AI and Cancer Research ── Lung ── 2026-11-09
Lung cancer: treatment, trials and AI
- Abstract
Background: Intratumor heterogeneity in lung adenocarcinoma (LUAD) profoundly influences tumor progression and immune evasion. However, current prognostic models largely fail to capture the complex cellular dynamics of the tumor microenvironment (TME) at single-cell resolution.
Methods: We integrated single-cell RNA sequencing with bulk transcriptomics to derive cell-type-specific signatures. A systematic evaluation of 101 machine learning algorithm combinations was performed across multiple independent cohorts to construct a robust prognostic model. Immune landscape, mutational features, intercellular communication, and immunotherapy responsiveness were comprehensively evaluated. The biological functions of the identified hub gene, transmembrane 4 L six family member 1 (TM4SF1), were rigorously validated in vitro .
Results: The optimal StepCox[forward]-Ridge model showed strong prognostic stratification in five cohorts. Low-risk tumors exhibited an immune-inflamed phenotype, suggesting greater predicted good responses to immune checkpoint blockade. In contrast, high-risk tumors had an immunosuppressive TME with overactive cell communication driven by fibroblasts and myeloid cells. TM4SF1 was a key driver in this high-risk signature. TM4SF1 silencing significantly inhibited LUAD cell proliferation, migration, and invasion, and strongly induced apoptosis. TM4SF1 depletion reversed tumor immune evasion by reducing PD-L1, increasing MHC-I, and lowering the secretion of CCL5.
Conclusions: This single-cell-guided machine-learning framework provides strong prognostic stratification in LUAD and captures immune phenotypes associated with potential immunotherapy responsiveness. TM4SF1 emerged as a candidate mediator associated with malignant progression and immune-related phenotypes in LUAD, warranting further mechanistic and in vivo investigation. This suggests it could be a useful therapeutic target.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Li J, Liu T, Li L, Lu W, Huang J, Li X, et al. A single-cell transcriptomics-based prognostic model identifies TM4SF1 as a key promoter of malignant progression in lung adenocarcinoma. Front Oncol. 2026 Nov 6;16:1948068. doi:10.3389/fonc.2026.1948068.Checked: Abstract only - Abstract
Real-time detection of rare cancer cells holds potential for guiding more informed treatment decisions at the bedside. However, conventional enrichment methods are limited by low sensitivity, a high risk of cell damage, and prolonged processing times. To address these challenges, this study developed an intelligent micro-total analysis platform toward H1299 cells (a metastatic non-small cell lung cancer cell line). The system integrates a cascaded dual-microfluidic chip architecture with a high-resolution microscope driven by a deep learning-based detection framework. Finite element simulations were performed to optimize droplet generation conditions for the cross-junction microfluidic chip. A dataset of 749 high-resolution droplet images (1536 × 1024 pixels) was constructed, comprising 2619 live H1299 lung cancer cells distributed across 5127 droplets. Among the algorithms evaluated, the YOLOv7 model was selected and optimized, achieving a detection accuracy above 98% and a real-time inference speed exceeding 55 frames per second. As a proof of concept, the platform was applied to quantitatively analyze single-cell level responses to anticancer drugs in undiluted whole blood and processed biopsy samples within 30-40 min, yielding results consistent with those of the standard Cell Counting Kit-8 (CCK-8) assay. The minimal sample requirement, label-free operation, and stable performance of the system underscore its potential for treatment monitoring. Moreover, its capacity for continuous image acquisition and autonomous analysis may facilitate dynamic monitoring in future cancer research.
Journal IF-equivalent: 11.9 (OpenAlex 2-year mean citedness, value as of 2026-11-07, retrieved 2026-11-08; not the official Clarivate IF)Reference: Zhang S, Bo B, Gu H, Chen M, Wu D, Li Y, et al. A deep monitoring microfluidic platform for label-free identification and nanodrug evaluation in lung cancer. Biosensors and Bioelectronics. 2027 Feb;318:119395. doi:10.1016/j.bios.2026.119395.Checked: Abstract only - Abstract
BACKGROUND: Neoadjuvant chemoimmunotherapy improves outcomes for resectable non-small cell lung cancer (NSCLC), but can cause severe treatment-related adverse events (TRAEs). Pretreatment models for predicting grade ≥3 TRAEs during neoadjuvant chemoimmunotherapy remain limited. This study aimed to develop and externally validate an explainable machine learning model for predicting grade ≥3 TRAEs in patients with resectable NSCLC receiving neoadjuvant chemoimmunotherapy. METHODS: This multicenter retrospective study included a discovery cohort (n=953) and an independent external validation cohort (n=143). The discovery cohort was stratified into training (n=668) and internal test (n=285) sets. Pretreatment candidates included routine clinical and laboratory variables and first-cycle treatment characteristics. Stability-oriented Boruta selection was used for variable selection, followed by variance inflation factor (VIF) assessment. Eight machine learning (ML) algorithms were tuned using 10-fold cross-validation in the training set. Discrimination, calibration, decision curve analysis (DCA), and explanations based on SHapley Additive exPlanations (SHAP) were evaluated. Secondary analyses assessed hematologic, nonhematologic, and immune-related adverse event (irAE) endpoints. RESULTS: Grade ≥3 TRAEs occurred in 496 of 1,096 patients (45.3%); 464 patients (42.3%) experienced grade ≥3 hematologic toxicity, and 446 (89.9% of patients with grade ≥3 TRAEs) reached the composite endpoint exclusively because of hematologic events. The final random forest model incorporating 12 pretreatment predictors achieved areas under the receiver operating characteristic curve (AUCs) of 0.782 and 0.742 in the internal test and external validation cohorts, respectively. When the locked overall-TRAE model was applied to toxicity subtypes, discrimination was maintained for grade ≥3 hematologic toxicity (AUCs, 0.764 and 0.732) but was substantially weaker for grade ≥3 nonhematologic toxicity (0.638 and 0.631) and any-grade irAEs (0.435 and 0.535). The hematologic toxicity-specific model achieved AUCs of 0.777 and 0.711 in the internal test and external validation cohorts, respectively. CONCLUSIONS: The pretreatment random forest model showed moderate external discrimination for overall grade ≥3 TRAEs, primarily reflecting hematologic toxicity.
Journal IF-equivalent: 2.5 (OpenAlex 2-year mean citedness, value as of 2026-10-13, retrieved 2026-10-15; not the official Clarivate IF)Reference: Zhi FH, Zhang Y, Ge T, Yang L, Dong Y, Wen J, et al. Explainable machine learning for predicting grade ≥3 treatment-related adverse events in resectable non-small cell lung cancer receiving neoadjuvant chemoimmunotherapy: predominant contribution of hematologic toxicity. Transl Lung Cancer Res. 2026 Oct;15(10):311-1. doi:10.21037/tlcr-2026-0984.Checked: Abstract only - Abstract
BACKGROUND: Despite the efficacy of immune checkpoint inhibitors (ICIs) in advanced non-small cell lung cancer (NSCLC), reliable non-invasive predictors of response remain lacking, partly due to intratumoral heterogeneity.This multicenter study aimed to develop and validate a habitat-based radiomics model to predict ICI response in patients with advanced NSCLC by integrating clinical, whole-tumor radiomics, and habitat features. METHODS: This multicenter retrospective study included 330 patients with stage IIIB-IV NSCLC who received ICIs between January 2019 and November 2024 across three Chinese hospitals. Patients from the First Affiliated Hospital of Bengbu Medical University were allocated to the training cohort (n=191); those from Shandong Cancer Hospital and the First Affiliated Hospital of Anhui Medical University formed the external validation cohorts (test1, n=70; test2, n=69). The primary endpoint was 3-month progression status per RECIST 1.1. Following feature extraction and a rigorous selection process, seven machine learning algorithms were employed to construct six model types: clinical, radiomics, habitat-based, and three fusion models [clinical-habitat (FCH), radiomics-habitat (FRH), and clinical-radiomics-habitat (FCRH)]. Model performance was assessed by receiver operating characteristic (ROC) analysis, calibration metrics, and decision curve analysis (DCA); DeLong's test, integrated discrimination improvement, and net reclassification improvement were used for model comparisons. Feature interpretability was assessed using SHapley Additive exPlanations (SHAP). RESULTS: The training cohort had 22 progressive disease events (11.52%), test1 had 14 (20.00%), and test2 had 16 (23.19%). The final FCRH model, using extreme gradient boosting (XGBoost), achieved areas under the curve of 0.937 [95% confidence interval (CI): 0.903-0.971] in training, 0.862 (95% CI: 0.773-0.951) in test1, and 0.823 (95% CI: 0.710-0.936) in test2. Sensitivities/specificities were 87.6%/88.9%, 76.8%/78.6%, and 83.0%/72.7%, respectively. Calibration was good (Hosmer-Lemeshow P>0.05 in all cohorts). SHAP analysis identified that habitat-derived features constituted 70.83% of the top 24 predictors. CONCLUSIONS: Habitat-based radiomics substantially improves prediction of ICI response in advanced NSCLC, with promising performance in external validation. Prospective studies are warranted before clinical translation.
Journal IF-equivalent: 2.3 (OpenAlex 2-year mean citedness, value as of 2026-11-08, retrieved 2026-11-09; not the official Clarivate IF)Reference: Li S, Jin L, Zhu Q, Zhu D, Meng Y, Lu X, et al. Development and validation of a habitat-based radiomics fusion model for predicting immunotherapy response in advanced NSCLC: a multicenter retrospective study. J Thorac Dis. 2026 Oct;18(10):1123-3. doi:10.21037/jtd-2026-1443.Checked: Abstract only