Browse Articles
Discover research articles across all indexed journals
A CRISPR-assisted passive microfluidic chip for rapid, visual detection of multiple respiratory viruses
A novel accessible in vitro model of proliferative vitreoretinal diseases shows facilitated epithelial mesenchymal transition through aquaporin-1
An XGBoost-based cellular automata for modeling PM₂.₅ concentration using ground and satellite data
Population dynamics of seed and seedlings of Albizia procera (Roxb.) in Mizoram, India
Deep learning-based disease detection in potato and mango leaves: a comparative study of CNN, AlexNet, ResNet, and EfficientNet
Abstract Timely and precise detection of diseases on plants is crucial for minimizing losses during crop production in order to sustain food supply demands worldwide. In this work, deep learning (DL) was used to develop an automatic disease identification system for the leaves of potato and mango plants using two publicly available datasets, the PlantVillage Potato Leaf Disease (2,152 images) dataset and the Kaggle Mango Leaf Disease dataset (4,000 images). Images were pre-processed, augmented, and split into training and testing datasets (80:20), to enable better model generalization. Four deep learning architectures, namely Convolutional Neural Networks (CNN), AlexNet, Residual Networks (ResNet), and EfficientNet, were evaluated in the context of multi-class disease classification. The baseline CNN achieved a training accuracy of 93.67% and a testing accuracy of 92.61%, with balanced precision and recall (92.5%), thus providing a very strong feature extraction and classification capability. AlexNet showed moderate performance (91.3% training, 90.2% validation), and a very small overfitting was observed. ResNet had an efficient convergence, and attained 96.7% validation accuracy in just a few epochs, thus pointing out the advantage of residual connections in the context of deeper learning. EfficientNet surpassed all the other architectures, since it reached a training accuracy of 98.2% and a validation accuracy of 97.8%, with very small loss (≈ 0.015) and no overfitting, thus proving to have the best generalization ability. The models demonstrated stability and discriminative ability with the support of confusion matrices and accuracy and loss plots produced on an epoch-wise basis. Therefore, the findings indicate that DL models can be adapted for real-time and accurate plant disease diagnosis, establishing a pathway for early remediation, and supporting precision agriculture. The research establishes the opportunity for EfficientNet to be considered a promising solution for scalable smart farming.
Design and analysis of high-k wrapped underlap induced GaN multi-channel GAA nanosheet FET for enhanced performance with cut-off frequency in THz range
A scalable scheduling and resource management framework for cloud-native B2B applications
Melatonin attenuates glandular cystitis via modulation of THBS1 and TLR4/p38 MAPK/STAT3 signaling
The body mass-maximum speed relationship and the athletic capability of giant proboscideans and sauropods
Abstract Estimating the maximum speed capability (athletic capability) of very large fossil animals is challenging. While large animals have limbs favoring longer stride lengths and higher speeds, their body mass imposes limitations on bones, joints, available forces, and physiology, resulting in the fastest animals not being the largest. Here we use the well-known relationship between body mass and potential maximum speed to calculate upper limits for the athletic capability of fossil giant proboscideans and sauropods. First, we assess the reliability of two different fits to maximum speed data for mammals. Subsequently, we analyze masses and speeds of live proboscideans, finding they consistently move below the lower bounds predicted by model fits. Finally, we propose maximum speed for representative fossil proboscideans and sauropods. Although calculated maximum speeds for larger proboscideans are comparatively lower, there is substantial overlap with those observed in live proboscideans. For the largest sauropods approaching or exceeding 50 Tn our calculated maximum speeds are around 10 km/h or lower. These findings suggest that immense body size and graviportal structure of sauropods were key factors likely restricting their locomotion to a single, steady gait, consistent with fossil trackway evidence.
Multi-scale soil moisture dynamics in arid mined Loess Plateau arise from crack evolution and microstructure collapse
Critical role of the CMGC insert sequence for tyrosine autophosphorylation in the protein kinase DYRK1B
Abstract Many CMGC group kinases require phosphorylation of a conserved tyrosine residue in the activation loop to achieve catalytic activity. DYRK family members use a distinctive mechanism involving constitutive cis -autophosphorylation of this tyrosine. The structural basis of this process has remained unclear, as it occurs while the kinase is still in an inactive conformation, and the tyrosine does not match the known substrate consensus of DYRKs. Here, we exploited the different autophosphorylation capacities of the paralogs DYRK1A and DYRK1B to define structural determinants of this process. DYRK1A efficiently autophosphorylates even in cell-free systems, whereas DYRK1B does not. Using domain swaps and point mutations, we identify the CMGC insert in the C-terminal lobe and two adjacent proline residues (P332/P333 in DYRK1B) as critical for proper folding and activation. Mutation of either proline impaired DYRK1B autophosphorylation and nuclear localization but had no effect in DYRK1A. Substitution of the DYRK1B CMGC insert with that of DYRK1A rescued the maturation defect, demonstrating functional interplay between the insert and flanking prolines. Furthermore, the pathogenic R349W mutation in DYRK1B, associated with monogenic obesity and type 2 diabetes, also disrupted autophosphorylation. These findings highlight the role of the CMGC insert and adjacent prolines in DYRK kinase maturation and autoactivation.
SentXFormer: a transformer-enhanced hybrid deep learning framework for cross-domain sentiment analysis of customer reviews
Abstract Cross domain sentiment analysis is still a difficult task because of vocabulary changes, context change and domain specific sentiment. Conventional models are not known to generalize over unknown areas leading to decreased accuracy and unreliable transfer performance. This paper presents a deep learning model named SentXFormer, which is a transformer-based hybrid framework that enhances the sentiment classification in heterogeneous domains. SentXFormer is based on the hybrid SentiConGRU-Net architecture, which consists of CNN and GRU layers, and contextual embeddings of BERT, RoBERTa, and Domain-Adaptive BERT (DABERT). A domain adaptation module based on an adversarial training with a Gradient Reversal Layer (GRL) also encourages domain-invariant representations to be learned. The experiments are done on 23,440 sentiment-labeled reviews across three publicly available datasets including Amazon (7,550 samples), Yelp (8,450 samples), and IMDB (7,440 samples). SentXFormer performed well in in-domain, reaching 98.7% (Amazon), 97.67% (Yelp) and 98.8% (IMDB) accuracies. The model is stable in terms of transferability in cross-domain settings with an accuracy of 91–93% on all train-test combinations. The comparative analysis with LSTM, CNN, GRU, and latest transformer-based adaptation models demonstrates that SentXFormer is always better than the current baselines. The findings indicate that SentXFormer is an efficient, strong and scalable tool to sentiment analysis in heterogeneous and real-life customer review contexts.