BMC Med Imaging. 2025 May 7;25(1):155. doi: 10.1186/s12880-025-01683-4.

ABSTRACT

BACKGROUND: Cardiovascular diseases are the number one cause of death globally, making cardiac magnetic resonance image segmentation a popular research topic. Existing schemas relying on manual user interaction or semi-automatic segmentation are infeasible when dealing thousands of cardiac MRI studies. Thus, we proposed a full automatic and robust algorithm for large-scale cardiac MRI segmentation by combining the advantages of deep learning localization and 3D-ASM restriction.

MATERIAL AND METHODS: The proposed method comprises several key techniques: 1) a hybrid network integrating CNNs and Transformer as a encoder with the EFG (Edge feature guidance) module (named as CTr-HNs) to localize the target regions of the cardiac on MRI images, 2) initial shape acquisition by alignment of coarse segmentation contours to the initial surface model of 3D-ASM, 3) refinement of the initial shape to cover all slices of MRI in the short axis by complex transformation. The datasets used are from the UK BioBank and the CAP (Cardiac Atlas Project). In cardiac coarse segmentation experiments on MR images, Dice coefficients (Dice), mean contour distances (MCD), and mean Hausdorff distances (HD95) are used to evaluate segmentation performance. In SPASM experiments, Point-to-surface (P2S) distances, Dice score are compared between automatic results and ground truth.

RESULTS: The CTr-HNs from our proposed method achieves Dice coefficients (Dice), mean contour distances (MCD), and mean Hausdorff distances (HD95) of 0.95, 0.10 and 1.54 for the LV segmentation respectively, 0.88, 0.13 and 1.94 for the LV myocardium segmentation, and 0.91, 0.24 and 3.25 for the RV segmentation. The overall P2S errors from our proposed schema is 1.45 mm. For endocardium and epicardium, the Dice scores are 0.87 and 0.91 respectively.

CONCLUSIONS: Our experimental results show that the proposed schema can automatically analyze large-scale quantification from population cardiac images with robustness and accuracy.

PMID:40335966 | DOI:10.1186/s12880-025-01683-4

BMC Med Imaging. 2025 May 7;25(1):156. doi: 10.1186/s12880-025-01701-5.

ABSTRACT

Medical images occupy the largest part of the existing medical information and dealing with them is challenging not only in terms of management but also in terms of interpretation and analysis. Hence, analyzing, understanding, and classifying them, becomes a very expensive and time-consuming task, especially if performed manually. Deep learning is considered a good solution for image classification, segmentation, and transfer learning tasks since it offers a large number of algorithms to solve such complex problems. PRISMA-ScR guidelines have been followed to conduct the scoping review with the aim of exploring how deep learning is being used to classify a broad spectrum of diseases diagnosed using an X-ray, MRI, or Ultrasound image modality.Findings contribute to the existing research by outlining the characteristics of the adopted datasets and the preprocessing or augmentation techniques applied to them. The authors summarized all relevant studies based on the deep learning models used and the accuracy achieved for classification. Whenever possible, they included details about the hardware and software configurations, as well as the architectural components of the models employed. Moreover, the models that achieved the highest accuracy in disease classification were highlighted, along with their strengths. The authors also discussed the limitations of the current approaches and proposed future directions for medical image classification.

PMID:40335965 | DOI:10.1186/s12880-025-01701-5

BMC Bioinformatics. 2025 May 7;26(1):123. doi: 10.1186/s12859-025-06072-w.

ABSTRACT

The integration of deep learning, particularly AI-Generated Content, with high-quality data derived from ab initio calculations has emerged as a promising avenue for transforming the landscape of scientific research. However, the challenge of designing molecular drugs or materials that incorporate multi-modality prior knowledge remains a critical and complex undertaking. Specifically, achieving a practical molecular design necessitates not only meeting the diversity requirements but also addressing structural and textural constraints with various symmetries outlined by domain experts. In this article, we present an innovative approach to tackle this inverse design problem by formulating it as a multi-modality guidance optimization task. Our proposed solution involves a textural-structure alignment symmetric diffusion framework for the implementation of molecular optimization tasks, namely 3DToMolo. 3DToMolo aims to harmonize diverse modalities including textual description features and graph structural features, aligning them seamlessly to produce molecular structures adhere to specified symmetric structural and textural constraints by experts in the field. Experimental trials across three guidance optimization settings have shown a superior hit optimization performance compared to state-of-the-art methodologies. Moreover, 3DToMolo demonstrates the capability to discover potential novel molecules, incorporating specified target substructures, without the need for prior knowledge. This work not only holds general significance for the advancement of deep learning methodologies but also paves the way for a transformative shift in molecular design strategies. 3DToMolo creates opportunities for a more nuanced and effective exploration of the vast chemical space, opening new frontiers in the development of molecular entities with tailored properties and functionalities.

PMID:40335938 | DOI:10.1186/s12859-025-06072-w

BMC Med. 2025 May 7;23(1):267. doi: 10.1186/s12916-025-04104-z.

ABSTRACT

BACKGROUND: CT images and circulating tumor cells (CTCs) are indispensable for diagnosing the mediastinal lesions by providing radiological and intra-tumoral information. This study aimed to develop and validate a deep multimodal fusion network (DMFN) combining CTCs and CT images for the multi-classification of mediastinal lesions.

METHODS: In this retrospective diagnostic study, we enrolled 1074 patients with 1500 enhanced CT images and 1074 CTCs results between Jan 1, 2020, and Dec 31, 2023. Patients were divided into the training cohort (n = 434), validation cohort (n = 288), and test cohort (n = 352). The DMFN and monomodal convolutional neural network (CNN) models were developed and validated using the CT images and CTCs results. The diagnostic performances of DMFN and monomodal CNN models were based on the Paraffin-embedded pathologies from surgical tissues. The predictive abilities were compared with thoracic resident physicians, attending physicians, and chief physicians by the area under the receiver operating characteristic (ROC) curve, and diagnostic results were visualized in the heatmap.

RESULTS: For binary classification, the predictive performances of DMFN (AUC = 0.941, 95% CI 0.901-0.982) were better than the monomodal CNN model (AUC = 0.710, 95% CI 0.664-0.756). In addition, the DMFN model achieved better predictive performances than the thoracic chief physicians, attending physicians, and resident physicians (P = 0.054, 0.020, 0.016) respectively. For the multiclassification, the DMFN achieved encouraging predictive abilities (AUC = 0.884, 95%CI 0.837-0.931), significantly outperforming the monomodal CNN (AUC = 0.722, 95%CI 0.705-0.739), also better than the chief physicians (AUC = 0.787, 95%CI 0.714-0.862), attending physicians (AUC = 0.632, 95%CI 0.612-0.654), and resident physicians (AUC = 0.541, 95%CI 0.508-0.574).

CONCLUSIONS: This study showed the feasibility and effectiveness of CNN model combing CT images and CTCs levels in predicting the diagnosis of mediastinal lesions. It could serve as a useful method to assist thoracic surgeons in improving diagnostic accuracy and has the potential to make management decisions.

PMID:40335930 | DOI:10.1186/s12916-025-04104-z

Med Biol Eng Comput. 2025 May 8. doi: 10.1007/s11517-025-03365-3. Online ahead of print.

ABSTRACT

Despite the extensive use of biomarkers like PSA, AMACR, and PCA3, prostate cancer (PCa) is still a major clinical challenge, demanding the development of more precise and specific methods for diagnosis. In this study, a deep learning model was applied to identify ten key genes from a pool of 68 common differentially expressed genes in the three transcriptomic datasets. The model demonstrated high performance, with the accuracy of 0.969, R2 of 0.88, and PR-AUC of 0.98. Notably, selected genes have been previously reported as functionally important in various cancers. Among them, SCUBE2 stands out as a novel potential diagnostic biomarker in prostate cancer, showing a strong diagnostic performance in the TCGA dataset with AUC = 0.84, sensitivity = 0.76, and specificity = 0.84. SCUBE2 is a secreted glycoprotein known for its ability to suppress tumor growth, cell migration, and epithelial-mesenchymal transition (EMT) in several cancer types, including gliomas, breast, and colorectal cancers, mainly through its regulation of signaling pathways such as Hedgehog (Shh). Although its role in prostate cancer (PCa) has not been previously explored, its consistent downregulation across multiple PCa datasets in this study suggests it may act as a tumor suppressor, warranting further investigation. Another candidate, SLC16A5, showed moderate performance individually (AUC = 0.62, SP = 0.81, SE = 0.42 in GSE88808), but its combination with SCUBE2 significantly enhanced diagnostic accuracy (combined AUC = 0.76, SE = 0.75, SP = 0.71). SLC16A5 is a monocarboxylate transporter involved in metabolic reprogramming, and prior studies have linked its downregulation to immune infiltration and poor prognosis in PCa. Functional enrichment analysis of the ten identified genes revealed strong involvement of these genes in cancer-related processes, including gap junction assembly, tight junction formation, efflux transporter activity, and pathways such as Hedgehog signaling, leukocyte transendothelial migration, and cell-cell adhesion. Hub gene analysis further confirmed the central roles of identified genes such as CAV1, GJA1, AMACR, and CLDN8, which are well-documented in cancer progression, metastasis, or therapeutic resistance. In summary, this study identifies SCUBE2 as a novel potential diagnostic biomarker for prostate cancer and supports the use of AI-driven gene discovery in identifying key players in tumor biology. The combination of SCUBE2 with SLC16A5 not only enhances diagnostic precision but also opens new avenues for functional and clinical validation, ultimately contributing to the development of more accurate, multi-gene diagnostic panels for PCa.

PMID:40335872 | DOI:10.1007/s11517-025-03365-3

Interdiscip Sci. 2025 May 7. doi: 10.1007/s12539-025-00703-9. Online ahead of print.

ABSTRACT

Enhancers are short DNA fragments capable of significantly increase the frequency of gene transcription. They often exert their effects on targeted genes over long distances, either in cis or in trans configurations. Identifying enhancers poses a challenge due to their variable position and sensitivities. Genetic variants within enhancer regions have been implicated in human diseases, highlighting critical importance of enhancers identification and strength prediction. Here, we develop a two-layer predictor named iEnhancer-GDM to identify enhancers and to predict enhancer strength. To address the challenges posed by the limited size of enhancer training dataset, which could cause issues such as model overfitting and low classification accuracy, we introduce a Wasserstein generative adversarial network (WGAN-GP) to augment the dataset. We employ a dna2vec embedding layer to encode raw DNA sequences into numerical feature representations, and then integrate multi-scale convolutional neural network, bidirectional long short-term memory network and multi-head attention mechanism for feature representation and classification. Our results validate the effectiveness of data augmentation in WGAN-GP. Our model iEnhancer-GDM achieves superior performance on an independent test dataset, and outperforms the existing models with improvements of 2.45% for enhancer identification and 11.5% for enhancer strength prediction by benchmarking against current methods. iEnhancer-GDM advances the precise enhancer identification and strength prediction, thereby helping to understand the functions of enhancers and their associations on genomics.

PMID:40335860 | DOI:10.1007/s12539-025-00703-9

Nat Biotechnol. 2025 May 7. doi: 10.1038/s41587-025-02655-3. Online ahead of print.

ABSTRACT

Naturally existing enzymes have been adapted for a variety of molecular technologies, with enhancements or modifications to the enzymes introduced to improve the desired function; however, it is difficult to engineer variants with enhanced activity while maintaining specificity. Here we engineer the compact Obligate Mobile Element Guided Activity (OMEGA) RNA-guided endonuclease IscB and its guiding RNA (ωRNA) by combining ortholog screening, structure-guided protein domain design and RNA engineering, and deep learning-based structure prediction to generate an improved variant, NovaIscB. We show that the compact NovaIscB achieves up to 40% indel activity (~100-fold improvement over wild-type OgeuIscB) on the human genome with improved specificity relative to existing IscBs. We further show that NovaIscB can be fused with a methyltransferase to create a programmable transcriptional repressor, OMEGAoff, that is compact enough to be packaged in a single adeno-associated virus vector for persistent in vivo gene repression. This study highlights the power of combining natural diversity with protein engineering to design enhanced enzymes for molecular biology applications.

PMID:40335752 | DOI:10.1038/s41587-025-02655-3

Plant Cell Rep. 2025 May 8;44(6):115. doi: 10.1007/s00299-025-03498-7.

ABSTRACT

Using deep learning-based image restoration, we achieved high-resolution 4D imaging with minimal photodamage, revealing distinct localization and suggesting Lifeact-RFP-labeled actin microfilaments play a role in initiating cell plate formation. Phragmoplasts are plant-specific intracellular structures composed of microtubules, actin microfilaments (AFs), membranes, and associated proteins. Importantly, they are involved in the formation and the expansion of cell plates that partition daughter cells during cell division. While previous studies have revealed the important role of cytoskeletal dynamics in the proper functioning of the phragmoplast, the localization and the role of AFs in the initial phase of cell plate formation remain controversial. Here, we used deep learning-based image restoration to achieve high-resolution 4D imaging with minimal laser-induced damage, enabling us to investigate the dynamics of AFs during the initial phase of cell plate formation in transgenic tobacco BY-2 cells labeled with Lifeact-RFP or RFP-ABD2 (actin-binding domain 2). This computational approach overcame the limitation of conventional imaging, namely laser-induced photobleaching and phototoxicity. The restored images indicated that RFP-ABD2-labeled AFs were predominantly localized near the daughter nucleus, whereas Lifeact-RFP-labeled AFs were found not only near the daughter nucleus but also around the initial cell plate. These findings, validated by imaging with a long exposure time, highlight distinct localization patterns between the two AF probes and suggest that Lifeact-RFP-labeled AFs play a role in initiating cell plate formation.

PMID:40335746 | DOI:10.1007/s00299-025-03498-7

Nature. 2025 May 7. doi: 10.1038/s41586-025-08985-1. Online ahead of print.

ABSTRACT

The information-processing capability of the brain's cellular network depends on the physical wiring pattern between neurons and their molecular and functional characteristics. Mapping neurons and resolving their individual synaptic connections can be achieved by volumetric imaging at nanoscale resolution1,2 with dense cellular labelling. Light microscopy is uniquely positioned to visualize specific molecules, but dense, synapse-level circuit reconstruction by light microscopy has been out of reach, owing to limitations in resolution, contrast and volumetric imaging capability. Here we describe light-microscopy-based connectomics (LICONN). We integrated specifically engineered hydrogel embedding and expansion with comprehensive deep-learning-based segmentation and analysis of connectivity, thereby directly incorporating molecular information into synapse-level reconstructions of brain tissue. LICONN will allow synapse-level phenotyping of brain tissue in biological experiments in a readily adoptable manner.

PMID:40335689 | DOI:10.1038/s41586-025-08985-1

Int J Pediatr Otorhinolaryngol. 2025 May 4;194:112369. doi: 10.1016/j.ijporl.2025.112369. Online ahead of print.

ABSTRACT

BACKGROUND: Artificial Intelligence (AI) and machine learning (ML) have transformative potential in enhancing diagnostics, treatment planning, and patient management. However, their application in pediatric otolaryngology remains limited as the unique physiological and developmental characteristics of children require tailored AI applications, highlighting a gap in knowledge.

PURPOSE: To provide a narrative review of current literature on the application of AI in pediatric otolaryngology, highlighting knowledge gaps, associated challenges and future directions.

RESULTS: ML models have demonstrated efficacy in diagnosing conditions such as otitis media, adenoid hypertrophy, and pediatric obstructive sleep apnea through deep learning-based image analysis and predictive modeling. AI systems also show potential in surgical settings such as landmark identification during otologic surgery and prediction of middle ear effusion during tympanostomy tube placement. Telemedicine solutions and large language models have shown potential to improve accessibility to care and patient education. The principal challenges include flawed generalization of adult training data and the relative lack of pediatric data.

CONCLUSIONS: AI holds significant promise in pediatric otolaryngology. However, its widespread clinical integration requires addressing algorithmic bias, enhancing model interpretability, and ensuring robust validation across pediatric population. Future research should prioritize federated learning, developmental trajectory modeling, and psychosocial integration to create patient-centered solutions.

PMID:40334638 | DOI:10.1016/j.ijporl.2025.112369

Artif Intell Med. 2025 Apr 29;166:103116. doi: 10.1016/j.artmed.2025.103116. Online ahead of print.

ABSTRACT

Multiple Sclerosis (MS) is a chronic neuroinflammatory disease of the Central Nervous System (CNS) in which the body's immune system attacks and destroys the myelin sheath that protects nerve fibers, leading to a wide range of debilitating symptoms and causing disruption of axonal signal transmission. Accurate prediction, diagnosis, monitoring and treatment (PDMT) of MS are essential to improve patient outcomes. Recent advances in neuroimaging technologies, particularly electroencephalography (EEG), combined with machine learning (ML) techniques - including Deep Learning (DL) models - offer promising avenues for enhancing MS management. This systematic review synthesizes existing research on the application of ML and DL models to EEG data for MS. It explores the methodologies used, with a focus on DL architectures such as Convolutional Neural Networks (CNNs) and hybrid models, and highlights recent advancements in ML techniques and EEG technologies that have significantly improved MS diagnosis and monitoring. The review addresses the challenges and potential biases in using ML-based EEG analysis for MS. Strategies to mitigate these challenges, including advanced preprocessing techniques, diverse training datasets, cross-validation methods, and explainable Artificial Intelligence (AI), are discussed. Finally, the paper outlines potential future applications and trends in ML for MS management. This review underscores the transformative potential of ML-enhanced EEG analysis in improving MS management, providing insights into future research directions to overcome existing limitations and further improve clinical practice.

PMID:40334524 | DOI:10.1016/j.artmed.2025.103116

Food Chem. 2025 Apr 28;486:144311. doi: 10.1016/j.foodchem.2025.144311. Online ahead of print.

ABSTRACT

Resistant starch (RS) is a vital dietary component with notable health benefits, but tradition quantification methods are labor-intensive, costly, and unsuitable for large-scale applications. This study introduced an innovative data-driven framework integrating Near-Infrared (NIR) spectroscopy with Convolutional Neural Networks (CNN) and data augmentation to achieve rapid, cost-effective RS prediction. Achieving exceptional accuracy (Rp2 = 0.992), the CNN model outperformed traditional methods like Partial Least Squares Regression (PLSR) and Support Vector Machine Regression (SVMR). To overcome the "black-box" limitation of deep learning, SHapley Additive exPlanations (SHAP) were innovatively employed, pinpointing critical wavelengths (2000-2500 nm), significantly narrowing the spectral range while providing meaningful insights into the contribution of specific wavelengths to RS prediction. This optimized spectral enhanced data acquisition efficiency, reduces analytical costs, and simplifies operational complexity, establishing a practical and scalable solution for deploying NIR spectroscopy in food quality assessment and production-line applications.

PMID:40334489 | DOI:10.1016/j.foodchem.2025.144311

Comput Methods Programs Biomed. 2025 May 6;267:108830. doi: 10.1016/j.cmpb.2025.108830. Online ahead of print.

ABSTRACT

OBJECTIVES: Achieving efficient analysis of multiple pathological indicators has great significance for breast cancer prognosis and therapeutic decision-making. In this study, we aim to explore a deep multi-task learning (MTL) framework for collaborative prediction of histological grade and proliferation marker (Ki-67) status in breast cancer using multi-phase dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI).

METHODS: In the novel design of hybrid multi-task architecture (HMT-Net), co-representative features are explicitly distilled using a feature extraction backbone. A customized prediction network is then introduced to perform soft-parameter sharing between two correlated tasks. Specifically, task-common and task-specific knowledge is transmitted into tower layers for informative interactions. Furthermore, low-level feature maps containing tumor edges and texture details are recaptured by a hard-parameter sharing branch, which are then incorporated into the tower layer for each subtask. Finally, the probabilities of two histological indicators, predicted in the multi-phase DCE-MRI, are separately fused using a decision-level fusion strategy.

RESULTS: Experimental results demonstrate that the proposed HMT-Net achieves optimal discriminative performance over other recent MTL architectures and deep models based on single image series, with the area under the receiver operating characteristic curve of 0.908 for tumor grade and 0.694 for Ki-67 status.

CONCLUSIONS: Benefiting from the innovative HMT-Net, our proposed method elucidates its strong robustness and flexibility in the collaborative prediction task of breast biomarkers. Multi-phase DCE-MRI is expected to contribute valuable dynamic information for breast cancer pathological assessment in a non-invasive manner.

PMID:40334302 | DOI:10.1016/j.cmpb.2025.108830

Comput Biol Med. 2025 May 6;192(Pt B):110242. doi: 10.1016/j.compbiomed.2025.110242. Online ahead of print.

ABSTRACT

Brain tumours continue to be a primary cause of worldwide death, highlighting the critical need for effective and accurate diagnostic tools. This article presents MK-YOLOv8, an innovative lightweight deep learning framework developed for the real-time detection and categorization of brain tumours from MRI images. Based on the YOLOv8 architecture, the proposed model incorporates Ghost Convolution, the C3Ghost module, and the SPPELAN module to improve feature extraction and substantially decrease computational complexity. An x-small object detection layer has been added, supporting precise detection of small and x-small tumours, which is crucial for early diagnosis. Trained on the Figshare Brain Tumour (FBT) dataset comprising (3,064) MRI images, MK-YOLOv8 achieved a mean Average Precision (mAP) of 99.1% at IoU (0.50) and 88.4% at IoU (0.50-0.95), outperforming YOLOv8 (98% and 78.8%, respectively). Glioma recall improved by 26%, underscoring the enhanced sensitivity to challenging tumour types. With a computational footprint of only 96.9 GFLOPs (representing 37.5% of YOYOLOv8x'sFLOPs) and utilizing 12.6 million parameters, a mere 18.5% of YOYOLOv8's parameters, MK-YOLOv8 delivers high efficiency with reduced resource demands. Also, it trained on the Br35H dataset (801 images) to guarantee the model's robustness and generalization; it achieved a mAP of 98.6% at IoU (0.50). The suggested model operates at 62 frames per second (FPS) and is suited for real-time clinical processes. These developments establish MK-YOLOv8 as an innovative framework, overcoming challenges in tiny tumour identification and providing a generalizable, adaptable, and precise detection approach for brain tumour diagnostics in clinical settings.

PMID:40334297 | DOI:10.1016/j.compbiomed.2025.110242

PLoS One. 2025 May 7;20(5):e0322540. doi: 10.1371/journal.pone.0322540. eCollection 2025.

ABSTRACT

Knee osteoarthritis (KOA) is a leading cause of disability globally. Early and accurate diagnosis is paramount in preventing its progression and improving patients' quality of life. However, the inconsistency in radiologists' expertise and the onset of visual fatigue during prolonged image analysis often compromise diagnostic accuracy, highlighting the need for automated diagnostic solutions. In this study, we present an advanced deep learning model, OA-HybridCNN (OHC), which integrates ResNet and DenseNet architectures. This integration effectively addresses the gradient vanishing issue in DenseNet and augments prediction accuracy. To evaluate its performance, we conducted a thorough comparison with other deep learning models using five-fold cross-validation and external tests. The OHC model outperformed its counterparts across all performance metrics. In external testing, OHC exhibited an accuracy of 91.77%, precision of 92.34%, and recall of 91.36%. During the five-fold cross-validation, its average AUC and ACC were 86.34% and 87.42%, respectively. Deep learning, particularly exemplified by the OHC model, has greatly improved the efficiency and accuracy of KOA imaging diagnosis. The adoption of such technologies not only alleviates the burden on radiologists but also significantly enhances diagnostic precision.

PMID:40334259 | DOI:10.1371/journal.pone.0322540

PLoS One. 2025 May 7;20(5):e0322978. doi: 10.1371/journal.pone.0322978. eCollection 2025.

ABSTRACT

BACKGROUND: Predicting protein-DNA binding sites in vivo is a challenging but urgent task in many fields such as drug design and development. Most promoters contain many transcription factor (TF) binding sites, yet only a few have been identified through time-consuming biochemical experiments. To address this challenge, numerous computational approaches have been proposed to predict TF binding sites from DNA sequences. However, current deep learning methods often face issues such as gradient vanishing as the model depth increases, leading to suboptimal feature extraction.

RESULTS: We propose a model called CBR-KAN (where C represents Convolutional Neural Network (CNN), B represents Bidirectional Long Short Term Memory (BiLSTM), and R represents Residual Mechanism) to predict transcription factor binding sites. Specifically, we designed a multi-scale convolution module (ConvBlock1, 2, 3) combined with BiLSTM network, introduced KAN network to replace traditional multilayer perceptron, and promoted model optimization through residual connections. Testing on 50 common ChIP seq benchmark datasets shows that CBR-KAN outperforms other state-of-the-art methods such as DeepBind, DanQ, DeepD2V, and DeepSEA in predicting TF binding sites.

CONCLUSIONS: The CBR-KAN model significantly improves prediction accuracy for transcription factor binding sites by effectively integrating multiple neural network architectures and mechanisms. This approach not only enhances feature extraction but also stabilizes training and boosts generalization capabilities. The promising results on multiple key performance indicators demonstrate the potential of CBR-KAN in bioinformatics applications.

PMID:40334196 | DOI:10.1371/journal.pone.0322978

PLoS One. 2025 May 7;20(5):e0319476. doi: 10.1371/journal.pone.0319476. eCollection 2025.

ABSTRACT

In the context of consumer economics, it is imperative to consider the functionality of sports venues based on customer demand. However, traditional survey methods are time-consuming, resource-intensive, and coverage-limited. This paper conducted sentiment mining based on Internet big data, deep learning, topic analysis, and social network analysis to capture the satisfaction of consumers and its influencing factors. Findings indicate that activity, courses, and facilities are core factors driving positive comments. Coaches, environment, and activities are key determinants influencing neutral evaluations. Attitude, integrity, and qualifications can trigger negative reviews. The findings offer insights into developing consumer-friendly service for sports venues.

PMID:40333946 | DOI:10.1371/journal.pone.0319476

PLoS One. 2025 May 7;20(5):e0322936. doi: 10.1371/journal.pone.0322936. eCollection 2025.

ABSTRACT

Identifying relevant plant parts is one of the most significant tasks in the pharmaceutical industry. Correct identification minimizes the risk of mis-identification, which might have unfavorable effects, and it ensures that plants are used medicinally. Traditional methods for plant part identification are often time-consuming and require specific expertise. This study proposed a Depth-wise Separable Convolutional Neural Network (DWS-CNN) to enhance the accuracy of medicinal plant part identification. Furthermore, we incorporated the tuned pre-trained models such as VGG16, Res Net-50, and Inception V3 which are designed by Standard convolutional neural network (S-CNN) for comparative purposes. We trained variants of the Standard convolutional neural network (S-CNN) model with high-resolution images of medicinal plant leaves which contains 15,100 leaf images. The study used supervised learning by which leaf images are used as an identity for the other parts of the plants. We used transfer learning to tune training and model parameters. Experimental results showed that our DWS-CNN model achieved better performance compared to S-CNN models, with an accuracy of 99.84% for training data, 99.44% for F1-score and 99.44% for testing data, which improves in both accuracy and training speed. The presence of depth-wise separable convolution and batch normalization at the fully connected layer of the model made the model achieved a good classification performance.

PMID:40333881 | DOI:10.1371/journal.pone.0322936

Molecules. 2025 Apr 9;30(8):1683. doi: 10.3390/molecules30081683.

ABSTRACT

While natural products and derivatives have been crucial in drug discovery, the current databases are limited to known compounds. There is a need for tools that can automatically generate and assess novel derivatives of natural products to enhance early-stage drug discovery. We present DerivaPredict (v1.0), a user-friendly tool that generates novel natural product derivatives through chemical and metabolic transformations. It predicts binding affinities using pretrained deep learning models and assesses drug-likeness via ADMET profiling. DerivaPredict is freely accessible with a source code on GitHub.

PMID:40333643 | DOI:10.3390/molecules30081683

Invest Ophthalmol Vis Sci. 2025 May 1;66(5):17. doi: 10.1167/iovs.66.5.17.

ABSTRACT

PURPOSE: Amid efforts to understand spaceflight associated neuro-ocular syndrome (SANS), uncovering the role of the choroid in its etiology is challenged by the accuracy of image segmentation. The present study extended deep learning-based choroid quantification from optical coherence tomography (OCT) to the characterization of pulsatile and topological changes in the macular plane and investigated changes in response to prolonged microgravity exposure.

METHODS: We analyzed OCT macular videos and volumes acquired from astronauts before, during, and after long-duration spaceflight. Deep learning models were fine-tuned for choroid segmentation and combined with further image processing toward vascularity quantification. Statistical analysis was performed to determine changes in time-dependent and spatially averaged variables from preflight baseline.

RESULTS: For 12 astronauts with a mean age of 47 ± 9 years, there were significant increases in choroid thickness and luminal area (LA) averaged over OCT macular video segments. There was also a significant increase in pulsatile LA. For a subgroup of six astronauts for whom inflight imaging was available, choroid volume, luminal volume, and the choroidal vascularity index over the macular region all increased significantly during spaceflight.

CONCLUSIONS: The findings suggest that localized choroid pulsatile changes occur following prolonged microgravity exposure. They show that the choroid vessels expand in a manner similar to the choroid layer across the macular region during spaceflight, with a relative increase in the space they occupy. The methods developed provide new tools and avenues for studying and establishing effective countermeasures to risks associated with long-duration spaceflight.

PMID:40332907 | DOI:10.1167/iovs.66.5.17