Accurate prediction of drug-target interactions (DTIs) plays a pivotal role in drug discovery, facilitating optimization of lead compounds, drug repurposing and elucidation of drug side effects. However, traditional DTI prediction methods are often limited by incomplete biological data and insufficient representation of protein features. In this study, we proposed KG-CNNDTI, a novel knowledge graph-enhanced framework for DTI prediction, which integrates heterogeneous biological information to improve model generalizability and predictive performance. The proposed model utilized protein embeddings derived from a biomedical knowledge graph via the Node2Vec algorithm, which were further enriched with contextualized sequence representations obtained from ProteinBERT. For compound representation, multiple molecular fingerprint schemes alongside the Uni-Mol pre-trained model were evaluated. The fused representations served as inputs to both classical machine learning models and a convolutional neural network-based predictor. Experimental evaluations across benchmark datasets demonstrated that KG-CNNDTI achieved superior performance compared to state-of-the-art methods, particularly in terms of Precision, Recall, F1-Score and area under the precision-recall curve (AUPR). Ablation analysis highlighted the substantial contribution of knowledge graph-derived features. Moreover, KG-CNNDTI was employed for virtual screening of natural products against Alzheimer's disease, resulting in 40 candidate compounds. 5 were supported by literature evidence, among which 3 were further validated in vitro assays.
Alzheimer Disease/drug therapy* ; Biological Products/therapeutic use* ; Humans ; Neural Networks, Computer ; Machine Learning ; Drug Discovery/methods* ; Algorithms ; Drug Evaluation, Preclinical/methods*

As a strategic emerging industry, biomanufacturing faces core challenges in achieving precise optimization and efficient scale-up of fermentation processes. This review focuses on two critical aspects of fermentation-real-time sensing and intelligent control-and systematically summarizes the advancements in online monitoring technologies, artificial intelligence (AI)-driven optimization strategies, and digital twin applications. First, online monitoring technologies, ranging from conventional parameters (e.g., temperature, pH, and dissolved oxygen) to advanced sensing systems (e.g., online viable cell sensors, spectroscopy, and exhaust gas analysis), provide a data foundation for real-time microbial metabolic state characterization. Second, conventional static control relying on expert experience is evolving toward AI-driven dynamic optimization. The integration of machine learning technologies (e.g., artificial neural networks and support vector machines) and genetic algorithms significantly enhances the regulation efficiency of feeding strategies and process parameters. Finally, digital twin technology, integrating real-time sensing data with multi-scale models (e.g., cellular metabolic kinetics and reactor hydrodynamics), offers a novel paradigm for lifecycle optimization and rational scale-up of fermentation. Future advancements in closed-loop control systems based on intelligent sensing and digital twin are expected to accelerate the industrialization of innovative achievements in synthetic biology and drive biomanufacturing toward higher efficiency, intelligence, and sustainability.
Artificial Intelligence ; Fermentation ; Bioreactors/microbiology* ; Neural Networks, Computer ; Algorithms ; Biotechnology/methods*

Accurately obtaining the crop quantity and density is not only crucial for the demand-based input of water and fertilizer in the field but also vital for ensuring the yield and quality of crops. Aerial photography by unmanned aerial vehicles (UAVs) can quickly acquire the distribution image information of crops over a large area. However, the accurate recognition of a single type of dense targets is a huge challenge for most recognition algorithms. Taking banana seedlings as an example in this study, we captured the images of banana plantations by UAVs from high altitudes to explore an efficient recognition method for dense targets. We proposed a strategy of "cut-recognition-stitch" and constructed a counting method based on the improved Faster R-CNN algorithm. First, the images containing highly dense targets were cropped into a large number of image tiles according to different sizes (simulating different flight altitudes), and the Contrast Limited Adaptive Histogram Equalization (CLAHE) algorithm was adopted to improve the image quality. A banana seedling dataset containing 36 000 image tiles was constructed. Then, the Faster R-CNN network with optimized parameters was used to train the banana seedling recognition model. Finally, the recognition results were reversely stitched together, and a boundary deduplication algorithm was designed to correct the final counting results to reduce the repeated recognition caused by image cropping. The results show that the recognition accuracy of the Faster R-CNN with optimized parameters for banana image datasets of different sizes can reach up to 0.99 at most. The deduplication algorithm can reduce the average counting error for the original aerial images from 1.60% to 0.60%, and the average counting accuracy of banana seedlings reaches 99.4%. The proposed method effectively addresses the challenge of recognizing dense small objects in high-resolution aerial images, providing an efficient and reliable technical solution for intelligent crop density monitoring in precision agriculture.
Musa/growth & development* ; Crops, Agricultural/growth & development* ; Algorithms ; Neural Networks, Computer ; Unmanned Aerial Devices ; Seedlings/growth & development* ; Image Processing, Computer-Assisted/methods* ; Photography ; Agriculture/methods*

OBJECTIVE: To propose a new protocol for personalized mandibular reconstruction assisted by three-dimensional (3D) retrieval model based on fully connected neural network (FCNN) and a database of mandibles, and to verify clinical feasibility of the protocol. METHODS: A database of mandibles of 300 normal northern Chinese Han people was established. On the basis of cephalometry, the mandible landmarks with good stability were further screened. Mandibular landmarks were selected and geometric features of the mandible were extracted. A 3D retrieval algorithm was developed, which could retrieve the mandible most similar to a given mandible from the database. A FCNN was built to train the algorithm to improve accuracy of the 3D retrieval model. Using Geomagic Control 2014 software, matching accuracy of the 3D retrieval model was based on aforementioned mandible database and algorithm. From December 2019 to March 2021, a total of 5 patients underwent personalized mandibular reconstruction assisted by a 3D retrieval model based on mandible database and FCNN in the Department of Oral and Maxillofacial Surgery, Peking University School and Hospital of Stomatology. The most similar mandible was retrieved from mandible database through 3D retrieval algorithm. It was used to restore the premorbid morphology of defect area and guide mandibular reconstruction. For the 5 patients, mandible was reconstructed with iliac flap. Virtual surgical plan was transformed using individual surgical guides. RESULTS: Through screening, mandibular landmarks with high reproducibility and stability were identified and composed of mandibular landmarker protocols. After training, the average deviation between most similar mandible retrieved from the 300-case mandible database through 3D retrieval model based on FCNN and given mandible was (1.77±0.44) mm. And the root-mean-square deviation between the most similar mandible retrieved from the database and given mandible was (2.58±0.86) mm. The mandibular reconstruction surgery was successful in all the 5 patients. Their facial symmetry and occlusion were restored. All the patients were satisfied with postoperative appearance. The mean deviation between postoperative mandible and preoperative design was (0.98±0.17) mm. The area with a deviation ≤1 mm accounted for 61.34%±14. 13%, ≤2 mm accounted for 83.82%±7.35%, and ≤3 mm accounted for 93.94%± 2.87%. CONCLUSION The personalized mandibular reconstruction assisted by 3D retrieval model based on the 300-case mandible database and FCNN is feasible clinically.
Humans ; Neural Networks, Computer ; Mandibular Reconstruction/methods* ; Mandible/diagnostic imaging* ; Imaging, Three-Dimensional/methods* ; Adult ; Databases, Factual ; Female ; Male ; Algorithms ; Middle Aged ; Cephalometry

Medical robotic arm often encounters multi-source and nonlinear errors during the calibration process, making it difficult for traditional mathematical modeling methods to fully characterize system error features, thereby limiting further improvement in calibration accuracy. In this study, a robotic arm parameter error identification model is established, and a calibration method based on an adaptive long short-term memory (ALSTM) neural network is proposed. The method incorporates a particle swarm optimization (PSO) algorithm to optimize the weights of each layer of the LSTM neural network, enabling more effective fitting of robotic arm kinematic errors and ultimately yielding more accurate Denavit-Hartenberg (D-H) parameters. To validate the proposed approach, 110 sets of experimental data are collected using the HSR-JR680 robotic arm calibration system. Experimental results demonstrate that the ALSTM model reduces the root mean square error (RMSE) by 23.07%-80.39% compared to traditional calibration methods, and shortens the convergence time by 32.44% compared to a standard LSTM model. The optimized D-H parameters obtained meet the high-precision calibration requirements of medical robotic arm, confirming the effectiveness of the proposed method.
Calibration ; Neural Networks, Computer ; Algorithms ; Robotics ; Robotic Surgical Procedures ; Models, Theoretical

OBJECTIVES: Non-small cell lung cancer (NSCLC) is associated with poor prognosis, with 30% of patients diagnosed at an advanced stage. Mutations in the EGFR and KRAS genes are important prognostic factors for NSCLC, and targeted therapies can significantly improve survival in these patients. Although tissue biopsy remains the gold standard for detecting gene mutations, it has limitations, including invasiveness, sampling errors due to tumor heterogeneity, and poor reproducibility. This study aims to develop machine learning models based on radiomic features to predict EGFR and KRAS gene mutation status in NSCLC patients, thereby providing a reference for precision oncology. METHODS: Imaging and mutation data from eligible NSCLC patients were obtained from the publicly available Lung-PET-CT-Dx dataset in The Cancer Imaging Archive (TCIA). A three-dimensional-convolutional neural network (3D-CNN) was used to extract imaging features from the regions of interest (ROI). The LightGBM algorithm was employed to build classification models for predicting EGFR and KRAS gene mutation status. Model performance was evaluated using 5-fold cross-validation, with receiver operator characteristic (ROC) curves, area under the curve (AUC), accuracy, sensitivity, and specificity used for validation. RESULTS: The models effectively predicted EGFR and KRAS mutations in NSCLC patients, achieving an AUC of 0.95 for EGFR mutations and 0.90 for KRAS. The models also demonstrated high accuracy (EGFR 89.66%; KRAS 87.10%), sensitivity (EGFR 93.33%; KRAS 87.50%), and specificity (EGFR 85.71%; KRAS 86.67%). CONCLUSIONS A radiogenomics-machine learning predictive model can serve as a non-invasive tool for anticipating EGFR and KRAS gene mutation status in NSCLC patients.
Humans ; Carcinoma, Non-Small-Cell Lung/diagnostic imaging* ; Lung Neoplasms/diagnostic imaging* ; Mutation ; Proto-Oncogene Proteins p21(ras)/genetics* ; ErbB Receptors/genetics* ; Machine Learning ; Positron Emission Tomography Computed Tomography ; Female ; Male ; Neural Networks, Computer ; Middle Aged ; Aged

OBJECTIVES: The presence of multi-scale skin lesion regions and image noise interference and limited resources of auxiliary diagnostic equipment affect the accuracy of skin disease detection in skin disease detection tasks. To solve these problems, we propose a highly efficient and lightweight skin disease detection model using an improved RT-DETR model. METHODS: A lightweight FasterNet was introduced as the backbone network and the FasterNetBlock module was parametrically refined. A Convolutional and Attention Fusion Module (CAFM) was used to replace the multi-head self-attention mechanism in the neck network to enhance the ability of the AIFI-CAFM module for capturing global dependencies and local detail information. The DRB-HSFPN feature pyramid network was designed to replace the Cross-Scale Feature Fusion Module (CCFM) to allow the integration of contextual information across different scales to improve the semantic feature expression capacity of the neck network. Finally, combining the advantages of Inner-IoU and EIoU, the Inner-EIoU was used to replace the original loss function GIOU to further enhance the model's inference accuracy and convergence speed. RESULTS: The experimental results on the HAM10000 dataset showed that the improved RT-DETR model, as compared with the original model, had increased mAP@50 and mAP@50:95 by 4.5% and 2.8%, respectively, with a detection speed of 59.1 frames per second (FPS). The improved model had a parameter count of 10.9 M and a computational load of 19.3 GFLOPs, which were reduced by 46.0% and 67.2% compared to those of the original model, validating the effectiveness of the improved model. CONCLUSIONS The proposed SD-DETR model significantly improves the performance of skin disease detection tasks by effectively extracting and integrating multi-scale features while reducing both parameter count and computational load.
Humans ; Skin Diseases/diagnosis* ; Skin/pathology* ; Neural Networks, Computer ; Algorithms

OBJECTIVES: We propose a novel deep learning segmentation algorithm (DSRF) based on multi-scale supervision and residual feedback strategy for precise segmentation of the optic chiasm and optic nerves in CT images of nasopharyngeal carcinoma (NPC) patients. METHODS: We collected 212 NPC CT images and their ground truth labels from SegRap2023, StructSeg2019 and HaN-Seg2023 datasets. Based on a hybrid pooling strategy, we designed a decoder (HPS) to reduce small organ feature loss during pooling in convolutional neural networks. This decoder uses adaptive and average pooling to refine high-level semantic features, which are integrated with primary semantic features to enable network learning of finer feature details. We employed multi-scale deep supervision layers to learn rich multi-scale and multi-level semantic features under deep supervision, thereby enhancing boundary identification of the optic chiasm and optic nerves. A residual feedback module that enables multiple iterations of the network was designed for contrast enhancement of the optic chiasm and optic nerves in CT images by utilizing information from fuzzy boundaries and easily confused regions to iteratively refine segmentation results under supervision. The entire segmentation framework was optimized with the loss from each iteration to enhance segmentation accuracy and boundary clarity. Ablation experiments and comparative experiments were conducted to evaluate the effectiveness of each component and the performance of the proposed model. RESULTS: The DSRF algorithm could effectively enhance feature representation of small organs to achieve accurate segmentation of the optic chiasm and optic nerves with an average DSC of 0.837 and an ASSD of 0.351. Ablation experiments further verified the contributions of each component in the DSRF method. CONCLUSIONS The proposed deep learning segmentation algorithm can effectively enhance feature representation to achieve accurate segmentation of the optic chiasm and optic nerves in CT images of NPC.
Humans ; Tomography, X-Ray Computed/methods* ; Optic Chiasm/diagnostic imaging* ; Optic Nerve/diagnostic imaging* ; Algorithms ; Nasopharyngeal Carcinoma ; Deep Learning ; Nasopharyngeal Neoplasms/diagnostic imaging* ; Neural Networks, Computer ; Image Processing, Computer-Assisted/methods*

OBJECTIVES: To design a deep learning model that balances model complexity and performance to enable its integration into wearable ECG monitoring devices for automated diagnosis of atrial fibrillation. METHODS: This study was performed based on data from 84 patients with atrial fibrillation, 25 patients with atrial fibrillation, and 18 subjects without obvious arrhythmia collected from the publicly available datasets LTAFDB, AFDB, and NSRDB, respectively. A lightweight attention network based on depthwise separable convolution and fusion of channel-spatial information, namely DSC-AttNet, was proposed. Depthwise separable convolution was introduced to replace standard convolution and reduce model parameters and computational complexity to realize high efficiency and light weight of the model. The multilayer hybrid attention mechanism was embedded to compute the attentional weights of the channels and spatial information at different scales to improve the feature expression ability of the model. Ten-fold cross-validation was performed on LTAFDB, and external independent testing was conducted on AFDB and NSRDB datasets. RESULTS: DSC-AttNet achieved a ten-fold average accuracy of 97.33% and a precision of 97.30% on the test set, both of which outperformed the other 4 comparison models as well as the 3 classical models. The accuracy of the model on the external test set reached 92.78%, better than those of the 3 classical models. The number of parameters of DSC-AttNet was 1.01M, and the computational volume was 27.19G, both smaller than the 3 classical models. CONCLUSIONS This proposed method has a smaller complexity, achieves better classification performance, and has a better generalization ability for atrial fibrillation classification.
Atrial Fibrillation/diagnosis* ; Humans ; Electrocardiography ; Deep Learning ; Wearable Electronic Devices ; Neural Networks, Computer

OBJECTIVES: We propose a multi-scale jaw cyst segmentation model, AConvLSTM U-Net, which is based on bidirectional dense connections and attention mechanisms to achieve accurate automatic segmentation of mandibular cyst images. METHODS: A dataset consisting of 2592 jaw cyst images was used. AConvLSTM U-Net designs a MBC on the encoding path to enhance feature extraction capabilities. A DPD was used to connect the encoder and decoder, and a bidirectional ConvLSTM was introduced in the jump connection to obtain rich semantic information. A decoding block based on scSE was then used on the decoding path to enhance the focus on important information. Finally, a DS was designed, and the model was optimized by integrating a joint loss function to further improve the segmentation accuracy. RESULTS: The experiment with AConvLSTM U-Net for jaw cyst lesion segmentation showed a MCC of 93.8443%, a DSC of 93.9067%, and a JSC of 88.5133%, outperforming all the other comparison segmentation models. CONCLUSIONS The proposed algorithm shows a high accuracy and robustness on the jaw cyst dataset, demonstrating its superior performance over many existing methods for automatic segmentation of jaw cyst images and its potential to assist clinical diagnosis.
Humans ; Jaw Cysts/diagnostic imaging* ; Algorithms ; Image Processing, Computer-Assisted/methods* ; Neural Networks, Computer