I spent the weekend traveling back from San Francisco to Sweden, so this issue is a little leaner than usual.

But I’m coming home with a lot of energy. I was in California for the Vitalist Bay conference, where friends and people I’ve interviewed for years now increasingly overlap with the AI × bio world BAIO covers. Several people came up to tell me they read and like the newsletter.

And since we’re already flirting with shameless self-promotion, I’ve also been getting emails from readers with encouragement. People inside large AI × bio organizations have reached out to say they appreciate the newsletter.

BAIO is growing organically, reader by reader. So if you find it useful, please forward this issue to one person who works in AI, biology, biotech, pharma, investing, or just wants to understand where this field is going. That helps me a lot.

And feel free to reach out by replying to this email. I read the replies, and I’ll likely respond as well. All feedback is welcome.

Okay, over to today’s issue:

☑️ Retro raises at a $1.8 billion valuation to move aging biology toward the clinic.

☑️ Incyte puts proprietary data behind Genesis’ molecular AI platform.

☑️ Deep learning designs tricky RNA structures - without waiting for an RNA AlphaFold moment.

☑️ And Benchling Inference gives biotech teams GPU capacity for scientific AI models without managing the infrastructure.

Let’s dive in.

Retro Biosciences raised its next round at a $1.8 billion pre-money valuation, giving one of the most ambitious longevity startups fresh capital as it moves from lab promise into human testing.

The Redwood City company, backed early by Sam Altman, wants to add ten years to healthy human lifespan. The round is led by 4P Capital, and Retro says it will support several aging programs: autophagy enhancement, iPSC-derived cell therapies, tissue reprogramming, and AI-enabled protein engineering.

The clinical anchor is RTR242, a first-in-class oral drug designed to restart autophagy - the cell’s waste-clearing system - for Alzheimer’s disease. Retro says the program moved from indication selection to first-in-human dosing in 15 months and is now in Phase I.

The AI angle here is (so far at least) narrower than the valuation headline. Retro’s autophagy drug is not being presented as AI-discovered. But the company does say it has built AI-enabled protein engineering programs, and its earlier OpenAI collaboration on reprogramming gives it a clear AI × aging connection.

That said, AI for aging is starting to show up more often on BAIO’s map: in Issue 14 with MaxToki and Longevity-LLM, and in Issue 25 with Morgan Levine arguing that aging AI needs richer maps across biological scales, and Martin Borch Jensen saying that AI needs task-shaped data and verification at the right biological layer.

Why it matters: In the press release Retro says that “aging biology is reaching a translational inflection point”. It acknowledges there is “an enormous amount” of work ahead. “But we believe the coming decade of aging therapeutics will look very different from the last, and we intend to help define it”.

Did you know? Retro is hiring.

Incyte is putting $120 million and its own experimental data behind Genesis Molecular AI’s drug-design platform, expanding a 2025 collaboration into at least five new targets.

Incyte, a Delaware biopharma company known for drugs in oncology and inflammation, will let Genesis use proprietary results to train the next generation of GEMS, short for Genesis Exploration of Molecular Space.

Genesis Molecular AI, out of San Mateo, California, builds AI for small-molecule drug discovery. GEMS combines foundation models, agents, and chemistry tools so drug hunters can generate molecules, predict how they bind to protein targets, and decide what to synthesize.

The deal includes $80 million in cash, a $40 million equity investment, recurring funding for model training and compute, and more than $1 billion in possible milestones across the first five targets.

“There are a lot of companies that make very strong claims about what their capabilities are, and then you come out of the meeting unconvinced. The opposite was true here,” Pablo Cagnoni, Incyte’s president and global head of R&D, said in an interview with Forbes.

In Issue 26, we covered Incyte deploying Kosmos, Edison Scientific’s AI scientist platform, across target discovery, target validation, and translational biology. Genesis adds another layer: small-molecule design, with Incyte’s own experimental data feeding the GEMS model.

Why it matters: Genesis CTO Sergey Edunov says most pharma-AI deals are limited to small, target-specific datasets. “Our expanded collaboration with Incyte takes this to the next level, greatly scaling the extent to which we are leveraging Incyte’s significant volume of internal data to train AI models for drug discovery.”

Did you know? Genesis is hiring.

RNA never had its AlphaFold moment. Protein design took off once scientists could predict protein structures with high accuracy. RNA design has been stuck with a harder problem: structure prediction is still much less reliable, especially for synthetic RNA molecules.

A new preprint from Stanford’s Rhiju Das, Eterna’s RNA-design community, Nobel laureate David Baker, and collaborators asks whether RNA design can move forward anyway.

The target here is so-called pseudoknots - RNA shapes where one part of the molecule folds back and pairs with another, making a kind of molecular knot. Nature uses these shapes in RNA systems that switch genes on and off, catalyze reactions, and help some viruses make different proteins from the same stretch of genetic code. The question: can AI design new ones, not just predict what existing ones look like?

The team ran OpenKnot, an Eterna challenge where human RNA designers and AI systems were asked to design sequences for 57 pseudoknot targets. Then the designs were made and tested in the lab. One test checked whether the RNA folded into the expected pattern of paired and loose regions. Another changed individual RNA letters to see whether predicted pairings broke - and whether matching changes could restore them. Cryo-EM then imaged several of the folded structures.

The result: after initially trailing the experienced Eterna players, AI methods caught up fast. Across the later OpenKnot rounds, both AI and human approaches solved over 95% of the 57 pseudoknot targets, with roughly 50,000 sequences used in validation.

Why it matters: RNA is no longer just biology’s messenger molecule. It is a vaccine platform, a gene-regulation tool, and a therapeutic platform in its own right. Better tools for predicting and designing RNA shapes could make that more programmable - closer to what AlphaFold and protein design did for proteins.

Did you know? The team says the folding readouts and modeled structures are public, while cryo-EM structures and maps are being made available through PDB and EMDB.

Benchling Inference, powered by Baseten, gives biotech teams GPU capacity for scientific AI models without managing the infrastructure. Benchling frames the problem it is solving simply: wet-lab data arrives in bursts, then teams may need to run 100,000 predictions in hours before going quiet again. The service supports third-party or internal models, runs through Benchling or notebooks, and can deploy in cloud, VPC, or hybrid setups.

☑️ Skape Bio / Baker lab designs GPCR-targeting miniproteins.

☑️ Bristol Myers Squibb brings Claude to 30,000 employees.

☑️ Lilly TuneLab models come to CDD Vault.

☑️ QIAGEN launches QIA Agent.

☑️ Noetik links routine pathology images to immunotherapy response.