From sequence to structure: A comprehensive review of deep learning models for RNA structure prediction.

Application of artificial intelligence in colonoscopy imaging for polyp analysis-A systematic review.

Image Processing-Based Automatic Tooth Segmentation and Age Estimation in Sheep Using Deep Learning.

Adaptive differential privacy mechanism for enhanced deep learning model utility and privacy.

Artificial intelligence in animal anatomy: Exploring the technologies, applications, benefits, and challenges.

Predicting the active sites of quinolone antibiotics interacting with organisms by deep learning and molecular docking.

A Deep Learning Framework for Predicting Teprotumumab Treatment Response in Thyroid Eye Disease.

Estimating proximity to muscular failure using surface EMG and deep learning.

Bridging causality and deep learning for harmful algal bloom prediction.

Deep learning for psychiatric genomics: from tools to applications.

Optimization of drug diffusion in drug-eluting stents for coronary artery based on deep reinforcement learning.

FTIR spectroscopy combined with machine learning reveals molecular signatures distinguishing three phenotypes of endometriosis.

Bangladesh has diverse and vibrant cultural sports, some of which have gained international recognition in recent years. However, there is a lack of standardized datasets for deep learning and computer vision tasks. To address this gap, BD Sports-10 was developed as a comprehensive dataset for Bangladeshi sports. It consists of ten unique sports categories, with a total of 3000 videos, 300 per class, with a resolution of 1920×1080 pixels and 30 frames per second (FPS). Each sport in the dataset features distinct rules, viewing angles, playground setups, and actions, which also depend on players' skills. The dataset captures a diverse range of actions, including jumping, running, tagging, throwing, attempting to hit a clay pot, and capturing an opponent before they cross a designated line. BD Sports-10 includes ten traditional and culturally significant sports: Kabaddi, Nouka Baich, Lathi Khela, Kho Kho, Kanamachi, Toilakto Kolagach Arohon (Kolagach), Hari Vanga, Morog Lorai, Lathim, and Joldanga. This standardized and balanced dataset is not only suitable for classification tasks but also for object detection, player tracking, and automated scoring systems. The dataset supports research in deep learning, machine learning, and computer vision by providing ready-to-use scripts, datasets, and preprocessing pipelines that facilitate diverse AI-based experimental workflows.

Medium- and long-term precipitation prediction has always been a major challenge in precipitation prediction. This research proposes a generalizable Physics-guided Artificial Intelligence (PAI) framework for precipitation prediction. First, By using multifractal detrended fluctuation analysis(MFDFA)method, multifractal characteristic of precipitation is analyzed to identify complexity of the precipitation series. Second, for each precipitation regime, a BP Neural Network prediction model is trained by employing precipitation metrics at monthly scale combining the multifractal characteristic of precipitation as input and subsequently used to predict estimation precipitation for IMERG(Integrated Multi-satellite Retrievals for GPM). The PAI framework is demonstrated in the 18 cities in Sichuan province using the monthly precipitation over 2000--2023.And as the comparison, BP neural network and LSTM(Long Short-Term Memory) neural network were used to predict monthly precipitation in various cities in Sichuan Province. Results show that compared with other machine learning precipitation prediction model the MFDFA-BP prediction model performs better. The MFDFA-BP prediction model can then be used for hydrologic simulations and precipitation prediction.

The conventional drug discovery pipeline is labour-intensive, time-consuming, and costly, involving target identification, hit discovery, lead optimization, and extensive preclinical and clinical evaluation. To overcome these limitations, artificial intelligence (AI) has emerged as a transformative tool in drug discovery, gaining widespread adoption in the pharmaceutical industry during the 2010s due to advances in computing power, data availability, and deep learning. AI-based approaches, including molecular property prediction, protein structure modelling, natural language processing, and ADME/Tox prediction, have enhanced efficiency, reduced costs, and improved decision-making across multiple stages of drug development. Several AI-guided molecules have progressed into clinical trials, with encouraging early-phase success rates, highlighting the potential of AI to accelerate innovation. However, despite more than a decade of intensive research, no AI-only originated drug has yet achieved full regulatory approval, reflecting persistent challenges consistent with Eroom's law. Key limitations include poor data quality and accessibility, lack of model interpretability, gaps between computational predictions and chemical feasibility, and the inherent complexity of biological systems that limit translational success. Furthermore, AI-driven hypothesis generation does not replace the need for scientific reasoning and experimental validation. Overall, while AI has significantly accelerated early drug discovery stages, it remains a supportive tool rather than a standalone solution, underscoring the continued need for human expertise and experimental research.

Magnetic field sensing bolstered by deep learning on scattering images from random and conventional laser illumination

LK-Road3R: Road point cloud mapping via UAV-based video and deep learning

Deep learning with geographical post-processing optimization: an integrated framework for detecting qanat activity states

A comprehensive survey of image clustering based on deep learning

The ConcreteCARB dataset provides a comprehensive repository of 903 high-resolution images of concrete surfaces evaluated using the phenolphthalein test for carbonation detection. This data was collected under controlled laboratory conditions and aims to support artificial intelligence applications in civil engineering, especially in structural health monitoring tasks. The images are systematically organized into two distinct classes: "Carbonated Samples" and "No Carbonation Presence," enabling binary classification approaches. All samples were manually tested, split, and visually labelled by expert engineers to ensure reliable ground-truth classification, in accordance with standardized procedures. The dataset includes images of concrete prism elements fabricated with varying mix designs, incorporating different water-cement ratios and additives, such as industrial silica waste and natural admixtures derived from Opuntia ficus-indica. The specimens were subjected to natural atmospheric carbonation conditions for 180 days, and their carbonation fronts were revealed by phenolphthalein staining. The samples were then split manually with a chisel and hammer, and photographic documentation was performed with a Samsung SM-S901U1 smartphone using predefined settings to ensure consistency and quality across the dataset. ConcreteCARB is intended for researchers, engineers, and data scientists working on machine learning, deep learning, and computer vision solutions for concrete diagnostics. It provides valuable training and benchmarking data for the development of automated detection, classification, and segmentation models for carbonation damage assessment. Furthermore, the dataset can serve as a foundational tool for cross-comparative studies on the efficacy of AI techniques in materials degradation analysis. The openly accessible nature of the dataset through a public repository supports reproducibility and encourages the extension of AI applications in concrete durability and sustainability studies.