Multimodal tumor foundation model trained on paired H&E histology and spatial transcriptomics to infer whole-transcriptome and tumor-microenvironment signal from routine H&E alone.
TARIO-2 is a multimodal tumor foundation model from Noetik, an AI biotech focused on building causal, data-driven models of cancer biology. The model is pretrained on paired hematoxylin and eosin (H&E) histology images and matched spatial transcriptomics, learning the statistical link between tissue morphology and the underlying molecular state of the tumor and its microenvironment. The central goal is to infer whole-transcriptome and tumor-microenvironment (TME) signal from routine H&E slides alone — the inexpensive, ubiquitous stain already generated for nearly every cancer biopsy — without requiring the costly spatial assays at inference time.
This positions TARIO-2 in the rapidly growing space of computational-pathology foundation models, but with a distinguishing emphasis on recovering full transcriptomic and spatial context rather than only producing slide-level embeddings or coarse classifications. By treating histology as a readout of molecular biology, Noetik aims to make H&E a proxy for assays that are otherwise expensive, tissue-limited, or unavailable in routine practice.
TARIO-2 is at an announcement and conference stage. The available information comes entirely from Noetik's own materials: a company blog post, the ASCO 2026 program page, and an ASCO 2026 poster. There is no peer-reviewed paper or preprint, and the model is proprietary — no weights, code, or training data have been released, and access is gated and commercial. The claims below should be read as developer-reported and not yet independently validated.
Noetik describes TARIO-2 as a multimodal foundation model pretrained on paired H&E histology and spatial transcriptomics. Detailed architecture specifications, parameter counts, training-corpus size, and quantitative benchmark results have not been publicly disclosed in a technical paper; the company's blog and ASCO 2026 poster are the primary sources. The headline application reported by Noetik is using H&E-derived predictions to stratify response to the BOT+BAL checkpoint-inhibitor combination in MSS metastatic colorectal cancer — a setting that has historically responded poorly to immune checkpoint blockade and where reliable predictive biomarkers are scarce. Because no peer-reviewed evaluation or independent benchmark is yet available, performance figures should be treated as preliminary, developer-reported results.
The intended use cases center on extracting molecular and microenvironmental information from routine pathology slides. Potential beneficiaries include translational and clinical oncology researchers seeking inexpensive surrogates for spatial and transcriptomic assays, biomarker-discovery and drug-development teams working on immunotherapy stratification, and trial designers who need to enrich for likely responders. The ASCO 2026 use case — predicting BOT+BAL checkpoint-inhibitor response in MSS metastatic colorectal cancer from H&E — illustrates the target workflow: applying the model to standard biopsy images to forecast therapy response where conventional biomarkers underperform.
As an announcement-stage, proprietary model, TARIO-2's broader impact is not yet established and cannot be assessed through independent benchmarks, citations, or external adoption. Its significance, if the reported results hold up, would lie in demonstrating that whole-transcriptome and tumor-microenvironment signal can be recovered from inexpensive H&E images and translated into actionable immunotherapy-response predictions for a historically refractory cancer setting. For now, the entry reflects Noetik's own claims; peer-reviewed validation, technical disclosure, and third-party evaluation would be needed to confirm the model's performance and clinical utility.