The Psoriasis Pill an Algorithm Built Just Humiliated Its Biggest Rival

Think of it like two locksmiths trying to pick the same lock. Both figured out which lock to pick (the JH2 domain). But zasocitinib's key fits with almost absurd precision.

How absurd? Zasocitinib binds TYK2 with a dissociation constant in the low picomolar range (around 0.004-0.009 nM). For context, its affinity for the next closest family member, JAK1, is roughly 4,975 nM. That translates to selectivity of more than one million fold for TYK2 over other JAK enzymes.

This matters because traditional JAK inhibitors (the older cousins in this family) hit multiple JAKs at once. That broad inhibition works, but it also causes side effects: blood cell problems, lipid changes, cardiovascular concerns. Zasocitinib's extreme selectivity is designed to deliver the therapeutic punch without the collateral damage.

The AI Angle Is Real This Time

Let's be honest: "AI-designed drug" has become one of the most overused phrases in biotech. Half the time it means someone ran a few compounds through a machine learning model and slapped "AI" on the press release. Zasocitinib's story is more substantive.

The drug was discovered through a collaboration between Nimbus Therapeutics and Schrödinger, a company built on physics-based computational modeling. The team used AI-driven molecular simulations to explore huge chemical spaces of potential TYK2-binding molecules, predict how they'd interact with the JH2 domain, and optimize for selectivity, potency, and drug-like properties before ever synthesizing a compound.

The critical design element they landed on: a methoxycyclobutyl ring that fits perfectly into a pocket on TYK2's JH2 domain (defined by two amino acids, Val603 and Lys642). In JAK1 and JAK2, that pocket is slightly different; a bulkier amino acid blocks the ring from fitting. The AI models identified this subtle structural difference and exploited it.

Takeda later acquired the program through a $4 billion purchase of Nimbus's subsidiary and ran it through clinical development. The result is a molecule whose mechanism is identical to Sotyktu's, but whose precision is dramatically higher. Same concept, better execution, with computational design doing the heavy lifting.

A Watershed Moment for AI Drug Discovery

The broader AI drug discovery track record is promising but incomplete. Phase 1 success rates for AI-designed molecules run around 80-90%, nearly double the historical industry average of roughly 50-60%. That tells you AI is good at designing molecules that are safe and well-behaved in humans.

But Phase 2 success (where you actually prove a drug works) has been less impressive, hovering around 40%, roughly in line with traditional approaches. Until recently, the most advanced AI-designed molecule was Insilico Medicine's rentosertib for lung fibrosis, which showed efficacy in a Phase 2a trial published in Nature Medicine in 2025. That was encouraging but small-scale.

No AI-designed novel small molecule has received regulatory approval yet.

Zasocitinib's head-to-head Phase 3 win changes the conversation. This isn't "AI drug survives Phase 1" or "AI drug shows a signal in 50 patients." This is an AI-designed molecule beating the marketed best-in-class competitor in a registration-quality trial. If you were waiting for proof that computational drug design can produce clinically superior medicines, this is the strongest evidence to date.

The $4 Billion Oral Psoriasis War

Zasocitinib isn't entering a quiet market. The moderate-to-severe plaque psoriasis space is a roughly $20-22 billion global market in 2026. And the oral treatment segment is getting crowded fast.

Sotyktu currently owns the oral TYK2 space, with about $291 million in global sales in 2025. J&J just launched Icotyde (icotrokinra), the first oral IL-23 receptor peptide, approved in March 2026 with sales projections climbing to $1.6 billion by 2028. Alumis has ESK-001, another oral TYK2 inhibitor, posting strong 52-week Phase 2 data.

FierceBiotech has framed zasocitinib as a "$4 billion psoriasis bet" for Takeda, and the head-to-head data give that number some teeth. Takeda plans to file for FDA approval starting in its fiscal year 2026 (which began in April), with a potential U.S. launch in the first half of 2027.

The unmet need driving all this competition is real. Many psoriasis patients still cycle through treatments without achieving full clearance. Injectable biologics (IL-17 and IL-23 antibodies) offer the highest efficacy but require needles, cold storage, and specialty pharmacy logistics. Patients want pills. Increasingly, the pills are catching up to the injections.

What BMS Should Be Worried About

Sotyktu's five-year durability data are genuinely impressive: PASI 90 rates held steady around 46% through year five, with no new safety signals over 5,000-plus patient-years of exposure. BMS also recently expanded Sotyktu into psoriatic arthritis, giving it a dual-indication advantage.

But durability data don't matter much if a competitor beats you on the metric patients and doctors care about most: clearance. In clinical practice, a drug that clears skin completely in more than 35% of patients at four months versus one that likely clears skin completely in roughly 13-14% is a different value proposition entirely.

BMS can argue safety track record and years of real-world experience. That's a legitimate card to play. But if zasocitinib's safety profile holds up through regulatory review (and so far, it looks clean), Sotyktu could find itself playing defense in a market it was supposed to own.

The Bigger Picture

Zoom out and the implications extend well beyond psoriasis. Takeda is also testing zasocitinib in Crohn's disease, ulcerative colitis, vitiligo, and hidradenitis suppurativa. If the molecule's selectivity advantage translates across inflammatory conditions, this becomes a franchise story, not just a dermatology story.

More broadly, this trial answers a question the industry has been asking for years: can AI-designed drugs actually win against human-designed drugs in head-to-head clinical combat?

The answer, at least this once, is a convincing yes. An algorithm looked at TYK2's molecular structure, found a pocket that human chemists also found, and designed a key that fits better. Not by a little. By a lot.

Drug discovery just got its AlphaGo moment. And the game is still early.

Hackers Got Into Novo Nordisk's Clinical Trials. That's Worse Than It Sounds.

Hackers didn't just hit Novo Nordisk's corporate systems — they got into clinical trial patient data. With claims of 1.3 terabytes stolen and a $25 million ransom demand, this breach exposes how vulnerable drug development's most sensitive data really is.