Isomorphic Labs Is Designing Drugs on a Computer. Now It Has to Prove They Work.

Isomorphic Labs Is Designing Drugs on a Computer. Now It Has to Prove They Work. | Omniscient Media

The most consequential test in artificial intelligence may not involve a language model or a reasoning benchmark. It may take place in a hospital oncology ward, sometime before the end of this year, when a patient receives a drug whose molecular architecture was designed entirely on a computer by an AI system that never once ran a lab experiment.

That is the wager Isomorphic Labs has made. Founded in 2021 as a spinout of Google DeepMind, the London-based company has spent four years building what it believes is a fundamentally new approach to pharmaceutical research: rather than screening millions of physical compounds and observing what sticks, its platform models the atomic-level geometry of biological systems and designs molecules computationally, from first principles. The science is elegant. The industrial proof remains outstanding.

What does Isomorphic Labs actually do?

Isomorphic's core claim is that AI can replace much of the trial-and-error laboratory work that defines early-stage drug discovery. Traditional drug development begins with a biological target - usually a protein implicated in disease - and then searches, often by brute force, for small molecules or biologics that interact with it beneficially. The process is expensive, slow, and fails at extraordinary rates: roughly 90 percent of drugs that enter human clinical trials never reach the market.^[1]

Isomorphic's approach starts with structure. Its platform, built on top of AlphaFold 3 - the protein-structure prediction model co-developed with Google DeepMind - extends the lineage of work recognized by the 2024 Nobel Prize in Chemistry, awarded to Hassabis and DeepMind Director John Jumper jointly with protein design pioneer David Baker of the University of Washington. The Nobel recognized AlphaFold 2; IsoDDE builds on the subsequent AlphaFold 3.^[2] The goal: design the right molecule computationally before synthesizing anything in a lab.

In February 2026, the company published technical details of its next-generation platform, the Isomorphic Labs Drug Design Engine (IsoDDE), which it describes as a unified computational drug-design system that goes meaningfully further than AlphaFold 3 alone.^[3] The distinction matters. AlphaFold 3 solved the structure prediction problem with remarkable accuracy; IsoDDE is an attempt to close the gap between knowing what a molecule looks like and knowing whether it will make a good drug.

How does the IsoDDE platform work?

Structure prediction, Isomorphic argues, was always necessary but never sufficient. A drug designer needs to know not just where a molecule sits on a protein, but how strongly it binds, whether it will find pockets that are hidden in the unbound protein state, and whether it interacts differently with antibody targets than with small molecules. IsoDDE integrates all four of these tasks into a single system.^[3]

The technical claims are specific. On a generalization benchmark designed to test how well models handle protein-ligand systems highly dissimilar to their training data - the category that matters most for genuinely novel drug targets - IsoDDE reportedly more than doubles AlphaFold 3's accuracy in the hardest difficulty tier.^[3] For antibody-antigen structure prediction, the company reports IsoDDE outperforms AlphaFold 3 by 2.3 times and competing model Boltz-2 by nearly 20 times in the high-fidelity regime. On binding affinity prediction, IsoDDE claims to surpass not just competing deep-learning methods but also physics-based simulation approaches like Free Energy Perturbation, which require expensive experimental crystal structures as inputs and take far longer to run.^[3]

Perhaps the most commercially significant capability is what the company calls blind pocket identification: the ability to detect ligandable binding sites on a protein using only its amino acid sequence, without knowing in advance where a drug might attach. The company demonstrated this with cereblon, a protein for which researchers believed for 15 years there was only one druggable site. A 2026 experimental study discovered a second, hidden allosteric pocket; IsoDDE, the company says, predicted both sites from sequence alone, without specifying the ligands.^[3] The implications for so-called "undruggable" targets - proteins that have resisted therapeutic intervention for decades - are potentially large.

What do the Eli Lilly and Novartis partnerships tell us?

The clearest external signal of confidence in Isomorphic's platform is commercial. In early 2024, the company signed research collaboration agreements with both Eli Lilly and Novartis - deals structured around milestone payments that could collectively reach nearly $3 billion.^[4] In February 2025, Novartis expanded that agreement, adding up to three additional research programs on the same financial terms, citing progress in the first year of collaboration.^[5] In January 2026, Johnson & Johnson added a cross-modality, multi-target collaboration to the roster.^[9]

Large pharma partnerships are often treated as validation events in biotech. They deserve some skepticism here. The headline "biobucks" figure - the $3 billion in potential milestone payments - almost entirely comprises payments that will only materialize if specific clinical milestones are hit. The actual upfront payments are not publicly disclosed, and as Drug Target Review noted in a February 2026 industry analysis, a "50:1 ratio between announced biobucks and actual upfront payments reveals appropriate industry caution."^[6] That Novartis expanded rather than walked away from its collaboration is meaningful; that both partners have maintained silence on specific pipeline candidates is also notable.

Why has the clinical trial timeline slipped?

Hassabis had stated publicly in 2024 that Isomorphic would have AI-designed drugs in human clinical trials by the end of 2025. That deadline passed without a trial. In January 2026, Reuters reported the target had been pushed back to the end of 2026.^[7] No specific reason was given for the delay.

One-year slippages are routine in pharmaceutical development, and a company running preclinical programs in novel target categories has every reason to move carefully. But the optics are worth examining. Isomorphic raised $600 million in its first-ever external funding round in March 2025 - led by Thrive Capital with participation from GV and Alphabet - on an explicit mandate to bring its discovered drugs to clinical trials.^[8] Hassabis told the New York Times the company did not need the capital but wanted the funds to hire top research scientists. The combination of a funding round premised on clinical progress and a subsequent timeline slip invites scrutiny, however ordinary the underlying delay may be.

The broader context matters. Multiple smaller AI drug discovery companies - including Exscientia, which merged with Recursion Pharmaceuticals in 2024 in an all-stock deal announced at $688 million but worth approximately $193 million at closing prices - have faced clinical failures or significant pipeline deprioritizations, contributing to an industry mood that is increasingly skeptical of headline claims. One CEO's assessment, quoted in Drug Target Review's 2026 analysis, captured that sentiment bluntly: "AI has really let us all down in the last decade when it comes to drug discovery - we've just seen failure after failure."^[6]

Is Isomorphic Labs in a stronger position than its AI-biotech peers?

The comparison table below captures where the major players stand today. The most striking data point is clinical depth: Insilico Medicine, which listed on the Hong Kong Stock Exchange in December 2025, already has 10 programs in clinical trials and became the first AI-native company to publish Phase IIa proof-of-concept data in Nature Medicine for an AI-discovered drug (Rentosertib, targeting idiopathic pulmonary fibrosis).^[10] Recursion - which absorbed Exscientia's AI chemistry capabilities in its 2024 merger - has consolidated to five internal clinical programs and reported a 43 percent median reduction in polyp burden in a Phase 1b/2 study for familial adenomatous polyposis (FAP), a rare inherited colorectal cancer predisposition condition with no approved therapies, while sitting on $754 million in cash through early 2028.^[11]

	Isomorphic Labs	Recursion (+ Exscientia)	Insilico Medicine
			AI Drug Discovery: Company Comparison
Founded	2021 (DeepMind spinout)	2013 (merged with Exscientia, 2024)	2014
Core AI approach	Structure-based design (AlphaFold/IsoDDE)	Phenomics + AI chemistry + ClinTech	Generative AI end-to-end (Pharma.AI)
Capital position	$600M (Series A, 2025); Alphabet backing	Public (NASDAQ: RXRX); $754M cash, runway to 2028	$393M cash post-IPO (HKEX: 3696, Dec. 2025)
Clinical programs	0 in trials (first targeted end-2026)	5 internal; ~15 partnered/discovery programs	10 programs in clinical trials
Key partnerships	Eli Lilly, Novartis, J&J (2026)	Sanofi ($500M+ cumulative), Roche ($150M upfront)	Eli Lilly, Servier, MSK; 13 of top-20 pharma
Clinical proof-of-concept	None yet	FAP: 43% median polyp burden reduction (Phase 1b/2)	Rentosertib (IPF): Phase IIa data in Nature Medicine
Unique structural advantage	AlphaFold/Nobel pedigree; Alphabet compute	Merged biology + chemistry + real-world clinical data	First published Phase IIa proof-of-concept for AI drug

Structurally, Isomorphic benefits from vertical integration that smaller AI-biotech startups cannot replicate: Google's compute infrastructure, DeepMind's research talent pipeline, and the foundational scientific credibility conferred by the AlphaFold lineage and its Nobel recognition. Where peers like Recursion must manage public market expectations quarterly, Isomorphic operates with Alphabet's balance sheet as a backstop and no obligation to disclose results.^[8]

That insulation from public market pressure cuts both ways. It allows Isomorphic to pursue genuinely long-horizon science without being forced into premature clinical decisions. It also means the company faces fewer accountability mechanisms than a listed entity. The table above makes the competitive context plain: Isomorphic's peers have already produced clinical data - some of it published in top-tier journals. The first Isomorphic trial will arrive later, with more accumulated expectation, and against a benchmark that has moved.

What can AI actually fix in drug development?

The honest accounting of AI's impact on drug discovery in 2026 is more granular than either boosters or skeptics typically acknowledge. Early discovery timelines are being compressed: AI-enabled workflows have reduced preclinical candidate development to roughly 13 to 18 months, compared with a traditional three to four years, and antibody design hit rates have improved from around 0.1 percent computationally to 16 to 20 percent in reported benchmarks.^[6] These are genuine, measurable gains.

What AI cannot yet compress are the rate-limiting steps: clinical trial duration, patient enrollment, regulatory review, and manufacturing scale-up. Biology imposes non-negotiable timelines that computational acceleration cannot bypass. Structure prediction, even at IsoDDE's reported fidelity, does not resolve "off-target" effects - the ways a molecule interacts with proteins other than its intended target - which remain among the most common causes of late-stage clinical failure. A separate Nature commentary published in 2025 argued that the field's core problem is not algorithmic sophistication but data quality: accurate biological, pharmacological, and clinical annotation datasets remain scarce, and 68 percent of tech executives in one survey identified poor data quality as the primary reason AI initiatives fail in drug development contexts.^[6]

Isomorphic's IsoDDE is an attempt to bridge the prediction-to-drug gap more completely than any prior platform. But the gap between a high-quality computational prediction and a molecule that clears Phase I safety trials in humans remains substantial, and Isomorphic has not yet publicly disclosed which targets its lead candidates address, what stage they are at in IND-enabling studies, or what their mechanisms of action are.

What happens if the trials succeed - or fail?

The stakes extend well beyond a single company. A successful Phase I readout from an Isomorphic candidate would represent the first real-world validation that AI-native molecular design - not AI-assisted screening, not AI-accelerated chemistry, but design from first principles - produces therapeutically viable compounds. That outcome would fundamentally alter the investment thesis for the entire AI-biotech sector and likely accelerate consolidation around the few players with genuine deep-tech foundations.

Failure, or continued delay, would not invalidate the science; drug development is a long game in which individual trial outcomes are weak evidence about platform quality. But given the scale of financial and reputational capital accumulated around AI drug discovery, further slippage would intensify calls for more rigorous disclosure standards across the field - and would hand ammunition to critics who argue that the gap between computational elegance and clinical utility has been systematically undersold.

Isomorphic Labs is not short of resources, talent, or institutional validation. What it does not yet have is a patient who got better because of a drug its AI designed. That remains the only number that ultimately matters.

AI Research