Frontiers in bioinformatics最新文献

{"title":"LNMGAT: a laplacian regularized pseudo-negative mining graph attention network for robust drug-target interaction prediction under multi-scenario cold-start settings.","authors":"Shuai Guo, Weichi Liu, Jie Zou, Tao Ban, Gaifang Dong","doi":"10.3389/fbinf.2026.1882476","DOIUrl":"https://doi.org/10.3389/fbinf.2026.1882476","url":null,"abstract":"<p><p>Computational drug-target interaction (DTI) prediction provides a scalable alternative to costly and time-consuming experimental screening, but its reliability is limited by the scarcity of experimentally verified negative interactions. In public DTI databases, most unobserved drug-target pairs are unlabeled rather than true non-interactions. Randomly treating these unlabeled pairs as negatives can introduce label noise and reduce model reliability, particularly in cold-start scenarios involving unseen drugs or targets. To address this issue, we propose LNMGAT, a LapRLS-guided reliable pseudo-negative mining framework coupled with dual graph attention encoders. Instead of relying on experimentally confirmed negative labels or randomly sampled negatives, LNMGAT first applies Laplacian regularized least squares to drug and target similarity graphs to identify low-confidence unlabeled pairs as reliable pseudo-negatives. Drug and target representations are then learned separately on similarity-based k-nearest-neighbor graphs using graph attention networks, and their embeddings are concatenated for MLP-based interaction prediction. Across Yamanishi, Davis, KIBA, and BindingDB benchmarks, LNMGAT achieved the best AUPR in 10 of 16 evaluation settings and ranked within the top two in 14 of 16 settings. In the 12 cold-start settings, LNMGAT obtained the best AUPR in 9 cases, with absolute AUPR gains over the strongest baseline of up to 0.016 in pair cold-start prediction. External evaluation on DrugBank positive interactions and SwissDock-based molecular docking further provided database-level and <i>in silico</i> support for the plausibility of high-ranked predictions. Nevertheless, the biological validation in this study remains computational and database-based; no wet-lab binding assay was performed. LNMGAT therefore provides a competitive and interpretable framework for DTI prediction under negative-label uncertainty, while further experimental validation is required for its top-ranked candidates.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1882476"},"PeriodicalIF":3.6,"publicationDate":"2026-07-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13473418/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148762268","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":"A linked independent component analysis framework for characterizing site-effect patterns in multi-site structural and functional MRI.","authors":"Huashuai Xu, Yuge Xing, Weiya Guo","doi":"10.3389/fbinf.2026.1902380","DOIUrl":"10.3389/fbinf.2026.1902380","url":null,"abstract":"<p><strong>Introduction: </strong>Large-scale multi-site magnetic resonance imaging (MRI) improves population coverage and statistical power, but scanner- and protocol-related variability can obscure biological effects. Most harmonization methods aim to reduce site-related variance for downstream analysis, whereas less attention has been paid to where site effects are spatially expressed, whether they are reproducible across site compositions, and which acquisition parameters contribute to them.</p><p><strong>Methods: </strong>We developed a modality-wise Linked Independent Component Analysis (LICA) framework to identify and interpret site-effect patterns in structural and resting-state functional MRI. Grey matter (GM) volume, amplitude of low-frequency fluctuation (ALFF), and regional homogeneity (ReHo) maps were analyzed separately. For each imaging measure, LICA decomposed voxel-wise maps into spatial components and subject-level loadings. Components were classified according to their associations with site labels and biological covariates, their spatial reproducibility was assessed using stepwise site-inclusion analyses, and their technical attribution was evaluated using cross-validated models based on site labels and recorded acquisition parameters. The framework was applied to ABIDE II GM maps from 913 participants across 18 sites and ALFF and ReHo maps from 795 participants across 16 sites.</p><p><strong>Results: </strong>LICA identified site-related components across all three imaging measures. Site effects were not limited to uniform global shifts, but formed modality-specific spatial patterns. GM volume showed a dominant and highly stable whole-brain site-effect pattern, together with site-specific and regional components. In contrast, ALFF and ReHo showed more heterogeneous functional patterns, including global, focal, and scattered configurations. Site labels explained the largest proportion of loading variance, whereas recorded acquisition parameters showed modality-dependent contributions: TR and TE were more prominent for structural site effects, while FA, voxel size, TR, and scanner model contributed more strongly to functional site effects.</p><p><strong>Discussion: </strong>The proposed framework provides a component-level diagnostic approach for multi-site MRI analysis. By mapping, stabilizing, and technically interpreting site-effect patterns, it complements conventional harmonization methods and may improve the transparency and reproducibility of multi-site structural and functional MRI studies.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1902380"},"PeriodicalIF":3.6,"publicationDate":"2026-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13461629/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148726769","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":"Computational design of immunogenic peptide-ligand conjugates for targeted therapy against Nipah virus infection.","authors":"Sudip Prasad Jena, Prateek Nayak, M A Adithyan, Manoviraj Morey, R Vihaas Tharsini, Abilash Valsala Gopalakrishnan, Kuntal Pal, Sabina Evan Prince","doi":"10.3389/fbinf.2026.1846380","DOIUrl":"10.3389/fbinf.2026.1846380","url":null,"abstract":"<p><strong>Introduction: </strong>Nipah virus (NiV) poses a major risk to global public health due to its high infectivity and associated mortality rates. Currently, no licensed vaccines or antiviral medications are available for NiV infection, leaving clinical management limited to supportive care. The viral receptor glycoprotein responsible for binding NiV to host cell receptors (ephrin-B2/B3) represents an ideal therapeutic target. This study proposes a novel peptide-ligand conjugate (PLC) immunotherapeutic approach that exploits pre-existing immune responses in NiV-endemic populations to selectively target and eliminate infected cells.</p><p><strong>Methods: </strong>We employed biomolecular modeling (<i>in silico</i>) to establish binding affinities and perform docking studies using a compound library obtained from the MolProphet database. A non-cleavable oxime linker was selected to enhance physical stability and ensure robust conjugation between ligand and peptide components. The peptide was engineered to contain immunogenic minimal epitope regions derived from measles, mumps, and rubella vaccines, selected based on their high immunization rates and long-lived memory responses in individuals residing in NiV-endemic areas.</p><p><strong>Results: </strong>The PLC design demonstrated selective binding capacity to a transmembrane protein present on NiV-infected cells. The oxime linker provided enhanced stability, and the peptide epitope design successfully incorporated regions associated with established long-term immunity.</p><p><strong>Discussion: </strong>This PLC system represents a promising framework for antiviral therapeutic development by harnessing pre-existing immune recognition to promote selective clearance of NiV-infected cells. The findings highlight critical structural components and functional roles of PLCs in therapeutic development, including drug target screening and rational design strategies for enhancing targeting specificity and molecular stability. Future work should focus on experimental validation of the computational predictions and <i>in vitro</i>/<i>in vivo</i> efficacy studies.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1846380"},"PeriodicalIF":3.6,"publicationDate":"2026-07-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13454029/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148708863","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":"Instability of LLM text embeddings for unsupervised dimension reduction of tabular data.","authors":"Jun Li, Yixuan Gou, Shawn Su, Mujun Xu, Brendan Chen","doi":"10.3389/fbinf.2026.1851023","DOIUrl":"10.3389/fbinf.2026.1851023","url":null,"abstract":"<p><p>Large language model (LLM) text embeddings have recently been used for supervised learning on tabular data by serializing each observation into text and then converting the text into a dense fixed-length vector. This strategy is appealing for biomedical tabular data, which are often mixed-type and contain missing values, because it produces a complete numeric representation even when some original entries are missing. However, its suitability for unsupervised tasks remains unclear. Here we evaluate the use of LLM-derived text embeddings for dimension reduction of tabular data, focusing on biological and clinical datasets. We compare an LLM embedding-based approach with a direct tabular approach that computes dissimilarities directly from the original variables. Because unsupervised dimension reduction has no ground-truth low-dimensional target, we assess performance through stability under perturbation. Across multiple datasets and analysis settings, the LLM embedding-based approach is consistently less stable than the direct tabular approach. In particular, small amounts of additional missingness and random permutation of feature order can substantially alter the resulting low-dimensional representation. These results suggest that the straightforward use of LLM text embeddings is not reliable for unsupervised dimension reduction of tabular data.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1851023"},"PeriodicalIF":3.6,"publicationDate":"2026-07-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13454057/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148708851","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>In silico</i> drug discovery pipelines targeting antibiotic resistance: from genomes to leads.","authors":"Rohan Gupta, Pallavi Singh, Karthikeyan Ravi, Mosleh Mohammad Abomughaid, Sorabh Lakhanpal, Ravi Sharma, Naveen Kumar, Niraj Kumar Jha","doi":"10.3389/fbinf.2026.1839606","DOIUrl":"10.3389/fbinf.2026.1839606","url":null,"abstract":"<p><p>Antimicrobial resistance (AMR) is currently one of the leading global health threats. The evolution of drug-resistant bacterial pathogens is rapid, and there is a growing number of bacterial pathogens that have developed resistance to multiple antibiotics, and the rate at which new antibiotics are being developed is lagging far behind these two issues. In addition, historical drug discovery processes rely on conducting traditional <i>in vitro</i>-based studies to determine new antibiotics to use in practice. However, this process is becoming increasingly constricted due to high costs of conducting traditional <i>in vitro</i> research, lengthy timeframes to bring products to market, a high attrition rate of research projects in traditional wet laboratory environments, and the need for more effective and efficient ways of developing new drugs. For this reason, Drug discovery continues to evolve from a wet lab-based approach to an <i>in silico</i> (i.e., computational) based approach, which takes advantage of the advances made in various fields, such as bacterial genomics, structural bioinformatics, machine learning (ML), and systems biology to enable researchers to rationally design, discover, and develop new antibiotics to combat drug-resistant pathogens. This review aims to provide a comprehensive and critical overview of contemporary <i>in silico</i> antibiotic discovery strategies and their potential to accelerate the development of novel, resistance-resilient, and clinically relevant antimicrobial agents. It examines genome-informed approaches ranging from genomic data generation, resistome analysis, and computational target identification to structure-based drug design, ligand-based and fragment-based discovery, drug repurposing, and the expanding applications of ML and artificial intelligence (AI) in activity prediction, <i>de novo</i> antibiotic design, and resistance evolution modeling. The review also highlights the importance of <i>in silico</i> ADMET prediction in lead optimization and discusses representative case studies demonstrating successful translation of computational predictions into experimental validation. Overall, the integration of digital-first, data-driven, and genome-guided discovery pipelines with experimental validation offers a powerful framework to overcome current challenges in antibiotic development and represents a promising strategy for addressing the global threat of AMR. Lastly, the review was conducted using a structured literature search across major biomedical and computational databases with emphasis on experimentally validated case studies and translational relevance.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1839606"},"PeriodicalIF":3.6,"publicationDate":"2026-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13449674/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148698496","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":"Functional restoration of conformational (R175H) and contact (R273H) mutant p53 by <i>Bacillus</i>-derived Hydroxymycotrienin A in non-small cell lung cancer: a computational approach.","authors":"Adeline Celina Rufus, Sidharth Kumar N, Magesh Ramasamy, Elavarashi Elangovan","doi":"10.3389/fbinf.2026.1875897","DOIUrl":"10.3389/fbinf.2026.1875897","url":null,"abstract":"<p><strong>Introduction: </strong>The TP53 mutations in Non-Small Cell Lung Cancer (NSCLC) remain a formidable clinical challenge. Current strategies, using the covalent binder APR-246, are limited by off-target toxicity and resistance.</p><p><strong>Methods: </strong>This study screens 1,580 <i>Bacillus</i>-derived metabolites to identify novel non-covalent pharmacological chaperones for native, conformational (R175H), and contact (R273H) p53 mutants. Virtual screening, ADMET profiling, molecular docking, extended 500 ns Molecular Dynamics (MD) simulations and Principal Component Analysis, against experimental anti-cancer drug APR-246, and top hit compounds.</p><p><strong>Results: </strong>Hydroxymycotrienin A emerged as the potential candidate, demonstrating thermodynamic superiority with binding affinities of -6.63 kcal/mol (R175H) and -6.57 kcal/mol (R273H), demonstrated significant superior non covalent docking affinity compared to APR-246 parent scaffold (approximately -3.5 kcal/mol) in the mutated p53 protein, suggesting a direct pharmacological chaperone activity. Unlike the covalent alkylating mechanism of APR-246, Hydroxymycotrienin A utilizes a non-covalent network to chaperone the mutant p53. MD simulations revealed that Hydroxymycotrienin A acted as a structural stabilizer for the conformational mutant R175H by suppressing atomic fluctuations within the L2 loop and reducing overall structural deviations. In the contact mutant R273H, the ligand stabilized the DNA-binding interface while maintaining favorable conformational dynamics without introducing steric clashes.</p><p><strong>Discussion: </strong>ADMET profiling predicts high bioavailability and a non-toxic safety profile, characterizing Hydroxymycotrienin A as a promising, bioavailable 'privileged scaffold' that offers a non-covalent therapeutic strategy for NSCLC. By restoring p53 function, this study addresses Sustainable Development Goals Target 3.4 to reduce premature mortality from non-communicable diseases. However, future <i>in vitro</i> and <i>in vivo</i> studies are required to validate the efficacy of these compounds.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1875897"},"PeriodicalIF":3.6,"publicationDate":"2026-07-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13447244/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148690328","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":"Integrative phosphoproteomics reveals kinase-mediated regulation of OCIAD1 and its roles in mitochondrial quality control.","authors":"Amal Fahma, Suhail Subair, Fathimathul Lubaba, Athira Perunelly Gopalakrishnan, Prathik Basthikoppa Shivamurthy, Rajesh Raju","doi":"10.3389/fbinf.2026.1770712","DOIUrl":"10.3389/fbinf.2026.1770712","url":null,"abstract":"<p><strong>Background: </strong>The ovarian cancer immunoreactive antigen domain-containing protein 1 (OCIAD1) is a mitochondrial protein implicated in mitochondrial morphology, energy metabolism, and differentiation. Although understudied, recent studies position it as a critical player in carcinogenesis and neurodegenerative disorders, making it a potentially druggable node in cellular signaling networks. However, the phosphoregulatory networks and the upstream kinases governing OCIAD1 remain unknown.</p><p><strong>Methods: </strong>A large-scale literature mining and analysis of 177 phosphoproteomic datasets with differential expression of OCIAD1 was carried out to map its phosphoregulatory network. The predominant phosphosites were determined based on localization probability, detection frequency, and differential regulation. Multipronged computational approaches were employed to gather novel candidate kinases that may target OCIAD1 phosphosites. Co-differential phosphorylation analysis was conducted with other proteins, including interactors and candidate upstream kinases, to infer functional and regulatory associations.</p><p><strong>Results: </strong>The sites S108 and S123 emerged as predominant, together accounting for 70% of OCIAD1 phosphorylation. Co-differential phosphorylation analysis revealed associations with proteins involved in the cell cycle, DNA repair, autophagy, mitophagy, endocytosis, and apoptosis. Novel candidate kinases for OCIAD1 phosphosites were identified; notably, SRMS and YES1 emerged as potential upstream regulators of Y199. Furthermore, the phosphosites in the candidate kinases of sites, including PLK1 (T210), CDK13 (S383, S397), PRKD2 (S200), CIT (S1343), and RPS6KA3 (T577), showed strong positive co-differential regulation with OCIAD1 predominant sites, supporting their potential involvement as upstream kinases.</p><p><strong>Conclusion: </strong>This study presents the first systematic map of the OCIAD1 phosphoregulatory network and provides candidate upstream kinases that may contribute to its phosphorylation, which warrant further experimental validation. The strong co-differential regulation of proteins involved in autophagy, mitophagy, endocytosis, and neurodegenerative pathways, as well as of kinases that orchestrate these processes, suggests that OCIAD1 phosphoregulatory network maybe involved in mitochondrial quality control and mitochondria-associated neurodegeneration, establishing a foundation for therapeutic investigations targeting OCIAD1 signaling.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1770712"},"PeriodicalIF":3.6,"publicationDate":"2026-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13442898/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148686838","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":"Pan-cancer multi-omics analysis of pharmacogenomic alterations.","authors":"Aisha AlMulla, Simerpreet Kaur, Prachi Balyan, Sana Al-Saafin, Manoj Kumar Balyan, Dinesh Velayutham, Puthen Veettil Jithesh","doi":"10.3389/fbinf.2026.1858518","DOIUrl":"10.3389/fbinf.2026.1858518","url":null,"abstract":"<p><strong>Background: </strong>Pharmacogenomic (PGx) variation is an important determinant of inter-individual variability in drug response. While germline pharmacogenomics has been extensively studied, somatic alterations affecting these pharmacogenes in tumors remain less systematically characterised.</p><p><strong>Methods: </strong>We performed a pan-cancer multi-omics analysis integrating somatic mutation, copy number alteration, and transcriptomic data across 765 patients with primary tumors spanning seven The Cancer Genome Atlas (TCGA) cancer types. Exploratory survival analyses were restricted to 733 patients with complete clinical and overall survival information. Using a curated panel of 40 pharmacogenes, we characterized gene-level alterations, summarised recurrent events across tumor types, and aggregated these signals into pharmacological modules representing key pharmacological processes.</p><p><strong>Results: </strong>Pharmacogene alterations were generally infrequent at the individual gene level but showed recurrent, non-uniform patterns across cancers. A uniform-gene null model showed that descriptive combined scores were more concentrated than expected under a simplified uniform-gene background, with a global Gini index of 0.306 compared with a null mean of 0.248, and the top 10 gene-cancer pairs accounting for 10.1% of the total score compared with a null mean of 7.8%. Broad non-synonymous mutation, copy number amplification, and copy number deletion frequencies averaged 3.8%, 4.7%, and 0.3%, respectively, while the mean frequency of any integrated alteration was 8.6%. A descriptive prioritization score, defined as broad non-synonymous mutation frequency plus copy number amplification frequency, identified recurrent high-ranking alterations in transporter- and drug-handling genes, including <i>ABCB1</i>, <i>ABCC1</i>, and <i>ABCC2</i>. Sensitivity analyses supported cautious interpretation, including partial ranking robustness across stricter truncating/LoF-like mutation definitions, small-cohort exclusions, and cBioPortal GISTIC amplification sensitivity analyses. Systematic CNA-expression testing across 280 gene-cancer pairs identified 64 significant positive associations after FDR correction, with representative examples including <i>GGH</i>, <i>ERCC1</i>, and <i>SLC29A1</i>. Module-level pharmacogenomic alteration was not significantly associated with survival in exploratory treatment-agnostic survival models.</p><p><strong>Conclusion: </strong>Our findings provide an exploratory TCGA-based map of somatic alterations of pharmacogenes across cancers and identify recurrent alteration patterns that warrant validation in treatment-annotated and external datasets.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1858518"},"PeriodicalIF":3.6,"publicationDate":"2026-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13437544/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148681417","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":"Analysing open-source protein folding models for nanobody binding prediction.","authors":"Yannick Vogt, Rebekka Roßberg, Jan Habermann, Pia Kluge, Wendan Xu, Mehdi Naouar, Stefan Hirschberg, Cornelius Miething, Joschka Bödecker, Maria Kalweit, Evelyn Ullrich, Gabriel Kalweit","doi":"10.3389/fbinf.2026.1858891","DOIUrl":"10.3389/fbinf.2026.1858891","url":null,"abstract":"<p><strong>Introduction: </strong>Antibody-based therapeutics are a rapidly expanding class of treatments, with over 200 approved candidates and thousands in clinical trials. Computational pre-filtering using protein structure prediction models has the potential to reduce the cost of wet-lab screening, yet the relationship between model confidence measures and functional binding properties remains incompletely understood. Here, we evaluate whether confidence measures produced by contemporary open-source protein structure prediction models are suitable for <i>in silico</i> screening of nanobody-antigen interactions.</p><p><strong>Methods: </strong>We benchmark Boltz-2, Chai-1, IntFold, and AlphaFold3 using two complementary tasks: (i) ranking true nanobody-antigen binding complexes above non-binding bait pairs across 17 antigens, and (ii) detecting out-of-distribution sequences generated by alanine substitution of all complementarity-determining region residues. We further assess confidence measure sensitivity through progressive alanine mutagenesis on 13 nanobody-antigen complexes spanning the range of CDR3 lengths in our dataset and evaluate generalizability using data from a camelid immunization campaign against CD33.</p><p><strong>Results: </strong>Boltz-2-derived confidence measures achieved the highest median performance for identifying true binders. Local confidence measures, including pLDDT and interface- or CDR-focused metrics, were most effective at detecting out-of-distribution sequences and exhibited the greatest sensitivity to mutations. No single confidence measure performed best across both tasks, and all evaluated protein structure prediction models showed limited generalization to previously unseen antigens.</p><p><strong>Discussion: </strong>Our results suggest that robust <i>in silico</i> nanobody candidate selection should combine complementary global and local confidence measures rather than relying on a single metric. These findings provide practical guidance for integrating open-source protein structure prediction models into AI-driven nanobody discovery pipelines while highlighting the need for improved generalization across antigens.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1858891"},"PeriodicalIF":3.6,"publicationDate":"2026-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13437531/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148681227","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":"Clinically interpretable deep learning for breast cancer missense variant pathogenicity prediction.","authors":"Rahaf M Ahmad, Noura AlDhaheri, Mohd Saberi Mohamad, Bassam R Ali","doi":"10.3389/fbinf.2026.1844071","DOIUrl":"10.3389/fbinf.2026.1844071","url":null,"abstract":"<p><strong>Background: </strong>Missense variants in breast cancer remain diagnostically challenging due to their functional diversity and complex genomic contexts. Conventional laboratory assays for evaluating pathogenicity are labor-intensive, costly, and often impractical for large-scale screening, creating a pressing need for accurate, scalable, and clinically interpretable computational approaches.</p><p><strong>Methods: </strong>In this study, we present a novel deep learning framework for predicting the pathogenicity of breast cancer missense variants, integrating comprehensive preprocessing, advanced imputation, rigorous model benchmarking, and explainability. Genetic variants were curated from multiple genomic databases, annotated using the Ensembl Variant Effect Predictor (VEP), and processed with Variational Autoencoders (VAE) for missing-value imputation. Seven deep learning models, MLP, CNN, DNN, RNN, LSTM, GRU, and Transformer, were trained and evaluated across 11 performance metrics. To quantify performance stability, each model was trained across five random seeds; mean AUC ± SD across seeds is reported as the primary performance estimate, with the best-seed run used only for LIME and PMI interpretability analyses. Recursive feature elimination, permutation importance (PMI), and Local Interpretable Model-Agnostic Explanations (LIME) were employed to enhance transparency. Statistical analyses, including Z-tests, ANOVA, and calibration assessments, validated performance consistency and inter-model differences.</p><p><strong>Results: </strong>GRU achieved the highest internal AUC (0.9956 [95% CI 0.9936-0.9972]; mean across five seeds 0.9941 ± 0.0011), with precision 0.9967 and calibration ECE 0.0095. Externally, LSTM led with AUC 0.9457, exceeding all eleven standalone predictors benchmarked on the same set. Models showed strong alignment with conservation signals such as phyloP470way and Eigen-PC scores. Notably, the pipeline provides performance metrics with 95% confidence intervals and incorporates case-level LIME visualizations for true positive, true negative, false positive, and false negative predictions, bolstering interpretability and clinical relevance.</p><p><strong>Conclusion: </strong>This work delivers one of the most comprehensive evaluations of deep learning in breast cancer variant classification to date. By combining high-performance sequential models with interpretable AI tools, the proposed framework provides a reproducible, transparent benchmark for variant pathogenicity prediction and a foundation for future research use and translation in cancer genomics.</p>","PeriodicalId":73066,"journal":{"name":"Frontiers in bioinformatics","volume":"6 ","pages":"1844071"},"PeriodicalIF":3.6,"publicationDate":"2026-07-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13433856/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148674682","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}