Long non-coding RNAs (lncRNAs) play essential roles in various biological processes, including chromatin modification, cell cycle regulation, transcription, and translation. Recent studies have revealed that the biological functions of lncRNAs are closely associated with their subcellular localizations, making accurate localization prediction critical for understanding their biological roles in cellular regulation and disease mechanisms. However, most existing methods mainly rely on sequence features while neglecting structural information, and they are often limited to single-label predictions covering only a small number of subcellular compartments. In this study, we proposed an efficient deep learning framework, LncTracker, for multi-label prediction of lncRNA subcellular localizations across seven distinct compartments. LncTracker adopts a multi-channel architecture that integrates diverse input features into model training, including both primary sequence and secondary structure information. Secondary structures are converted into attributed graphs to capture spatial relationships among nucleotides, including adjacency and base-pairing connections. These structural features are then combined with sequence-based features to predict subcellular localization probabilities. Such a design enables LncTracker to learn joint representations of sequences and structures, thereby enhancing predictive performance and robustness. Benchmarking experiments demonstrated the superiority of LncTracker over state-of-the-art approaches, particularly in handling imbalanced localization scenarios. Furthermore, we leveraged LncTracker to identify sequence motifs critical for each subcellular localization and analysed key sub-structures contributing to predictions.

Cannulation is a common procedure in minimally invasive surgery. Force sensing during this procedure is essential for its safety and effectiveness. Achieving high-precision force sensing remains challenging due to the dimensional constraints of catheters and the need for hollow channels. This study presents a hollow triaxial force sensor based on fiber Bragg gratings, featuring a double-layer beam configuration that achieves both compact size and balanced stiffness. We develop an evolutionary design process of the sensor using the Freedom and Constraint Topology method. We also establish a mechanical model of the sensor, and optimize structural parameters through the condition number of the mapping matrix to suppress error amplification. In addition, we propose an improved tree Parzen estimator optimized random forest algorithm, integrated with temperature compensation, to enhance the precision of triaxial force sensing. The sensor achieves root mean square errors of 0.0214 and 0.0225 N for radial force components F-x and F-y with measurement ranges of $\pm$ 2N, and 0.0170N for axial force component F-z with measurement range of 0-2N. An ex vivo tissue experiment validates that the sensor effectively detects dynamic force variations and facilitates compliant operation with surgical instruments in a porcine duodenal papilla.

This paper presents an integrated design and Model Predictive Feedforward (MPFF) control method for a rotary direct-drive servo valve (RDDV) to address key challenges in robotic electro-hydraulic actuators, including internal leakage reduction, high-precision motion control, robustness against nonlinearities and contamination. A systematic design methodology is introduced for the valve spool geometry, with Computational Fluid Dynamics (CFD) employed to characterize and optimize the pressure-flow and torque-angle relationships. To mitigate static friction and eliminate dead-zone behavior near zero velocity, a high-frequency Micro-Vibration Friction Suppression (MVFS) strategy is proposed, achieving steady-state angular accuracy within +/- 0.005 rad. To improve force control performance under embedded computing constraints, a lightweight Model Predictive Feedforward (MPFF) strategy is designed based on internal dynamic prediction and single-step gradient descent. Compared with conventional PID control, MPFF increases the force control bandwidth from 22 Hz to 40 Hz (-3 dB cutoff) and maintains steady-state force error within +/- 15 N, while avoiding complex online optimization. Experimental results confirm that the proposed RDDV achieves internal leakage as low as 0.2 L/min at 7 MPa and supports precise, high-bandwidth force and position control. These contributions demonstrate the RDDV and its lightweight control framework as a compact, efficient, and scalable solution for next-generation intelligent hydraulic systems.

The rise of antimicrobial resistance has outpaced the discovery of antibiotics, creating a pressing global health crisis. Artificial intelligence (AI) offers tools to explore chemical and biological space more efficiently than traditional methods. Here, we review the use of AI in antibiotic research. We outline machine-learning models that have been applied to screen and optimize known compounds, including small molecules and peptides. We also summarize modern generative models leveraged to design antibiotic candidates. We cover approaches such as protein language models for advanced sequence and structural analysis, graph neural networks for modeling complex molecular interactions, and generative models for de novo generation of antimicrobial compounds. We discuss how these methods have accelerated hit identification in silico and sometimes in vitro and even in vivo, while also noting important challenges. Finally, we outline future directions, which could help AI fulfill its promise in discovering next-generation antibiotics.

While hydraulic robots excel in high load capacity and interference immunity, achieving precise force tracking is hindered by complex fluid dynamics, system nonlinearities, and time-varying disturbances. This paper introduces a Physics-Enhanced Neural Network (PENN) approach. This method employs an Extended Kalman Filter (EKF) as a teacher to guide the prediction of hydraulic force dynamics. By effectively leveraging both data and physical baselines, this method achieved significant performance improvements, with the MSE, RMSE, and MAE decreasing by 85.5%, 61.9%, and 65.3%, respectively, compared with the traditional EKF baseline. Consequently, we propose a Neural Input-Output Feedback Linearization (NFBL) Proportional-Integral (PI) controller to globally linearize the nonlinear dynamics and track the desired force. The online EKF-PENN identifies the autonomous response and control gain terms of the hydraulic affine nonlinear system, providing accurate control variables under designed operating conditions. The method compensates for the pressure drop during hydraulic cylinder piston movement via flow compensation, while feedforward techniques significantly enhance force control response. Experimental results verify the effectiveness of these proposed methods. The proposed method can significantly improve the locomotion performance of hydraulically actuated legged robots.

Semantic segmentation in colonoscopy images is pivotal in aiding healthcare professionals to interpret images and enhance diagnostic precision. Nonetheless, the detection of polyps and instruments is challenged by the difficulty in capturing the textures and edges of tiny lesions, and these challenges are exacerbated by low contrast, inconsistent illumination, and noise. To address these challenges, we introduce WDNet, a network adopting a multi-tiered feature extraction and fusion approach, with each encoder layer amalgamating local and global information to construct expressive high-level representations. The input of the network is derived from wavelet transform to dissect images into low- and high-frequency sub-bands, utilizing learnable soft-thresholding to diminish noise while maintaining essential features. High-frequency data are adept at capturing details and edges, whereas low-frequency data furnish a global context. Moreover, WDNet harnesses a diffusionbased decoding mechanism with adaptive step sizes to amplify target region features and mitigate background interference, achieving meticulous segmentation. Comprehensive experiments conducted on a new surgical dataset, along with public benchmarks underscore its remarkable performance. WDNet not only exhibits state-of-the-art performance of semantic segmentation in colonoscopy images with remarkable detail and boundary accuracy but also stands out in processing speed, facilitating the swift handling of extensive datasets. The dataset and source code are available at https://github.com/hedongdong6060/WDNet.

PurposeThe purpose of this paper is to design a flying wheel-legged humanoid robot (FWLR), endowing the robot with flight capability to improve the obstacle-crossing ability of the wheel-legged humanoid robot. A flight control method using thrust-vector-control (TVC) under constant thrust strength is proposed, which reduces the performance requirements on the response speed of thrusters.Design/methodology/approachTo endow the robot with flight capability, three sets of thrusters are installed at the robot's back and two arm ends to provide flight lift and the direction of thrust can be changed through the arm swing. According to the robot configuration, this paper established a linearized dynamic model and proposed a constant-strength-thrust-vector-control (CSTVC) framework enabling the robot to achieve flight without thrust intensity change.FindingsWith the proposed modeling method and CSTVC framework, FWLR can inhibit attitude and position drift during takeoff and hovering, and has certain adaptability to takeoff attitude. Finally, FWLR reached a flying height up to 1 m under a 30 kg large self-weight with fixed thrust strength.Originality/valueThe design, modeling and flight control method proposed in this paper enables a human-sized wheel-legged humanoid robot to achieve takeoff and hovering for the first time. The movement range of wheel-legged humanoid robot is extended to the air, thereby enhancing its application value in emergency tasks such as disaster search-and-rescue.

For biomedical and rescue applications, this paper presents the Hyd-U hand, a compact, hydraulically actuated underactuated robotic hand designed to enhance payload capability in dexterous manipulation. With a payload-to-weight ratio of 22.14 (16,kg payload, 722.6,g self-weight), it is ideal for applications requiring high force and low weight, such as field rescue and mobile robotics. The system integrates three key innovations: Electro-hydraulic underactuation: Combines a singleacting hydraulic cylinder with truss-pulley transmission and cable-driven fingers to enable sensor-free adaptive grasping Hybrid PID-NNIM control: Implements neural network inverse modeling (NNIM) with PWM current sensing to achieve +/- 0.05 mm displacement precision Embedded high-force manipulation: Miniaturized servo driver enables responsive operation in unstructured environments Experimental results demonstrate the Hyd-U hand's stable grasping of objects with varying shapes and weights, showcasing its repeatability and responsiveness in dynamic, unstructured environments.

MOTIVATION:Proteolysis-targeting chimeras (PROTACs) are heterobifunctional molecules that can degrade "undruggable" protein of interest by recruiting E3 ligases and hijacking the ubiquitin-proteasome system. Some efforts have been made to develop deep learning-based approaches to predict the degradation ability of a given PROTAC. However, existing deep learning methods either simplify proteins and PROTACs as 2D graphs by disregarding crucial 3D spatial information or exclusively rely on limited labels for supervised learning without considering the abundant information from unlabeled data. Nevertheless, considering the potential to accelerate drug discovery, it is critical to develop more accurate computational methods for PROTAC-targeted protein degradation prediction. RESULTS:This study proposes DegradeMaster, a semisupervised E(3)-equivariant graph neural network-based predictor for targeted degradation prediction of PROTACs. DegradeMaster leverages an E(3)-equivariant graph encoder to incorporate 3D geometric constraints into the molecular representations and utilizes a memory-based pseudolabeling strategy to enrich annotated data during training. A mutual attention pooling module is also designed for interpretable graph representation. Experiments on both supervised and semisupervised PROTAC datasets demonstrate that DegradeMaster outperforms state-of-the-art baselines, with substantial improvement of AUROC by 10.5%. Case studies show DegradeMaster achieves 88.33% and 77.78% accuracy in predicting the degradability of VZ185 candidates on BRD9 and ACBI3 on KRAS mutants. Visualization of attention weights on 3D molecule graph demonstrates that DegradeMaster recognizes linking and binding regions of warhead and E3 ligands and emphasizes the importance of structural information in these areas for degradation prediction. Together, this shows the potential for cutting-edge tools to highlight functional PROTAC components, thereby accelerating novel compound generation. AVAILABILITY AND IMPLEMENTATION:The source code and datasets are available at https://github.com/ABILiLab/DegradeMaster and https://zenodo.org/records/14715718.

The paper focuses on incorporating the surgeon’s motion intention into impedance control, which is one of the most commonly used control schemes in master manipulation, to reduce the user’s motion efforts. To achieve this, the non-autoregressive spatio-temporal transformer network is applied to predict the surgeon’s motion trajectory, which is described by the control points of B-spline. The study investigates the effects of input data length, data feature, and feedforward data on trajectory prediction. Furthermore, a Cartesian variable impedance control strategy without force sensor is implemented by adjusting the damping parameters according to the future external force. The experiment shows a promising result, where the interactive force under this control strategy is lower than the one with the current external force, and the precise operation still performs well.

Surgical smoke removal is crucial for enhancing laparoscopic image quality in computer-assisted surgery. While existing methods utilize estimated smoke distribution to address non-homogeneous characteristics, most treat this information merely as prior input and often suffer from over-desmoking artifacts. To address these limitations, this study introduces a desmoking network that reconstructs smoke-free images by explicitly utilizing smoke distribution information. The network comprises two key modules: the Smoke Attention Estimator (SAE) and the Hybrid Guided Embedding (HGE). The SAE generates a smoke attention map via a channel-aware position embedding with lightness prior to improve accuracy. The HGE takes the predicted smoke attention map from the SAE as input and employs convolutional layers along with a novel field transformation method to generate residual terms. By combining these residual terms with the original image, the HGE preserves fine details in smoke-free regions, thereby preventing over-desmoking. Experimental results reveal that the proposed method achieves improvements of at least 3.71% in Peak Signal-to-Noise Ratio (PSNR) and 18.75% in Learned Perceptual Image Patch Similarity compared to state-of-the-art methods on the synthetic dataset, while attaining the lowest Perception-based Image Quality Evaluator score (24.55) on the Cholec80 dataset. It operates at around 174 frames per second, indicating strong real-time processing capability. The network achieves over 40 dB in PSNR for smoke-free regions, excelling in both color restoration and detail preservation. This work is available at https://homepage.hit.edu.cn/wpgao?lang=en.

Accurate identification of the subthalamic nucleus (STN) borders is time-consuming, relying heavily on the neurosurgeon expertise in manually interpreting the electrophysiological signals. Local field potentials (LFPs) have garnered imperative attention due to their strong correlation with the STN. However, existing detection models often face challenges with high computational complexity, hyperparameter optimization and lack of explainability, making them unreliable for clinicians. Therefore, this study introduces an explanatory framework using convolutional neural networks (CNN) for detecting the STN region from LFPs. Continuous wavelet transform is employed to convert LFPs signals into scalogram images, which are then processed by sixteen CNN models. We evaluated our framework by examining the impact of various limiting factors on the classification performance, including model size, learning rate (LR), optimizers and data scaling. Deep features are extracted from the top-performing CNN architectures to capture rich representations of the scalograms. These features are then fused and classified using k-nearest neighbour algorithm. Gradient-weighted class activation mapping is used to explain the decisions made by the proposed model. Our approach achieved an accuracy of 99.61

With the advent of the deep learning-based colonoscopy system, the need for a vast amount of high-quality colonoscopy image datasets for training is crucial. However, the generalization ability of deep learning models is challenged by the limited availability of colonoscopy images due to regulatory restrictions and privacy concerns. In this paper, we propose a method for rendering high-fidelity 3D colon models and synthesizing diversified colonoscopy images with abnormalities such as polyps, bleeding, and ulcers, which can be used to train deep learning models. The geometric model of the colon is derived from CT images. We employed dedicated surface mesh deformation to mimic the shapes of polyps and ulcers and applied texture mapping techniques to generate realistic, lifelike appearances. The generated polyp models were then attached to the inner surface of the colon model, while the ulcers were created directly on the inner surface of the colon model. To realistically model blood behavior, we developed a simulation of the blood diffusion process on the colon’s inner surface and colored vertices in the traversed region to reflect blood flow. Ultimately, we generated a comprehensive dataset comprising high-fidelity rendered colonoscopy images with the abnormalities. To validate the effectiveness of the synthesized colonoscopy dataset, we trained state-of-the-art deep learning models on it and other publicly available datasets and assessed the performance of these models in abnormal classification, detection, and segmentation. Notably, the models trained on the synthesized dataset exhibit an enhanced performance in the aforementioned tasks, as evident from the results.

Global population aging has led to a sharp increase in patients of upper limb motor dysfunction. Robot assisted virtual training, as a novel solution, can offer safe and precise assistance for upper limb rehabilitation. However, it remains a critical challenge to compensate virtual interaction force and realize joint synergy movement. In this paper, we design an upper limb rehabilitation robot for virtual training (ULRVT II) which is a cable driven exoskeleton with high compatibility controlled by a joint synergy method. Moreover, we establish a rehabilitation platform with a virtual training environment and evaluation system for experimental validation. Tests for the performance of joint synergy and virtual training are carried out to show the effectiveness of our robot.

BACKGROUND:The single-port surgical robot can reduce incision size and accelerate postoperative recovery. This paper analyses the dynamic model of the remote centre mechanism (RCM) of the proposed single-port robot for force control. METHODS:This paper proposes a dynamic model identification method for the RCM with a minimal parameter set derived from its tree structure. A nonlinear friction model for the prismatic joints and corresponding identification method are introduced, and the parameter set is iteratively refined using iterative reweighted least squares (IRLS), sequential quadratic programming (SQP) and an outlier detection algorithm. An adaptive Kalman filter (AKF) is applied to suppress noise in position differentiation, ensuring smooth velocity and acceleration. RESULTS:The proposed method improves fitting accuracy and provides low-deviation predictions for cross-validation trajectory data. CONCLUSIONS:The proposed method enhances modelling accuracy and noise suppression in single-port surgical robots. CLINICAL TRIAL REGISTRATION:The authors declare that this research is not a clinical trial and is not registered with any clinical trial registry.

PurposeThe purpose of this paper is to focus on the task of object-agnostic grasping in clutter with the aid of pushing or shifting, where pushing is more suitable for objects with enough operation space and shifting is more effective for restricted objects.Design/methodology/approachA deep reinforcement learning framework is proposed to learn synergetic push-shift-grasp (SPSG) policies from visual observations. The action-values (Q) are modeled by Q maps outputted by three separate branches on the top of a shared backbone and a multi-scale feature fusion module, resulting in finer Q predictions and better computation efficiency. The target network is integrated to reduce overestimation and instability in learning. Moreover, to improve the training sample efficiency, a rewarding strategy and a filter-mask function, both constructed upon the convex hull of objects, are proposed. The former makes sure that only the meaningful non-prehensile actions that disperse cluttered objects are rewarded, whereas the latter allows SPSG to focus action attempts on the object area with smoothed Q values.FindingsExperiments in simulation and the real world are both performed. SPSG outperforms baselines in task completion rate and grasp success rate, at challenging test scenarios with tightly packed objects and randomly arranged novel objects, especially when objects located at the corners of a bin where preferential shift actions are required.Originality/valueThis paper presents a novel SPSG policy learning framework to efficiently learn effective SPSG policies, which outperforms existing methods in task completion rate and grasp success rate.

This paper presents a wheel-leg hybrid gait planning and whole-body motion generation method for wheeled biped robots (WBR). This method utilizes momentum to evaluate the robot’s balance, and employs direct collocation to optimized generation the wheel-leg hybrid gait in real-time based on the environment elevation information. The trajectory optimization (TO) is decoupled into the forward and lateral motion to reduce the solution time for each subproblem, thereby enhancing the efficiency of the overall planning, enabling online trajectory generation at 5 Hz. An obstacle search algorithm is proposed to identify the obstacles along the robot’s reference path and construct terrain and safety constraints, which are incorporated into the optimization problem through exponential control barrier function. This facilitates the online generation of the robot’s contact sequence, ensuring that wheels movement remain within a safe region. Finally, a balance and trajectory tracking controller based on nonlinear model predictive control (NMPC) is proposed to generate the CoM’s spatial motion and contact trajectory at a higher update frequency (200Hz), which are then mapped into the joint space using inverse kinematic, enabling the robot to track the reference gait trajectory while maintaining balance. Experimental validation on a hydraulically driven WBR demonstrates that the method enables periodic hybrid gait generation and real-time obstacle search and traversal.

Achieving stable movement on uneven terrains for wheeled biped robots (WBR) is nontrivial due to their under-actuated and inherently unstable nature. To address this issue, this article proposes a whole-body motion control framework based on hierarchical model predictive control (HMPC). First, the wheeled linear inverted pendulum (WLIP) model is proposed to analyze the dynamic coupling mechanism of WBR. Based on this coupling, an optimal ground reaction force (GRF) location control policy is formulated, which serves as parameters for the single rigid body (SRB) dynamic model, enabling the spatial motion of the under-actuated SRB fully controllable. Finally, the inverse kinematics control is utilized to generate the whole-body motion of the robot. This method directly considers the effect of GRF on the robot system, and balances the performance with computational efficiency of MPC. Experiments on a real hydraulic WBR verify that the proposed method provides excellent performance and robustness for both indoor and outdoor motion.

Force control forms the foundation for achieving dexterous motion in hydraulic robots. Tracking the desired force with high accuracy and responsiveness faces persistent challenges from the nonlinearities of the valve-controlled cylinder system, as well as model inaccuracies due to model-plant mismatch, time-varying disturbances, and unmodeled effects. An input-output feedback linearization-based proportional-integral(PI) controller is introduced to globally linearize the nonlinear dynamics and track the desired force. This Controller is augmented with an online learning algorithm that fits the residuals between the nominal model and the actual plant to compensate for model inaccuracies. Additionally, current dither is introduced in the current loop control. Current dither effectively reduces spool startup friction, enhances spool dynamic response, and minimizes pressure hysteresis, thereby improving overall force control responsiveness performance. We validate these methods through comparisons with state-of-the-art approaches in terms of sinusoidal and step signal responses using simulations and a single joint testbed. The results demonstrate a 57% of the rise time improvement in step responses. This work advances hydraulic force control by synergizing model-based control with data-driven online adaptation, offering a deployable solution for high-dynamic robotic applications.