Insilico's 40% Response Rate Was Two of Five Patients

Two patients behind the percentage

Percentages hide their denominators at exactly the size where denominators matter most. At five patients, each patient is worth 20 percentage points. One fewer response at 130 mg and the headline rate is 20%. One more and it is 60%. Neither version would tell a reader much more than the current one does, because five patients cannot settle how often a drug will work in the next fifty.

The wider cohort gives the steadier reading. Across the three higher dose levels in mesothelioma, four of 22 evaluable patients responded, a rate of 18.2%. Insilico also reported a 68.2% disease control rate — 15 of 22 evaluable patients, by calculation from the reported percentage. In standard oncology usage, that includes complete responses, partial responses and stable disease. The release does not establish how long disease control lasted across that group. Those are the numbers to carry forward. They describe a drug showing activity in a small, heavily treated group, at a stage of testing designed first to find a tolerable dose.

Context cuts both ways here, and the release supplies some. The study's enrolled patients had a median of two prior lines of treatment, ranging from one to twelve, and responses appeared in patients who had already been through five and six prior lines. A response at that point in a treatment history is not nothing. It is also not a rate that can be projected onto earlier-stage patients, other cancers, or mesothelioma patients generally. The company also described lesion reductions in other tumor types, but supplied no patient-level denominator or response rate for those observations. Those should be read as observations, not results.

What a Phase 1 trial can show, and what this one has not

Phase 1 dose escalation answers a narrow set of questions: whether patients can tolerate the drug, at what dose, and whether there is enough activity to justify a larger study. On the questions this design can answer, the reported data are coherent. No dose-limiting toxicity appeared between 15 mg and 130 mg. No treatment-related side effect caused a patient to stop treatment. Grade 3 treatment-related events were reported in 8.3% of participants, three of the 36 enrolled, and treatment-related proteinuria in 11.1%, four of 36, described by the company as low-grade and reversible. Insilico interprets the reported reduction in blood CCN2 as evidence consistent with TEAD target engagement. It says the observed half-life of approximately 18–32 hours supports once-daily dosing. That biomarker reading does not independently prove clinical benefit.

What the design cannot do is separate the drug's effect from everything else in an uncontrolled group of 36. There is no comparison arm, no blinding, and no completed dose selection. The planned Phase 2a randomized dose study is the step that starts to address dose properly, and Insilico describes it as conditional on emerging data. Until then, efficacy language should stay where the trial is: responses were observed, in numbers small enough to name individually.

The two FDA designations in the release need the same restraint. Orphan Drug and Fast Track status change the regulatory path and the support a program receives. They are not findings that a drug works, and in this case both predate the human data reported today.

The AI claim is a separate claim

Insilico's larger argument is that generative AI designed this molecule, through its Chemistry42 platform, against the TEAD target. Insilico describes TEAD as a target historically considered difficult to drug. The trial reported today does not test that argument. Every patient received the same AI-designed drug, so the data can show whether ISM6331 shows activity and tolerability in humans. It cannot show whether AI design produced a better molecule than conventional methods would have, because no conventionally designed comparator exists in the study.

That distinction matters for how the result should travel. If later trials confirm a benefit, ISM6331 will be an AI-designed drug that worked, a meaningful milestone on its own. A randomized trial comparing ISM6331 with a control treatment could establish clinical benefit, but it would still not isolate the contribution of the AI design method; that would require a separate comparison, or a credible counterfactual, about the discovery process itself. For now, the honest summary is narrower: four responses among 22 evaluable patients, a tolerability profile the company can take into dose selection, and a top-dose percentage that should never be quoted without the two of five underneath it.

Sources

[1] Insilico Medicine, "Insilico Medicine Reports 40% ORR at 130 mg for Pan-TEAD Inhibitor ISM6331 Designed Using Generative AI in Heavily Pretreated Mesothelioma at ESMO 2026" (press release via PR Newswire, October 26, 2026)Read source
[2] TechNode Global, "Insilico Medicine Reports 40% ORR at 130 mg for Pan-TEAD Inhibitor ISM6331" (October 26, 2026; republication cross-check of the same release)Read source

THE ALFW VERDICT

Insilico has reported an early clinical signal worth investigating, not established efficacy. The 40% response rate at the highest dose represents two of five evaluable mesothelioma patients; across the reported dose cohorts, four of 22 responded. Those observations are meaningful enough to report, particularly in a previously treated population, but they come from a small, uncontrolled Phase 1 study with dose selection still underway. The interim safety findings are encouraging within that limited population, not a general safety conclusion. Insilico also says generative AI helped design ISM6331, but the trial cannot determine whether that design method was better than conventional discovery. The next evidence to watch is a larger, rigorously evaluated clinical dataset—not another percentage detached from its denominator.