Insilico's Brain Drug Isn't in the Brain Yet

ISM8969 is a credible clinical asset from one of the more serious AI drug companies, but the claims that make it interesting still need human data.

I take Insilico more seriously than most AI drug discovery companies, which is exactly why I do not want us to treat the ISM8969 announcement as a clean platform win.

Before the ISM8969 announcement, Rentosertib earned the company some credibility. Insilico tied that fibrosis program to its AI workflow, moved it into patients, and later published Phase 2a data in Nature Medicine. If you are looking for a company that can argue AI drug discovery has moved beyond demos, Insilico is one of the better places to start.

There’s no doubt that Insilico has been on an absolute streak with positive announcements this year, but this week’s announcement around the ISM8969 also makes it easy to overread.

Insilico says it has dosed the first participant in a Phase 1 study of ISM8969, an oral NLRP3 inhibitor being developed for Parkinson’s disease and other CNS (central nervous system) disorders [#1]. The FDA cleared the IND in January. Hygtia partnered on the program, taking half the worldwide rights and putting millions of dollars behind it.

The study will test single and multiple ascending oral doses in healthy adults, elderly participants, and obese adults at cardiovascular risk, with safety, tolerability, pharmacokinetics, and pharmacodynamics as the core readouts.

This is a real drug-development milestone, i.e., a candidate that has cleared enough preclinical and regulatory work to enter human testing. Anyone who has watched drug programs fail before this point should not dismiss that achievement.

But we should be careful about what first-in-human dosing can tell us. It shows the program has reached the clinic. It does NOT show that the drug is safe in people. It does not show meaningful target engagement. And it does not show that ISM8969 is better than other NLRP3 inhibitors already in development.

The key unresolved claim is CNS penetration. For a Parkinson’s program, that matters because the drug has to reach the compartment where the disease biology is being argued. Insilico says the Phase 1 study will collect cerebrospinal fluid (CSF) to assess CNS penetration. That is exactly the measurement we should want, but it also means the human evidence has not arrived yet. Until those data exist, “brain-penetrant” is still supported by the preclinical package, not by a human CSF result.

This is where AI drug discovery can get a little slippery. A company can be genuinely good at moving candidates into the clinic and still be a long way from proving that a specific drug has the properties its story depends on.

Speed through discovery and IND clearance is certainly important, but in drug development, though, speed only buys you the chance to ask harder questions in humans.

The AI claim should stay precise too. ISM8969 appears to be a legitimate AI-designed molecule. Insilico says Chemistry42 was used to discover and optimize it, and designing an oral NLRP3 inhibitor with a plausible CNS profile is not a small medicinal chemistry task.

If the platform helped the team get there faster, with fewer dead ends, we should care. However, the target itself was already known. NLRP3 had a large body of biology and industry interest before ISM8969. That makes this a narrower story than rentosertib, where Insilico could connect both target discovery and molecule design to its AI workflow [#7][#9]. Here, the important understanding is that AI helped design a candidate for a target field that they already cared about with a large body of research behind it.

The competitive context raises the bar. Lilly’s Ventyx deal put a large dollar sign on NLRP3, and BioAge has an oral NLRP3 program further along in Phase 2. That validates the space, but it also tells us Insilico is joining an active race.

So I read this as a credible clinical step being asked to carry more meaning than it can hold today. The next useful update is human tolerability, plasma exposure, pharmacodynamics, and the CSF data that show if ISM8969 reaches the relevant CNS compartment.

Until then, we can give Insilico credit for moving an AI-designed NLRP3 inhibitor into humans without treating the hardest claims as answered.

Merck's up-to-$510 million AI protein-design pact discloses a ceiling but not the upfront, the targets, or a single named program. Merck signed the multi-target collaboration on June 16 with venture-backed Protillion Biosciences, built on its Prot-MaP platform for generating protein-design AI training data. Protillion gets an undisclosed upfront and up to $510 million in milestones across unnamed therapies. [Link]

Jazz handed AbCellera a hard $56 million upfront for cancer antibodies, the kind of disclosed cash most AI discovery deals never reveal. The June 17 preclinical deal covers T-cell engaging multispecific antibodies for GI and other solid tumors, with another $28 million due on a committed third program and up to $792 million per program if Jazz exercises its options. AbCellera's own release calls it an "antibody discovery engine," not an AI platform. [Link]

Two brain-penetrant oral NLRP3 inhibitors reached clinical dosing the same week, and the one not designed by AI is a full phase ahead. BioAge dosed the first patient on June 16 in QUELL-CV, a roughly 160-patient Phase 2 testing BGE-102, discovered from human longevity data, with hsCRP at 12 weeks as the endpoint and topline due in H2 2026; Insilico's AI-designed ISM8969 started Phase 1 the next day. BioAge's CEO pitched the drug as potentially statin-like for inflammation. [Link]