NOETIK_AI unveils Perturb‑MARS platform
- Noetik said it has launched Perturb‑MARS, a platform that reads pooled mouse tumor perturbation experiments through a human cancer tissue model called TARIO‑2. - The key trick is pairing Perturb‑map — which tests hundreds of knockouts in one mouse — with TARIO‑2, which infers whole‑genome spatial transcriptomics from H&E. - It matters because drug teams want causal biology in vivo, but they need outputs that map onto human tissue, not just mouse readouts.
Cancer drug discovery keeps running into the same wall. Mouse experiments can show causality, but mouse readouts are still mouse readouts. Human tumor data is closer to what drug developers actually need, but you usually can’t perturb it at scale. Noetik’s new Perturb‑MARS platform is an attempt to fuse those two worlds — using multiplexed in vivo mouse experiments on one side and a human-tissue foundation model on the other. (vuink.com) ### What did Noetik actually launch? Perturb‑MARS is not a single model and not just a wet-lab assay. It is a combined system. The first half is Perturb‑map, Noetik’s in vivo functional genomics platform for running many tumor perturbations in parallel inside the same animal. The second half is TARIO‑2, a foundation model trained exclusively on human cancer tissue. Noetik says the combined setup lets it interpret mouse perturbation experiments in “human” terms. (vuink.com) ### Why is that a big deal? Because the usual tradeoff is brutal. Human data is the thing you care about for translation into drugs and trials, but it is hard to generate, hard to perturb, and often observational. Mouse data is easier to create and manipulate, so it is good for cause-and-effect. But it is still a proxy. Perturb‑MARS is trying to create something that normally does no(vuink.com)cer tissue. (vuink.com) ### What does Perturb‑map do? Perturb‑map is Noetik’s scaled mouse platform. The company has described it as a way to analyze hundreds of genetically modified tumor clones in a single experiment while preserving spatial context inside tissue. Earlier company materials said it was generating an initial dataset with more than 650 mutations in a lung cancer model, and an AACR 2025 updat(vuink.com)se to checkpoint blockade. That scale matters because immunology is spatial — where cells sit can matter as much as what genes they express. (businesswire.com) ### What does TARIO‑2 add? TARIO‑2 is the “human lens” part. Noetik says the model was trained only on human cancer tissue and can convert ordinary H&E pathology images into whole‑genome spatial transcriptomics predictions. In plain English, that means a standard stained slide can be translated into a much richer map of gene activity across the tissue. Noetik says it can apply TARIO‑2 directly to H&E images from Perturb‑map experiments without retraining the model. (vuink.com) ### Why does “no retraining” matter? That is the boldest claim here. Usually, crossing from one domain to another — human tissue to mouse experiments, or one assay to another — is where models break. Noetik is saying TARIO‑2 can read H&E from mouse perturbation studies and still produce useful, human-centric characterizations of the tumor microenvironment. Basically, the company is a(vuink.com)slational pitch, even if outside readers will still want more validation data. (vuink.com) ### How does this fit with Noetik’s earlier work? This launch looks like a synthesis of two tracks Noetik had already been building in public. One track was spatial foundation models like OCTO and TARIO‑2 for understanding human tumors. The other was Perturb‑map for high-throughput in vivo target discovery and immunophenotype modeling. AACR 2025 already framed Perturb‑map as a way to(vuink.com)e ideas. (noetik.ai) ### What is the catch? The catch is that this is a platform story, not a drug approval story. The promise is better target discovery, better patient stratification, and better experimental readouts before a therapy reaches the clinic. But the real test is whether these humanized in vivo readouts actually improve hit rates in downstream drug programs. That takes time — and eventually partner programs or internal candidates will have to prove it. (vuink.com) ### Bottom line? Noetik is making a specific bet about biotech’s next layer. Not just bigger models, and not just bigger screens. The bet is that the winning stack combines scaled perturbation biology with models trained directly on human tissue — so experiments stay causal, but the outputs get a lot closer to the biology drug developers actually need. (vuink.com)
