Understanding RNA conformational dynamics is essential to understand its roles in complex biological processes. While computational methods have revolutionized the prediction of static 3D RNA structures, predicting local flexibility directly from structure remains a significant challenge. We developed DeepRMSF, a deep learning-based method that leverages atomic-level descriptions of RNA to predict vibrational flexibility given a tertiary structure. Trained on MD-derived root-mean-square fluctuations(RMSF), DeepRMSF was benchmarked on 371 nonredundant RNAs, with 311 RNAs used for five-fold cross-validation (PCC = 0.7219-0.7464) and 60 RNAs as an independent test set (PCC = 0.734), ensuring minimal sequence/structural similarity between sets. DeepRMSF predicts the local flexibility of medium-sized RNAs (~75 nucleotides) in ~8.2 s, achieving >3000-fold speed-up over MD simulations while maintaining strong extrapolative accuracy. Rather than replacing MD, DeepRMSF offers a scalable and practical alternative for transcriptome-scale screening of RNA flexibility, facilitating studies on RNA structure-dynamics-function relationships and supporting computational modeling in RNA biology.

The wet slag removal system in thermal power plants is critical for boiler operation but suffers from severe monitoring challenges. High-temperature steam obstructs visiblelight cameras, making it difficult to detect large slag chunks that can cause costly conveyor blockages and unplanned downtime. To address this, we propose ISMF-Net (Intelligent Slag Monitoring Framework Network), an integrated framework that leverages multi-modal sensing and multi-task learning for robust, realtime automated monitoring. ISMF-Net fuses visible (RGB) and thermal (T) imaging, using thermal data to penetrate steam while retaining RGB details. An end-to-end network simultaneously performs slag volume estimation (semantic segmentation), large slag blockage warning (object detection), and scraper plate status monitoring (regression). Key innovations include a Real-time Steam Denoising Module (RSDM) and a Context-Aware Crossmodal Fusion (CACF) module. Experiments on our new multimodal dataset show that ISMF-Net significantly outperforms baseline methods, demonstrating its high value for industrial applications facing similar visual challenges.

Improving motion speed and efficiency remains a critical challenge in snake robots gait control. This paper introduces the Loop gait, a novel locomotion gait designed to enhance both speed and energy efficiency of snake robots without passive wheels. Compared to Crawler gait and S-pedal gait, which are more widely used, the Loop gait has a better motion speed (1.8 times of the Crawler gait in the same parameter) and a better motion efficiency (1.6 times of the Crawler gait in the same parameter) due to its more loop body morphology. A static stability model is developed to guide parameter optimization, addressing potential instability caused by elevated center of mass of snake robots. Experiments confirm the Loop gait’s exceptional energy efficiency and propulsion, validating the static stability model’s utility in selecting parameters.

MOTIVATION: Alzheimer's disease (AD) is a neurodegenerative disorder characterized by memory loss and cognitive decline. While graph convolutional networks (GCNs) have emerged as popular tools for AD diagnosis due to their ability to handle structural information and fuse multi-modal features, deep learning approaches face significant challenges including the requirement for large datasets and sensitivity to unbalanced label distributions in AD research. To address these limitations and enhance the flexibility of GCNs, we propose a graph convolutional network based on the meta-learning paradigm (ADMGCN) for early AD diagnosis. This approach incorporates weighting and dimensionality reduction to improve performance, storage, and training efficiency. By leveraging meta-learning, we sample subjects to create numerous label-balanced tasks, maximizing data utilization and mitigating the impact of label imbalance. Additionally, the meta-learning framework enables rapid adaptation to new tasks and facilitates independent testing of the GCN. RESULTS: Our model, ADMGCN, was extensively validated on the Alzheimer's Disease Neuroimaging Initiative datasets. It achieved a maximum accuracy of 73.7% in the multi-classification task for early AD diagnosis. In three binary classification tasks, the model also demonstrated strong performance, achieving accuracies of 92.8%, 88.0%, and 79.6%, respectively. These results confirm that the proposed method provides an effective approach and worthwhile support for the early diagnosis of Alzheimer's disease. AVAILABILITY AND IMPLEMENTATION: ADMGCN is freely available at https://github.com/WendySun16/ADMGCN.

The rapid progress in Deep Learning (DL) and Large Language Models (LLMs) has exponentially increased demands of computational power and bandwidth. This, combined with the high costs of faster computing chips and interconnects, has significantly inflated High Performance Computing (HPC) construction costs. To address these challenges, we introduce the Fire-Flyer AI-HPC architecture, a synergistic hardware-software co-design framework and its best practices. For DL training, we deployed the Fire-Flyer 2 with 10,000 PCIe A100 GPUs, achieved performance approximating the DGX-A100 while reducing costs by half and energy consumption by $40 \%$. We specifically engineered HFReduce to accelerate allreduce communication and implemented numerous measures to keep our Computation-Storage Integrated Network congestion-free. Through our software stack, including HaiScale, 3FS, and HAI-Platform, we achieved substantial scalability by overlapping computation and communication. Our system-oriented experience from DL training provides valuable insights to drive future advancements in AI-HPC.

MOTIVATION: Recovering the gene structures is one of the important goals of genome assembly. In low-quality assemblies, and even some high-quality assemblies, certain gene regions are still incomplete; thus, novel scaffolding approaches are required to complete gene regions. RESULTS: We developed an efficient and fast genome scaffolding method called PEP_scaffolder, using proteins to scaffold genomes. The pipeline aims to recover protein-coding gene structures. We tested the method on human contigs; using human UniProt proteins as guides, the improvement on N50 size was 17% increase with an accuracy of ∼97%. PEP_scaffolder improved the proportion of fully covered proteins among all proteins, which was close to the proportion in the finished genome. The method provided a high accuracy of 91% using orthologs of distant species. Tested on simulated fly contigs, PEP_scaffolder outperformed other scaffolders, with the shortest running time and the highest accuracy. AVAILABILITY AND IMPLEMENTATION: The software is freely available at http://www.fishbrowser.org/software/PEP_scaffolder/ CONTACT: [email protected] information: Supplementary data are available at Bioinformatics online.