Frontiers in bioinformatics最新文献

{"title":"All-atoms MD simulations study of newly designed ciprofloxacin derivatives as potential bacterial DHPS and DNA gyrase inhibitors.","authors":"Chita Ranjan Sahoo, Madhusmita Rout, Sudhir Kumar Paidesetty, Budheswar Dehury, Debdutta Bhattacharya, Sanghamitra Pati","doi":"10.3389/fbinf.2026.1876643","DOIUrl":"10.3389/fbinf.2026.1876643","url":null,"abstract":"<p><p>The widespread use of antibiotics, particularly ciprofloxacin, has accelerated the emergence of antibiotic-resistant bacteria, posing a major public health challenge. Limited investment and declining pharmaceutical interest have slowed the discovery of new antibiotics, contributing to the increasing prevalence of multidrug-resistant (MDR) infections, especially those caused by ESKAPE pathogens. Uropathogenic <i>Escherichia coli</i> (UPEC), a leading cause of urinary tract infections (UTIs), develops multidrug resistance through mechanisms including drug target modification and reduced antibiotic susceptibility. In the present study, a series of ciprofloxacin-sulfonamide hybrid molecules (CIS1-CIS15) was designed, in which each pharmacophore retains its established mechanism of action. Ciprofloxacin inhibits DNA gyrase and topoisomerase IV, whereas sulfonamides inhibit dihydropteroate synthase (DHPS) by competing with para-aminobenzoic acid (PABA) in the bacterial folate biosynthetic pathway. Accordingly, the designed hybrids were computationally evaluated against both enzymatic targets to investigate their potential dual-target antibacterial activity. Molecular docking and molecular dynamics (MD) simulations were performed to assess the binding affinity and stability of the ciprofloxacin-bearing sulfonamide derivatives (CIS1-CIS15) against DNA gyrase and <i>Staphylococcus aureus</i> DHPS. Principal component analysis (PCA) and free-energy landscape (FEL) analysis were used to characterize the conformational flexibility and low-energy states of the protein-ligand complexes, while MD simulations validated the stability of the highest-scoring docked complexes. Among the designed derivatives, CIS4 (ciprofloxacin-sulfamethoxazole), CIS5 (ciprofloxacin-sulfaguanidine), and CIS12 (ciprofloxacin-sulfamerazine) exhibited the strongest binding affinity toward both targets while maintaining conformational stability throughout the simulations. Structural analysis of the pre- and post-MD trajectories identified Arg52, Lys39, Arg88, Lys94, Lys120, Glu201, and Lys203 as key residues involved in ligand recognition. Overall, CIS4, CIS5, and CIS12 emerged as the most promising dual-target inhibitor of bacterial DHPS and DNA gyrase. These findings provide valuable structural insights for the rational design of novel fluoroquinolone-based antibacterial agents and demonstrate the utility of computational approaches in identifying potential therapeutics against drug-resistant bacterial pathogens.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1876643"},"PeriodicalIF":3.6,"publicationDate":"2026-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13582570/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148992935","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Software engineering for reproducible pipeline development in bioinformatics.","authors":"Daniel Pérez-Rodríguez, Alba Nogueira-Rodríguez, Jorge Vieira, Cristina P Vieira, Daniel Glez-Peña, Hugo López-Fernández","doi":"10.3389/fbinf.2026.1824590","DOIUrl":"10.3389/fbinf.2026.1824590","url":null,"abstract":"<p><p>Reproducibility in bioinformatics remains challenging despite the availability of workflow management systems and mature computational infrastructures. This work presents a software-engineering perspective for developing reproducible bioinformatics pipelines, with emphasis on pipeline-specific code. We reinterpret the SOLID principles in this context at two levels: workflow management systems (e.g., Nextflow, Snakemake, and Compi) and pipeline implementation. Our approach promotes pipeline designs based on well-defined task interfaces, explicit input/output specifications, and a clear separation between compute tasks and glue/adaptation tasks, in order to improve flexibility, reuse, and maintainability. The paper provides practical guidance for robust and reproducible pipeline development, including systematic validation checks (environment, inputs, and runtime), standardized project organization, and comprehensive testing strategies using both real and synthetic data within continuous integration workflows. It also discusses how modular ecosystems (such as nf-core modules and Snakemake wrappers) support these principles in community-driven environments. Finally, we relate these recommendations to FAIR-oriented research software guidelines (FAIR4RS and FAIRsoft), showing how core engineering practices strengthen robustness, portability, and long-term sustainability, thereby supporting reproducibility in bioinformatics pipelines.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1824590"},"PeriodicalIF":3.6,"publicationDate":"2026-09-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13582464/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148992941","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Partial enumeration of extreme rays in metabolic networks using bit pattern trees.","authors":"Wannes Mores, Satyajeet Sheetal Bhonsale, Filip Logist, Jan F M Van Impe","doi":"10.3389/fbinf.2026.1892684","DOIUrl":"10.3389/fbinf.2026.1892684","url":null,"abstract":"<p><p>Extreme ray analysis of metabolic networks, even though very powerful, is currently limited to smaller metabolic networks. Some approaches to generating partial sets of extreme rays exist, but the computational efficiency of the so-called double-description method is yet to be exploited. Previous work highlighted the possibility of sampling within its iterations, enabling partial enumeration for double-description based methods. However, these approaches severely lack computational efficiency to be a suitable alternative. In this work, the highly efficient bit pattern trees are used within the sampling framework to significantly enhance its output and speed. Combined with the recent revision of the Canonical Basis Approach (CBA), our approach outperforms the other tested methods under the reported benchmark conditions even for a full enumeration study, requiring only half the computation time. In addition, a filter setting allows the memory demand to be scaled down while retaining high efficiency. However, some issues with the combinatorial explosion of candidates still persist and are further investigated. This study therefore puts forward a novel, double description-based alternative to partial enumeration of extreme rays. Further improvements in memory efficiency would allow this promising approach to scale powerful extreme ray-based analyses to genome-scale metabolic networks.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1892684"},"PeriodicalIF":3.6,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13574982/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148978895","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"DBP-CanPred: a machine learning model for predicting cancer-causing mutations in DNA-binding proteins.","authors":"Amit Phogat, Sowmya Ramaswamy Krishnan, Medha Pandey, M Michael Gromiha","doi":"10.3389/fbinf.2026.1920228","DOIUrl":"10.3389/fbinf.2026.1920228","url":null,"abstract":"<p><strong>Introduction: </strong>The fundamental cellular processes, including transcriptional regulation, chromatin organization, and genome maintenance, are regulated by DNA-binding proteins (DBPs). Mutations in DBPs can alter protein-DNA interactions, leading to tumor development. However, identifying such driver mutations remains a major challenge due to limitations of experimental approaches.</p><p><strong>Methods: </strong>We have trained a machine learning model, DBP-CanPred, to identify driver mutations in DBPs. We used the sequence-derived evolutionary features, as well as structure-based features such as mutation-perturbed structural descriptors.</p><p><strong>Results: </strong>We evaluated DBP-CanPred using a curated test set, achieving an AU-ROC of 0.86 and a balanced accuracy of 0.79. Further analysis based on substitution-type showed consistent performance across different categories, especially higher performance on charged residues. In addition, we applied the model on an independent dataset and identified potential driver mutations with high confidence scores.</p><p><strong>Discussion: </strong>The study contributes to understanding mutation patterns in DNA-binding proteins and supports variant interpretation in cancer research.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1920228"},"PeriodicalIF":3.6,"publicationDate":"2026-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13562013/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148956755","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Assessment of the impact of manual curation in BioCyc.","authors":"R Caspi, B Wilson-Mortier, L R Moore, J N Karp, D Beesley, I T Paulsen, P D Karp","doi":"10.3389/fbinf.2026.1887950","DOIUrl":"10.3389/fbinf.2026.1887950","url":null,"abstract":"<p><strong>Introduction: </strong>BioCyc is an extensive collection of databases of genomic and pathway information for microorganisms and model eukaryotes. These organismal databases integrate diverse biological data by combining computationally inferred information, data imported from other databases, and, for selected organisms, literature-based manual curation. This study investigates the magnitude and significance of annotation changes performed during the curation of 10 prokaryotic genomes to better understand the rate of erroneous annotations and the value of BioCyc curation.</p><p><strong>Methods: </strong>We identified curation changes by finding cases where the annotation of the protein at the start of the curation process differed from its annotation at the end of the process.</p><p><strong>Results: </strong>We found that across a sample of curated databases (<i>n</i> = 10), the annotation of 6,753, or 25.6% of the proteins in the pooled protein dataset (n = 26,126) were modified. Assessment of considerable sampling fractions of these proteins found that a median of 62% (mean of 52.9%) represented functionally informative name changes, rather than stylistic annotation changes. These results were then extrapolated to total proteins with name changes with uncertainty quantified via finite population correction, indicating that most Tier 2 Biocyc PGDBs received hundreds of functionally informative name changes during manual curation. On average 363, or13% (±5.4% SD) of the proteins encoded in each genome received functionally informative annotation changes, ranging from 5.3% <i>(Streptococcus pneumoniae</i> D39V) to 22.7% (<i>Staphylococcus aureus</i> NCTC 8325).</p><p><strong>Discussion: </strong>These findings demonstrate a substantial improvement in the accuracy of manually curated BioCyc databases compared with automated annotation pipelines. This result is particularly impactful as the rate of downstream propagation of erroneous annotations across biological databases can significantly compromise scientific discovery.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1887950"},"PeriodicalIF":3.6,"publicationDate":"2026-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13558213/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148931566","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Targeting IDH-driven Oncometabolism in glioblastoma: a computational approach for the identification of potential inhibitors.","authors":"Nivedhitha Tamilazhagan, Sivakumar Arumugam","doi":"10.3389/fbinf.2026.1884553","DOIUrl":"10.3389/fbinf.2026.1884553","url":null,"abstract":"<p><strong>Background: </strong>Glioblastoma (GBM) is a malignant brain tumor frequently driven by mutations in isocitrate dehydrogenase (IDH1 and IDH2) enzymes, which promote neomorphic synthesis of the oncometabolite 2-hydroxyglutarate (2-HG) and subsequent metabolic dysfunction. Although these mutations are associated with different clinical outcomes, therapeutic intervention remains challenging due to tumor heterogeneity, cellular plasticity, restricted blood-brain barrier permeability, drug efflux mechanisms, and effective DNA repair pathways. Therefore, this study aims to identify selective inhibitors of mutant IDH proteins by overcoming these challenges.</p><p><strong>Methods: </strong>10,309 compounds from ChemFaces and MedChemExpress libraries are used for large-scale virtual screening and multi-level ADMET filtering, uncovering 19 lead compounds with favourable drug-likeness. Further, molecular docking followed by 300 ns of molecular dynamics simulations, PCA analysis, and MM-PBSA calculations were performed to explore conformational dynamics.</p><p><strong>Results: </strong>Molecular docking revealed strong binding affinities of PubChem ID 91457 (I1Hit1) as -7.57 kcal/mol and PubChem ID 4946 (I1Hit2) as -6.82 kcal/mol against IDH1, and PubChem ID 71550939 (I2Hit1) as -11.23 kcal/mol and PubChem ID 4946 (I2Hit2) as -7.87 kcal/mol against IDH2, by outperforming the standard (-6.17 kcal/mol). Molecular dynamics simulations, PCA analysis, and MM-PBSA calculations confirmed complex stability and inhibitory potential. Notably, I1Hit1 and I2Hit1 exhibited enhanced binding free energies of -20.67 ± 2.32 kcal/mol against IDH1 and -28.78 ± 2.67 kcal/mol against IDH2 respectively, compared to the standard, as further supported by computational pharmacophore evaluation.</p><p><strong>Conclusion: </strong>Collectively, these findings highlight I1Hit1 and I2Hit1 as novel therapeutic compounds with efficient IDH-target inhibition to address epigenetic modification in GBM, and further experimental validation of these compounds is required to demonstrate potential inhibitors of IDH-driven metabolism in GBM.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1884553"},"PeriodicalIF":3.6,"publicationDate":"2026-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13553832/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148921939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Molecular biology and integrated strategies for activating cryptic biosynthetic gene clusters toward next-generation antibiotic discovery.","authors":"Veilumuthu Pattapulavar, Sathiyabama Ramanujam, Avanthi Yellampalli, Henu Sree Badiginchala, John Godwin Christopher","doi":"10.3389/fbinf.2026.1893206","DOIUrl":"10.3389/fbinf.2026.1893206","url":null,"abstract":"<p><p>Antimicrobial resistance (AMR) has been identified as one of the 21st century's severest global public health crises. AMR led to an estimated 4.95 million deaths in 2019 and will claim 10 million lives a year by 2050 in the absence of targeted interventions. During the same period, the number of novel antibiotics discovered has decreased drastically as many researchers are rediscovering known antibiotics, non-model microorganisms are poorly understood or difficult to culture and antibiotic research and development investment has declined drastically. However, high-throughput whole genome sequencing and the subsequent application of bioinformatics in bacterial and fungal genomes have shown that a numerous of cryptic or silent biosynthetic gene clusters (BGCs) remain latent at ambient laboratory conditions since their genes are transcriptionally inactive. Cryptic BGCs represent a vast source of unique secondary metabolites, many of which may yield novel antibacterial, antifungal, anti-cancer and other potentially valuable natural products. This review discusses the biological relevance of cryptic BGCs, the major limiting factors that restricts their activation and novel strategies that have been employed to activate them and exploit their potential to produce novel natural products. The review focuses on biological approaches including CRISPR-Cas mediation for the activation of cryptic BGCs, promoter engineering, pathway refactoring, and heterologous expression; biochemical strategies such as Osman, OsMAC, Precursor Feeding, Chemical Elicitation, Epigenetic Regulation and Co-cultivation and technology-based strategies such as Genome mining, Microfluidic Cultivation systems, High-Throughput Screening, Metabolomics, Molecular Networking and Artificial Intelligence and Machine Learning based prediction of BGCs and their metabolites. The use of multi-omics technologies combined with synthetic biology to achieve better discovery, characterization and large-scale production of novel natural products is also discussed herein. Finally, we will talk about the ecological significance and evolutionary advantage of cryptic BGCs' role in interactions between microorganisms, such as competition, communication, symbiosis and environmental adaptability, so as to provide a useful background for accelerating next-generation antibiotics.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1893206"},"PeriodicalIF":3.6,"publicationDate":"2026-08-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13550855/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148914535","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"<i>Sesamum indicum</i>-derived valdiate as a novel neuroprotective agent targeting PDE10A2 and SIRT1 in Huntington's disease.","authors":"Mukul Shyam, M D Wafi Ismail, Deepak Sharma, Prathap Srirangan, Rajiniraja Muniyan, Sabina Evan Prince","doi":"10.3389/fbinf.2026.1868337","DOIUrl":"10.3389/fbinf.2026.1868337","url":null,"abstract":"<p><strong>Background: </strong>Huntington's disease is prevalent globally, with approximately 4.88 cases per 100,000 people, based on a systematic review and meta-analysis of 33 studies published between 2010 and 2022. Despite its significant prevalence, no proper treatment is available that directly addresses Huntington's disease. The existing treatments focus on symptom management, such as controlling chorea and psychiatric symptoms. The drugs used for this purpose may also cause side effects, including depression and Parkinson's disease.</p><p><strong>Methods: </strong>An integrated experimental and computational approach was employed, involving cold maceration extraction of <i>Sesamum indicum</i>, LC-MS/MS based phytochemical profiling, ADMET screening, molecular docking, molecular dynamics simulations, and MMBPSA binding free energy analysis to identify potential inhibitors of Huntington's diseases associated targets.</p><p><strong>Results: </strong>Among the LC-MS/MS identified compounds, eight compounds satisfied the ADMET criteria. Valdiate (PubChem CID: 129715809) demonstrated favourable multi-target binding with docking score of -7.3 kcal/mol against HDAC4 and -8.9 kcal/mol against HDAC7. Molecular dynamics simulations confirmed stable protein-ligand interactions, while MMPBSA analysis yielded binding free energies of -23.22 ± 2.45 kcal/mol (HDAC4) and -28.27 ± 2.28 kcal/mol (HDAC7), identifying Valdiate as the most promising potential inhibitor.</p><p><strong>Conclusion: </strong>Valdiate may serve as a potential inhibitor of mutant huntingtin-associated pathological pathways, pending further <i>in-vitro</i> and <i>in-vivo</i> validation.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1868337"},"PeriodicalIF":3.6,"publicationDate":"2026-08-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13547276/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148908225","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluation of population-specific polygenic risk scores for blood lipids: insights from Taiwanese cohorts and multiancestry meta-analysis.","authors":"Yu-Chia Chen, Ting-Yuan Liu, Chi-Chou Liao, Jai-Sing Yang, Wei-De Lin, Wen-Ling Liao, Shih-Yin Chen, Fuu-Jen Tsai","doi":"10.3389/fbinf.2026.1823599","DOIUrl":"10.3389/fbinf.2026.1823599","url":null,"abstract":"<p><strong>Background: </strong>Blood lipids are heritable risk factors for cardiovascular disease (CVD), a leading cause of mortality worldwide. However, the genetic architecture of lipid traits and the performance of polygenic risk scores (PRSs) remain underexplored in East Asian (EAS) populations, including Taiwanese Han individuals.</p><p><strong>Methods: </strong>We conducted genome-wide association studies and PRS analyses for five lipid traits: total cholesterol, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, triglycerides, and the ratio of low-density lipoprotein cholesterol to total cholesterol. Lipid profile data were obtained from the China Medical University Hospital cohort. PRSs were evaluated on the basis of their correlations with measured lipid levels. To evaluate trans-ancestry PRS transferability, localized models were systematically compared against models derived from discovery-stage GWAS meta-analyses incorporating five ancestry groups from the Global Lipids Genetics Consortium. The performance of the PRS models in predicting lipid-related diseases was evaluated through receiver operating characteristic curve analyses.</p><p><strong>Results: </strong>The population-specific PRS models explained 11%-40% of the variance in lipid levels within the target cohort. Models leveraging global multiancestry GWAS meta-analysis weights revealed limited predictive performance (<i>r</i> ² = 0.04-0.19), whereas analyses incorporating EAS-specific data yielded higher correlations (<i>r</i> ² = 0.13-0.30), although these correlations did not exceed those derived from the hospital-based cohort alone. When combined with age and sex, the PRS models demonstrated strong predictive performance for coronary artery disease, atherosclerosis, and ischemic stroke, with area under the curve values of 0.910, 0.926, and 0.854, respectively.</p><p><strong>Conclusion: </strong>Population-specific PRS models derived from a Taiwanese population outperformed meta-analysis-derived frameworks in predicting lipid levels and demonstrated substantial potential for predicting CVD risk, indicating the importance of ancestry-matched genetic studies in precision medicine.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1823599"},"PeriodicalIF":3.6,"publicationDate":"2026-08-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13547480/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148908250","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TransCat: a hybrid CNN-transformer network with KAN for medical image segmentation.","authors":"Jin Wang, Zhenghua Yang, Dongming Zhou, Gang Chen, Sihan Wang, Yuhong Liu","doi":"10.3389/fbinf.2026.1849059","DOIUrl":"10.3389/fbinf.2026.1849059","url":null,"abstract":"<p><p>Medical image segmentation is a fundamental task in computer vision and plays an important role in clinical diagnosis and treatment planning. Existing methods are mainly built on either convolutional neural networks (CNNs) or Transformer-based encoder-decoder architectures. CNNs are effective at capturing local patterns, whereas Transformers are better suited to modeling long-range dependencies. However, current hybrid designs still face three persistent challenges: insufficient exploitation of multi-scale CNN features for global modeling, rapidly increasing attention cost after token expansion, and ineffective fusion between CNN and Transformer representations. To address these issues, we propose TransCat, a hybrid CNN-Transformer architecture for medical image segmentation. First, we introduce Patch Concat, which converts multi-scale CNN features into patch tokens and feeds them jointly into the Transformer, enabling global reasoning across multiple scales. Second, to control the computational burden caused by the enlarged token set, we develop an extended deformable attention mechanism with attentive value identification. Third, we design a Kolmogorov-Arnold Network (KAN) attention module for adaptive cross-branch fusion. The module projects high-level CNN and Transformer features into a shared space and uses a KAN layer to estimate data-dependent pixel-wise weights. These weights selectively modulate the globally modeled Transformer features, allowing complementary local structural evidence from the CNN branch to be incorporated according to the image content rather than through fixed addition or concatenation. Experiments on multiple medical image segmentation benchmarks show that TransCat is competitive under the reported protocols, with its clearest advantage observed on unseen polyp test data.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1849059"},"PeriodicalIF":3.6,"publicationDate":"2026-08-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13541688/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148898511","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}