1) The single biomarker you’d actually gate treatment on in 2026 (setting + cutoff) — and why.

In my opinion, the single most reliable biomarker for anti-PD-1-based therapy, among three that are currently FDA-approved and many that are still in testing, is MSI-H/dMMR. This marker, based on a cancer’s genetic profile, can reliably identify a small subset of patients (~4–5%) across all solid tumor types for whom anti-PD-1 is likely to confer durable antitumor benefit.

2) What must be true for PD-1 + personalized vaccines to become standard, and which non-melanoma tumor is likeliest to get there first — why.

To enter standard-of-care, a combination of anti-PD-(L)1 plus a personalized vaccine must be shown to be feasible and provide significant clinical benefit compared to anti-PD-(L)1 alone in a randomized trial. The greatest opportunity may be for cancer types that show relatively low or no response to anti-PD-(L)1 monotherapy, for example pancreatic cancer.

3) As personalized cancer vaccines move into combination trials, how do you see immune checkpoint blockade (ICB) evolving to complement these approaches?

Personalized cancer vaccines are most likely to be feasible and effective in the adjuvant (post-surgical) setting, which is the context for most current clinical trials. Vaccine-reactive immune cells typically upregulate immune checkpoint molecules such as PD-1 and others, so combining vaccines with ICB may provide a synergistic antitumor effect.

Several cancer types already have FDA-approved ICB in the adjuvant setting based on relapse-free survival benefit in high-risk populations. This provides a framework for randomized comparisons of vaccine plus ICB versus ICB alone. In other cancers, vaccine plus ICB may lead to new disease indications for ICB in the adjuvant setting.

By interrogating which checkpoint molecules are upregulated in T cells following vaccination, novel vaccine-ICB combinations can be nominated for clinical testing. Finally, recognizing that vaccine-induced immune activation engages checkpoint pathways should enable more effective sequencing of agents in multi-drug combination regimens.

4) What biomarkers or clinical settings will best define which patients benefit most from vaccines, checkpoint inhibitors, or their combination?

The adjuvant (post-surgical) setting is most likely to yield the most informative signals regarding the efficacy of personalized cancer vaccines. Because surgery alone can cure a proportion of patients with early-stage resectable cancers, and adjuvant therapies carry potential toxicity, the risk-benefit ratio must be optimized.

This can be achieved by using accepted clinicopathologic criteria to identify patients at high risk for relapse. Vaccine combinations with ICB should be guided by immune profiling to determine which checkpoint pathways are induced by a given vaccine, and then selectively targeting those pathways with specific inhibitors in rational combination strategies.

Responses to interview questions from Tal Behar, Precision Medicine World Conference

1. What unique role does Noetik’s “virtual patient” platform play in reshaping how we approach clinical trials or drug development?
The fundamental failure mode in our industry isn’t that we can’t design good molecules; it’s that we test them in models that don’t look anything like the human disease. We’ve spent decades relying on reductionist cell lines that lack context. Noetik is flipping that paradigm by building virtual cells and patients directly from high quality, multimodal human tissue data.

Our platform, powered by OCTO, doesn’t just look at a cell in isolation. It understands spatial context, tissue architecture, the immune microenvironment, and how cells communicate with one another. Our world models allow us to simulate how a virtual patient’s tissue responds to a perturbation in silico, shifting experimentation from the wet lab to the GPU cluster.

We also use large pretrained foundation models to predict, from a simple H&E image, whether a patient is likely to respond to a given drug. This allows us to mathematically define the patient population where a drug will work, effectively solving the translation problem before the trial even begins.

2. As AI adoption in biotech accelerates, where do you see the biggest impact in the next 2–3 years?
AI in biology is moving from the era of molecule design to the era of simulating biology. The last few years were defined by breakthroughs like AlphaFold and structure based design. The next 2–3 years will focus on simulating biological systems using foundation models trained on real patient biology.

The biggest impact will come when models move beyond static predictions and begin functioning as dynamic simulators. These systems won’t just identify targets, they will predict what happens downstream when you perturb those targets in a messy, complex human environment.

The winners in biology won’t simply have the best algorithms. They’ll have the best data. Public datasets aren’t enough to train frontier biological models. We need to generate data that is fit for purpose.

3. What’s one insight or preview you’re excited to share with the PMWC audience in your upcoming talk?
We’re excited to show what happens when biology is treated not as a static snapshot, but as a programmable system. We’ve been developing OCTO-vc, our Virtual Cell capability, which allows us to take a virtual cell, such as a T cell, and place it into different regions of a patient’s tumor to observe how it behaves.

This extends to larger multicellular structures, and we can run these simulations at scale across hundreds of unique patient samples, testing hypotheses directly in human tissue.

We’ve also built industry first capabilities that turn a simple clinical H&E into a powerful predictive biomarker across multiple therapeutic mechanisms using multimodal foundation models. We’re validating this work with partners like Agenus, while also applying it internally to build our own clinical stage programs.

We’re excited to share these capabilities at PMWC, and potentially preview new ones we haven’t shared publicly before.