DeepMind's Nobel Playbook: From AlphaFold to IsoDDE and the $3B Drug-Discovery Bet

From Nobel medal to IsoDDE: what DeepMind did next

Most Nobel Prizes are terminal. They reward a piece of work that's already in the textbooks. AlphaFold was different — it shipped as a product before it got a medal. The AlphaFold Database now hosts over 200 million predicted protein structures from more than a million organisms. SKYCovione, a coronavirus vaccine co-designed by Baker's lab at Washington, became the first de novo-designed medicine approved for human use.

So when DeepMind won the Nobel, the question wasn't “what next?” — it was “how fast?”

The answer is Isomorphic Labs, Alphabet's biotech spin-out, now run by Hassabis himself in parallel with his DeepMind CEO role. Isomorphic stopped publishing. It stopped open-sourcing. And in February 2026 it released a model — IsoDDE — that independent computational biologists have publicly called “a major advance, on the scale of an AlphaFold 4.”

IsoDDE: the “AlphaFold 4” that Lilly and Novartis are paying $3B for

IsoDDE is a unified drug-design engine. Where AlphaFold 2 predicted protein structure and AlphaFold 3 predicted protein–ligand, protein–DNA and protein–RNA interactions, IsoDDE does end-to-end drug design: simulate the target, screen millions of candidate molecules in silico, optimise for binding affinity, solubility and toxicity, and hand a pharma partner a list of preclinical candidates before a single physical compound is synthesised.

The numbers from the February report are what got the industry's attention:

• IsoDDE more than doubles AlphaFold 3's accuracy on protein-ligand structure and binding affinity prediction.
• It has already been deployed on the Lilly and Novartis partnerships, moving those programmes from target identification to multiple preclinical candidates.
• The commercial structure: $45M upfront plus up to $1.7B in milestones from Lilly; $37.5M upfront plus up to $1.2B from Novartis — roughly $3B in potential contract value on top of royalties.

The first Isomorphic-designed oncology drug is on track for Phase 1 clinical trials by the end of 2026 — a delay from Hassabis's earlier target of late 2025, but a real trial with a real IND, not a demo.

The Nobel itself: Baker, Hassabis, Jumper — and why it mattered

It's worth naming the thing. The 2024 Nobel Prize in Chemistry went, in one half, to David Baker “for computational protein design,” and in the other half jointly to Demis Hassabis and John M. Jumper “for protein structure prediction.” Baker built Rosetta and then RFdiffusion — the tools that let you design a protein that does not exist in nature. Hassabis and Jumper built the AI system that cracked a 50-year-old problem: folding any protein sequence to its 3D structure with experimental-grade accuracy.

Nobel committees don't usually reward software. They did here because AlphaFold didn't just solve a problem — it published 200 million answers to it and handed them to every biologist on earth for free. That was the old DeepMind playbook: do Nobel-class science, open the weights, let the field run. IsoDDE is the first big signal that the new playbook is different.

Beyond biology: AlphaProof, AlphaGeometry and AlphaEvolve

The protein story is the headline, but it's not the only axis DeepMind is compounding on. In mathematics, AlphaProof and AlphaGeometry 2 hit silver-medal standard at the 2024 International Mathematical Olympiad, and an advanced Gemini Deep Think model went on to hit gold-medal standard at IMO 2025, solving five of six problems perfectly. AlphaProof is an AlphaZero-style agent that proves theorems inside the Lean proof assistant; AlphaGeometry 2 is a neuro-symbolic hybrid two orders of magnitude faster than its predecessor.

Then there's AlphaEvolve — an evolutionary-search system built on top of Gemini that, when tested on 50 open mathematical problems, matched state-of-the-art algorithms on 75% of them and improved on the state of the art on 20%. (We covered AlphaEvolve's marketing implications in detail in our cousin piece on AlphaEvolve and AI algorithm discovery.)

The pattern across AlphaFold, AlphaProof, AlphaGeometry and AlphaEvolve is consistent: take a hard formal domain, build a specialist model, let it compound. Protein folding, formal maths, algorithm discovery — each one a moat in a field that used to take decades to move.

Open science vs proprietary moat: the fight DeepMind just picked

IsoDDE's most controversial feature isn't its accuracy — it's the lack of open weights. AlphaFold 2's weights were released. AlphaFold 3 was scaled back to a web server with usage limits. IsoDDE is fully proprietary: partners see results, not the recipe.

This matters commercially and culturally. Commercially, because the moat is now the model, not the publication. Culturally, because a material chunk of AlphaFold's original impact came from academics being able to download, fine-tune, and build on it. The AI-for-science community has started to split between the “open” camp (Baker's lab still ships weights) and the “closed” camp (Isomorphic, and increasingly the frontier labs downstream of it). Expect this to be the central fight of 2026–2027 in computational biology.

The 5-step playbook to convert Nobel momentum into productised breakthroughs

DeepMind's Nobel-to-IsoDDE arc isn't luck. It's a repeatable pattern any research-heavy organisation — biotech, fintech, materials, climate, legal — can borrow from. Five steps:

1. Pick a domain with a formal benchmark. Protein folding had the CASP competition. Maths has the IMO. Your field has something. No benchmark, no loop.
2. Ship the open thing first. AlphaFold 2 and the 200M-structure database built the credibility, the user base, and the hiring magnet. Trying to start closed means nobody trusts the model when you eventually go commercial.
3. Spin the commercial vehicle early. Isomorphic was founded in 2021 — three years before the Nobel. By the time IsoDDE launched, the corporate structure, the pharma relationships (Lilly, Novartis signed January 2024) and the regulatory posture were in place.
4. Land milestone-rich partnerships, not one-off licenses. The $3B Lilly/Novartis structure is mostly milestones: you get paid as candidates progress. That aligns the model's improvements with real cash flow, not just vanity.
5. Let the model compound across adjacent domains. AlphaFold → AlphaProof → AlphaGeometry → AlphaEvolve. Same architectural DNA, new domains every 12–18 months. The research org is the product.

What this means for marketers (and why your stack is the bottleneck)

You are reading this on a marketing blog for a reason.

The DeepMind story is, at surface level, about science. But the mechanic underneath — Nobel-grade specialist models compounding into commercial products in 18 months — is now playing out in every field with a formal benchmark, and marketing is one of them. SEO has Google's quality signals. Paid has ROAS. Content has engagement curves. Each of these is a benchmark a specialist AI system can now learn to optimise end-to-end.

The marketing teams that will be left behind in 2026 aren't the ones without access to frontier models. Everyone has access. The ones getting crushed are the ones whose operational stack can't keep up — who still need three days and four freelancers to ship what a single AI-native Anjin-grade competitor ships in an hour.

That's the IsoDDE lesson for marketing: the moat isn't the model, it's the system around the model. The briefs, the assets, the approvals, the distribution, the analytics — all needs to flow at model speed, or the model's advantage evaporates in the handoffs.

Anjin: The Marketing Operating System for the AlphaFold-speed era

Anjin is the Marketing Operating System. It's the operational layer that lets a small team run marketing the way Isomorphic runs drug discovery: specialist AI agents doing the generative, exploratory work; humans doing the judgement and the relationships; a single orchestration layer that collapses the time from insight to output.

• Research agents that ingest a market the way AlphaFold ingests a genome — fast, structured, exhaustively.
• Content agents that draft, variant-test and ship brand-aligned material without the three-round freelance handoff.
• Distribution agents that publish, optimise and re-deploy across channels at pipeline speed.
• One operating system, not fifteen SaaS tabs taped together.

Agencies are our launch audience because they feel the margin pressure first. But Anjin is for any founder, in-house marketer, or operator who has watched the gap between “we should do that” and “we shipped that” widen every quarter for two years.

Sources: Nature, Scientific American, Isomorphic Labs, NobelPrize.org, EMBL, TechCrunch, MarketScreener, Google DeepMind (Deep Think IMO), Google DeepMind (AlphaProof), Google DeepMind (AlphaEvolve)