Biodata Mining最新文献

{"title":"Disease- and gene-specific deep learning for pathogenicity prediction of rare missense variants in cancer predisposition genes.","authors":"Da-Bin Lee, Hyun-Uk Kang, Kyu-Baek Hwang","doi":"10.1186/s13040-026-00533-5","DOIUrl":"10.1186/s13040-026-00533-5","url":null,"abstract":"<p><p>BACKGROUND: Hereditary cancers frequently arise from germline pathogenic variants, yet only a small proportion of reported variants have been clinically classified, leaving most missense variants unresolved as variants of uncertain significance (VUS). Although recent machine-learning approaches have explored disease-specific or gene-specific contexts to improve pathogenicity prediction, these models remain fundamentally limited by the scarcity of labeled data and the underutilization of abundant VUS. RESULTS: We propose a deep-learning framework that integrates autoencoder pretraining with a deep ensemble strategy to improve variant pathogenicity prediction, effectively leverage unlabeled VUS during pretraining, and reduce uncertainty arising from limited training samples. To validate each component of our framework, we evaluated its performance under both disease-specific and gene-specific training setups. Experiments on ClinVar variants from BRCA1, BRCA2, MLH1, and MSH2 showed that our framework achieved the best performance in the gene-specific setup for BRCA1—likely because BRCA1 contains substantially more gene-specific training data than the other genes—whereas the disease-specific setup yielded superior results for the remaining genes, which had comparatively limited gene-specific samples. Overall, our method significantly outperformed existing approaches. We also introduce an interpretability approach that provides variant-level importance profiles across pathogenicity classes, thereby enhancing transparency and clinical applicability. Moreover, by projecting feature-level importance scores into a two-dimensional space, we demonstrate that pretraining enables the model to learn distinctly different feature representations, illustrating how pretraining and ensemble learning synergistically contribute to improved predictive performance. CONCLUSIONS: Our framework preserves the specificity of disease- and gene-specific approaches, overcomes data scarcity through VUS-guided pretraining and ensembling, and offers interpretable outcomes that may be helpful for clinical decision support. Moreover, our results suggest a promising direction for pathogenicity prediction of rare missense variants and indicate that the proposed framework may be extendable to additional genes under appropriate data and modeling conditions. </p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12964711/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147357108","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep learning-enabled ECG system for detecting left ventricular hypertrophy and predicting cardiovascular prognoses.","authors":"Zhe-Yu Yang, Shi-Chue Hsing, Dung-Jang Tsai, Chin Lin, Chin-Sheng Lin, Chih-Hung Wang, Wen-Hui Fang","doi":"10.1186/s13040-026-00536-2","DOIUrl":"10.1186/s13040-026-00536-2","url":null,"abstract":"<p><p>Left ventricular hypertrophy (LVH) is a common condition with a prevalence of 15%-20% in general population. Prior studies have suggested that deep learning model (DLM)-enabled electrocardiogram (ECG) systems can aid LVH detection and cardiovascular risk assessment; however, conventional manual ECG criteria have limited sensitivity and their prognostic utility remains suboptimal. Therefore, this study aimed to develop a DLM-enabled ECG system to detect LVH and evaluate its prognostic associations with incident cardiovascular outcomes. A total of 40,736 patients from hospital A were used for model development (training and tuning) and internal validation (29,595/5,935/5,206 patients, respectively), and 6,271 patients from hospital B were used for external validation. LVH was defined by left ventricular mass index (LVMI) derived from echocardiography. Prognostic outcomes included new-onset acute myocardial infarction (AMI), heart failure (HF), and atrial fibrillation (AFib). In the external validation set, our AI-ECG-LVH model achieved area under the receiver operating characteristic curve (AUC) values of 0.82 in males and 0.77 in females. Furthermore, the hazard ratios for incident AMI, HF, and AFib were 2.67, 3.15, and 2.23 for AI-ECG-LVH, compared with 2.76, 3.78, and 2.25 for echocardiography-defined LVH (ECHO-LVH). Our AI-ECG-LVH model may provide a straightforward, affordable, and noninvasive approach for LVH screening and first-contact risk stratification.</p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13069749/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147357169","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"From retina to heart: explainable machine learning using OCT and Clinical covariates for heart failure screening.","authors":"Sona M Al Younis, Samit Kumar Ghosh, Feryal A Alskafi, Siamak Yousefi, Namareq Widatalla, Ahsan H Khandoker","doi":"10.1186/s13040-026-00529-1","DOIUrl":"10.1186/s13040-026-00529-1","url":null,"abstract":"<p><p>Heart failure affects over 64 million individuals globally, contributing to elevated mortality rates and substantial healthcare costs. This study investigates the potential of retinal optical coherence tomography features combined with routine clinical variables as biomarkers for the detection of heart failure, exploring a potential avenue for improved risk assessment and screening support using explainable machine-learning tools. A comprehensive dataset of normal and heart failure patients' demographic and medical records including retinal measurements from both eyes was used. Among the machine learning models employed, the Extreme Gradient Boosting model demonstrated the best performance, achieving an accuracy of 73.31%, a precision of 71.81%, and an area under the receiver operating characteristic curve of 0.837. Explainability analyses further revealed that macular thickness metrics, particularly in the inner temporal subfield, inner nasal subfield, and outer superior subfields of the left eye, along with key clinical indicators such as age, body mass index, and glycated hemoglobin, were the most influential predictors of heart failure status. Local explanation methods also provided patient-level reasoning consistent with overall cohort patterns. To our knowledge, this is the first study to use an integrated, explainable approach incorporating bilateral retinal optical coherence tomography measurements with routine clinical indicators for heart failure detection, providing an interpretable and accessible alternative to black-box models while helping address the cost, invasiveness, and limited accessibility of existing heart failure diagnostic tools.</p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-02-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13097962/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147345564","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"KeySDL: sparse dictionary learning for keystone microbe identification from steady-state observations using a dynamical-systems model.","authors":"Max Gordon, Turgut Yigit Akyol, B Amos, Stig U Andersen, Cranos Williams","doi":"10.1186/s13040-026-00527-3","DOIUrl":"10.1186/s13040-026-00527-3","url":null,"abstract":"<p><p>BACKGROUND: Identification of microbes with large impacts on their microbial communities, known as keystone microbes, is a topic of long-standing interest in microbiome research. However, many approaches to identify keystone microbes are limited by the inherent nonlinearity and state-dependence of microbial dynamics. Machine learning approaches have been applied to address these shortcomings but often require more data than is available for a given microbial system. RESULTS: We propose a keystone identification approach called KeySDL which reduces the amount of data required by incorporating assumptions about the type of microbial dynamics present in the experimental system. The data are modeled as originating from a Generalized Lotka-Volterra (GLV) model, an architecture commonly used to simulate microbial systems. The parameters of this model are then estimated using Sparse Dictionary Learning (SDL).We also propose a self-consistency score to help evaluate whether the assumption of GLV dynamics is reasonable for a given dataset, either through the application of KeySDL or other analysis tools validated using GLV simulation. CONCLUSION: Compared to existing methods, this approach allows accurate prediction of keystone microbes from small numbers of samples and provides an output interpretable as reconstructed system dynamics. </p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":"19 1","pages":"18"},"PeriodicalIF":7.9,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12922210/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146229489","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Computational insights into the natural phage endolysin linker landscape.","authors":"Emma Cremelie, Alexandre Boulay, Roberto Vázquez, Yves Briers","doi":"10.1186/s13040-026-00526-4","DOIUrl":"10.1186/s13040-026-00526-4","url":null,"abstract":"<p><p>Phage endolysins are increasingly investigated as novel protein-based antibiotics, offering solutions to the antibiotic resistance crisis. Endolysins targeting Gram-positive bacteria come in a variety of modular architectures, combining domains that bind the bacterial cell wall or enzymatically degrade it. While much research has focused on either understanding this multidomain architecture or leveraging it to create custom engineered lysins, far less is known about the oligopeptide linkers connecting their domains. Nevertheless, several engineering studies have observed a remarkable influence of the linker on lysin activity. In this work, we computationally investigated a broad set of Gram-positive endolysin linkers to bridge this knowledge gap. Relying on AlphaFold2-generated protein structural models, we collected 1072 linker sequences by finely delineating the domain limits using the SPAED tool, and described these through sixteen physicochemical and structural properties. Initial data exploration showed that endolysin linkers are highly diverse and feature similar amino acid compositions as previously described, general protein linkers. Subsequently, data mining and interpretable machine learning approaches were adopted to uncover the relationships between linkers and their endolysin domain architectures, as well as the associated phage host genus. These analyses revealed that such relationships do exist and are multidimensional in nature. Therefore, our findings provide evidence that the evolutionary pressure put on phages to adapt their lysis system to ever-changing environments and host requirements is not limited to the endolysin domains, but extends to the linkers connecting them. For instance, certain domain architectures were consistently associated with longer linkers, while others were highly stable. In summary, this work presents the first in-depth exploration of phage endolysin linkers, shedding light on their phage host- or domain architecture-specific design rules, and offering new perspectives for engineering endolysins as novel antimicrobials.</p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-02-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13019800/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146229465","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Multimodal deep learning for survival prediction and biomarker discovery in non-small cell lung cancer.","authors":"Yiqing Wang, Pinghui Xia, Yang Xu, Jianxin Xu, Wenzhen Xu, Wang Lv, Junrong Yan, Qiuxiang Ou, Hua Bao, Luming Wang, Jian Hu","doi":"10.1186/s13040-026-00522-8","DOIUrl":"10.1186/s13040-026-00522-8","url":null,"abstract":"<p><p>BACKGROUND: Deep learning (DL) is revolutionizing high-dimensional analysis for clinical applications, with multimodal approaches showing promise in integrating diverse datasets and improving predictive accuracy. However, its use in prognostic assessment, particularly for non-small-cell lung cancer (NSCLC), remains limited. METHODS: Whole-slide imaging (WSI), next-generation sequencing (NGS), and clinical features from 2,366 NSCLC patients sourced from various databases were analyzed. A Sequential-Adapted Attention (SeAttn) model was developed to process WSI, while multiple DL models analyzed genetic and clinical data. Multimodal features were combined using a COX-based DL approach to predict long-term prognosis. Validation (N = 100) and test (N = 101) cohorts were used to confirm model performance. RESULTS: The SeAttn model achieved an area under the curve (AUC) of 0.98 for NSCLC histologic subtype prediction in the test cohort (P < 0.0001). Individually, WSI, clinical, and genetic features predicted prognosis with concordance indices (C-indices) of 0.55–0.67 using the test cohort. Combined multimodal features improved the test cohort C-index to 0.71 (P = 0.002). The model accurately predicted prognosis up to 5 years post-diagnosis (P < 0.05 for all time-dependent AUCs) and stratified patients by overall survival (P = 0.012). SeAttn demonstrated attention shifts when comparing subtyping and prognosis models and revealed immuno-hot and immuno-cold stroma enrichment underlying survival differences. Novel prognostic markers, including TPTE mutation, microRNA cluster amplification, and RIPK4-IL10RB co-deletion, were identified and validated. CONCLUSIONS: This multimodal DL framework demonstrates robust prognostic capabilities for NSCLC, integrating clinical, genetic, and imaging data to yield accurate survival estimates and identify biomarkers. These findings highlight the scalability of multimodal DL for broader cancer applications. </p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":""},"PeriodicalIF":7.9,"publicationDate":"2026-02-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13014861/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146214253","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Early differentiation between paroxysmal and persistent atrial fibrillation based on interpretable machine learning: a multicenter retrospective study.","authors":"Sijin Li, Yuqi Zhang, Weijie Wu, Guang Li, Tucheng Huang, Kuan Zeng, Chao Tong, Heng Li, Hui Huang","doi":"10.1186/s13040-026-00525-5","DOIUrl":"10.1186/s13040-026-00525-5","url":null,"abstract":"<p><p>AIMS: Atrial fibrillation (AF) is a common arrhythmia associated with increased risks of stroke and heart failure. Early differentiation between paroxysmal and persistent AF at first diagnosis is critical for guiding treatment decisions. This study aimed to develop an interpretable machine learning model based on structured electronic health records (EHR) to distinguish AF subtypes and identify key contributing factors. METHODS AND RESULTS: In this multicenter, retrospective cohort study, data were collected from three tertiary hospitals in China between January 2013 and January 2023. A total of 11,986 patients with suspected AF were screened, of whom 4155 patients with first-diagnosed AF were included (paroxysmal: 2565 [61.3%]; persistent: 1620 [38.7%]). Structured EHR variables, including clinical demographics, serological indicators, and echocardiographic parameters, were extracted for analysis. Variable selection was performed using Spearman correlation and least absolute shrinkage and selection operator regression. Three machine learning algorithms were trained and externally validated. The CatBoost model achieved the best performance, with an area under the receiver operating characteristic curve of 0.876 (95% CI: 0.871–0.880) and accuracy of 0.808 (95% CI: 0.803–0.816). Sensitivity and specificity ranged from 0.802 to 0.811. Shapley additive explanations (SHAP) were used to interpret model outputs and identify variables most associated with AF subtype classification. CONCLUSION: This multicenter study demonstrates that interpretable machine learning models based on structured EHR data can accurately distinguish paroxysmal from persistent AF at first diagnosis. The proposed model may facilitate early subtype-specific risk stratification and personalized treatment, potentially improve outcomes, and reduce disparities in AF care across different medical conditions. </p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":"17"},"PeriodicalIF":7.9,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12892771/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146133350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A crisis of overconfidence: Why confidence, not accuracy, is the real risk in clinical AI.","authors":"Jacob S Berkowitz, Jake R Patock, Asma Nawaz, Graciela Gonzalez-Hernandez, Nicholas P Tatonetti","doi":"10.1186/s13040-026-00518-4","DOIUrl":"10.1186/s13040-026-00518-4","url":null,"abstract":"<p><p>Language models today are trained to convey confidence in their outputs, regardless of whether those outputs are correct. The alignment methods we use to make them helpful also push them toward unwarranted certainty, rewarding decisive answers over appropriate hedging. As these foundation models enter high-stakes domains such as science and medicine, this disconnect between how sure they sound and how accurate they are can become dangerous. Here, we examine why post-training degrades a model’s sense of uncertainty, and we review techniques that can bring expressed confidence back in line with actual reliability. Through this, we argue that trustworthy AI means treating calibration as a core design goal.</p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":"19 1","pages":"10"},"PeriodicalIF":7.9,"publicationDate":"2026-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12874690/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146127166","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Network analysis of longitudinal electronic health records using linear mixed models.","authors":"Marina Vargas-Fernández, Juan de Dios Luna, Jordi Martorell-Marugán, Pedro Carmona-Sáez","doi":"10.1186/s13040-025-00508-y","DOIUrl":"10.1186/s13040-025-00508-y","url":null,"abstract":"<p><p>BACKGROUND: The accelerating development of healthcare data stored in electronic health records (EHRs) has created novel opportunities for biomedical research. Clinical data contains rich, heterogeneous, and longitudinal information of diverse cohorts of population. These data can be analyzed to uncover complex patterns that describe disease progression, comorbidities, and patient trajectories. Significant analytical challenges arise when identifying relationships among clinical variables, often due to high dimensionality and interdependence of temporal data, which requires adaptable methodologies. Classical approaches like Gaussian Graphical Modeling and Vector Autoregression often fall short in addressing these complexities because of the strict required assumptions of independence and stationarity, limiting their applicability to real-world EHRs data. To address this, we present MariNET, a novel approach based on linear mixed models designed to build networks based on clinical variables interactions from longitudinal EHRs. The methodology can effectively handle correlated observations and confounding variables, providing a robust framework for analyzing dynamic interactions between clinical variables over time, offering a scalable and plausible model which is adaptable to researchers’ questions. RESULTS: We show the potential of our method in the analysis of three different datasets. When applied to a COVID-19 mental health cohort, the model successfully captured symptom interactions aligned with previous knowledge. This result demonstrates robustness in handling repeated measures and homogeneous data. In evaluations using Parkinson’s Disease data, we effectively modeled temporal interactions among key clinical variables, aligning with known PD symptom progression and outperforming partial correlation and autoregressive vector models in handling heterogeneous and missing data. In a modified PD dataset by incorporating sex as a random factor, novel method also excelled in correcting biased interactions, unlike partial correlation, which produced misleading results. CONCLUSIONS: By using a linear mixed model-based approach, we can effectively capture dynamic interactions in longitudinal EHR data, outperforming traditional methods in analyzing complex clinical datasets. The different use cases here presented support the application of network science in healthcare, enabling more accurate predictions, evidence-based decision-making and improved disease management. </p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":"7"},"PeriodicalIF":7.9,"publicationDate":"2026-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12870267/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146114509","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Automatic clear cell renal cell carcinoma grading framework using histopathological images via artificial intelligence: a benchmarking study.","authors":"Hongde Song, Meng Yuan, Zhixi Li, Meijia Wang, Mengfan Xu, Haishan Lin, Huimin Lu, Hui Zhan, Ting Luan, Caifang Guo, Hong Fan, Kai Zhang, Haifeng Wang","doi":"10.1186/s13040-026-00524-6","DOIUrl":"10.1186/s13040-026-00524-6","url":null,"abstract":"<p><p>BACKGROUND: Renal cell carcinoma (RCC) is a common malignant tumor of the urinary system. Postoperative patients face the risk of recurrence or metastasis, and those who experience disease progression often have a poor prognosis. Therefore, investigating prognostic factors for RCC is crucial for guiding individualized patient treatment. Current studies demonstrate that RCC grading is significantly associated with prognosis, but pathological grading requires high expertise while being time-consuming. METHODS: Convolutional neural networks (CNNs) have shown significant advantages in tumor tissue recognition. Based on this, our study employed different CNNs (ResNet101, DenseNet121, EfficientNetb6, Wide ResNet101 32 × 8d, ResNeXt101) to automatically distinguish non-tumor tissue from tumor tissue and determine tumor grade. RESULTS: The results demonstrated that at 100× magnification (actual magnification of the pathological images, likewise for the following), in patch-level, the CNN achieved an accuracy of 0.8565 for identifying non-tumor tissues; at 400× magnification, the accuracy reached 0.8919. For tumor tissues of different grades, in patch-level, the CNNs performed best for Grade 1 tissues, with accuracies of 0.8616 at 100× magnification and 0.8182 at 400× magnification. Besides, at 100× magnification, in patch-level, the accuracy in identifying for Grade 2, Grade 3, Grade 4 tissues was 0.6225, 0.7604, and 0.8361. At 400× magnification, the accuracy was 0.6883, 0.6546 and 0.5127 for Grade 2, Grade 3 and Grade 4 tissues. Compared to pathologists, who can not distinguish between Grade 1 and Grade 2 tissues at 100× magnification, nor between Grade 2 and Grade 3 tissues at 400× magnification, the CNN can distinguish between Grade 1 and Grade 2 tissues at 100×magnification, achieving an accuracy of 0.6225 for Grade 2 tissue identification. They are also capable of differentiating Grade 2 and Grade 3 tissues at 400x magnification, with an accuracy of 0.6546 for Grade 3 tissue, which demonstrated superior performance. An integrated model constructed from patch-level predictions was used to determine the grade at the whole-slide level. The results revealed a clear tendency towards left-skewed or right-skewed distributions in the grading predictions causing by the manner constructed the integrated model emphasizing whether the highest-grade tissue is present in the tumor tissue. This approach requires further refinement. CONCLUSION: These findings indicate that CNNs can identify tumor and non-tumor tissue of clear cell renal cell carcinoma (ccRCC) and determine tumor grade, exhibiting advantages over pathologists. However, the model’s performance at the whole-slide level requires further improvement. Owing to the generalizability of this study, its findings can be considered as a potential benchmark for similar research, providing a reference for subsequent related studies. </p>","PeriodicalId":48947,"journal":{"name":"Biodata Mining","volume":" ","pages":"16"},"PeriodicalIF":7.9,"publicationDate":"2026-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12892658/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146097579","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"生物学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}