastructure modules: at the interfaces between the model, the data, and the physical experiment. If this loop works, the speed of science will start to depend less on the talent of any single lab and more on how fast the whole machine can run the validation cycle.