A multimodal foundation model for molecular and morphologic data in oncology
In brief
THREADS outperforms leading models across 107 cancer pathology tasks
Pretrained on more than 47,000 tissue sections by aligning slide images with genomic or transcriptomic profiles, THREADS outperformed leading whole-slide models and supervised approaches across 107 oncology tasks. It also transferred well to external cohorts and performed with fewer labels, but the abstract does not quantify the gains; broader testing with additional clinical and molecular data remains ahead.
- Journal
- Nature cancer (Q1)
- Published
- 20 November 2026
- Study design
- Unclassified
- Evidence level
- Level 5, Expert Opinion (CEBM 5)
- Authors
- Anurag Vaidya, Andrew Zhang, Guillaume Jaume, Andrew H Song, Tong Ding, Sophia J Wagner, et al.
- PMID
- 43064800
- DOI
- 10.1038/s43018-026-01252-w
Why clinicians should know about it
- Picked for Pathology and Forensic Medicine (top studies of the week, 22 November 2026): Whole-slide foundation model evaluated for grading and subtyping
Abstract
Foundation models are advancing computational pathology by transferring knowledge from large datasets to diagnostic, prognostic and treatment response tasks. Most existing models, however, either operate at the patch level or learn slide-level representations using image-based objectives, coarse tissue labels or pathology reports that vary across institutions and reflect the subjectivity inherent in narrative reporting. Here we introduce THREADS, which instead learns general-purpose whole-slide representations by aligning tissue morphology with genomic or transcriptomic profiles. THREADS was pretrained on over 47,000 hematoxylin-and-eosin-stained tissue sections and evaluated across 107 oncology tasks spanning cancer subtyping and grading, tumor microenvironment characterization, treatment response prediction and survival analysis. THREADS consistently outperformed leading whole-slide foundation models and supervised baselines while demonstrating strong transfer across external cohorts and high label efficiency. These findings establish molecularly guided pretraining as an effective strategy for learning general-purpose whole-slide representations and motivate future extensions incorporating additional molecular, spatial and clinical modalities.
Abstract as published, via PubMed.
For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.