Sci Rep. 2025 Apr 19;15(1):13614. doi: 10.1038/s41598-025-97883-7.

ABSTRACT

Early-stage glaucoma diagnosis is crucial for preventing permanent structural damage and irreversible vision loss. While various machine-learning approaches have been developed for glaucoma diagnosis, only a few specifically address early-stage detection. Moreover, existing early-stage detection methods rely on unimodal information and exclude subjects with high myopia, which contradicts clinical practice and overlooks the adverse effect of high myopia on prediction performance. To develop a clinically practical tool, this study proposes a deep-learning-based, end-to-end early-stage glaucoma detection framework designed for a cohort likely with high myopia. This framework uniquely integrates functional information from visual field (VF) parameters of standard automated perimetry (SAP) and Pulsar perimetry (PP) with structural information derived from optical coherence tomography (OCT) thickness maps. It comprises three key components: 3D OCT ganglion cell complex (GCC) layer segmentation, thickness map generation, and early-stage glaucoma detection. Evaluated on 394 subjects using five-time, 10-fold cross-validation, the proposed system achieved a mean area under the receiver operating characteristic (ROC) curve of 0.887 ± 0.006, outperforming the Asaoka method without transfer learning and nine models based solely on VF parameters. Results further confirmed that incorporating SAP and PP parameters was essential for mitigating the adverse effects of high myopia.

PMID:40253455 | DOI:10.1038/s41598-025-97883-7

Sci Rep. 2025 Apr 19;15(1):13605. doi: 10.1038/s41598-025-96827-5.

ABSTRACT

The most dangerous form of cancer is breast cancer. This disease is life-threatening because of its aggressive nature and high death rates. Therefore, early discovery increases the patient's survival. Mammography has recently been recommended as diagnosis technique. Mammography, is expensive and exposure the person to radioactivity. Thermography is a less invasive and affordable technique that is becoming increasingly popular. Considering this, a recent deep learning-based breast cancer diagnosis approach is executed by thermography images. Initially, thermography images are chosen from online sources. The collected thermography images are being preprocessed by Contrast Limited Adaptive Histogram Equalization (CLAHE) and contrasting enhancement methods to improve the quality and brightness of the images. Then, the optimal binary thresholding is done to segment the preprocessed images, where optimized the thresholding value using developed Rock Hyraxes Dandelion Algorithm Optimization (RHDAO). A newly implemented deep learning structure StackVRDNet is used for further processing breast cancer diagnosing using thermography images. The segmented images are fed to the StackVRDNet framework, where the Visual Geometry Group (VGG16), Resnet, and DenseNet are employed for constructing this model. The relevant features are extracted usingVGG16, Resnet, and DenseNet, and then obtain stacked weighted feature pool from the extracted features, where the weight optimization is done with the help of RHDAO. The final classification is performed using StackVRDNet, and the diagnosis results are obtained at the final layer of VGG16, Resnet, and DenseNet. A higher scoring method is rated for ensuring final diagnosis results. Here, the parameters present within the VGG16, Resnet, and DenseNet are optimized via the RHDAO to improve the diagnosis results. The simulation outcomes of the developed model achieve 97.05% and 86.86% in terms of accuracy and precision, respectively. The effectiveness of the designed methd is being analyzed via the conventional breast cancer diagnosis models in terms of various performance measures.

PMID:40253418 | DOI:10.1038/s41598-025-96827-5

Sci Rep. 2025 Apr 19;15(1):13565. doi: 10.1038/s41598-025-95079-7.

ABSTRACT

Increasing the number of organ donations after circulatory death (DCD) has been identified as one of the most important ways of addressing the ongoing organ shortage. While recent technological advances in organ transplantation have increased their success rate, a substantial challenge in increasing the number of DCD donations resides in the uncertainty regarding the timing of cardiac death after terminal extubation, impacting the risk of prolonged ischemic organ injury, and negatively affecting post-transplant outcomes. In this study, we trained and externally validated an ODE-RNN model, which combines recurrent neural network with neural ordinary equations and excels in processing irregularly-sampled time series data. The model is designed to predict time-to-death following terminal extubation in the intensive care unit (ICU) using the history of clinical observations. Our model was trained on a cohort of 3,238 patients from Yale New Haven Hospital, and validated on an external cohort of 1,908 patients from six hospitals across Connecticut. The model achieved accuracies of [Formula: see text] and [Formula: see text] for predicting whether death would occur in the first 30 and 60 minutes, respectively, with a calibration error of [Formula: see text]. Heart rate, respiratory rate, mean arterial blood pressure (MAP), oxygen saturation (SpO2), and Glasgow Coma Scale (GCS) scores were identified as the most important predictors. Surpassing existing clinical scores, our model sets the stage for reduced organ acquisition costs and improved post-transplant outcomes.

PMID:40253393 | DOI:10.1038/s41598-025-95079-7

Acad Radiol. 2025 Apr 18:S1076-6332(25)00304-6. doi: 10.1016/j.acra.2025.04.005. Online ahead of print.

ABSTRACT

RATIONALE AND OBJECTIVES: The research aims to examine how CT-derived habitat radiomics can be used to predict lymphovascular invasion (LVI) in patients with T1-stage lung adenocarcinoma (LUAD), and compare its effectiveness to traditional radiomics and deep learning (DL) models.

MATERIALS AND METHODS: We retrospectively analyzed 349 T1-stage LUAD patients from three centers from January 2021 to March 2024. The K-means algorithm was utilized to cluster CT images and apparent diffusion coefficient maps. Following features selection, we constructed three types of models, namely radiomics, habitat, and DL to identify patients with LVI. The evaluation of all models was conducted by employing the area under the receiver operating characteristic curve (AUC), calibration curves and decision curve analysis.

RESULTS: 349 eligible patients were divided into an internal training set of 210 and an external test set of 139. We identified four distinct habitats, with the AUC for the overall habitat area outperforming that of the four sub-areas. Within the test set, the habitat model reached a higher AUC of 0.941 in contrast to the radiomics model at 0.918 and the deep learning model at 0.896.

CONCLUSION: CT-based habitat radiomics shows promise in predicting LVI in T1-stage LUAD patients, with the habitat signature demonstrating superior performance and significant advantages in identifying patients who are LVI-positive.

PMID:40253221 | DOI:10.1016/j.acra.2025.04.005

Food Res Int. 2025 May;209:116285. doi: 10.1016/j.foodres.2025.116285. Epub 2025 Mar 17.

ABSTRACT

The growing concern over food safety, driven by threats such as food contaminations and adulterations has prompted the adoption of advanced technologies like electronic nose (e-nose) and hyperspectral imaging (HSI), which are increasingly enhanced by machine learning innovations. This paper aims to provide a comprehensive review on food safety, by combining insights from both e-nose and HSI technologies alongside machine learning algorithms. First, the basic principles of e-nose, HSI, and machine learning, with particular emphasis on artificial neural network (ANN) and deep learning (DL) are briefly discussed. The review then examines how machine learning enhances the performance of e-nose and HSI, followed by an exploration of recent applications in detecting food hazards, including drug residues, microbial contaminants, pesticide residues, toxins, and adulterants. Subsequently, key limitations encountered in the applications of machine learning, e-nose and HSI, along with future perspectives on the potential advancements of these technologies are highlighted. E-nose and HSI technologies have shown their great potential for applications in food safety assessment through machine learning assistance. Despite this, their use is primarily limited to laboratory environments, restricting their real-world applications. Additionally, the lack of standardized protocols hampers their acceptance and the reproducibility of tests in food safety assessments. Thus, further research is essential to address these limitations and enhance the effectiveness of e-nose and HSI technologies in practical applications. Ultimately, this paper offers a detailed understanding of both technologies, highlighting the pivotal role of machine learning and presenting insights into their innovative applications within food safety evaluation.

PMID:40253192 | DOI:10.1016/j.foodres.2025.116285

Food Res Int. 2025 May;209:116267. doi: 10.1016/j.foodres.2025.116267. Epub 2025 Mar 25.

ABSTRACT

The increasing prevalence of muscle aging, exacerbated by an aging population, poses a significant threat to public health, necessitating the development of more effective interventions. This study primarily aimed to elucidate the mechanism by which bovine bone collagen facilitates muscle differentiation and regeneration. Initially, peptide sequences within bovine bone collagen hydrolysate were identified using peptidomics. Molecular docking and dynamics simulations subsequently demonstrated that the peptide AGPPGPPGPAGK could form a stable complex with Myomaker, suggesting its potential to regulate myoblast differentiation by targeting Myomaker. The physicochemical properties of AGPPGPPGPAGK were predicted using various deep learning tools, providing insight into its functional capabilities. Further molecular and cellular experiments confirmed that the peptide could enhance myoblast differentiation by regulating energy metabolism. Transcriptome analysis further supported these findings, revealing that the peptide modulated energy metabolism during myoblast differentiation. Finally, a combined bioinformatic and transcriptomic analysis indicated a potential regulatory role of Hrh1 in energy metabolism during cell differentiation, a finding that warrants further investigation.

PMID:40253143 | DOI:10.1016/j.foodres.2025.116267

Drug Discov Today. 2025 Apr 17:104360. doi: 10.1016/j.drudis.2025.104360. Online ahead of print.

ABSTRACT

Artificial intelligence (AI) is reshaping preclinical drug research offering innovative alternatives to traditional animal testing. Advanced techniques, including machine learning (ML), deep learning (DL), AI-powered digital twins (DTs), and AI-enhanced organ-on-a-chip (OoC) platforms, enable precise simulations of complex biological systems. AI plays a critical role in overcoming the limitations of DTs and OoC, improving their predictive power and scalability. These technologies facilitate early-stage, reliable evaluations of drug safety and efficacy, addressing ethical concerns, reducing costs, and accelerating drug development while adhering to the 3Rs principle (Replace, Reduce, Refine). By integrating AI with these advanced models, preclinical research can achieve greater accuracy and efficiency in drug discovery. This review examines the transformative impact of AI in preclinical research, highlighting its advancements, challenges, and the critical steps needed to establish AI as a cornerstone of ethical and efficient drug discovery.

PMID:40252989 | DOI:10.1016/j.drudis.2025.104360

Sci Data. 2025 Apr 19;12(1):665. doi: 10.1038/s41597-025-04973-2.

ABSTRACT

Tackling the current global biodiversity crisis requires large-scale spatially accurate biodiversity data to rapidly assess knowledge gaps and set conservation priorities. Obtaining such data is often challenging because surveying biodiversity across broad spatial scales requires massive logistical and economic efforts. Here, we provide high-resolution (0.81 to 81 km2, depending on species ecology) habitat suitability raster maps for all 225 widespread breeding bird species in Italy. Maps were generated by means of species distribution models based on ~2.5 million spatially accurate (≤1 km-scale) and expert-validated occurrence records. Occurrence data were collected during the breeding seasons 2010-2016 by over 3000 skilled observers, mostly through the Ornitho.it web platform, with the aim of realizing the second Atlas of Breeding Birds in Italy, released in 2022. These raster maps will be useful to ecologists, conservation scientists and practitioners for investigating broad spatial patterns in avian diversity and identifying conservation priorities. We discuss potential applications of this dataset for inferring the composition of ecological communities and species distributions at the Italian scale.

PMID:40253483 | DOI:10.1038/s41597-025-04973-2

Nat Commun. 2025 Apr 19;16(1):3718. doi: 10.1038/s41467-025-58982-1.

ABSTRACT

Tissue-resident immune cells such as innate lymphoid cells (ILC) are known to reside in the parenchymal compartments of tissues and modulate local immune protection. Here we use intravascular cell labeling, parabiosis and multiplex 3D imaging to identify a population of group 3 ILCs in mice that are present within the intravascular space of lung blood vessels (vILC3). vILC3s are distributed broadly in alveolar capillary beds from which inhaled pathogens enter the lung parenchyma. By contrast, conventional ILC3s in tissue parenchyma are enriched in lymphoid clusters in proximity to large veins. In a mouse model of pneumonia, Pseudomonas aeruginosa infection results in rapid vILC3 expansion and production of chemokines including CCL4. Blocking CCL4 in vivo attenuates neutrophil recruitment to the lung at the early stage of infection, resulting in prolonged inflammation and delayed bacterial clearance. Our findings thus define the intravascular space as a site of ILC residence in mice, and reveal a unique immune cell population that interfaces with tissue alarmins and the circulating immune system for timely host defense.

PMID:40253407 | DOI:10.1038/s41467-025-58982-1

Virol J. 2025 Apr 19;22(1):107. doi: 10.1186/s12985-025-02727-5.

ABSTRACT

To be successful, a virus must maintain high between-host transmissibility while also effectively adapting within hosts. The impact of these potentially conflicting demands on viral genetic diversity and adaptation remains largely unexplored. These modes of adaptation can induce uncorrelated selection, bring mutations that enhance certain fitness aspects at the expense of others to high freqency, and contribute to the maintenance of genetic variation. The vast wealth of SARS-CoV-2 genetic data gathered from within and across hosts offers an unparalleled opportunity to test the above hypothesis. By analyzing a large set of SARS-CoV-2 sequences (~ 2 million) collected from early 2020 to mid-2021, we found that high frequency mutations within hosts are sometimes detrimental during between-host transmission. This highlights potential inverse selection pressures within- versus between-hosts. We also identified a group of nonsynonymous changes likely maintained by pleiotropy, as their frequencies are significantly higher than neutral expectation, yet they have never experienced clonal expansion. Analyzing one such mutation, spike M1237I, reveals that spike I1237 boosts viral assembly but reduces in vitro transmission, highlighting its pleiotropic effect. Though they make up about 2% of total changes, these types of variants represent 37% of SARS-CoV-2 genetic diversity. These mutations are notably prevalent in the Omicron variant from late 2021, hinting that pleiotropy may promote positive epistasis and new successful variants. Estimates of viral population dynamics, such as population sizes and transmission bottlenecks, assume neutrality of within-host variation. Our demonstration that these changes may affect fitness calls into question the robustness of these estimates.

PMID:40253323 | DOI:10.1186/s12985-025-02727-5

Brain Stimul. 2025 Apr 17:S1935-861X(25)00090-7. doi: 10.1016/j.brs.2025.04.009. Online ahead of print.

NO ABSTRACT

PMID:40252968 | DOI:10.1016/j.brs.2025.04.009

N Biotechnol. 2025 Apr 17:S1871-6784(25)00041-X. doi: 10.1016/j.nbt.2025.04.003. Online ahead of print.

ABSTRACT

Kirsten rat sarcoma viral oncogene homolog (KRAS) is the most mutated oncogene in human cancers, found in approximately 30% of tumors. These mutations primarily consist of single-base missense alterations in codon G12. While extensive efforts have focused on developing allele-specific inhibitors for KRAS mutations, mutation-specific antibodies (Abs) remain largely unexplored, with only a few research-use-only catalog Abs available. In this study, we employed the proprietary Epivolve technology to develop site-directed monoclonal Abs (mAbs) that target KRAS oncogenic driver mutation KRAS G12D. These site-directed mAbs demonstrate high binding affinity, with equilibrium dissociation constants (KD) in the nanomolar range, showing over 1,000-fold greater affinity for KRAS G12D compared to wild-type KRAS. Western blot analyses using both purified KRAS protein variants and tumor cell lines harboring G12D mutations confirmed the high specificity of these mAbs. Furthermore, immunocytochemistry revealed co-localization of the site-directed mAbs with endogenously expressed KRAS in cancer cells bearing G12D mutations. The validated high affinity and specificity of these site-directed mAbs highlight their potential for diagnostic applications and therapeutic development targeting KRAS driver mutations.

PMID:40252917 | DOI:10.1016/j.nbt.2025.04.003

J Insect Physiol. 2025 Apr 17:104813. doi: 10.1016/j.jinsphys.2025.104813. Online ahead of print.

ABSTRACT

Natural selection has favoured the incorporation of ions, including transition metals, in materials of various biological structures susceptible to mechanical fracture to enhance their failure and wear resistance. With regards to insects, only a few taxa have been investigated. The objective of this study was to analyse the biomechanical properties of the ovipositor in a damselfly Calopteryx splendens Harris, 1780 (Odonata, Zygoptera, Calopterygidae) through nanoindentation and to ascertain the elemental composition gradient within the cuticle using energy-dispersive X-ray spectroscopy. This research represents the first report indicating that the damselfly ovipositor exhibits a gradient in the mechanical properties of the cuticle, with Young's modulus ranging from approximately 3.0 to 7.0 GPa and hardness from 0.1 to 0.3 GPa. These properties are shown to highly correlate with the contents of copper and magnesium, both of which increase in the distal direction. The results also suggests that the mechanical properties of the cuticle are significantly influenced by the degree of sclerotization revealed by confocal laser scanning microscopy. These findings propose that the material properties of the ovipositor cuticle in C. splendens may have adapted to enhance piercing capability and to reduce the risk of structural failure during insertion of eggs in plant substrates.

PMID:40252915 | DOI:10.1016/j.jinsphys.2025.104813

Cell Host Microbe. 2025 Apr 15:S1931-3128(25)00126-X. doi: 10.1016/j.chom.2025.03.015. Online ahead of print.

ABSTRACT

The gut microbiome plays a crucial role in modulating human immunity. Previously, we reported that antibiotic-induced microbiome perturbation affects influenza vaccine responses, depending on pre-existing immunity levels. Here, we employed a systems biology approach to analyze the impact of antibiotic administration on both primary and secondary immune responses to the rabies vaccine in humans. Antibiotic administration reduced the gut bacterial load, with a long-lasting reduction in commensal diversity. This alteration was associated with reduced rabies-specific humoral responses. Multi-omics profiling revealed that antibiotic administration induced (1) an enhanced pro-inflammatory signature early after vaccination, (2) a shift in the balance of vaccine-specific T-helper 1 (Th1) to T-follicular-helper response toward Th1 phenotype, and (3) profound alterations in metabolites, particularly in secondary bile acids in the blood. By integrating multi-omics datasets, we generated a multiscale, multi-response network that revealed key regulatory nodes, including the microbiota, secondary bile acids, and humoral immunity to vaccination.

PMID:40252648 | DOI:10.1016/j.chom.2025.03.015

Bioorg Chem. 2025 Apr 9;160:108454. doi: 10.1016/j.bioorg.2025.108454. Online ahead of print.

ABSTRACT

Altered histone deacetylase 6 (HDAC6) expression and function have been linked to cancer progression, positioning it as a promising therapeutic target for cancer treatment. Herein, we introduce HDAC6 inhibitors based on the tetrahydro-β-carboline scaffold, with compound 18d exhibiting the strongest HDAC6 inhibitory potency, achieving an IC50 of 1.3 nM. Compound 18d exhibited significant growth inhibitory activity against an NCI panel of 60 human cancer cell lines with a minimal cytotoxic effect on non-tumor cells. In vitro mechanistic investigations were conducted in HCT-116 colorectal cancer cells where the capability of 18d to enhance the acetylation of α-tubulin (HDAC6 substrate) rather than nuclear H3 histone (HDAC1 substrate) confirmed selective inhibition of HDAC6 subtype. Additionally, compound 18d was observed to suppress the S phase and promote accumulation in the apoptotic sub-G1 phase, potentially through increasing cleaved caspase 3 and reducing Bcl-2 levels in HCT-116 cells. A wound healing assay also elicited the ability of 18d to hinder cell migration. Notably, 18d could suppress the phosphorylation of extracellular signal-regulated kinase (ERK)1/2, a crucial signaling pathway implicated in cancer cell proliferation, migration and apoptosis. Moreover, downregulation of the critical immune checkpoint protein programmed death-ligand 1 (PD-L1) revealed a potential role of 18d in augmenting immune response towards tumor cells. In summary, these findings highlight 18d's dual role in direct tumor growth suppression and immune system sensitization, highlighting a broader cancer therapeutic potential beyond conventional HDAC inhibition.

PMID:40252366 | DOI:10.1016/j.bioorg.2025.108454

Neurotherapeutics. 2025 Apr 18:e00581. doi: 10.1016/j.neurot.2025.e00581. Online ahead of print.

ABSTRACT

Cognitive impairment is common in people with multiple sclerosis (PwMS). There is an urgent need to identify/develop novel therapies that can help cognitive function in MS. Insulin is critical for helping with regulation of multiple CNS functions, including learning and memory. Insulin administrated intranasally has shown to improve memory and learning in healthy people and in those with some neurodegenerative disorders. Hence, there was rationale for investigating intranasal insulin in PwMS who experience cognitive impairment. We completed a phase Ib/II, randomized, double-blind, placebo-controlled trial; participants were randomized in a 1:1:1 fashion, stratified by relapsing versus progressive MS, to intranasal insulin 10 ​international units (IU) twice a day, 20 IU twice a day, or placebo for 24 weeks. One-hundred and five PwMS were enrolled, 69 of whom had at least one follow up visit during the active treatment phase of the trial (baseline to week 24). The cohort's mean age was 52.4 ​± ​9.7years, 62 ​% were female, and ∼60 ​% had relapsing-remitting MS. The most common side effects amongst treatment groups included headache, rhinorrhea, and dizziness. There were 13 SAEs which were not deemed study drug related; there were no deaths. The main clinical outcome measure, SDMT, did not demonstrate any difference between intranasal insulin and placebo. Similar findings were noted for all secondary outcome measures. Intranasal insulin appeared safe and well-tolerated in PwMS. However, it was not superior to placebo in any of the clinical outcome measures assessed, which could have been impacted by the duration of the trial, small sample size for a three-arm trial design, data missingness (particularly during COVID-19), outcome measure insensitivity to change, baseline cognitive reserve, or other factors. Nonetheless, intranasally-administered therapeutics may be of interest to develop further as a way to get across the blood brain barrier.

PMID:40253245 | DOI:10.1016/j.neurot.2025.e00581

Eur J Cancer. 2025 Apr 15;222:115427. doi: 10.1016/j.ejca.2025.115427. Online ahead of print.

ABSTRACT

AIM OF THE STUDY: The antibody-drug conjugate enfortumab vedotin (EV) received approval in patients with metastatic urothelial carcinoma (mUC). EV-related cutaneous toxicities are frequently reported, whether EV-related AEs association with survival may exist is still unknown. We aim to report the association between cutaneous toxicities and survival in patients receiving EV.

METHODS: This retrospective study enrolled patients treated with monotherapy EV from two oncology centers, followed up for at least 3-months, and data collection demographics, treatments, toxicities, and outcomes. The primary endpoint was progression-free survival (PFS) in patients experiencing cutaneous toxicities or not. Overall survival (OS) was the secondary endpoint.

RESULTS: Data from 63 patients treated with EV from July 19, 2019, to March 12, 2024, were collected. Among them, the 18 (28.6 %) patients experiencing any-grade cutaneous toxicities during EV treatment showed significantly longer median PFS (mPFS: 9.2 vs. 4.7 months, hazard ratio [HR] 0.35; p = 0.0041) and OS (mOS: not reached vs. 8.4 months, HR 0.38; p = 0.0253). The multivariate analysis showed a significant association of cutaneous toxicities with improved PFS (HR 0.40, p = 0.0319), and did not demonstrate significant association with OS even if tendency was kept (HR 0.41, p = 0.067).

CONCLUSION: These results support that patients experiencing any-grade cutaneous toxicity (skin rash) had a prolonged PFS. With the recent expansion of combined treatment using EV plus pembrolizumab in first-line in mUC patients, cutaneous toxicities need to be carefully monitored and optimized dedicated management provided, considering that cutaneous toxicity may be predictive of patient outcome.

PMID:40252634 | DOI:10.1016/j.ejca.2025.115427

J Proteome Res. 2025 Apr 19. doi: 10.1021/acs.jproteome.4c00304. Online ahead of print.

ABSTRACT

With the advantage of extensive coverage, predicted spectral libraries are becoming an attractive alternative in proteomic data analysis. As a popular false discovery rate estimation method, target decoy search has been adopted in library search workflows. While existing decoy methods for curated experimental libraries have been tested, their performance in predicted library scenarios remains unknown. Current methods rely on perturbing real spectra templates, limiting the diversity and number of decoy spectra that can be generated for a given library. In this study, we explore the shuffle-and-predict decoy library generation approach, which can generate decoy spectra without the need for template spectra. Our experiments shed light on decoy method performance for predicted library scenarios and demonstrate the quality of predicted decoys in FDR estimation.

PMID:40252226 | DOI:10.1021/acs.jproteome.4c00304

Interdiscip Sci. 2025 Apr 19. doi: 10.1007/s12539-025-00707-5. Online ahead of print.

ABSTRACT

Multi-modal medical image registration aims to align images from different modalities to establish spatial correspondences. Although deep learning-based methods have shown great potential, the lack of explicit reference relations makes unsupervised multi-modal registration still a challenging task. In this paper, we propose a novel unsupervised dual-stream multi-modal registration framework (DSMR), which combines a dual-stream registration network with a refinement module. Unlike existing methods that treat multi-modal registration as a uni-modal problem using a translation network, DSMR leverages the moving, fixed and translated images to generate two deformation fields. Specifically, we first utilize a translation network to convert a moving image into a translated image similar to a fixed image. Then, we employ the dual-stream registration network to compute two deformation fields respectively: the initial deformation field generated from the fixed image and the moving image, and the translated deformation field generated from the translated image and the fixed image. The translated deformation field acts as a pseudo-ground truth to refine the initial deformation field and mitigate issues such as artificial features introduced by translation. Finally, we use the refinement module to enhance the deformation field by integrating registration errors and contextual information. Extensive experimental results show that our DSMR achieves exceptional performance, demonstrating its strong generalization in learning the spatial relationships between images from unsupervised modalities. The source code of this work is available at https://github.com/raylihaut/DSMR .

PMID:40252168 | DOI:10.1007/s12539-025-00707-5

Mol Divers. 2025 Apr 19. doi: 10.1007/s11030-025-11178-7. Online ahead of print.

ABSTRACT

Predicting molecular and quantum material properties, especially the band gap, is crucial for accelerating discoveries in drug design and material science. Although graph neural networks and probabilistic encoders are well established in molecular data analysis, their targeted integration and application for band-gap prediction remain an active research area. This paper introduces QMGBP-DL, a deep learning approach that combines a molecular graph encoder with machine learning models to improve the prediction accuracy of molecular and material band-gap energy. The encoder uses graph convolutional networks to derive latent representations of chemical structures from SMILES strings, optimized via Kullback-Leibler divergence loss. These representations serve as inputs for training various machine learning models to predict properties. QMGBP-DL's effectiveness is assessed using the QM9, PCQM4M, and OPV datasets, demonstrating significant improvements, particularly with a random forest model for property prediction. A comparative analysis against established approaches DenseGNN, MEGNet, and ALIGNN reveals that QMGBP-DL excels in predicting HOMO, LUMO, and band gap, achieving notably lower MAE values. The integration of GCN-derived latent spaces with traditional machine learning models, especially Random Forest, provides a powerful approach for band-gap prediction. The results highlight the efficacy of our integrated approach, showcasing that graph-based molecular encoding combined with machine learning, particularly Random Forest, is highly effective for accurate band-gap prediction, thereby facilitating material discovery and design.

PMID:40252145 | DOI:10.1007/s11030-025-11178-7