Today in Biology — Oct 10

[SOFIA] Welcome back to The Dish! Today in Biology, we're diving straight into some massive shifts happening at the national level, especially around how we're thinking about AI and biology. And honestly, it feels like a really pivotal moment.

[DANIEL] Pivotal, yes, but also one where we need to be clear-eyed about what these initiatives actually promise versus what they deliver. The dollars are certainly big, but the underlying mechanisms and the immediate impact on day-to-day research are still forming.

[SOFIA] Totally. And at the top of that list, the Department of Energy just dropped a huge announcement about their new Genesis Mission. They're making a big push into what they’re calling "Super Intelligence for Science" — and that includes biology. We're talking new awards, new graduate fellowships, all geared towards this.

[DANIEL] Right, and the term "Super Intelligence for Science" — or SI-ready data, as the NIH is calling it — is important to unpack. It's not just about applying existing AI models to biological data. It's about designing data from the ground up, generating it in ways that are specifically structured and organized to be maximally useful for advanced AI.

[SOFIA] Exactly. So, for engineers like me, this is huge because it means we're moving beyond just trying to *find* patterns in messy, existing biological datasets. We're talking about actively *designing* experiments and data collection protocols so that the output is immediately digestible and actionable for AI models. Think about building a robust chassis for your genetic circuits – now we're building a robust *data chassis* for AI. It could really accelerate our ability to, say, predict how a novel genetic circuit will behave in a non-model organism, or even design new metabolic pathways.

[DANIEL] And that's where the rigor comes in. If the goal is truly predictive models, then the foundational data needs to be impeccably controlled, well-annotated, and robust. We're not just talking about larger datasets, but *smarter* datasets. The NIH has joined this effort, working with DOE and Biohub, specifically to build these SI-ready datasets for human biology. The question is, how will they ensure that the data isn't just voluminous, but also carries the necessary orthogonal measurements and replications to build truly predictive, rather than merely correlative, models?

[SOFIA] That's the million-dollar question. But the intent, at least, is clearly to push for that predictive power. And it's not just the DOE and NIH. The NSF is also laying groundwork for the "next generation of technologies," aligning with the White House's "Science: A New Golden Age" report. It’s a comprehensive rethinking of how we fund and support science.

[DANIEL] Which, historically, can mean anything from genuinely transformative new programs to a rebranding of existing initiatives. The devil will be in the details of those specific programs and how they prioritize funding towards these "SI-ready" approaches. Will we see more emphasis on standardized assays, or platforms that automatically generate metadata compatible with AI models?

[SOFIA] I hope so. Because ultimately, for engineering biology, this could mean a future where iterating on designs for things like novel enzyme production or bioremediation pathways isn't just faster, but also much more predictable. Imagine designing a bacterial strain, feeding its genetic blueprint and environmental conditions into an AI, and getting a highly accurate prediction of its performance *before* you even hit the lab. That would be the good stuff.

[DANIEL] That would certainly be a step change. But until we see the specifics of those data standards and the validation pipelines, my skepticism remains. Are we truly building foundational data for *predictive* models, or just larger datasets that AI can sift through for correlations? The difference is significant when it comes to engineering applications.

[SOFIA] Absolutely. But the momentum behind this cross-agency push for "Super Intelligence for Science" feels genuinely different. It's not just about applying AI; it's about reshaping how we *do* science to *enable* AI. It’s going to be fascinating to watch how this unfolds.