pith. sign in

arxiv: 2606.09898 · v1 · pith:T5JKEEQSnew · submitted 2026-06-05 · 💻 cs.LG · cs.MA· q-bio.QM

TRAPS: Therapeutic Response Analysis via Pathway-informed Stratification

Sujoy Banik , Sayantan Chakraborty , Boishakhi Das Toma , Zainab Ghafoor , Ushashi Bhattacharjee , Koushik Howlader , Tirtho Roy This is my paper

Pith reviewed 2026-06-27 22:59 UTC · model grok-4.3

classification 💻 cs.LG cs.MAq-bio.QM
keywords pathway-informed deep learningtherapy response predictionunified benchmarkTCGA cohortsReactome pathwaysGraphPathtargeted molecular therapysurvival prediction
0
0 comments X

The pith

A unified benchmark shows pathway models have task-specific strengths, with GraphPath achieving an AUROC of 0.92 on prostate targeted molecular therapy prediction.

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

The paper introduces the first unified benchmark for pathway-guided therapy response modeling by evaluating three architectures—BINN, GraphPath, and PATH—jointly on targeted molecular therapy, radiation therapy, and six-month survival prediction across five TCGA cancer cohorts with 2,622 patients using Reactome pathway activity scores. Results indicate that no single model excels across all tasks, with PATH performing best overall for targeted therapy, BINN for survival, and none useful for radiation therapy due to reliance on clinical variables absent from gene expression data. Notably, GraphPath attains the highest score of 0.92 AUROC on prostate targeted therapy despite only 11% positive prevalence, highlighting the value of lateral co-regulation structure in cohorts with narrow targetable drivers.

Core claim

We present the first unified benchmark for pathway-guided therapy response modeling, evaluating BINN, GraphPath, and PATH across five cancer cohorts from The Cancer Genome Atlas representing 2,622 patients encoded using Reactome pathway activity scores. Each model is trained jointly on targeted molecular therapy, radiation therapy, and survival prediction under identical conditions, showing task-dependent performance with GraphPath reaching 0.92 AUROC on prostate targeted molecular therapy.

What carries the argument

The unified benchmark that jointly evaluates biologically informed deep learning architectures on multiple clinical outcomes using Reactome pathway activity scores.

If this is right

  • Model performance varies by clinical task, requiring task-specific architecture selection.
  • Lateral co-regulation in pathways enables high performance for targeted therapies in cancers with focused driver programs even under class imbalance.
  • Radiation therapy prediction requires clinical variables beyond gene expression pathway scores.
  • Joint multi-outcome training allows fair comparison of pathway-informed models.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Different models could be deployed depending on the therapy decision at hand in clinical settings.
  • Future models might benefit from combining gene expression with clinical data for radiation decisions.
  • The benchmark could be extended to additional cancer types or pathway databases to test generalizability.
  • High AUROC in imbalanced data suggests utility for identifying rare responsive patients.

Load-bearing premise

Reactome pathway activity scores provide a sufficient representation of the biological information needed for therapy response modeling across the evaluated tasks.

What would settle it

A test showing whether adding clinical variables significantly improves radiation therapy prediction performance beyond what pathway scores alone achieve.

Figures

Figures reproduced from arXiv: 2606.09898 by Boishakhi Das Toma, Koushik Howlader, Sayantan Chakraborty, Sujoy Banik, Tirtho Roy, Ushashi Bhattacharjee, Zainab Ghafoor.

Figure 1
Figure 1. Figure 1: Workflow of the proposed method Pathway Adjacency Graph. A symmetric binary adjacency matrix 𝐴 ∈ {0, 1} 𝑁 ×𝑁 is constructed such that: 𝐴𝑝𝑞 = 1 ⇐⇒ pathways 𝑝 and 𝑞 share a Reactome parent-child or sibling relationship. (10) This yields 4,672 undirected edges with a mean node degree of 5.5. Self-loops are added so that each node attends to its own embedding during message passing. Input Projection. Each scal… view at source ↗
Figure 2
Figure 2. Figure 2: Held-out test-set F1, AUPRC, and AUROC per clinical head (TMT, RT, OS) across all five TCGA cohorts. BINN blue, [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Confusion-matrix counts on the test split at the 0.5 [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Normalised pathway importance scores for BINN and GraphPath across all three clinical heads. Red arrows mark [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

Cancer treatment planning requires decisions across multiple clinical dimensions at once. Clinicians must determine whether a patient should receive targeted molecular therapy, radiation therapy, and whether they are likely to survive beyond six months. Existing pathway-informed deep learning models have been developed and tested in isolation, making fair comparison across architectures impossible. We present the first unified benchmark for pathway-guided therapy response modeling, evaluating three biologically informed architectures, BINN, GraphPath, and PATH, across five cancer cohorts drawn from The Cancer Genome Atlas, representing 2,622 patients encoded using Reactome pathway activity scores. Each model is trained jointly on all three clinical outcomes under identical data and evaluation conditions, the first study to treat pathway-structured deep learning as a combined therapy and survival prediction problem. Our results show that no single architecture wins across all tasks: PATH performs best for targeted molecular therapy prediction overall, BINN is most reliable for survival prediction, and no model produces useful predictions for radiation therapy, as the key drivers of that decision are clinical variables not captured in gene expression data. Most strikingly, GraphPath achieves an AUROC of 0.92 on prostate targeted molecular therapy prediction, the highest score in the entire benchmark, demonstrating that lateral co-regulation structure produces exceptional discriminative power when matched to a cohort with a narrow targetable driver programme, even under conditions of extreme class imbalance at only 11\% positive prevalence.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces TRAPS, the first unified benchmark evaluating three pathway-informed architectures (BINN, GraphPath, PATH) on joint prediction of targeted molecular therapy response, radiation therapy response, and 6-month survival. It uses Reactome pathway activity scores from 2,622 TCGA patients across five cancer cohorts, with all models trained under identical multi-task conditions. Results indicate no architecture dominates all tasks (PATH strongest overall for targeted therapy, BINN for survival), no model is useful for radiation therapy (attributed to missing clinical variables), and GraphPath reaches 0.92 AUROC on prostate targeted therapy despite 11% positive prevalence, which the authors attribute to the benefits of modeling lateral co-regulation structure.

Significance. If the empirical results and data-processing pipeline hold under full scrutiny, the work supplies a reproducible, standardized multi-task benchmark that enables fair head-to-head comparison of biologically structured models—an explicit strength. The explicit acknowledgment that radiation decisions lie outside gene-expression data is a useful negative result. The GraphPath prostate result, if robust, supplies a concrete, falsifiable data point for the utility of graph-based lateral co-regulation inductive biases under narrow-driver, imbalanced regimes.

major comments (1)
  1. [Abstract] Abstract: The headline claim that GraphPath’s 0.92 AUROC demonstrates the discriminative power of lateral co-regulation structure under narrow driver programmes rests on the untested premise that Reactome pathway activity scores alone encode the biologically relevant signals for targeted molecular therapy response. The abstract correctly states that radiation decisions hinge on clinical variables absent from gene expression, yet supplies no parallel confirmation, ablation, or external validation (e.g., addition of mutation status or clinical covariates) that targeted-therapy labels are adequately represented by the Reactome features. This assumption is load-bearing for interpreting the result as architectural evidence rather than dataset artifact or leakage.
minor comments (2)
  1. [Abstract] Abstract: The five cancer cohorts are not enumerated; explicit listing would aid readers in assessing generalizability.
  2. [Abstract] Abstract: No mention is made of the statistical procedure used to establish that 0.92 is meaningfully higher than the other models or of how class imbalance was handled during training and evaluation.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback on our manuscript. We address the single major comment below and indicate where revisions will be made to strengthen the presentation.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The headline claim that GraphPath’s 0.92 AUROC demonstrates the discriminative power of lateral co-regulation structure under narrow driver programmes rests on the untested premise that Reactome pathway activity scores alone encode the biologically relevant signals for targeted molecular therapy response. The abstract correctly states that radiation decisions hinge on clinical variables absent from gene expression, yet supplies no parallel confirmation, ablation, or external validation (e.g., addition of mutation status or clinical covariates) that targeted-therapy labels are adequately represented by the Reactome features. This assumption is load-bearing for interpreting the result as architectural evidence rather than dataset artifact or leakage.

    Authors: We appreciate the referee’s observation that the manuscript draws an explicit parallel for radiation therapy but does not supply an equivalent statement or experiment for targeted therapy. The core contribution of the work is a controlled, multi-task benchmark that holds data, preprocessing, and training protocol fixed while varying only the model architecture; performance differences are therefore attributable to architectural inductive biases rather than differences in input representation. Nevertheless, we agree that the absolute claim that Reactome pathway scores are sufficient to represent the targeted-therapy signal would be strengthened by an explicit caveat or limited ablation. In the revised manuscript we will (i) insert a qualifying clause in the abstract stating that results are conditioned on Reactome-derived features and (ii) add a short paragraph in the Discussion acknowledging that incorporation of mutation status or clinical covariates could further validate the feature set and is left for future work. These textual changes will make the scope and limitations of the interpretation transparent without altering the benchmark design. revision: partial

Circularity Check

0 steps flagged

Empirical benchmark evaluation with no derivation chain

full rationale

The paper reports results from a unified benchmark comparing three architectures (BINN, GraphPath, PATH) on TCGA cohorts using Reactome pathway activity scores. All performance metrics (e.g., AUROC 0.92) are obtained via direct evaluation on held-out data under joint training on multiple outcomes. No equations, first-principles derivations, or parameter-fitting steps are described that reduce predictions to inputs by construction. The abstract explicitly notes limitations for radiation therapy due to missing clinical variables, confirming the work is empirical rather than deductive. No self-citation chains or ansatzes are invoked as load-bearing for the central claims.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no information on free parameters, axioms, or invented entities provided.

pith-pipeline@v0.9.1-grok · 5814 in / 1090 out tokens · 27515 ms · 2026-06-27T22:59:30.319847+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

62 extracted references · 36 canonical work pages

  1. [1]

    Barbie, Pablo Tamayo, Jesse S

    David A. Barbie, Pablo Tamayo, Jesse S. Boehm, So Young Kim, Susan E. Moody, Ian F. Dunn, Anna C. Schinzel, Peter Sandy, Etienne Meylan, Claudia Scholl, et al

  2. [2]

    doi:10.1038/nature08460

    Systematic RNA interference reveals that oncogenic KRAS-driven cancers require TBK1.Nature462, 7269 (2009), 108–112. doi:10.1038/nature08460

    work page doi:10.1038/nature08460 2009

  3. [3]

    Rajamanickam Baskar, Kuo Ann Lee, Richard Yeo, and Kheng-Wei Yeoh. 2012. Cancer and radiation therapy: current advances and future directions.Interna- tional Journal of Medical Sciences9, 3 (2012), 193–199. doi:10.7150/ijms.3635

    work page doi:10.7150/ijms.3635 2012

  4. [4]

    Harald Binder et al. 2020. Adaptive ERK Signalling Activation in Response to Therapy and In Silico Prognostic Evaluation of EGFR-MAPK in HNSCC.British Journal of Cancer(2020). doi:10.1038/s41416-020-0892-9

    work page doi:10.1038/s41416-020-0892-9 2020

  5. [5]

    Jim Clauwaert, Gerben Menschaert, and Willem Waegeman. 2021. Explainability in transformer models for functional genomics.Briefings in Bioinformatics22, 5 (2021), bbab060. doi:10.1093/bib/bbab060

    work page doi:10.1093/bib/bbab060 2021

  6. [6]

    Dhillon, Suzanne Hagan, Oliver Rath, and Walter Kolch

    Amardeep S. Dhillon, Suzanne Hagan, Oliver Rath, and Walter Kolch. 2007. MAP Kinase Signalling Pathways in Cancer.Oncogene26, 22 (2007), 3279–3290. doi:10.1038/sj.onc.1210421

    work page doi:10.1038/sj.onc.1210421 2007

  7. [7]

    Yifan Dou and Golrokh Mirzaei. 2025. MO-GCAN: multi-omics integration based on graph convolutional and attention networks.Bioinformatics41, 8 (2025), btaf405. doi:10.1093/bioinformatics/btaf405

    work page doi:10.1093/bioinformatics/btaf405 2025

  8. [8]

    Vijay Prakash Dwivedi and Xavier Bresson. 2021. A Generalization of Trans- former Networks to Graphs.AAAI Workshop on Deep Learning on Graphs: Methods and Applications(2021). https://arxiv.org/abs/2012.09699

    arXiv 2021

  9. [9]

    Elmarakeby, Justin Hwang, Rand Arafeh, Jett Crowdis, Sydney Gang, David Liu, Saud H

    Haitham A. Elmarakeby, Justin Hwang, Rand Arafeh, Jett Crowdis, Sydney Gang, David Liu, Saud H. AlDubayan, Keyan Salari, Steven Kregel, Camden Richter, et al

  10. [10]

    Nature598, 7880 (2021), 348–352

    Biologically informed deep neural network for prostate cancer discovery. Nature598, 7880 (2021), 348–352. doi:10.1038/s41586-021-03922-4

    work page doi:10.1038/s41586-021-03922-4 2021

  11. [11]

    Fisch et al

    Alexander S. Fisch et al. 2022. Precision Drugging of the MAPK Pathway in Head and Neck Cancer.npj Genomic Medicine(2022). doi:10.1038/s41525-022-00293-1

    work page doi:10.1038/s41525-022-00293-1 2022

  12. [12]

    Yan-Li Ge et al. 2013. MiR-200c Increases the Radiosensitivity of NSCLC Cell Line A549 by Targeting VEGF-VEGFR2 Pathway.PLOS ONE(2013). https: //www.ncbi.nlm.nih.gov/pmc/articles/PMC3813610/ PMC3813610

    2013

  13. [13]

    Marc Gillespie, Bijay Jassal, Ralf Stephan, Marija Milacic, Karen Rothfels, Andrea Senff-Ribeiro, Johannes Griss, Cristoffer Sevilla, Lisa Matthews, Chuqiao Gong, et al. 2022. The Reactome pathway knowledgebase 2022.Nucleic Acids Research 50, D1 (2022), D687–D692. doi:10.1093/nar/gkab1028

    work page doi:10.1093/nar/gkab1028 2022

  14. [14]

    Goldman, Brian Craft, Mim Hastie, Kristupas Repečka, Fran McDade, Akhil Kamath, Ayan Banerjee, Yunhai Luo, Dave Rogers, Angela N

    Mary J. Goldman, Brian Craft, Mim Hastie, Kristupas Repečka, Fran McDade, Akhil Kamath, Ayan Banerjee, Yunhai Luo, Dave Rogers, Angela N. Brooks, et al

  15. [15]

    Nature Biotechnology38 (2020), 675–678

    Visualizing and interpreting cancer genomics data via the Xena platform. Nature Biotechnology38 (2020), 675–678. doi:10.1038/s41587-020-0546-8

    work page doi:10.1038/s41587-020-0546-8 2020

  16. [16]

    Scott, Christofer Karlsson, Tirthankar Mohanty, Suvi T

    Erik Hartman, Aaron M. Scott, Christofer Karlsson, Tirthankar Mohanty, Suvi T. Vaara, Adam Linder, Lars Malmström, and Johan Malmström. 2023. Interpret- ing biologically informed neural networks for enhanced proteomic biomarker discovery and pathway analysis.Nature Communications14 (2023), 5359. doi:10.1038/s41467-023-41146-4

    work page doi:10.1038/s41467-023-41146-4 2023

  17. [17]

    Koushik Howlader, Md Tauhidul Islam, and Wei Le. 2026. Graph Transformer- Based Pathway Embedding for Cancer Prognosis. arXiv:2604.16685 [cs.LG] https://arxiv.org/abs/2604.16685

    Pith/arXiv arXiv 2026

  18. [18]

    Johnson, Bryan R

    Zhi Huang, Xiaohui Zhan, Shuo Xiang, Travis S. Johnson, Bryan R. Helm, Christina Y. Yu, Jie Zhang, Paul Salama, Monther Rizkalla, Zhi Han, and Kun Huang. 2019. SALMON: Survival Analysis Learning With Multi-Omics Neural Networks on Breast Cancer.Frontiers in Genetics10 (2019), 166. doi:10.3389/ fgene.2019.00166

    arXiv 2019

  19. [19]

    Yuexu Jiang, Manish Sridhar Immadi, Duolin Wang, Shuai Zeng, Yen On Chan, Jing Zhou, Dong Xu, and Trupti Joshi. 2024. IRnet: Immunotherapy response prediction using pathway knowledge-informed graph neural network.Journal of Advanced Research72 (2024), 319–331. doi:10.1016/j.jare.2024.07.036

    work page doi:10.1016/j.jare.2024.07.036 2024

  20. [20]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. InInternational Conference on Learning Representations (ICLR). https://arxiv.org/abs/1412.6980

    Pith/arXiv arXiv 2015

  21. [21]

    Hiroki Kudo et al . 2025. Androgen Receptor Contributes to Radioresistance Through DNA Repair and Autophagy in AR-Positive Prostate Cancer Cells. bioRxiv. doi:10.64898/2025.12.03.690226

    work page doi:10.64898/2025.12.03.690226 2025

  22. [22]

    Wei Lan, Haibo Liao, Qingfeng Chen, Ling-ling Zhu, Yi Pan, and Yi-Ping Phoebe Chen. 2024. DeepKEGG: a multi-omics data integration framework with biologi- cal insights for cancer recurrence prediction and biomarker discovery.Briefings in Bioinformatics25, 3 (2024), bbae185. doi:10.1093/bib/bbae185

    work page doi:10.1093/bib/bbae185 2024

  23. [23]

    Dohee Lee, Jaegyoon Ahn, and Jonghwan Choi. 2025. PathNetDRP: a novel biomarker discovery framework using pathway and protein–protein interaction networks for immune checkpoint inhibitor response prediction.BMC Bioinfor- matics26, 1 (2025), 119. doi:10.1186/s12859-025-06125-0

    work page doi:10.1186/s12859-025-06125-0 2025

  24. [24]

    Jin, Mingzhu Lou, Shaobo Deng, Lei Wang, and Hua Rao

    Min Li, M. Jin, Mingzhu Lou, Shaobo Deng, Lei Wang, and Hua Rao. 2025. Multiview-cooperated graph neural network enables novel multi-omics can- cer subtype classification.Computational Biology and Chemistry(2025). doi:10. 1016/j.compbiolchem.2025.108560

    arXiv 2025

  25. [25]

    Xiangmei Li, Bin Pan, Yao He, et al. 2025. PathHDNN: a pathway hierarchical- informed deep neural network framework for predicting immunotherapy re- sponse and mechanism interpretation.Genome Medicine17 (2025), 152. doi:10. 1186/s13073-025-01584-9

    2025

  26. [26]

    Cheng-Pei Lin, Yann-Jen Ho, Yen-Peng Chiu, Yun Tang, You Sheng Paik, Guan Chen, Wei-Chih Huang, and Tzong-Yi Lee. 2026. MoAGNN: a multi-omics hier- archical graph neural network for subtype classification and prognosis predic- tion in lung adenocarcinoma.Briefings in Bioinformatics27, 1 (2026), bbaf735. doi:10.1093/bib/bbaf735

    work page doi:10.1093/bib/bbaf735 2026

  27. [27]

    Xiaofan Liu, Yuhuan Tao, Zilin Cai, Pengfei Bao, Hongli Ma, Kexing Li, Mingyue Li, Yongchang Zhu, and Zhi John Lu. 2024. Pathformer: a biological pathway informed transformer for disease diagnosis and prognosis using multi-omics data. Bioinformatics40, 5 (2024), btae316. doi:10.1093/bioinformatics/btae316

    work page doi:10.1093/bioinformatics/btae316 2024

  28. [28]

    Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. InInternational Conference on Learning Representations (ICLR). https://arxiv.org/ abs/1711.05101

    Pith/arXiv arXiv 2019

  29. [29]

    Lundberg and Su-In Lee

    Scott M. Lundberg and Su-In Lee. 2017. A unified approach to interpret- ing model predictions. InAdvances in Neural Information Processing Sys- tems (NeurIPS), Vol. 30. https://papers.nips.cc/paper_files/paper/2017/hash/ 8a20a8621978632d76c43dfd28b67767-Abstract.html

    2017

  30. [30]

    Teng Ma and Jianxin Wang. 2024. GraphPath: a graph attention model for molecu- lar stratification with interpretability based on the pathway–pathway interaction network.Bioinformatics40, 4 (2024), btae165. doi:10.1093/bioinformatics/btae165

    work page doi:10.1093/bioinformatics/btae165 2024

  31. [31]

    Teng Ma, Haochen Zhao, Qichang Zhao, and Jianxin Wang. 2024. Cox-Path: Biological Pathway-Informed Graph Neural Network for Cancer Survival Pre- diction. InProceedings of the 15th ACM International Conference on Bioinfor- matics, Computational Biology and Health Informatics (BCB ’24). 70:1–70:6. doi:10.1145/3698587.3701397

    work page doi:10.1145/3698587.3701397 2024

  32. [32]

    Hye-Young Min and Ho-Young Lee. 2022. Molecular targeted therapy for anti- cancer treatment.Experimental & Molecular Medicine54, 10 (2022), 1670–1694. doi:10.1038/s12276-022-00864-3

    work page doi:10.1038/s12276-022-00864-3 2022

  33. [33]

    Masataka Okabe and Kei Ito. 2008. Color universal design (CUD): How to make figures and presentations that are friendly to colorblind people.Color Universal Design Organization technical note(2008). https://jfly.uni-koeln.de/color/

    2008

  34. [34]

    Parikh, Christopher Manz, Corey Chivers, et al

    Ravi B. Parikh, Christopher Manz, Corey Chivers, et al. 2019. Machine learning approaches to predict 6-month mortality among patients with cancer.JAMA Network Open2, 10 (2019), e1915997. doi:10.1001/jamanetworkopen.2019.15997

    work page doi:10.1001/jamanetworkopen.2019.15997 2019

  35. [35]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, et al. 2019. PyTorch: An imperative style, high-performance deep learning library.Advances in Neural Information Processing Systems32 (2019). https://papers.nips.cc/paper_ files/paper/2019/hash/bdbca288fee7f92f2...

    2019

  36. [36]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, et al. 2011. Scikit-learn: Machine learning in Python.Journal of Machine Learning Research12 (2011), 2825–2830. https://www.jmlr.org/papers/v12/pedregosa11a.html

    2011

  37. [37]

    Ethan Perez, Florian Strub, Harm de Vries, Vincent Dumoulin, and Aaron Courville. 2018. FiLM: Visual reasoning with a general conditioning layer. In Proceedings of the AAAI Conference on Artificial Intelligence, Vol. 32. doi:10.1609/ aaai.v32i1.11671

    2018

  38. [38]

    Olivier B Poirion, Zheng Jing, Kuldeep Chaudhary, Sijia Huang, and Lana X Garmire. 2021. DeepProg: an ensemble of deep-learning and machine-learning models for prognosis prediction using multi-omics data.Genome Medicine13, 1 (2021), 112. doi:10.1186/s13073-021-00930-x

    work page doi:10.1186/s13073-021-00930-x 2021

  39. [39]

    Rikard Rosenbacke, Åsa Melhus, Martin McKee, and David Stuckler. 2024. How Explainable Artificial Intelligence Can Increase or Decrease Clinicians’ Trust in AI Applications in Health Care: Systematic Review.JMIR AI3 (2024), e53207. doi:10.2196/53207 9 ICCA ’26

    work page doi:10.2196/53207 2024

  40. [40]

    Decoding of the speech envelope from EEG using the VLAAI deep neural network,

    Chris J. Sidey-Gibbons, Charlotte Sun, Alina Schneider, et al. 2022. Predicting 180- day mortality for women with ovarian cancer using machine learning and patient- reported outcome data.Scientific Reports12, 1 (2022), 20614. doi:10.1038/s41598- 022-22614-1

    work page doi:10.1038/s41598- 2022

  41. [41]

    Srinivasa Prasad Sisinthy and Dhruv L. Bhatt. 2021. Everything Old Is New Again: Drug Repurposing Approach for NSCLC Targeting MAPK Signaling Pathway. Frontiers in Oncology11 (2021). doi:10.3389/fonc.2021.741326

    work page doi:10.3389/fonc.2021.741326 2021

  42. [42]

    Mootha, Sayan Mukherjee, Ben- jamin L

    Aravind Subramanian, Pablo Tamayo, Vamsi K. Mootha, Sayan Mukherjee, Ben- jamin L. Ebert, Michael A. Gillette, Amanda Paulovich, Scott L. Pomeroy, Todd R. Golub, Eric S. Lander, and Jill P. Mesirov. 2005. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Scienc...

    work page doi:10.1073/pnas.0506580102 2005

  43. [43]

    The Cancer Genome Atlas Network. 2012. Comprehensive molecular portraits of human breast tumours.Nature490, 7418 (2012), 61–70. doi:10.1038/nature11412

    work page doi:10.1038/nature11412 2012

  44. [44]

    Petar Veličković, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. 2018. Graph attention networks. InInternational Confer- ence on Learning Representations (ICLR). https://arxiv.org/abs/1710.10903

    Pith/arXiv arXiv 2018

  45. [45]

    Di Wang, Chupei Tang, Junxiao Kong, Jixiu Zhai, Moyu Tang, and Tianchi Lu

  46. [46]

    arXiv:2604.24371 [cs.LG] https://arxiv.org/abs/2604

    PathMoG: A Pathway-Centric Modular Graph Neural Network for Multi- Omics Survival Prediction. arXiv:2604.24371 [cs.LG] https://arxiv.org/abs/2604. 24371

    Pith/arXiv arXiv

  47. [47]

    Johnson, Paul A

    Asim Waqas, Aakash Tripathi, Sabeen Ahmed, Ashwin Mukund, Hamza Farooq, Joseph O. Johnson, Paul A. Stewart, Mia Naeini, Matthew B. Schabath, and Ghulam Rasool. 2025. Self-Normalizing Multi-Omics Neural Network for Pan- Cancer Prognostication.International Journal of Molecular Sciences26, 15 (2025),

    2025

  48. [48]

    doi:10.3390/ijms26157358

    work page doi:10.3390/ijms26157358

  49. [49]

    Weinstein, Eric A

    John N. Weinstein, Eric A. Collisson, Gordon B. Mills, Kenna R. Mills Shaw, Brad A. Ozenberger, Kyle Ellrott, Ilya Shmulevich, Chris Sander, and Joshua M. Stuart. 2013. The Cancer Genome Atlas Pan-Cancer analysis project.Nature Genetics45, 10 (2013), 1113–1120. doi:10.1038/ng.2764

    work page doi:10.1038/ng.2764 2013

  50. [50]

    Gang Wen and Limin Li. 2023. FGCNSurv: dually fused graph convolutional network for multi-omics survival prediction.Bioinformatics39, 8 (2023), btad472. doi:10.1093/bioinformatics/btad472

    work page doi:10.1093/bioinformatics/btad472 2023

  51. [51]

    Bang Wong. 2011. Points of view: Color blindness.Nature Methods8, 6 (2011),

    2011

  52. [52]

    doi:10.1038/nmeth.1618

    work page doi:10.1038/nmeth.1618

  53. [53]

    Magdalena Wysocka, Oskar Wysocki, Marie Zufferey, Dónal Landers, and André Freitas. 2023. A systematic review of biologically-informed deep learning models for cancer: fundamental trends for encoding and interpreting oncology data. BMC Bioinformatics24 (2023), 198. doi:10.1186/s12859-023-05262-8

    work page doi:10.1186/s12859-023-05262-8 2023

  54. [54]

    Hongxi Yan, Dawei Weng, Dongguo Li, Yu Gu, Wenjie Ma, and Qingjie Liu

  55. [55]

    doi:10.1093/bib/bbae184

    Prior knowledge-guided multilevel graph neural network for tumor risk prediction and interpretation via multi-omics data integration.Briefings in Bioinformatics25, 3 (2024), bbae184. doi:10.1093/bib/bbae184

    work page doi:10.1093/bib/bbae184 2024

  56. [56]

    Jiayang Zhang, Yilin Che, Rongrong Liu, Zhicheng Wang, and Weiwu Liu. 2025. Deep learning–driven multi-omics analysis: enhancing cancer diagnostics and therapeutics.Briefings in Bioinformatics26, 4 (2025), bbaf440. doi:10.1093/bib/ bbaf440

    work page doi:10.1093/bib/ 2025

  57. [57]

    Tinghe Zhang, Md Hasib, Yu-Chiao Chiu, Zhizhong Han, Yu-Fang Jin, Mario Flo- res, Yidong Chen, and Yufei Huang. 2022. Transformer for Gene Expression Mod- eling (T-GEM): An Interpretable Deep Learning Model for Gene Expression-Based Phenotype Predictions.Cancers14, 19 (2022), 4763. doi:10.3390/cancers14194763

    work page doi:10.3390/cancers14194763 2022

  58. [58]

    Lianhe Zhao, Qiongye Dong, Chunlong Luo, Yang Wu, Dechao Bu, Xiaoning Qi, Yufan Luo, and Yi Zhao. 2021. DeepOmix: A scalable and interpretable multi-omics deep learning framework and application in cancer survival analysis. Computational and Structural Biotechnology Journal19 (2021), 2719–2725. doi:10. 1016/j.csbj.2021.04.067

    2021

  59. [59]

    Yue Zhao et al. 2015. Prognostic Values of ERK1/2 and p-ERK1/2 Expressions for Poor Survival in Non-Small Cell Lung Cancer.Tumor Biology(2015). doi:10. 1007/s13277-015-3048-4

    2015

  60. [60]

    Rong Zheng et al. 2015. TAT-ODD-p53 Enhances Radiosensitivity of Hypoxic Breast Cancer Cells by Inhibiting Parkin-Mediated Mitophagy.Oncotarget(2015). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4627318/ PMC4627318

    2015

  61. [61]

    Oh, Anish K

    Jiening Zhu, John H. Oh, Anish K. Simhal, Rena Elkin, Larry Norton, Joseph O. Deasy, and Allen Tannenbaum. 2023. Geometric graph neural networks on multi-omics data to predict cancer survival outcomes.Computers in Biology and Medicine163 (2023), 107117. doi:10.1016/j.compbiomed.2023.107117

    work page doi:10.1016/j.compbiomed.2023.107117 2023

  62. [62]

    Payam Zohari and Mostafa Haghir Chehreghani. 2025. Graph Neural Net- works in Multi-Omics Cancer Research: A Structured Survey.arXiv(2025). arXiv:2506.17234 https://arxiv.org/abs/2506.17234 10

    arXiv 2025