Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Genome-Anchored Foundation Model Embeddings Improve Molecular Prediction from Histology Images

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read PathLUPI, a training-time privileged-information framework, claims that anchoring whole-slide-image embeddings to transcriptomic pathway signatures improves slide-only molecular prediction across 49 oncology tasks, with a mean AUC gain of…

desk verdict A well-executed LUPI benchmark for WSI molecular prediction; the causal claim is not isolated without a no-privilege control, and the CIs are overconfident, but the scale and consistency warrant serious refereeing. read the letter →

arxiv 2506.19681 v1 pith:PIQT46ZF submitted 2025-06-24 cs.CV

Cheng Jin , Fengtao Zhou , Yunfang Yu , Jiabo Ma , Yihui Wang , Yingxue Xu , Huajun Zhou , Hao Jiang
show 8 more authors
Luyang Luo Luhui Mao Zifan He Xiuming Zhang Jing Zhang Ronald Chan Herui Yao Hao Chen
This is my paper · ORCID
classification cs.CV
keywords computationalpathologywhole-slideimageslearningusingprivilegedinformationtranscriptomicpathwaysignaturesHallmarkgenesetsbiomarkerpredictionmolecularsubtypingsurvivalprognosis
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to prove that routine H&E-stained whole-slide images alone can carry enough molecular information to predict mutations, subtypes, and survival, provided the histology model is trained with genomic guidance it will not see at inference. PathLUPI does this by giving the model, during training, a second input: the patient's bulk RNA-seq profile split into 50 Hallmark pathways, then forcing a slide-only branch to imitate the pathway-aware representations. Across 49 tasks, 20 cohorts, and 11,257 cases, the slide-only branch consistently beats models trained without the privileged transcriptomic signal, including a mean 4.48% AUC improvement over the best baseline in biomarker prediction. The paper also argues the resulting attention maps reveal tissue-level morphological signatures tied to specific mutations and pathways, which would make the method useful for hypothesis generation as well as prediction. If this holds, molecular-style screening could be delivered from the existing pathology workflow rather than from sequencing.

What carries the argument

The load-bearing object is a dual-branch learning-using-privileged-information architecture with a shared cross-attention block. The privileged branch encodes each patient's gene expression as 50 pathway-level vectors, one per Hallmark gene set, each produced by a dedicated multilayer perceptron, and lets those vectors attend over re-embedded WSI patch features, producing pathway-aware slide representations. The distilled branch constructs pseudo-pathway vectors from the WSI features alone, passes them through the same shared cross-attention, and is aligned to the privileged branch by three losses: reconstruction, attention-map alignment, and representation consistency. A region-aware re-embedding transformer refines patch features before attention. At inference the privileged branch is discarded, and the distillation branch alone produces genome-anchored embeddings from WSI input.

What would settle it

Train PathLUPI twice on the same tasks: once with the privileged branch's gene-expression vectors randomly permuted across patients and once with the branch removed but the WSI branch given equal parameters and the same alignment losses. If either variant keeps the reported AUC and C-index, the transcriptomic supervision is not causally load-bearing; if both drop toward the WSI-only baselines, the paper's attribution is supported.

Watch

Extended reading notes

Core claim

PathLUPI's central claim is that transcriptomic data should be treated not as another prediction target but as privileged training information. During training a privileged branch fuses pathology-foundation patch features with RNA-seq embeddings organized by 50 Hallmark pathways through cross-attention, while a distilled branch sees only the whole-slide image and is trained, through reconstruction, attention-alignment, and representation-consistency losses, to reproduce the privileged branch's outputs. At inference only the distilled branch runs, so predictions use the slide alone. The paper reports that this design beats four WSI-only baselines on nearly all of 25 biomarker tasks, 8 subtyping tasks, and 16 survival tasks, with a mean AUC gain of 4.48% over the strongest baseline in biomarker prediction, an internal mean subtyping AUC of 0.856, an internal mean survival C-index of 0.693, AUC at least 0.80 in 14 biomarker and subtyping tasks, and smaller but consistent gains on external cohorts. It further claims the learned attention is interpretable: high-attention patches are enriched in cell types and pathways known to accompany specific mutations such as BRAF in colorectal cancer and EGFR in lung adenocarcinoma.

Load-bearing premise

The load-bearing premise is that the privileged transcriptomic branch, and not the extra architecture, losses, or parameters, is what causes the reported gains, yet the paper never trains PathLUPI without transcriptomic input or with scrambled transcriptomic profiles.

Editorial extensions

If this is right

  • If PathLUPI's claim is right, a deployed system can predict actionable biomarkers, molecular subtypes, and prognosis from the standard H&E slide after a one-time training phase that needs paired transcriptomics, removing sequencing cost and turnaround time from the routine workflow.
  • The external-cohort results imply the genome-anchored embeddings transfer across institutions and staining conditions better than WSI-only baselines, so the benefit should not be confined to the training distribution.
  • The interpretability analyses imply that slide-only attention can localize tissue regions whose morphology tracks specific pathway activities, giving pathologists a visual hypothesis-generation tool for genotype-phenotype links.
  • The backbone ablation implies that future pathology foundation models with better pretraining will directly improve PathLUPI's molecular prediction, making the framework a compounding rather than one-off gain.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Treating the framework as a recipe, the same privileged-pathway design could be applied to methylation, proteomics, or spatial transcriptomics as the privileged modality, since nothing in the architecture is RNA-specific beyond the per-pathway encoders.
  • The attention maps could be tested as a cheap pre-screen for spatial transcriptomics: tissue regions PathLUPI weights most heavily should, if the biological claim is right, be the same regions where the corresponding pathway programs are spatially enriched.
  • A permutation control, where gene-expression vectors are shuffled across patients while all other training conditions stay fixed, would isolate whether pathway content or merely the presence of a second input drives the reported AUC gains.
  • The same dual-branch distillation could transfer other structured priors into slide encoders, such as copy-number signatures or tumor microenvironment cell fractions, to produce slide-only predictors for endpoints not tested here.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The manuscript introduces PathLUPI, a learning-using-privileged-information (LUPI) framework that uses transcriptomic pathway signatures (Hallmark gene sets) as privileged supervision during training to produce genome-anchored histological embeddings. At inference, only whole-slide images are used. The authors evaluate PathLUPI on 49 molecular oncology tasks spanning 13 cancer types and 20 cohorts with 11,257 cases, including biomarker prediction, molecular subtyping, and survival prognosis. They report consistent improvements over ABMIL, CLAM, DTFD, and TransMIL, with a headline mean AUC increase of 4.48% over the best baseline in biomarker prediction and improved C-index in survival analysis. The paper also presents interpretability analyses linking attention patterns to known genotype-phenotype associations, and an ablation study comparing CONCH with PLIP, UNI, and ResNet50 as feature extractors.

Significance. If the central claim is causally valid, this is a substantial contribution: it provides a general recipe for using paired transcriptomic data at training time to improve WSI-only molecular prediction, it is evaluated on an unusually broad and clinically relevant benchmark, and it includes external cohort validation and interpretability analyses. The scale of the evaluation, the public data sources, and the promised code release are strengths. However, the causal attribution that the improvements come specifically from the privileged transcriptomic signal is not yet isolated, and the statistical presentation overstates precision. The contribution is therefore significant but needs additional experimental and statistical work before the main claim is fully supported.

major comments (3)
  1. [Methods: Genome-anchored representation learning; Results: Validating the impact of foundation model embeddings] The central causal claim that transcriptomic privileged information drives the reported gains is not isolated. Figure 6 ablates only the visual backbone (ResNet50, PLIP, UNI, CONCH). No experiment removes, shuffles, or permutes the transcriptomic branch. Relative to the baselines, PathLUPI adds a region-aware re-embedding transformer, shared cross-attention, 50 pathway-specific MLPs, and three alignment losses, so an equal-capacity WSI-only model or a permuted-transcriptome control (e.g., shuffling gene expression vectors across patients while preserving the architecture) is required to attribute the mean 4.48% AUC improvement to genome anchoring rather than to added capacity or optimization changes.
  2. [Methods: Implementation details; Extended Data Tables 1-4] The reported 95% confidence intervals are implausibly narrow. For example, Extended Data Table 2 reports AUC values such as 0.849 (0.848-0.850) for a five-fold cross-validated estimate. Bootstrapping out-of-fold predictions with 1,000 resamples captures uncertainty of the metric for a fixed set of fits, not variability across the five training runs and random splits. The one-sided Wilcoxon signed-rank test is appropriate for paired task-level comparisons, but the manuscript does not specify whether the test is over tasks, folds, or bootstrap replicates, and the reported P<0.001 for molecular subtyping is not attainable with a task-wise test on 8 tasks (minimum one-sided p = 1/256 ≈ 0.0039). The statistical methods and interval estimates need to be clarified and, if necessary, corrected.
  3. [Methods: Multi-level alignment and training objective; Extended Data Table 8] The alignment loss weight λ is introduced in the total loss L_total = L_sup + λ(L_rec + L_attn + L_rep), but no value or range for λ is reported in Extended Data Table 8 or anywhere else. The number of latent spatial regions R is stated as 50, and the top attention patch fraction is used in the interpretability analysis, but the loss weight and the threshold choice for 'top 1%' patches are not justified or varied. Since the contribution is specifically about the privileged alignment mechanism, the sensitivity of the main result to λ should be reported.
minor comments (5)
  1. [Introduction and Results] The number of tasks is inconsistent: the Abstract and Results state 49 tasks, while the Introduction states 48 tasks. The Extended Data Tables sum to 49 tasks, so the Introduction should be corrected.
  2. [Results: PathLUPI enhances biomarker prediction] The sentence 'By integrating transcriptomic priors via the LUPI paradigm, PathLUPI significantly improves biomarker prediction accuracy' is duplicated verbatim. One copy should be removed.
  3. [Results: PathLUPI advances survival prognosis] The sentence beginning 'Evaluation across these diverse cohorts revealed PathLUPI’s strength...' is followed by a partial repetition ('revealed PathLUPI’s strength in survival prognosis, marked by...'). This should be cleaned up.
  4. [Results: Validating the impact of foundation model embeddings] The phrase 'depicted in pretraining strategies Figure 6b revealed that no single foundation model was universally optimal' is grammatically broken and appears to contain a stray heading fragment. The sentence should be rewritten.
  5. [Methods: Patient cohorts and ethics] The ethics statement should clarify whether the local institutional review board approvals cover the use of de-identified WSIs from Center-1 and Center-2, and whether the same approvals cover both the private cohorts and the public TCGA/CPTAC/EBRAINS data.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: inference is strictly WSI-only, external cohorts provide no transcriptomes, and no equation reduces the predicted output to the privileged input by construction.

full rationale

PathLUPI's derivation chain is not circular. The central claim is that transcriptomic pathway signatures, used only during training through a LUPI-style distillation objective, produce WSI-only embeddings that outperform WSI-only baselines. The Methods state that 'During inference, only the distilled branch is activated' and that for a new case 'transcriptomic data is unavailable,' so the test-time predictor consumes only patch features. External cohorts (CPTAC, EBRAINS, Center-1, Center-2) consist of 'only WSIs,' meaning the reported external AUC/C-index gains cannot be produced by feeding transcriptomes or labels into the inference model. No equation in the paper reduces the predicted quantity to the privileged input: the reconstruction loss L_rec aligns pseudo-pathway embeddings to pathway embeddings, but the supervised loss L_sup = L^priv_sup + L^distill_sup is evaluated on held-out slides without g_i, and the reported improvements are empirical rather than algebraically forced. The absence of a no-privilege control or a shuffled-transcriptome permutation experiment is a legitimate experimental-design limitation—it leaves open whether added capacity rather than transcriptomic supervision drives the gains—but that is a causal-attribution concern, not circularity. Self-citations (refs 70, 71, 83, 84) concern related multimodal survival methods and significance-testing conventions; none is load-bearing for the paper's central derivation. The interpretability analyses confirm known BRAF/EGFR morphology associations and are validated against independent external literature rather than being defined into existence by the framework.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central empirical claim depends on a small number of modelling choices: the loss balance lambda, the number of spatial regions R, and the top-attention patch fraction for interpretability. The main domain assumptions are that matched TCGA RNA-seq and WSIs represent the same tumor state and that Hallmark pathway grouping is a useful transcriptomic organisation. No invented entities are introduced.

free parameters (3)
  • lambda (alignment loss weight) = not reported
    Total loss L_total = L_sup + lambda(L_rec + L_attn + L_rep); the value is chosen by hand and not reported in Methods or Extended Data Table 8.
  • number of latent spatial regions R = 50
    R=50 is chosen to match the number of Hallmark pathways P=50; the paper does not provide an optimization or justification for this choice.
  • top attention patch fraction = 1%
    Used for global interpretability; the paper states 'This proportion was chosen to ensure selected patches are abundant and representative', a manual analytical choice.
assumptions (3)
  • domain assumption TCGA WSI and matched bulk RNA-seq from the same case reflect the same underlying molecular state, allowing pathway-level transcriptomic embeddings to supervise WSI morphology.
    The privileged branch aligns WSI patches to transcriptomic pathway embeddings during training; if the RNA-seq is from a different tumor region or time than the slide, the alignment target is noisy. Invoked in 'Genome-anchored representation learning with privileged supervision'.
  • domain assumption The Hallmark gene set collection is a valid biological structure for representing transcriptomic state relevant to morphology.
    Transcriptomic features are partitioned into P=50 Hallmark pathways from MSigDB; the method's performance depends on this grouping being informative.
  • domain assumption Each evaluated task is treated as an independent paired observation in the one-sided Wilcoxon signed-rank test.
    Tasks share TCGA cohorts, CONCH features, and the same training procedure, so cross-task P-values are likely overconfident; cited by 'Performance metrics were estimated using ... one-sided Wilcoxon signed-rank test'.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Genome-Anchored Foundation Model Embeddings Improve Molecular Prediction from Histology Images." pith.science (2026). https://pith.science/paper/PIQT46ZF

@misc{pith2026250619681,
  author       = {Pith},
  title        = {Pith review of: Genome-Anchored Foundation Model Embeddings Improve Molecular Prediction from Histology Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/PIQT46ZF}},
  note         = {Machine review of arXiv:2506.19681}
}
abstract

Precision oncology requires accurate molecular insights, yet obtaining these directly from genomics is costly and time-consuming for broad clinical use. Predicting complex molecular features and patient prognosis directly from routine whole-slide images (WSI) remains a major challenge for current deep learning methods. Here we introduce PathLUPI, which uses transcriptomic privileged information during training to extract genome-anchored histological embeddings, enabling effective molecular prediction using only WSIs at inference. Through extensive evaluation across 49 molecular oncology tasks using 11,257 cases among 20 cohorts, PathLUPI demonstrated superior performance compared to conventional methods trained solely on WSIs. Crucially, it achieves AUC $\geq$ 0.80 in 14 of the biomarker prediction and molecular subtyping tasks and C-index $\geq$ 0.70 in survival cohorts of 5 major cancer types. Moreover, PathLUPI embeddings reveal distinct cellular morphological signatures associated with specific genotypes and related biological pathways within WSIs. By effectively encoding molecular context to refine WSI representations, PathLUPI overcomes a key limitation of existing models and offers a novel strategy to bridge molecular insights with routine pathology workflows for wider clinical application.

Figures

Figures reproduced from arXiv: 2506.19681 by the authors.

Figure 1
Figure 1. Overview of the study. a. Scope of PathLUPI framework development and validation phases, covering 13 distinct cancer types among essential molecular oncology tasks: Investigational biomarker prediction, actionable biomarker prediction, molecular subtyping, and survival prognosis. b. Schematic of PathLUPI paradigm. By leveraging the learning using privileged information (LUPI) paradigm, the model integrates transcrip… view at source ↗
Figure 2
Figure 2. Results of biomarker prediction across 25 tasks. a. Bar charts of mean AUC for each method. In each subfigure, the relative improvement of PathLUPI over the best baseline method and corresponding statistical significance is indicated. Error bars represent 95% confidence intervals, and the centers correspond to the mean AUC values. b. Receiver operating characteristic (ROC) curves for each task. ROC curves were plott… view at source ↗
Figure 3
Figure 3. Results of molecular subtyping on 8 tasks. a. Bar charts of mean AUC for each method. In each subfigure, the relative improvement of PathLUPI over the best baseline method and corresponding statistical significance is indicated. Error bars represent 95% confidence intervals, and the centers correspond to the mean AUC values. b. Receiver operating characteristic (ROC) curves for each task. ROC curves were plotted by … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Results of survival prognosis on 16 tasks. a. Bar charts of mean AUC for each method. In each subfigure, the relative improvement of PathLUPI over the best baseline method and corresponding statistical significance is indicated. Error bars represent 95% confidence inte…
Figure 5
Figure 5. Figure 5: Local and global interpretability analyses of PathLUPI for two representative biomarker prediction tasks: colorectal cancer with BRAF mutation and lung adenocarcinoma with EGFR mutation. a. Spatial attention heatmaps highlighting histological regions most relevant for …
Figure 6
Figure 6. Figure 6: Ablation study evaluating the impact of different foundation models on PathLUPI performance across all internal tasks. a. Bridge plots summarizing model performance (AUC or C-index) for each task category using various feature extractors. The overall performance improv…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Clinically Validated Foundation Model for Comprehensive Lung Pathology Interpretation

    eess.IV 2026-05 unverdicted novelty 7.0 of 10

    PulmoFoundation achieves 92.3% average AUC on 32 lung pathology tasks in prospective validation and raises pathologist accuracy from 83.8% to 91.7% in a crossover RCT.

Reference graph

Works this paper leans on

82 extracted references · 70 canonical work pages · cited by 1 Pith paper

  1. [1]

    Signal transduction targeted therapy6, 1–48 (2021)

    Zhong, L.et al.Small molecules in targeted cancer therapy: advances, challenges, and future perspectives. Signal transduction targeted therapy6, 1–48 (2021)

    2021

  2. [2]

    & Chen, Y

    Cheng, Z., Li, M., Dey, R. & Chen, Y . Nanomaterials for cancer therapy: current progress and perspectives.J. hematology & oncology14, 1–27 (2021)

    2021

  3. [3]

    N., Halama, N

    Kather, J. N., Halama, N. & Jaeger, D. Genomics and emerging biomarkers for immunotherapy of colorectal cancer. InSeminars in cancer biology, vol. 52, 189–197 (Elsevier, 2018). 17/35

    work page 2018

  4. [4]

    Vasudevan, S., Saha, A., Tarver, M. E. & Patel, B. Digital biomarkers: convergence of digital health technologies and biomarkers.NPJ digital medicine5, 36 (2022)

    work page 2022

  5. [5]

    Medicine1–12 (2024)

    De Laere, B.et al.Androgen receptor pathway inhibitors and taxanes in metastatic prostate cancer: an outcome-adaptive randomized platform trial.Nat. Medicine1–12 (2024)

    work page 2024

  6. [6]

    P.et al.Quantifying the expanding landscape of clinical actionability for patients with cancer.Cancer Discov.14, 49–65 (2024)

    Suehnholz, S. P.et al.Quantifying the expanding landscape of clinical actionability for patients with cancer.Cancer Discov.14, 49–65 (2024)

    work page 2024

  7. [7]

    Medicine27, 1553–1563 (2021)

    van de Haar, J.et al.Limited evolution of the actionable metastatic cancer genome under therapeutic pressure.Nat. Medicine27, 1553–1563 (2021)

    work page 2021

  8. [8]

    H.et al.A multilineage screen identifies actionable synthetic lethal interactions in human cancers.Nat

    Fong, S. H.et al.A multilineage screen identifies actionable synthetic lethal interactions in human cancers.Nat. genetics57, 154–164 (2025)

    work page 2025

Show all 82 references
  1. [9]

    N.et al.Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer.Nat

    Kather, J. N.et al.Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer.Nat. medicine25, 1054–1056 (2019)

    2019

  2. [10]

    medicine30, 85–97 (2024)

    Amgad, M.et al.A population-level digital histologic biomarker for enhanced prognosis of invasive breast cancer.Nat. medicine30, 85–97 (2024)

    2024

  3. [11]

    N.et al.Pan-cancer image-based detection of clinically actionable genetic alterations.Nat

    Kather, J. N.et al.Pan-cancer image-based detection of clinically actionable genetic alterations.Nat. cancer1, 789–799 (2020)

    2020

  4. [12]

    Qu, H.et al.Genetic mutation and biological pathway prediction based on whole slide images in breast carcinoma using deep learning.NPJ precision oncology5, 87 (2021)

    2021

  5. [13]

    Liang, J.et al.Deep learning supported discovery of biomarkers for clinical prognosis of liver cancer. Nat. Mach. Intell.5, 408–420 (2023)

    2023

  6. [14]

    Medicine1–12 (2024)

    V olinsky-Fremond, S.et al.Prediction of recurrence risk in endometrial cancer with multimodal deep learning.Nat. Medicine1–12 (2024)

    2024

  7. [15]

    J.et al.Towards a general-purpose foundation model for computational pathology.Nat

    Chen, R. J.et al.Towards a general-purpose foundation model for computational pathology.Nat. Medicine30, 850–862 (2024)

    2024

  8. [16]

    Xu, H.et al.A whole-slide foundation model for digital pathology from real-world data.Nature630, 181–188 (2024)

    2024

  9. [17]

    Y .et al.A visual-language foundation model for computational pathology.Nat

    Lu, M. Y .et al.A visual-language foundation model for computational pathology.Nat. Medicine30, 863–874 (2024)

    2024

  10. [18]

    E.et al.Prevalence and heterogeneity of kras, braf, and pik3ca mutations in primary colorectal adenocarcinomas and their corresponding metastases.Clin

    Baldus, S. E.et al.Prevalence and heterogeneity of kras, braf, and pik3ca mutations in primary colorectal adenocarcinomas and their corresponding metastases.Clin. Cancer Res.16, 790–799 (2010)

    2010

  11. [19]

    M.et al.Generalizable biomarker prediction from cancer pathology slides with self- supervised deep learning: A retrospective multi-centric study.Cell reports Medicine4(2023)

    Niehues, J. M.et al.Generalizable biomarker prediction from cancer pathology slides with self- supervised deep learning: A retrospective multi-centric study.Cell reports Medicine4(2023)

    2023

  12. [20]

    & Vashist, A

    Vapnik, V . & Vashist, A. A new learning paradigm: Learning using privileged information.Neural networks22, 544–557 (2009)

    2009

  13. [21]

    & Vapnik, V

    Lopez-Paz, D., Bottou, L., Schölkopf, B. & Vapnik, V . Unifying distillation and privileged information. InThe F ourth International Conference on Learning Representations(2016). 22.Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation.cell144, 646–674 (2011)

    2016

  14. [23]

    N.et al.Comprehensive molecular characterization of urothelial bladder carcinoma.Nature 507, 315–322 (2014)

    Weinstein, J. N.et al.Comprehensive molecular characterization of urothelial bladder carcinoma.Nature 507, 315–322 (2014)

    2014

  15. [24]

    C.et al.Comprehensive molecular portraits of human breast tumours.Nature490, 61–70 (2012)

    Koboldt, D. C.et al.Comprehensive molecular portraits of human breast tumours.Nature490, 61–70 (2012). 18/35

    2012

  16. [25]

    medicine21, 1350–1356 (2015)

    Guinney, J.et al.The consensus molecular subtypes of colorectal cancer.Nat. medicine21, 1350–1356 (2015)

    2015

  17. [26]

    Liu, Y .et al.Comparative molecular analysis of gastrointestinal adenocarcinomas.Cancer cell33, 721–735 (2018)

    2018

  18. [27]

    S.et al.Comprehensive genomic characterization of head and neck squamous cell carcinomas.Nature517, 576–582 (2015)

    Lawrence, M. S.et al.Comprehensive genomic characterization of head and neck squamous cell carcinomas.Nature517, 576–582 (2015)

    2015

  19. [28]

    J.et al.Comprehensive molecular characterization of clear cell renal cell carcinoma

    Creighton, C. J.et al.Comprehensive molecular characterization of clear cell renal cell carcinoma. Nature499, 43–49 (2013)

    2013

  20. [29]

    Ceccarelli, M.et al.Molecular profiling reveals biologically discrete subsets and pathways of progression in diffuse glioma.Cell164, 550–563 (2016)

    2016

  21. [30]

    Cell169, 1327–1341 (2017)

    Ally, A.et al.Comprehensive and integrative genomic characterization of hepatocellular carcinoma. Cell169, 1327–1341 (2017)

    2017

  22. [31]

    Network, C. G. A. R.et al.Comprehensive molecular profiling of lung adenocarcinoma.Nature511, 543 (2014)

    2014

  23. [32]

    Network, C. G. A. R.et al.Comprehensive genomic characterization of squamous cell lung cancers. Nature489, 519 (2012). 33.Akbani, R.et al.Genomic classification of cutaneous melanoma.Cell161, 1681–1696 (2015)

    2012

  24. [34]

    A.et al.Integrated genomic characterization of endometrial carcinoma.Nature497, 67–73 (2013)

    Levine, D. A.et al.Integrated genomic characterization of endometrial carcinoma.Nature497, 67–73 (2013)

    2013

  25. [35]

    J.et al.Connecting genomic alterations to cancer biology with proteomics: the nci clinical proteomic tumor analysis consortium.Cancer discovery3, 1108–1112 (2013)

    Ellis, M. J.et al.Connecting genomic alterations to cancer biology with proteomics: the nci clinical proteomic tumor analysis consortium.Cancer discovery3, 1108–1112 (2013)

    2013

  26. [36]

    Data 9, 55 (2022)

    Roetzer-Pejrimovsky, T.et al.The digital brain tumour atlas, an open histopathology resource.Sci. Data 9, 55 (2022)

    2022

  27. [37]

    J., Chowell, D

    Havel, J. J., Chowell, D. & Chan, T. A. The evolving landscape of biomarkers for checkpoint inhibitor immunotherapy.Nat. Rev. Cancer19, 133–150 (2019)

    2019

  28. [38]

    Commun.15, 4690 (2024)

    Juan Ramon, A.et al.Development and deployment of a histopathology-based deep learning algorithm for patient prescreening in a clinical trial.Nat. Commun.15, 4690 (2024)

    2024

  29. [39]

    & Welling, M

    Ilse, M., Tomczak, J. & Welling, M. Attention-based deep multiple instance learning. InInternational conference on machine learning, 2127–2136 (PMLR, 2018)

    2018

  30. [40]

    Y .et al.Data-efficient and weakly supervised computational pathology on whole-slide images

    Lu, M. Y .et al.Data-efficient and weakly supervised computational pathology on whole-slide images. Nat. biomedical engineering5, 555–570 (2021)

    2021

  31. [41]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 18802–18812 (2022)

    Zhang, H.et al.Dtfd-mil: Double-tier feature distillation multiple instance learning for histopathology whole slide image classification. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 18802–18812 (2022)

    2022

  32. [42]

    neural information processing systems34, 2136–2147 (2021)

    Shao, Z.et al.Transmil: Transformer based correlated multiple instance learning for whole slide image classification.Adv. neural information processing systems34, 2136–2147 (2021)

    2021

  33. [43]

    Dawson, S.-J.et al.Analysis of circulating tumor DNA to monitor metastatic breast cancer.New Engl. J. Medicine368, 1199–1209 (2013)

    2013

  34. [44]

    Mossé, C.et al.Pan-cancer precision medicine trial AcSé shows that molecular screening is feasible and provides clinical benefit in patients with advanced solid tumours.Annals Oncol.32, 626–636 (2021)

    2021

  35. [45]

    C.et al.Discrimination and calibration of clinical prediction models: users’ guides to the medical literature.Jama318, 1377–1384 (2017)

    Alba, A. C.et al.Discrimination and calibration of clinical prediction models: users’ guides to the medical literature.Jama318, 1377–1384 (2017). 19/35

    2017

  36. [46]

    Y .et al.From whole-slide image to biomarker prediction: end-to-end weakly supervised deep learning in computational pathology.Nat

    Lu, M. Y .et al.From whole-slide image to biomarker prediction: end-to-end weakly supervised deep learning in computational pathology.Nat. Protoc.19, 381–407 (2024)

    2024

  37. [47]

    K.et al.Deep learning in cancer genomics and histopathology.Genome Medicine16, 49 (2024)

    Wagner, S. K.et al.Deep learning in cancer genomics and histopathology.Genome Medicine16, 49 (2024)

    2024

  38. [48]

    D.et al.Oncogenic braf, unrestrained by tgf β-receptor signalling, drives right-sided colonic tumorigenesis.Nat

    Leach, J. D.et al.Oncogenic braf, unrestrained by tgf β-receptor signalling, drives right-sided colonic tumorigenesis.Nat. communications12, 3464 (2021)

    2021

  39. [49]

    cancer17, 1–14 (2018)

    Román, M.et al.Kras oncogene in non-small cell lung cancer: clinical perspectives on the treatment of an old target.Mol. cancer17, 1–14 (2018)

    2018

  40. [50]

    Russell, P. A.et al.Correlation of mutation status and survival with predominant histologic subtype according to the new iaslc/ats/ers lung adenocarcinoma classification in stage iii (n2) patients.J. Thorac. Oncol.8, 461–468 (2013)

    2013

  41. [51]

    medicine24, 1559–1567 (2018)

    Coudray, N.et al.Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning.Nat. medicine24, 1559–1567 (2018)

    2018

  42. [52]

    Image Analysis94, 103143 (2024)

    Hörst, F.et al.Cellvit: Vision transformers for precise cell segmentation and classification.Med. Image Analysis94, 103143 (2024)

    2024

  43. [53]

    Hörst, F.et al.Cellvit++: Energy-efficient and adaptive cell segmentation and classification using foundation models.arXiv preprint arXiv:2501.05269(2025)

    2025 arXiv

  44. [54]

    F.et al.Intra-tumour diversification in colorectal cancer at the single-cell level.Nature556, 457–462 (2018)

    Roerink, S. F.et al.Intra-tumour diversification in colorectal cancer at the single-cell level.Nature556, 457–462 (2018)

    2018

  45. [55]

    Tian, J.et al.Combined pd-1, braf and mek inhibition in brafv600e colorectal cancer: a phase 2 trial. Nat. medicine29, 458–466 (2023)

    2023

  46. [56]

    J.et al.Single-cell analysis reveals prognostic fibroblast subpopulations linked to molecular and immunological subtypes of lung cancer.Nat

    Hanley, C. J.et al.Single-cell analysis reveals prognostic fibroblast subpopulations linked to molecular and immunological subtypes of lung cancer.Nat. communications14, 387 (2023)

    2023

  47. [57]

    Han, G.et al.An atlas of epithelial cell states and plasticity in lung adenocarcinoma.Nature627, 656–663 (2024)

    2024

  48. [58]

    & Yan, Q

    Sundararajan, M., Taly, A. & Yan, Q. Axiomatic attribution for deep networks. InInternational conference on machine learning, 3319–3328 (PMLR, 2017)

    2017

  49. [59]

    C.et al.Non-canonical functions of snail drive context-specific cancer progression.Nat

    Paul, M. C.et al.Non-canonical functions of snail drive context-specific cancer progression.Nat. Commun.14, 1201 (2023)

    2023

  50. [60]

    & Virshup, D

    Zhong, Z. & Virshup, D. M. Recurrent mutations in tumor suppressor fbxw7 bypass wnt/ β-catenin addiction in cancer.Sci. Adv.10, eadk1031 (2024)

    2024

  51. [61]

    E.et al.Determining the erk-regulated phosphoproteome driving kras-mutant cancer.Science 384, eadk0850 (2024)

    Klomp, J. E.et al.Determining the erk-regulated phosphoproteome driving kras-mutant cancer.Science 384, eadk0850 (2024)

    2024

  52. [62]

    Xu, J.-Y .et al.Integrative proteomic characterization of human lung adenocarcinoma.Cell182, 245–261 (2020)

    2020

  53. [63]

    for Immunother

    Xiao, G.et al.Heterogeneity of tumor immune microenvironment of egfr/alk-positive tumors versus egfr/alk-negative tumors in resected brain metastases from lung adenocarcinoma.J. for Immunother. Cancer11, e006243 (2023)

    2023

  54. [64]

    Huang, Z., Bianchi, F., Yuksekgonul, M., Montine, T. J. & Zou, J. A visual–language foundation model for pathology image analysis using medical twitter.Nat. medicine29, 2307–2316 (2023)

    2023

  55. [65]

    Cristofanilli, M.et al.Circulating tumor cells, disease progression, and survival in metastatic breast cancer.New Engl. J. Medicine351, 781–791 (2004). 20/35

    2004

  56. [66]

    medicine25, 1534–1539 (2019)

    Chemi, F.et al.Pulmonary venous circulating tumor cell dissemination before tumor resection and disease relapse.Nat. medicine25, 1534–1539 (2019)

    2019

  57. [67]

    Lin, D.et al.Circulating tumor cells: biology and clinical significance.Signal transduction targeted therapy6, 404 (2021)

    2021

  58. [68]

    Milacic, M.et al.The reactome pathway knowledgebase 2024.Nucleic acids research52, D672–D678 (2024)

    2024

  59. [69]

    & Ishiguro-Watanabe, M

    Kanehisa, M., Furumichi, M., Sato, Y ., Matsuura, Y . & Ishiguro-Watanabe, M. Kegg: biological systems database as a model of the real world.Nucleic Acids Res.53, D672–D677 (2025)

    2025

  60. [70]

    & Chen, H

    Zhou, F. & Chen, H. Cross-modal translation and alignment for survival analysis. InProceedings of the IEEE/CVF International Conference on Computer Vision, 21485–21494 (2023)

    2023

  61. [71]

    & Chen, H

    Xu, Y . & Chen, H. Multimodal optimal transport-based co-attention transformer with global structure consistency for survival prediction. InProceedings of the IEEE/CVF international conference on computer vision, 21241–21251 (2023)

    2023

  62. [72]

    Cui, S.et al.Harnessing multimodal data integration to advance precision oncology.Nat. Rev. Cancer 22, 141–154, DOI: 10.1038/s41568-021-00408-3 (2022)

    2022 doi

  63. [73]

    Zhu, Y .et al.Multimodal data fusion for cancer biomarker discovery with deep learning.Nat. Mach. Intell.5, 345–357, DOI: 10.1038/s42256-023-00633-5 (2023)

    2023 doi

  64. [74]

    K., Ovchinnikov, S

    Wayment-Steele, H. K., Ovchinnikov, S. & Colwell, L. J. Towards multimodal foundation models in molecular cell biology.NatureDOI: 10.1038/s41586-024-07499-z (2024). Published online 29 May 2024

    2024 doi

  65. [75]

    Cerami, E.et al.The cbio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data.Cancer discovery2, 401–404 (2012)

    2012

  66. [76]

    J.et al.Visualizing and interpreting cancer genomics data via the xena platform.Nat

    Goldman, M. J.et al.Visualizing and interpreting cancer genomics data via the xena platform.Nat. biotechnology38, 675–678 (2020). 77.Mendiratta, G.et al.Cancer gene mutation frequencies for the us population.Nat. communications12, 5961 (2021)

    2020

  67. [78]

    InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 11343–11352 (2024)

    Tang, W.et al.Feature re-embedding: Towards foundation model-level performance in computational pathology. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 11343–11352 (2024)

    2024

  68. [79]

    Liberzon, A.et al.The molecular signatures database hallmark gene set collection.Cell systems1, 417–425 (2015)

    2015

  69. [80]

    neural information processing systems32(2019)

    Paszke, A.et al.Pytorch: An imperative style, high-performance deep learning library.Adv. neural information processing systems32(2019)

    2019

  70. [81]

    M.et al.Explainable machine-learning predictions for the prevention of hypoxaemia during surgery.Nat

    Lundberg, S. M.et al.Explainable machine-learning predictions for the prevention of hypoxaemia during surgery.Nat. Biomed. Eng.2, 749 (2018)

    2018

  71. [82]

    Individual comparisons by ranking methods

    Wilcoxon, F. Individual comparisons by ranking methods. InBreakthroughs in statistics: Methodology and distribution, 196–202 (Springer, 1992)

    1992

  72. [83]

    arXiv preprint arXiv:2407.18449(2024)

    Ma, J.et al.Towards a generalizable pathology foundation model via unified knowledge distillation. arXiv preprint arXiv:2407.18449(2024)

    2024 arXiv

  73. [84]

    Xu, Y .et al.A multimodal knowledge-enhanced whole-slide pathology foundation model.arXiv preprint arXiv:2407.15362(2024). 21/35

    2024 arXiv

  74. [85]

    N.et al.The cancer genome atlas pan-cancer analysis project.Nat

    Weinstein, J. N.et al.The cancer genome atlas pan-cancer analysis project.Nat. genetics45, 1113–1120 (2013). 22/35 Supplementary BLCA-FGFR3 TCGA-CF-A7I0 Whole Slide Image Attention Heatmap TCGA-BT-A0S7 Attention Weight 0.0 1.0 Patch Visualization Attention Weight 0.0 1.0 Cell ...

    2013

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.