Browse Articles
Discover research articles across all indexed journals
Exploring DNA methylation age and the influence of physical performance, and hypertension on frailty in elderly women
Improved survival with high albumin leakage in patients with protein-energy wasting and inflammation on hemodialysis and online hemodiafiltration
How machine learning can help us understand what we have grown in the dish
Nitrogen-doped mesoporous carbon as an efficient metal-free catalyst for biodiesel production
Efficacy of imipenem combined with dimercaptosuccinic acid in a murine sepsis model using Pseudomonas aeruginosa
Novel dual gland GAN architecture improves human protein localization classification using salivary and pituitary gland inspired loss functions
Abstract Cellular classification is essential for understanding biological processes and disease mechanisms. This paper introduces a novel approach that employs two complementary loss functions within a Generative Adversarial Network (GAN) framework for processing images from the Human Protein Atlas dataset. Our method introduces the “Salivary Gland” loss function (SG-Loss), which addresses missing pixel imputation through a unique computational mechanism that models the graded secretion patterns of acinar cells, incorporating multi-scale contextual information to reconstruct incomplete cellular features. This is paired with our innovative “Pituitary Gland” loss function (PG-Loss), which preserves structural integrity through a novel homeostatic regularization approach that adaptively weights pixel relationships based on subcellular compartment boundaries, unlike conventional smoothing techniques. The SG-Loss specifically targets discontinuities in protein expression patterns, while PG-Loss maintains biological plausibility by enforcing organelle-specific constraints learned from annotated training data. Our proposed Dual-Gland GAN demonstrates superior performance with an Inception Score of 9.83 (± 0.31) and MS-SSIM Diversity of 0.187 (± 0.021). The model achieves impressive precision and recall metrics (0.872 and 0.835, respectively), resulting in an F1-score of 0.853. Training stability is reflected in minimal generator and discriminator loss variance (0.028 and 0.032) with convergence achieved in 78 epochs. Comprehensive evaluation shows high quality and diversity scores (0.912 and 0.894), yielding a combined score of 0.903, demonstrating the effectiveness of our biologically inspired approach for cellular image generation and classification. The results also prove the efficiency of the architecture in enhancing the classification results.
Enabling RNA-compatible synthetic receptors through RNA editing
On statistical analysis of topological indices and heat of formation for titanium diboride network
Associated factors and clinical outcomes of feeding intolerance in preterm extremely low birthweight infants
Computational screening identifies selective aldose reductase inhibitors with strong efficacy and limited off target interactions
Abstract Diabetes mellitus is characterized by persistent hyperglycemia that triggers micro-vascular complications in organs such as the eyes and kidneys; a pivotal enzymatic driver is aldose reductase (AR), which reduces glucose to sorbitol. Because existing AR inhibitors often cause off-target toxicity, we implemented an integrative in-silico workflow to discover selective, safer compounds. A library of 4 975 small molecules was docked against AR and, in parallel, against five clinically relevant antitarget proteins or proteins whose unintended inhibition is associated with adverse pharmacological or toxicological effects (CYP2A6, CYP2C9, CYP3A4, SULT1A3 and the pregnane X receptor), retaining 236 ligands whose binding energies to every antitarget were weaker than those of the reference drug tolrestat. These survivors were redocked to five high-resolution human AR crystal structures, and the ten best-scoring ligands underwent 100 ns molecular-dynamics simulations followed by MM-PBSA free-energy calculations to refine affinity estimates and probe complex stability. Ligand 4934, a benzo[a]anthracene–pyrene polyphenol, displayed the strongest predicted affinity for while showing poor affinity for the antitarget panel, outperforming tolrestat by more than 2 kcal mol⁻¹ and adopting a stable plug-like pose that occludes the catalytic pocket through extensive π–π and hydrophobic contacts with Trp111, Phe123 and Lys22. These findings highlight ligand 4934 as a promising scaffold for selective AR inhibition and demonstrate the effectiveness of the stepwise computational strategy in prioritizing lead compounds with reduced off-target liabilities.
Prediction of coal mine water conduction fracture zone height based on integrated learning model
Abstract To enhance the accuracy of predicting the height of water-conducting fracture zones (WCFZ) in coal mines, this study proposes a novel stacked ensemble learning model. The model integrates XGBoost and Support Vector Regression (SVR) as base learners, with CatBoost serving as the meta-learner, forming a two-layer architecture. Key geological and mining features—such as mining height, burial depth, working face length, and lithologic proportion coefficient—are used as input variables to better capture the complex influencing factors. Validation using data from the No. 3 coal seam of Husheng Coal Mine demonstrates that the proposed model achieves a predicted WCFZ height of 50.79 m, closely aligning with the measured value and outperforming empirical formulas (61.4 m), standalone SVR (58.14 m), XGBoost (56.62 m), and FLAC3D simulation (55 m). The model also achieves an R² of 0.98 and RMSE of 2.08, indicating excellent predictive performance. This research is the first to introduce stacked ensemble learning for WCFZ height prediction, overcoming the limitations of single-model and simulation-based methods. The proposed approach offers a more accurate and intelligent tool for mine water hazard assessment and represents a significant advancement in applying machine learning to underground geological engineering.