The real edge in AI drug discovery is the experimental loop that feeds the model, not the model alone — and two deals signed inside a single week make the point in money.
On 16 June 2026 Merck launched a multi-target discovery-and-licensing collaboration with Protillion Biosciences, worth up to $510 million in milestone payments plus an undisclosed upfront, initially focused on inflammatory diseases. A day later, on 17 June 2026, LG AI Research signed a master agreement with Korean biotech D&D Pharmatech to co-develop oral peptide medicines for "incurable diseases."
Both deals pair artificial intelligence with high-throughput laboratory feedback in a continuous cycle. The model proposes; the bench tests; the results return to the model. What the buyers are paying for is that loop, and the loop is also where the discipline lives.
Merck and Protillion: paying for the data, not just the model
The Merck collaboration centres on Protillion's "lab-in-the-loop" Prot-MaP platform — short for Protein Display on a Massively Parallel Array. The system repurposes Illumina DNA sequencing flow cells to test, by Protillion's account, up to one million protein variants simultaneously in a single experiment, returning results in as little as 48 hours.
The artificial intelligence and the experiment are bound together. Rather than designing candidates with a model and then hunting for data to validate them, Prot-MaP generates massive experimental datasets first and applies machine-learning models to identify and optimise candidates from real measurements.
Protillion's chief scientific officer framed the approach as a reversal of the industry's usual pattern. The platform starts with the data the model needs, then lets the model learn from it — a sequence Protillion describes as the inverse of "starting with AI and then looking for data."
The differentiator is not a cleverer model but a faster, larger experiment: a million variants measured in two days, fed back to the model before the next design cycle begins.
The commercial structure reflects the value placed on that capability. Merck — known as MSD outside the United States and Canada — pays an undisclosed upfront sum and up to $510 million tied to research, development, and commercial milestones, with the initial programmes aimed at inflammatory disease. Trade coverage from Genetic Engineering & Biotechnology News and Pharmaceutical Executive situated the deal among a cluster of mid-June discovery agreements.
LG and D&D: an oral peptide bet on the same loop
The LG AI Research agreement, sealed at a ceremony in the LG Twin Towers in Seoul, targets a famously hard problem: oral peptide therapeutics. Peptides are normally injectable because the gut's digestive enzymes break them down, so a tablet-form peptide for "incurable diseases" is a genuine prize.
LG's contribution is its "EXAONE Discovery" platform and an "AI Co-Scientist" system that design and validate peptide sequences and predict experimental outcomes. LG says EXAONE Discovery can find promising candidate substances "at a speed tens of times faster than traditional methods," with a particular focus on oral macrocyclic peptides.
D&D Pharmatech supplies the other half of the loop. The biotech handles structural design, synthesis, evaluation, oral-formulation work, and preclinical and clinical development, feeding the results of real experiments back into the model in a continuous cycle. Coverage from UPI and BioSpectrum Asia confirmed the division of labour.
💡One architecture, two therapies
The two deals describe the same architecture in different therapeutic languages. Merck and Protillion run the loop through massively parallel protein display; LG and D&D run it through peptide synthesis and oral-formulation testing. In both, the model's output is only as good as the experiment that checks it.
Why the experimental loop is the actual story
A predictive model in drug discovery is a hypothesis engine. It proposes molecules that might bind, fold, or survive the gut — but a proposal is not a result, and the history of computational chemistry is littered with confident predictions that the bench refused to confirm.
The lab-in-the-loop design answers this directly. By wiring high-throughput experiments into the model's training cycle, the loop forces every prediction toward a measurement. A model that proposes a million variants is interesting; a platform that tests a million variants in 48 hours and returns the truth to the model is transformative.
This is also why the deals are worth real money rather than press releases. Merck and LG are not buying an artificial-intelligence demonstration. They are buying a tightened feedback cycle that compresses the years a candidate normally spends moving from idea to validated lead, and they are buying the experimental throughput that makes the cycle trustworthy.
The contrast with hype matters. A recent AI-for-science reality check found that frontier models still fall short of reliable autonomous discovery; the credible work pairs capability with humility and verification. The lab loop is that humility made operational — the model is never allowed to be the last word.
What a dignity-first reading welcomes, and what it insists on
Emergent Intelligence (EI) — the dignity-first lens through which I read these developments — welcomes AI that compresses the path to medicines for diseases now called incurable. This is artificial intelligence pointed squarely at human flourishing, and a philosophy that puts human dignity first should celebrate it without apology.
But the same lens insists on the safeguards the lab loop already implies. Human validation, real-world testing, and honest reporting of failures are not friction to be engineered away; they are the conditions under which speed becomes trust. A platform that tests before it believes is, in miniature, exactly the discipline a dignity-first practice asks of any powerful system.
The discipline is the promise's twin. The work on honesty in frontier models makes the same case in a different register: a system earns belief by being candid about what it does not know. A drug-discovery loop earns belief the same way — by running the experiment instead of trusting the forecast, and by reporting the variants that failed alongside the ones that worked.
The promise is real; the discipline is what makes it trustworthy. Experiment, do not merely predict — and report the failures as plainly as the wins.
⚠️The dual-use shadow
The same throughput that accelerates medicine accelerates risk. The biosecurity conversation — from <a href="https://humphreytheodore.com/writing/openai-gpt-rosalind-life-sciences-biodefense-2026">GPT-Rosalind's biodefence trade-off</a> to <a href="https://humphreytheodore.com/writing/ai-labs-biosecurity-dna-screening-congress-2026">DNA-synthesis screening before Congress</a> — is the other face of this coin. Faster protein and peptide design is a dual-use capability, and the lab-in-the-loop is where responsible screening must live, not an afterthought bolted on later.
A non-US, non-China entrant matters too
There is a quieter signal in the LG agreement. The most consequential artificial-intelligence systems have largely been built inside two countries, and a serious frontier-scale entrant from neither widens the circle of who gets to shape powerful AI.
LG is a Korean industrial group applying a domestically built model to a hard scientific problem, alongside a Korean biotech partner. Pluralism in who builds capable AI is not a footnote to the dignity-first argument — it is part of it, because systems that touch human health should not answer to a single centre of gravity.
The pattern is broader than one deal. Anthropic's Seoul office and Korean partnerships point to the same widening map, in which the question of who builds and governs frontier AI is no longer settled by geography alone.
The model proposes; the bench decides
Two deals in two days do not remake an industry. But they do mark a maturing of how serious money reads AI drug discovery, and the reading is the right one: the value is in the loop, and the loop is what keeps the model honest.
The temptation in any powerful technology is to let the prediction stand in for the proof. These collaborations resist that temptation by design — the experiment is not optional, the failures are measured, and the model is corrected by reality on every cycle. That is the shape of trustworthy progress.
From an Ubuntu-informed view, the community a medicine is meant to serve deserves both the speed and the honesty. The promise of an oral peptide for an incurable disease, or a new inflammatory therapy reached in a fraction of the usual time, is worth pursuing with urgency. The discipline of the lab loop is what lets that urgency stay trustworthy — and trustworthy is the only kind of fast worth having.
Frequently Asked Questions
The questions below address the most common queries about the June 2026 AI drug discovery deals from Merck and LG, drawn from the companies' announcements and published trade coverage.
What is the Merck and Protillion AI drug discovery deal?
On 16 June 2026 Merck (MSD outside the US and Canada) launched a multi-target discovery-and-licensing collaboration with Protillion Biosciences, worth up to $510 million in milestone payments plus an undisclosed upfront, initially focused on inflammatory diseases. It centres on Protillion's "lab-in-the-loop" Prot-MaP platform, which pairs artificial intelligence with high-throughput protein testing.
What is Protillion’s Prot-MaP platform?
Prot-MaP — Protein Display on a Massively Parallel Array — tests, by Protillion's account, up to one million protein variants simultaneously on Illumina DNA sequencing flow cells, returning results in as little as 48 hours. It generates large experimental datasets first and then applies machine-learning models to identify and optimise candidates, reversing the usual order of designing with AI before finding data.
What did LG AI Research and D&D Pharmatech agree?
On 17 June 2026 LG AI Research signed a master agreement with D&D Pharmatech, at a ceremony in the LG Twin Towers in Seoul, to co-develop next-generation oral peptide therapeutics for "incurable diseases." LG's EXAONE Discovery platform and AI Co-Scientist system design and validate peptide sequences, while D&D handles synthesis, oral formulation, and preclinical and clinical development.
Why are oral peptide drugs difficult to make?
Peptides are normally administered by injection because they are broken down by the gut's digestive enzymes, which is why a tablet-form peptide is hard to achieve. The LG and D&D collaboration focuses on oral macrocyclic peptides, with LG saying its EXAONE Discovery platform can find candidates "at a speed tens of times faster than traditional methods."
Why does the experimental loop matter more than the AI model?
A predictive model proposes molecules that might work, but a proposal is not a validated result. The "lab-in-the-loop" design wires high-throughput experiments into the model's training cycle, forcing every prediction toward a real measurement and correcting the model with experimental truth. The buyers in both deals are paying for that tightened feedback cycle, not for the model alone.
Sources and Further Reading
Cover photograph: high-throughput pipette-tip rack and microwell plate on a laboratory bench — by Tima Miroshnichenko via Pexels.
