Anqi Lin

Background: Tumour‐node‐metastasis staging does not fully explain prognostic heterogeneity in non‐small cell lung cancer. We evaluated whether haematoxylin‐and‐eosin whole‐slide images could estimate histological subtype, pathological stage probabilities, survival risk and spatially grounded biological associations. Methods: SparseAGE‐MTL, a weakly supervised multi‐task multiple‐instance learning model with a shared projection‐topology encoder and endpoint‐specific heads, was trained and benchmarked in 954 The Cancer Genome Atlas cases using seven pathology feature spaces and 18 comparator models. External evaluation used 948 tissue‐microarray and 324 whole‐slide cases. Attention maps were co‐registered with 10x Visium spatial transcriptomics and integrated with bulk transcriptomics, immune‐infiltration estimates and ESTIMATE scores. Analyses included paired model comparisons, false‐discovery‐rate correction, Cox models, calibration assessment and decision curve analysis. Results: In the CONCH feature space, SparseAGE‐MTL achieved 93.73% accuracy, 98.19% area under the receiver‐operating‐characteristic curve and 93.08% F1‐score for adenocarcinoma/squamous cell carcinoma classification in internal benchmarking; external area‐under‐the‐curve values were approximately .91 and .82. Stage estimation had lower discrimination, with external overall area under the curve approximately .70 and cohort‐dependent calibration. Risk‐score‐defined groups differed in overall survival in both histological subtypes and showed similar external trends. High‐attention regions were enriched at tumour–stroma or tumour–immune interfaces and were associated with B‐cell, fibroblast, C1QC, COL1A1, epithelial–mesenchymal transition, metastasis and hypoxia signals. Higher risk cases showed malignant pathway activation, lower immune/stromal scores, higher tumour purity and subtype‐specific immune/stromal differences. Adding the risk score to the clinical model increased external pooled concordance index from approximately .620 to .672. Conclusions: In retrospective cohorts, SparseAGE‐MTL generated subtype‐classification, stage‐probability and survival‐risk outputs from routine pathology images. Subtype classification had higher numerical performance than stage estimation. Survival‐risk and attention outputs were associated with outcome and spatial/transcriptomic features, but prospective, treatment‐annotated validation is required before clinical use. Key points: SparseAGE‐MTL is a weakly supervised multi‐task MIL framework that jointly performs NSCLC subtype classification, stage prediction, and survival risk estimation from routine H&E slides using only slide-level labels. The model achieves stable competitive performance across seven feature spaces and 18 comparators, with external validation demonstrating robust generalization to independent WSI and TMA cohorts. Attention hotspots co‐localize with tumor‐stroma/immune interfaces and spatial transcriptomic signatures of EMT, hypoxia, and C1QC/COL1A1 enrichment, providing biologically grounded interpretability. High‐risk groups exhibit activated malignant pathways, lower immune/stromal scores, higher tumor purity, and incremental prognostic value beyond clinical variables.

Protein design has numerous applications in synthetic biology, drug discovery, and bioengineering. Recently, there has been a revolution in this field due to the emergence of artificial intelligence. At the forefront are deep learning models (DLMs). The impact of these models on the design pipeline is so transformative that it leads to radical advances in prediction accuracy, functional enrichment, and de novo synthesis of proteins. We performed bibliometric analysis on a Web of Science data-set (2006–2025) obtained through a targeted keyword search. Our analysis method orchestrated the use of standard tools such as CiteSpace and VOSviewer for co-citation, keyword co-occurrence, and burst detection, while we used custom python scripts to generate more nuanced plots for collaboration networks and intellectual overlap. Based on the bibliometric analysis, we observe a sudden jump in publication counts after the year 2018. This period also coincides with the publication of many methodological breakthroughs such as AlphaFold2 and RoseTTAFold. The analysis revealed three major intellectual clusters focused around protein structure prediction, directed evolution, and de novo design of proteins. DLMs play a major role in the publication impact of generating functional protein sequences, with frameworks such as ProteinMPNN having a notable citation footprint. Although China and the United States dominated in raw publication volume, citation impact told a different story—Switzerland ranked second in per-publication influence, suggesting that research programs combining computational design with systematic experimental validation tend to generate disproportionate scholarly impact. International collaborative efforts were notable after 2020. Citation and co-citation analysis highlighted works that have formed the bedrock of this field. Notable mentions include seminal works on AlphaFold2, ProteinMPNN, and RFdiffusion. Protein design stands at the center of an AI revolution that has drawn database mining, computational generation, and experimental validation into an increasingly rapid and interconnected engineering loop. Yet sustaining this momentum requires confronting limitations that the field’s most cited literature has largely left unspoken: models trained on stable, crystallizable structures struggle to generalize beyond their training distributions, static predictions remain blind to the conformational dynamics governing allostery and catalysis, and in silico confidence scores have repeatedly proven poor predictors of wet-lab outcomes. Closing this gap will demand not incremental refinement but the sustained, bidirectional coupling of high-throughput experimental feedback with iterative design-test-learn cycles, supported by next-generation models that natively represent PTMs, cellular context, and conformational ensembles—a convergence that holds genuine promise for translating the field’s computational ambitions into reliable therapeutic and synthetic biology outcomes.

Table S1 Characteristics of Specific Phenotype of (Transitional) Senescent T cells in cancers

Table S2 Current Clinical Trials and Approaches Targeting T-cell Senescence

DNA methylation levels are intimately associated with tumor development, progression, and therapeutic outcomes. Accurate analysis of the relationship between DNA methylation levels and tumor prognosis facilitates comprehensive investigation of tumor development mechanisms, enabling optimization of clinical decision‐making and subsequent enhancement of cancer patient survival rates. However, current web‐based tools for analyzing tumor methylation levels and survival prognosis exhibit significant limitations. We have developed a web‐based tool called Pan‐cancer DNA Methylation Survival Analysis (PDMSA) implemented in Shiny, which integrates DNA methylation data and clinical information from large public databases (TCGA and GEO). PDMSA currently encompasses tumor DNA methylation data from 30 TCGA datasets and 15 GEO datasets, consisting of 16 205 211 records that span 39 cancer types, 45 datasets, 19 909 genes, and 8369 samples. The tool executes prognostic Kaplan–Meier survival analysis and Cox regression analysis utilizing two distinct cutoff value grouping methods, offering customizable visualization options for the results. As a user‐friendly analytical platform, PDMSA serves as a comprehensive tool for biomedical researchers to investigate the relationship between methylation levels at specific gene loci and tumor survival outcomes, thereby facilitating the advancement of precision medicine in oncology. Access PDMSA at robinl‐lab.com/PDMSA.

Renal cell carcinoma (RCC) is a prevalent malignancy of the urinary system. Despite significant advances achieved through targeted therapies and immunotherapies, therapeutic resistance remains a major obstacle to sustained clinical efficacy. This review comprehensively examines the molecular mechanisms driving resistance to both targeted therapy and immunotherapy in RCC from a multi-omics perspective. By integrating findings across diverse omics layers, we underscore the pivotal role of multi-level regulatory networks in mediating drug resistance and immune evasion. Our objective is to provide an in-depth understanding of these resistance mechanisms and to establish a theoretical framework for developing innovative therapeutic strategies aimed at overcoming resistance, thereby facilitating the advancement of precision oncology in RCC.

Background: Non–small cell lung cancer (NSCLC), accounting for 80% of lung cancer cases, remains a leading cause of cancer‐related mortality globally. While immune checkpoint inhibitors (ICIs) have improved outcomes, their efficacy is limited to a subset of patients, necessitating robust biomarkers for personalized immunotherapy response prediction. Methods: We integrated transcriptomic data from 584 NSCLC patients across four cohorts treated with ICIs. Using 12,025 pathways from MSigDB, we applied 101 machine learning algorithm combinations (e.g., random survival forest [RSF], least absolute shrinkage and selection operator [Lasso], and Cox proportional hazards model with component‐wise likelihood‐based boosting [CoxBoost]) to identify prognostic signatures. OAK was used as the training set and Ravi, Jung, and Poplar as the validation set. The optimal pathway and algorithm combination was determined based on the average concordance index ( C ‐index) ranking in the validation sets, and a predictive model was generated. Performance was assessed by C ‐index, receiver operating characteristic (ROC) analysis, and survival analysis. Biological relevance was evaluated through gene set enrichment analysis (GSEA), immune infiltration profiling, and immunohistochemistry (IHC). Results: The FOXO‐mediated transcription pathway combined with Lasso‐RSF algorithms emerged as the top predictor. The derived FOXO‐related signature (FRS) stratified patients into high‐risk and low‐risk groups, with high‐risk patients showing significantly worse progression‐free survival (PFS) and overall survival (OS) across all cohorts ( p < 0.05). FRS outperformed clinical variables and 43 published models in predictive accuracy. IHC confirmed elevated expression of FRS‐associated genes (PCK1, IGFBP1) in nonresponders. Immune profiling revealed enriched antitumor immunity in low‐FRS patients. Conclusion: FRS, a machine learning–derived pathway signature, robustly predicts immunotherapy response and survival in NSCLC. Its integration of FOXO‐mediated immune regulation offers a clinically translatable tool for precision oncology.

Background: Using predicted drug sensitivity scores as survival biomarkers may improve precision medicine and overcome the limitations of genomically guided approaches in clinical trials. Methods: Pan-Cancer Drug Sensitivity Score Survival Analysis (DrugSurvPlot) is an interactive, login-free web analyzer built with R (v4.3.1), leveraging the Shiny package for interface/server logic, the DT package for data table queries/downloads, and the survival package for survival analysis. Data preprocessing was performed using OncoPredict, enabling users to export processed tables and results. Results: DrugSurvPlot integrates 189 GEO datasets (including 10 immune checkpoint inhibitor treatment datasets) and 33 TCGA datasets, totaling 85,531 records across 52 cancer types and 13 survival status data types, while incorporating 198 anticancer drugs from GDSC2. This tool supports two cutoff strategies for drug sensitivity scores, offers advanced survival analysis methods, and enables customizable high-definition visualization of results. Discussion: DrugSurvPlot represents a significant advancement in computational oncology by establishing predicted drug sensitivity scores as novel prognostic biomarkers for tumor survival analysis. This interactive platform integrates comprehensive datasets spanning 198 anticancer drugs and 52 cancer types, while providing researchers with intuitive tools for generating publication-ready Kaplan-Meier analyses. Current limitations in drug repertoire coverage and dataset diversity will be addressed through ongoing expansion of pharmacological databases and incorporation of emerging data modalities, including single-cell transcriptomics. Conclusions: In summary, DrugSurvPlot offers a no-code platform with comprehensive datasets, diverse cancer coverage, and customizable survival analysis, addressing critical research gaps. Continuous enhancements will improve predictive accuracy and clinical utility, establishing it as an evolving powerhouse in drug-survival investigations.

Therapeutic resistance remains a defining challenge in oncology, limiting the durability of current therapies and contributing to disease relapse and poor patient outcomes. This review systematically integrates recent progress in understanding the molecular, cellular, and ecological foundations of drug resistance across chemotherapy, targeted therapy, and immunotherapy. We delineate how genetic alterations, epigenetic reprogramming, post-translational modifications, and non-coding RNA networks cooperate with metabolic reprogramming and tumor microenvironment remodeling to sustain resistant phenotypes. The influence of the microbiome is highlighted as an emerging determinant of therapeutic response through immune modulation and metabolic cross-talk. By summarizing key regulatory circuits, We establishe a unified framework linking clonal evolution, metabolic adaptability, and tumor ecological dynamics. We further synthesizes novel therapeutic strategies that convert resistance mechanisms into therapeutic vulnerabilities, including synthetic lethality approaches, metabolic targeting, and disruption of stem cell and stromal niches. Advances in single-cell and spatial omics, liquid biopsy, and artificial intelligence are emphasized as transformative tools for early detection and real-time prediction of resistance evolution. This review also identifies major translational gaps in preclinical modeling and proposes precision oncology frameworks guided by evolutionary principles. By bridging mechanistic understanding with adaptive clinical design, this work provides an integrated roadmap for overcoming therapeutic resistance and achieving sustained, long-term cancer control.