Computational biology and chemistry最新文献

Disease comorbidity-the co-occurrence of multiple diseases in the same individual-is increasingly prevalent and poses major clinical and biological challenges. Computational approaches for studying disease relationships and predicting comorbidity have evolved from overlap-based similarity measures to molecular network modeling and graph deep learning. However, existing methods often (i) learn global or subgraph-based disease embeddings without modeling the topology of fragmented disease subgraphs in a comorbidity-adaptive manner, or (ii) incorporate Gene Ontology (GO) information in ways that underutilize GO's hierarchical ancestry and deeper functional abstractions. In this work, we propose DisSubFormer, a subgraph Transformer model for disease subgraph representation learning and comorbidity prediction. We first learn unified protein representations by integrating structural patterns from a PPI network with GO-aware functional information, explicitly incorporating GO's hierarchical ancestry. We next sample biologically informed anchor patches in a property-aware manner to prioritize disease-relevant regions of the PPI network, replacing full-graph attention with subgraph-to-subgraph attention between disease subgraphs and these anchor patches to improve scalability and relevance. Specifically, DisSubFormer introduces a learnable multi-head attention mechanism where each head attends over a distinct anchor-patch type, with head-specific relational terms to capture complementary positional, neighborhood, and structural properties within fragmented disease subgraphs for comorbidity prediction. Experiments on a benchmark comorbidity dataset demonstrate that DisSubFormer consistently outperforms state-of-the-art methods, achieving an AUROC of 0.97.

Dementia represents a rapidly rising global health challenge as a progressive neurodegenerative disease with few options for disease-modifyingtreatments. The present studyaimed to explore the leading phytochemicals from Crocus sativus (saffron) and Matricaria chamomilla (chamomile) and apply AI fragmentation on lead phytochemicals to target the aryl hydrocarbon receptor (AHR), an expertized target for dementia therapy. Bioactive compounds were screened from ISO 3632-2-2010 (E) specified for saffron and GC-MS specified for chamomile. Protein Network mapping, Density Functional Theory, Molecular docking, and molecular dynamics simulations were performed to determine thebinding affinity and interactions stability of key phytochemicals with AHR, such as safranal and bisabolone oxide A. In-silico ADMET predictions of pharmacokinetics and toxicity showed good properties for these molecules. In addition, their structuraland pharmacological properties were optimized to enhance drug-like features by using artificial intelligence (AI) generative model. Collectively, our findings highlight these AI-enhanced phytochemicals as promising AHR modulators with potentially therapeutic activities in pathological pathways that lead toneuroinflammation and oxidative stress involved in the pathogenesis of dementia. They offer an avenue for additional experimental validation and encourage further investigation of these leads as sources of new therapeutic modalities to treat neurodegenerativediseases.

With climate change, habitat destruction, and increased population ages, the incidence of both communicable and non-communicable diseases is rising, and managing these has become a growing concern. In recent years, generative artificial intelligence (AI) and large language models (LLMs) have ushered in a transformative era for smart healthcare applications. These models, built on advanced ML architectures like Generative Pre-trained Transformers (GPT) and Bidirectional Encoder Representations from Transformers (BERT), have demonstrated significant capabilities in various medical tasks. This review aims to provide an overview of the potential benefits of generative AI and LLMs in smart healthcare applications, as well as challenges and ethical considerations. A systematic literature review was conducted to identify relevant research papers published in peer-reviewed journals. Databases such as PubMed, PMC, Cochrane Library, Google Scholar, and Web of Science were searched using keywords related to generative AI, LLMs, and healthcare applications. The relevant papers were analyzed to extract key findings and contributions. Generative AI and LLMs are powerful tools that can process and analyze massive amounts of data. Researchers are actively exploring their potential to transform healthcare-powering intelligent virtual health assistants, crafting personalized patient care plans, and facilitating early detection and intervention for medical conditions. With ongoing research and development, the future of generative AI and LLMs in healthcare is promising; however, issues such as bias in AI models, lack of explainability, ethical concerns, and integration difficulties must be addressed.

Antigenic peptide (AP) prediction is one of the most important roles in improve vaccine design and interpreting immune responses. This paper develops a Multi-Level Pooling-based Transformer (MLPT) model, which improves the accuracy and efficiency of predicting T-cell epitopes (TCEs). The model has utilized peptide sequences from the Immune Epitope Database (IEDB) and utilized a refined Kolaskar & Tongaonkar algorithm for feature extraction as well as a Self-Improved Black-winged Kite optimization algorithm to optimize the scoring matrix. The MLPT architecture takes the input features from the Adaptive Depthwise Multi-Kernel Atrous Module (ADMAM) as inputs to the Swin Transformer, and the output of Swin block 1 is concatenated with the features extracted from the Kolaskar-Tongaonkar algorithm with the SA-BWK model. This hierarchical integration enhances feature representation and predictive capability. Advanced feature extraction, coupled with optimized feature selection for the MLPT model improves its performance over the conventional approach in the identification of reduced-complexity antigenic determinants.

Cytological examination of serous effusion is critical for diagnosing malignancies, yet it heavily relies on subjective interpretation by pathologists, leading to inconsistent accuracy and misdiagnosis, especially in regions with limited medical resources. To address this challenge, we propose a two-step deep learning framework to standardize and enhance the diagnostic process. First, we improved the YOLOv8 model by integrating the Online Convolutional Reparameterization (OREPA) module, achieving a 93.09 % sensitivity for detecting abnormal cells. Second, we employed the Dual Attention Vision Transformer (DaViT) to classify normal cells (lymphocytes, mesothelial cells, histiocytes, neutrophils) with 98.74 % accuracy. By jointly deploying these models, our approach reduces missed diagnoses and provides granular insights into cell composition, offering a robust tool for rapid and objective cytopathological diagnosis. This work bridges the gap between AI-driven automation and clinical needs, particularly in resource-constrained settings.

Antimicrobial resistance (AMR) poses a growing global threat, with antibiotic-resistant infections becoming a leading cause of death worldwide. The present study explores natural cyanobacterial compounds as possible inhibitors of Escherichia coli DNA gyrase B (GyrB) which is a verified antibacterial target that is not present in higher eukaryotes. Because of the urgent need for novel antibacterial drugs, we identified nine drug-like candidates using lipinski's rule of five and ADMET profiling. Molecular docking revealed that Biselyngbyaside B and Smenamide A exhibited greater binding affinities in comparison to the co-crystallized inhibitor EOF, with a binding energy of -9.03 kcal/mol. Further molecular dynamics simulations revealed that the Biselyngbyaside B-DNA gyrase B complex surpassed both EOF and Smenamide A in terms of structural stability, compactness, and strong hydrogen bonding. Umbrella sampling was employed to estimate the binding free energy from thirty sampling simulations, and Biselyngbyaside B exhibited a significantly favourable ΔG bind of -91.66 kJ/mol, outperforming EOF (-68.93 kJ/mol) and Smenamide A (-36.4 kJ/mol). These findings clearly indicate a stronger and more stable interaction between Biselyngbyaside B and GyrB. Biselyngbyaside B continuously showed better pharmacokinetic characteristics, non-hepatotoxicity, and a greater binding affinity than previously documented DNA gyrase B inhibitors. This study emphasizes the integration of molecular dockings, molecular dynamics simulation, umbrella sampling, and ADMET analysis provided crucial quantitative insights into the identification of potent drug-like candidates for further validation. Overall, the Biselyngbyaside B was found to be the most promising lead compound for novel antibacterial drug development targeting DNA gyrase B.

Acid sphingomyelinases (aSMases) are enzymes involved in the repair of the plasma membrane in eukaryotic cells. However, neutral sphingomyelinases (nSMases) have also been shown to possess other roles in bacteria and eukaryotic microorganisms, especially as virulence factors. These enzymes exhibit structural conservation but are characterized by elusive homology and the lack of sequence signatures or motifs. In a previous study, we reported the structural features of the complete set of sphingomyelinases (SMases) in Entamoeba histolytica and Trichomonas vaginalis, showing structural homology and functional differences in two aSMases from E. histolytica (EhSMase). However, the approach was limited due to the AlphaFold3 source code not being publicly available at the time. In this report, the structural transitions in the aSMases from T. vaginalis (TvSMase) were measured using open-source AlphaFold3 and collective motions of proteins via Normal Mode Analysis in internal coordinates. They compared them with the models from aSMase4 (EHI_100080) and aSMase6 (EHI_125660) from E. histolytica, containing different combinations of ligands. Using full-length sphingomyelin and the Mg²⁺ and Co²⁺ ions, where Co²⁺ was shown to inhibit the enzymes of both organisms, we demonstrate that the enzymes exhibit limited flexibility and deformability, except for the T. vaginalis TVAG_271580 enzyme, which displays high structural deformability. This contrasts with the inhibitory mechanism elicited by Co²⁺ as shown previously. TVSMase3 (TVAG_222460) could not be modelled with the sphingomyelin in the active site pocket, suggesting a regulatory role rather than a functional active enzyme. Additional physicochemical parameters calculated for T. vaginalis enzymes suggest unstable structures and high internal mobility (estimated using the Internal Coordinate method), which may be associated with the functional role of these enzymes. The results presented here open an avenue for searching for novel inhibitors of aSMases that target their physical properties, which could potentially complement treatment to control the parasite burden. These inhibitors could be valuable for further studying the role of these enzymes in parasite pathobiology and, potentially, as therapeutic targets.

Drug repurposing represents a promising approach towards drug discovery that has the potential to improve patient outcomes and address unmet medical needs. This study attempted to repurpose existing sulfonamide drugs in search of novel anticancer drugs because of their effectiveness in treating bacterial infections. A search was made in DrugBank for Sulfonamide, and 25 drugs with functional groups like SH, OSO, CS, and -S- were chosen for our study. The drug properties, such as dipole moment, volume, polarisability, highest occupied molecular orbital (HOMO), lowest unoccupied molecular orbital (LUMO), and electrostatic potential map, were analysed through a quantum mechanical approach at different functionals: M062X, M06HF, and B3LYP with basis sets (6-31 +G*, LANL2DZ). The electrostatic potential map was analyzed to determine the magnitude, size, and distribution of the electron cloud surrounding the sulfur atoms. Analysis of NBO (Natural Bond Orbital) and NCI (Non-Covalent Interaction) plots confirmed the presence of intramolecular hydrogen bonding in the sulfonamide drugs. Furthermore, the frontier molecular orbitals (HOMO and LUMO) and the band gap were thoroughly examined for all drugs to identify the best electron acceptors and donors. Docking analysis was performed to have a lock-and-key model of 25 sulfonamide drugs with the most promising cancer-targeted protein (1ZZ1): histone deacetylases (HDACs). The best drug orientation (optimal position) was discussed and compared with the control ligand SHH based on the analysis of binding affinity and root mean square deviation (RMSD). Binding affinity of control ligand SHH is -8.1 kcal/mol for the 2nd pose, which matches exactly with 1ZZ1 SHH ligand. The drugs Tolazamide, Fezolinetant, Ensulizole, Taurolidine, Acetohexamide, Isoxicam, Sulfamethizole, Sulfamethoxazole, Sulfapyridine, Sulfaphenazole, and Dodecyl sulphate were observed to exhibit high molecular volume, polarizability, dipole moment and significant HOMO, LUMO values, which are recommended for further quantum mechanical calculations. The findings of this study will be essential for evaluating the properties of sulfonamide drugs from a drugbank using a variety of analyses in order to repurpose them as novel anticancer drugs. Quantum mechanical calculations will be performed on the optimal docking poses in future work. Keywords: Sulfonamide drugs, Docking, Histone deacetylases, Lipinsk's rule, Binding affinity.

The human ether-a-go-go-related gene (hERG) encodes a voltage-gated potassium channel essential for cardiac action potential repolarization. Drug-induced hERG inhibition can prolong the QT interval, causing severe heart diseases like torsade de pointes and fatal arrhythmias. In pharmaceutical chemistry, early prediction of hERG blockers is crucial to mitigate cardiotoxicity risks, minimizing drug withdrawals and economic losses in discovery. To address this, an interpretable multi-modal molecular representation cross-learning framework (MMRCL) is developed, integrating multi-dimensional molecular fingerprints and molecular graphs to enrich structural features. MMRCL combines a dual-channel message passing neural network (MPNN) for atom- and bond-level structural features with a multi-layer perceptron for molecular fingerprint-based semantics. A multi-head cross-attention mechanism adaptively fuses features across modalities, enabling deep correlation modeling, followed by a fully connected neural network classifier. Extensive evaluation on an internal dataset (12,518 compounds with high-dimensional fingerprints and graph features) and three external test sets demonstrates MMRCL's superior performance compared to seven state-of-the-art baseline models, achieving the best AUC of 0.8895, PRC of 0.9073, and MCC of 0.6146 on the internal set. Interpretability analysis identifies key toxic substructures linked to hERG-blocking activity, aiding structure-activity relationship exploration. Ablation studies further confirm the contributions of multi-modal input and attention-based fusion. MMRCL achieves superior prediction accuracy and generalization, also enhances model interpretability, providing actionable insights for medicinal chemists.

Staphylococcus aureus is a leading cause of both community- and hospital-acquired infections, and the growing prevalence of antimicrobial resistance complicates clinical management worldwide. This study investigated the epidemiology, resistance trends, multidrug resistance (MDR) patterns, and the role of machine learning (ML) in predicting antibiotic susceptibility in Saudi Arabia. A total of 18,003 microbiology reports (2019-2024) were analyzed, identifying 2506 S. aureus isolates. Susceptibility testing included 31 antibiotics representing 11 pharmacological classes. Predictive ML models (Random Forest, Logistic Regression, Gradient Boosting) were trained and evaluated using accuracy, precision, recall, F1-score, and confusion matrices. Wound (24 %) and blood (23 %) were the most frequent sources of S. aureus. High resistance (>70 %) was observed for β-lactams, fluoroquinolones, and macrolides/lincosamides, while glycopeptides, oxazolidinones, and lipopeptides maintained excellent activity (<10 % resistance). MDR occurred in 30 % of isolates, XDR in 0.6 %, and no PDR isolates were detected. Among ML models, Random Forest achieved the best overall performance across most antibiotics, Logistic Regression was optimal for ampicillin, and Gradient Boosting for linezolid. Vancomycin, linezolid, penicillin, and SXT achieved precision and recall above 0.92, demonstrating strong predictive reliability. S. aureus remains a major clinical threat in Saudi Arabia, with high MDR rates but preserved efficacy of last-line antibiotics. This study highlights the value of combining multi-center surveillance with interpretable machine learning approaches to support antimicrobial stewardship, enhance early resistance prediction, and inform data-driven clinical decision-making, particularly in settings where rapid molecular diagnostics may be limited.