The protein at the center of one of immunology’s most crowded new races was considered undruggable until a few years ago.
VAV1, a guanine nucleotide exchange factor expressed almost exclusively in immune cells, relays signals from the T-cell and B-cell receptors into activation, proliferation, cytoskeletal remodeling, and cytokine release. “VAV1 sits at a choke point of adaptive immunity,” said Alex Zhavoronkov, founder and co-CEO of Insilico Medicine, the generative AI drug discovery company that nominated an oral VAV1 degrader as a preclinical candidate on October 13th. Because expression is restricted to immune cells, removing the protein could dampen pathogenic lymphocyte activity across several autoimmune diseases while sparing most other tissues.
Chemistry stood in the way for years. “For years, VAV1 was considered undruggable because it lacks a classic small-molecule binding pocket,” Zhavoronkov told DDN. Molecular glue degraders changed the arithmetic, and the target class has moved fast since. Novartis began a Phase 2 study of Monte Rosa’s MRT-6160 in Sjögren’s disease in September 2026, under a license that paid Monte Rosa $150 million upfront with up to $2.1 billion in milestones. InnoCare’s ICP-538 and Haisco’s HSK56630 are dosing healthy volunteers. No VAV1-directed therapy has reached approval anywhere.
Insilico’s ISM9546 enters that field at the preclinical stage, developed for inflammatory bowel disease (IBD) and rheumatoid arthritis (RA) first. Zhavoronkov read the competition as a signal rather than a threat. “The field is moving fast, which I see as validation,” he said. “In a race like this, the molecule has to differentiate.”
A target that came from academia and a human signal that came from a rival
For drug developers weighing how much of the AI discovery story to believe, Zhavoronkov drew the line himself. “Academic immunologists established VAV1 biology long before us, and the first clinical data in the class came from another company,” he said. Monte Rosa’s Phase 1 study in healthy volunteers reported sustained, dose-dependent VAV1 degradation above 90 percent in peripheral blood T cells, with the compound generally well tolerated.
That human signal did the work that target-discovery algorithms could not. “We chose VAV1 because the biology is strong, the human signal exists, and an oral, peripherally-restricted degrader can be a differentiated molecule in a large market,” Zhavoronkov said. Validation of the program followed conventional immunology practice, with degradation and functional assays across multiple cell systems, then several preclinical models of IBD.
The division of labor Zhavoronkov described tracks a broader pattern across the field. “Academics own the frontier of biology,” he said. “Most of the targets the industry works on, VAV1 included, were first characterized in university labs.” What industry contributes, by his account, is the closed loop — where every prediction gets tested against synthesis, assays, toxicology, and eventually patients. Plus, it’s “where a model is judged by whether its molecules survive that process,” he said.
Designing a three-body problem
Molecular glues have historically resisted rational design. “Historically, molecular glues were found by luck,” Zhavoronkov said, pointing to thalidomide, whose mechanism was explained decades after it reached patients. The difficulty is structural. “Rational design is hard because you are optimizing a three-body system, the degrader, the E3 ligase and the target protein, and small structural changes can switch degradation on or off or pull in unwanted proteins,” he said.
Insilico’s chemists worked from the structure of the target-ligand complex, defined a pharmacophore from the ligand and key binding-pocket residues, and focused generative models on de novo fragment growing. Each generative model design round scored candidates on several objectives at once, reinforcing the protein contacts that drive potency, holding drug-like properties, and lowering brain exposure. The company reported sub-nanomolar VAV1 degradation potency in cellular assays after multiple design-make-test-analyze cycles.
“For IBD and RA, the target sits in circulating and tissue lymphocytes, and there is no therapeutic reason to expose the brain,” Zhavoronkov said. “For an oral drug that patients may take for years, low CNS exposure removes a whole category of safety questions.”
Insilico reported confirming the profile two ways in preclinical work, through MDCKII-MDR1 assays that indicate whether the efflux transporter at the blood-brain barrier clears the molecule, and through in vivo tissue distribution studies showing low brain exposure. Proteomics profiling showed a clean degradation profile with no off-target protein depletion observed, and 14-day rodent and non-human primate studies showed favorable tolerability and broad safety margins, according to the company.
The questions that stay open until it’s tested in humans
Two reported design choices carry the most weight for anyone running an autoimmune program. The first is how quickly immunity returns. Insilico said it designed the molecule with a balanced half-life so target engagement stays sustained while, also by design, “immune function should be able to recover promptly when needed, for example if dosing is paused during an acute infection,” Zhavoronkov said. For a mechanism that quiets both T-cell and B-cell activation in patients dosed for years, that recovery window becomes a central safety question, and preclinical pharmacokinetics will leave it unresolved.
The second is ligase selectivity. Insilico has not disclosed which E3 ligase ISM9546 recruits, which leaves outside immunologists without the information needed to anticipate off-target degradation or to compare the compound against competitors. InnoCare has said ICP-538 works through cereblon, and cereblon carries most of the clinical degrader field, appearing in 13 of 35 programs in human trials as of late October and in six of the nine late-stage assets. Reviewers have raised the question of whether a field concentrated on one ligase may share resistance mechanisms.
Efficacy evidence so far comes from animal models. In an acute colitis model, ISM9546 significantly reduced the Disease Activity Index within a few days of treatment — an effect that held through the end of the study, with histopathology scores improving by about 60 percent. Zhavoronkov kept the caveats attached. “These are preclinical results, and human studies will show whether they translate,” he said.
Asked where discovery still moves slowly, he named the same constraint. “Animal models often predict human disease poorly,” he said, adding that the hardest open problems in AI drug discovery remain predicting translation and toxicity in humans. These problems are limited by data, “especially clinical and negative data that rarely leave company archives,” said Zhavoronkov.
An oral immunology race that pharma is paying to join
The competitive pressure around VAV1 reflects where large buyers have moved. Immunology and inflammation dealmaking through 2026 has concentrated on oral molecules that could match biologic efficacy with pill convenience, including Eli Lilly’s $3.2 billion acquisition of Morphic for an oral integrin inhibitor, Boehringer Ingelheim’s $500 million deal with Sitryx for preclinical immune-metabolism inhibitors, and Gilead’s $1.3 billion agreement for an oral IL-23 receptor peptide in IBD.
Degraders are following the same path into immunology, with roughly 10 of the 35 clinical degrader programs now in immunology or inflammation against about 23 in oncology, and Kymera’s oral STAT6 degrader in Phase 2b.
Insilico builds its programs for that market. “We de-risk the chemistry quickly, then partner where a larger company can run global trials faster than we can alone,” Zhavoronkov said, citing Exelixis licensing its USP1 inhibitor in 2023 for $80 million upfront and Menarini’s Stemline dosing the first patient with a KIF18A inhibitor in February 2026. Peripheral restriction, he said, “is the profile partners tell us they want.”
The platform numbers supporting that pitch have grown. ISM9546 is the company’s 35th preclinical candidate since 2021 and its 11th in the first 10 months of 2026, against what Zhavoronkov called a very different baseline. “Not long ago, one preclinical candidate a year was a good result for a biotech company,” he said. Insilico reports averaging 12 to 18 months per program with 60 to 200 molecules synthesized, where traditional early discovery takes 2.5 to four years, and 13 of its candidates had received investigational new drug clearance as of August 2026. Its lead compound, the TNIK inhibitor rentosertib, produced a Phase 2a result in idiopathic pulmonary fibrosis published in Nature Medicine in June 2025, with a mean forced vital capacity increase of 98.4mL at 60mg once daily over 12 weeks against a 20.3mL decline on placebo across 71 patients.
Zhavoronkov placed the limits of all of it where immunologists would. “It is still early, and a handful of clinical programs is a small sample, which we say openly,” he said. “AI can make a molecule better and faster, but a wrong target stays a wrong target, so biology decides most of translation.”
On VAV1, academic immunology supplied the target and a competitor supplied the human proof, which leaves ISM9546 to win or lose on chemistry. “All of this is preclinical data,” he said, “and the clinic will have the final word.”
