Today in Biology — Aug 17

[SOFIA] Alright, welcome back to The Dish. We've got a lot to unpack in the world of biology this week, and honestly, the biggest signal I'm seeing is coming straight from the top.

[DANIEL] Hmm. You mean the funding landscape, Sofia? Because it looks like the tectonic plates of scientific funding and infrastructure are definitely shifting.

[SOFIA] Exactly. And I think the biggest news, especially for anyone building molecular tools or working in engineering biology, is the Department of Energy’s Genesis Mission. They just announced a plan to create a whole new class of open-weight foundation models.

[DANIEL] Open-weight meaning the models themselves, the algorithms, are publicly accessible. It's not just an API you query; you can actually look under the hood.

[SOFIA] Right. They're collecting information from organizations that can provide these foundational systems, or the massive scientific datasets needed to train or fine-tune them. This isn't just about general-purpose AI; this is science-specific AI models. Think about what that could mean for designing new proteins, predicting molecular interactions, or even optimizing metabolic pathways in non-model organisms. It’s a huge push, and it's coming with significant investment.

[DANIEL] It’s a definite signal. The DOE has deep pockets and a history of large-scale infrastructure projects. If they're putting this kind of resource into open-weight AI for scientific discovery, it suggests a real belief that these models can accelerate research in ways traditional methods can't. My question is always, what's the actual empirical gain? What kind of problems are they hoping these models will solve that we’re currently stuck on?

[SOFIA] Well, for engineering biology, I'm thinking about the sheer complexity of biological systems. We're trying to design circuits, pathways, even whole organisms, and the design space is just massive. An AI that can process and learn from vast datasets—genomic, proteomic, metabolic—could dramatically cut down the iterative design-build-test cycle. Imagine designing a novel enzyme with specific activity for a non-model host, not through trial and error, but with AI-guided predictions. That's the promise.

[DANIEL] The promise is alluring, no doubt. But the rigor needs to be there. Will these foundation models truly capture the nuanced biological reality, or will they hallucinate solutions that don't hold up in the lab? The quality of those scientific datasets for pre-training or fine-tuning is going to be absolutely critical. Garbage in, garbage out, even with the most sophisticated AI.

[SOFIA] Agreed. But the intent is clear: to turbocharge scientific discovery. And it’s not just the DOE. We're seeing the NSF announcing $1.5 billion for foundational research to drive scientific breakthroughs for American technological leadership. It’s a really strong indicator of where the national priorities are landing. It’s all about tech leadership, and that includes biotech.

[DANIEL] And on the flip side of that coin, we're seeing some interesting shifts over at NIH. There's a proposal to remove peer review scores from grant applications. Instead of granular numerical scores, they're looking at broad categories for funding decisions.

[SOFIA] That's a big one. As someone who’s been through that process, those scores can feel… intense. The idea is to reduce the emphasis on hyper-specific numerical differences, maybe broaden the types of projects that get funded, or reduce the stress on reviewers.

[DANIEL] Or, perhaps, it's an attempt to streamline a system that's currently facing significant backlogs. There's been a fair bit of discussion online, and even news reports, about a 'graveyard for grants' at NIH, with hundreds of awards, including for critical areas like maternal health, being held up. Shifting to broader categories *could* potentially speed up decision-making, but it could also introduce more subjective bias if not managed carefully. The rationale, according to the agency, is to simplify and focus on impact over minute score differences.

[SOFIA] It’s a fascinating tension, isn't it? On one hand, massive investment in new AI tools to accelerate discovery, and on the other, a re-evaluation of how we fund the foundational human-driven research. Both are trying to push science forward, but from very different angles. It makes me wonder about the interplay between those two forces in the coming years.