AI and Cancer Research ── 2026-10-23
New papers on cancer and AI, collected from journal feeds, PubMed (searched across all journals) and OpenAlex (a public index used instead of Google Scholar), judged by Jev (an AI that makes language judgements): original research, AI central, which of 13 themes, treatment or trials, and then ranked by importance up to a daily limit. Journals are not filtered by impact factor; each paper shows its journal's IF-equivalent (OpenAlex two-year mean citedness) and the date of that value. Each entry shows the full original abstract.
Read by cancer type: Cancer biology（1） · Breast（0） · Lung（1） · Pancreatic（0） · Prostate（0） · ACC（0） · Brain（1） · Ovarian（1） · Endometrial（0） · Gastric（0） · Liver（0） · Kidney（0） · Bladder（1）
Cancer molecular biology and AI
- Abstract
Metabolic reprogramming is a hallmark of many diseases, particularly cancer, yet current methods for analyzing metabolic states from gene expression data often lack the resolution and interpretability needed for clinical applications. Gene-based signature approaches reduce the structure of metabolic networks, are susceptible to statistical biases, and eventually show poor reproducibility across studies. A network-structured, mechanistically grounded alternative is therefore needed. We developed CircumSpectrum, a computational framework that converts gene expression profiles into flux-based metabolic fingerprints by integrating genome-scale metabolic models with machine learning classifiers trained across 142 functional metabolic subsystems. The framework is validated using multiple independent transcriptomic cohorts of patients treated with metformin — a widely used anti-diabetes drug with growing relevance in oncology — spanning healthy individuals, type II diabetic patients, and cancer cell lines. Performance is benchmarked against conventional gene signature and pathway enrichment methods using bootstrapped classification accuracy. Here we show that CircumSpectrum surpasses conventional gene signature methods in distinguishing treatment effects and maintains reproducibility across datasets, platforms, and species. Applied to metformin-treated cohorts, the framework uncovers drug-induced metabolic adaptations at the phenotypic, subsystem, and individual reaction levels that are not detected by standard transcriptomic analyses, including disease-specific divergences between healthy and diabetic metabolic responses and conserved pathway rewiring across cancer types. CircumSpectrum enables multi-scale metabolic profiling from patient cohorts, experimental models, and single-cell data, providing structured and interpretable insights into complex phenotypes from gene expression data. This approach has broad potential for identifying metabolic vulnerabilities, characterizing drug mechanisms of action, and supporting precision therapeutic strategies across disease contexts. Cells constantly adapt their metabolism in response to disease or treatment, but current methods for studying these changes from gene activity data are often imprecise and hard to reproduce. We developed CircumSpectrum, a computational tool that converts gene expression measurements into detailed metabolic “fingerprints” by combining large-scale biological network models with machine learning. We tested it using metformin, a common anti-diabetes drug with emerging uses in cancer, across multiple patient studies and cancer types. CircumSpectrum identified metabolic changes that standard methods missed, and did so consistently across different datasets and even different species. These findings suggest that CircumSpectrum could help researchers better understand how drugs work at the metabolic level, potentially guiding more precise treatment strategies for diabetes, cancer, and other metabolic diseases. Polley et al. present CircumSpectrum, a framework transforming RNA-seq into flux-based metabolic fingerprints via genome-scale models and machine learning. They benchmark the framework on metformin across clinical and cancer cohorts, and find that it performs favorably against conventional gene-signature methods and suggests drug mechanisms at the phenotypic, subsystem, and reaction levels.
Journal IF-equivalent: 6.0 (OpenAlex 2-year mean citedness, value as of 2026-10-08, retrieved 2026-10-09; not the official Clarivate IF)Reference: Polley N, Kaoma T, Nazarov PV, Sarry JE. Inferring flux-based metabolic fingerprints from gene expression to predict drug response and facilitate patient stratification. Commun Med. 2026 Oct 21 [Epub ahead of print]. doi:10.1038/s43856-026-01970-3.Checked: Abstract only
See all 1 Cancer biology papers →
Breast cancer: treatment, trials and AI
No new qualifying paper for this issue.
Lung cancer: treatment, trials and AI
- Abstract
Background: This study aimed to investigate the associations of senescence-associated secretory phenotype (SASP)-related genes with the tumor immune microenvironment (TIME) and clinical outcomes in lung adenocarcinoma (LUAD) and to develop and evaluate a prognostic signature for risk stratification and exploratory treatment-response prediction.
Methods: Non-negative matrix factorization (NMF) was conducted to identify distinct molecular subtypes. Least absolute shrinkage and selection operator (LASSO), Random Forest, and extreme gradient boosting (XGBoost) were utilized to identify core prognostic SASP genes. The expression patterns of selected SASP-related genes were explored using multi-omics datasets, LUAD cell lines, and clinical specimens.
Results: Cluster 1 was associated with significantly shorter overall survival (OS), enhanced senescence pathway activity, and computationally inferred immunosuppressive and immune-exclusion features. Five intracellular SASP regulators— ANGPTL4 , CDKN2A , H2AX , MMP1 , and CCL20 —were selected through machine learning to construct the SASP Protein Regulator Score (SASP-PRS). A high SASP-PRS was associated with the activation of senescence/SASP and cell-cycle pathways, suppression of antitumor immune signatures, and distinct immune-checkpoint and chemokine expression profiles. A higher SASP-PRS was also associated with advanced disease stage, poorer OS, predicted lower responsiveness to immunotherapy, lower predicted sensitivity to erlotinib, and higher predicted sensitivity to selected chemotherapeutic agents. The nomogram incorporating the SASP-PRS demonstrated discriminatory ability and calibration for predicting 1-, 3-, and 5-year OS in the analyzed LUAD cohorts.
Conclusion: These findings suggest that SASP-related gene-expression patterns are associated with LUAD molecular subtypes, senescence signaling, computationally inferred immune features, and treatment-related outcomes. The SASP-PRS may serve as a potential biomarker for prognostic stratification in LUAD. However, its utility for treatment-response assessment remains exploratory. Prospective external validation and functional studies are required to establish its clinical applicability and clarify the mechanisms linking SASP-related networks to the LUAD immune microenvironment.
Journal IF-equivalent: 3.7 (OpenAlex 2-year mean citedness, value as of 2026-10-22, retrieved 2026-10-23; not the official Clarivate IF)Reference: Chang X, Jiang L, Zhang L, Ren T, Fan C, Li P. Machine learning–derived intracellular protein–based prognostic model reflecting a senescence-associated secretory phenotype-driven immunosuppressive microenvironment in lung adenocarcinoma. PeerJ. 2026 Oct 21;14:e21835. doi:10.7717/peerj.21835.Checked: Abstract only
Pancreatic cancer: treatment, trials and AI
No new qualifying paper for this issue.
Prostate cancer: treatment, trials and AI
No new qualifying paper for this issue.
Adenoid cystic carcinoma: treatment, trials and AI
No new qualifying paper for this issue.
Brain tumours: treatment, trials and AI
- Abstract
The Brain Tumor Reporting and Data System (BT-RADS) standardizes posttreatment MRI response assessment in patients with diffuse gliomas but requires complex integration of imaging trends, medication effects, and radiation timing. An end-to-end multiagent large language model and convolutional neural network system for automated BTRADS classification was developed and evaluated. Five hundred nine consecutive posttreatment glioma MRI examinations (March 2017-July 2021) from a single center were retrospectively evaluated. An extractor agent identified clinical variables from notes using gpt-oss (OpenAI, 20-billion parameters). A scorer agent applied BTRADS decision logic integrating these with quantitative volumetrics. System accuracy was compared against initial assessments, using expert annotation as the reference standard. Four hundred ninety-two examinations were evaluable. Overall system accuracy was 76.0% (374/492) versus 57.5% (283/492) for initial assessments, an absolute difference of +18.5 percentage points ( P < .001). Cohen kappa between the system and the reference standard was 0.71. The system achieved high sensitivity in context-dependent categories (BT-1b; 100%; BT-1a, 92.7%; BT-3a, 87.5%) and moderate sensitivity for threshold-dependent categories (BT-3c, 74.8%; BT-2, 69.2%; BT-4, 69.3%; BT-3b, 57.1%). The BT-4 positive predictive value was 92.9%. These results support further evaluation of this system as a clinical decision-support tool, with local open-weight inference preserving data privacy. ©RSNA, 2026
Journal IF-equivalent: 8.3 (OpenAlex 2-year mean citedness, value as of 2026-10-22, retrieved 2026-10-23; not the official Clarivate IF)Reference: Jabal MS, Zhang J, LaBella D, Houk JL, Zhang D, Rudie JD, et al. Agentic Automation of BT-RADS Scoring Using an End-to-End System for Standardized Glioma Follow-up Assessment. Radiology: Artificial Intelligence. 2026 Oct 21 [Epub ahead of print]. doi:10.1148/ryai.260234.Checked: Abstract only
Ovarian cancer: treatment, trials and AI
- Abstract
Homologous recombination deficiency (HRD) is a critical biomarker for guiding treatment decisions in high-grade serous tubo-ovarian carcinoma (HGSOC), a cancer with few reliable biomarkers. However, existing genomics-based tests for HRD are variable, expensive, and time-consuming. We develop HRDPath, a patient-level deep learning architecture that predicts HRD status from whole slide images, requiring no additional molecular testing. HRDPath is validated across three cohorts of female patients and achieves an AUC of 0.846, significantly outperforming previously reported results in HGSOC. By incorporating multiple slides per patient and targeted optimization strategies, we strengthen the model clinical utility, with both specificity and sensitivity reaching 0.90. Our interpretability pipeline enhances model trustworthiness and reveals important clinical and biological insights into HRD-positive cancers, highlighting the associated morphological and pathological changes at the cellular and tissue levels. HRDPath provides a foundation for an accessible and scalable digital biomarker to improve treatment stratification in ovarian cancer globally.
Journal IF-equivalent: 18.1 (OpenAlex 2-year mean citedness, value as of 2026-10-06, retrieved 2026-10-07; not the official Clarivate IF)Reference: Wu CI, Banda K, Arora R, Soong TR, Radke M, Dillon M, et al. HRDPath: an explainable multi-model deep learning architecture for predicting homologous recombination deficiency from histopathology images. Nat Commun. 2026 Oct 21;17(1):10870. doi:10.1038/s41467-026-78430-y.Checked: Abstract only
Endometrial cancer: treatment, trials and AI
No new qualifying paper for this issue.
Gastric cancer: treatment, trials and AI
No new qualifying paper for this issue.
Liver cancer: treatment, trials and AI
No new qualifying paper for this issue.
Kidney cancer: treatment, trials and AI
No new qualifying paper for this issue.
Bladder cancer: treatment, trials and AI
- Abstract
ABSTRACT Accurate prediction of lymph node metastasis (LNM) and survival outcomes is crucial for personalized treatment in bladder cancer, yet remains clinically challenging. This study aimed to develop a Cytology‐based multitask predictive Network (CytoNet) to simultaneously predict LNM and disease‐free survival (DFS) in bladder cancer, and to evaluate its generalizability in cervical and thyroid cancers. This study included a multicenter cohort of 5273 patients, with bladder cancer (n = 1134), cervical cancer (n = 598), and thyroid cancer (n = 3541). CytoNet was constructed through a three‐stage pipeline comprising self‐supervised pretraining, supervised fine‐tuning, and multitask prediction. In bladder cancer, CytoNet achieved an area under the curve (AUC) of 0.831‐0.881 for LNM prediction and a C index of 0.680‐0.710 for DFS prediction across all cohorts, outperforming single‐task models and standard radiological assessments. In cervical cancer, CytoNet demonstrated robust performance for both LNM prediction (AUC 0.847‐0.862) and DFS prediction (C index 0.690‐0.709), while in thyroid cancer, it achieved an AUC of 0.819‐0.855 for LNM prediction. In conclusion, CytoNet represents a reliable tool for predicting LNM and prognosis from cytopathology images across multiple cancer types, showing potential as a valuable adjunct to current clinical decision‐making. Further prospective validation is warranted to assess its clinical utility.
Journal IF-equivalent: 9.5 (OpenAlex 2-year mean citedness, value as of 2026-10-18, retrieved 2026-10-19; not the official Clarivate IF)Reference: Hong G, Jiang F, Xie Y, Huang X, Shen R, Wang Y, et al. Multitask Deep Learning Model for Predicting Lymph Node Metastasis and Disease‐Free Survival From Cytology Images in Bladder Cancer: A Multicenter Study With Pan‐Cytology Validation. Advanced Science. 2026 Oct 21 [Epub ahead of print]. doi:10.1002/advs.76577.Checked: Abstract only