Yulin Zhang

Hyperspectral unmixing (HU) aims to estimate endmembers and their corresponding abundances, a task commonly referred to as blind hyperspectral unmixing (BLU). Nonnegative matrix factorization (NMF) provides a unified framework for their joint estimation. It is widely assumed that more accurate endmember estimation leads to improved abundance estimation, enabling simultaneous optimization of both variables. However, this paper shows that, in practical noisy scenarios, the relationship between endmembers and abundances in NMF-based multi-variable joint optimization problems (NMF-based JOPs) is inherently coupled and significantly more complex, making it difficult to improve both estimation accuracies simultaneously. Furthermore, we demonstrate that the hard abundance sum-to-one constraint (ASC), commonly imposed in NMF-based JOPs, is inconsistent with realistic noisy conditions. To address these limitations, we propose a novel two-stage framework for BLU that decouples the refinement of endmembers and abundances. In the first stage, a strongly convex minimum-volume simplex model is employed to ensure robust and stable endmember extraction. In the second stage, we introduce a novel formulation, L₁_SoftASC, which promotes abundance sparsity and physical interpretability while improving convexity and robustness in abundance estimation. Experimental results on both synthetic and real benchmark datasets demonstrate that the proposed two-stage approach consistently outperforms existing single-stage NMF-based JOP methods in terms of both endmember and abundance estimation accuracy, while providing BLU with greater flexibility in handling ASC.

To support China’s dual-carbon goals and sustainability-oriented mitigation planning, this study develops an SSA–LSTM–Transformer framework for provincial carbon emission forecasting and interpretable driver analysis. Using panel data for 30 provinces from 2005 to 2022, SSA is employed for adaptive hyperparameter optimization, while the LSTM captures local temporal memory and the Transformer models long-range dependencies. Ablation tests and benchmarking against eight widely used models demonstrate that the proposed framework achieves the best overall performance on the held-out test set, with R² = 0.9911 and NRMSE = 0.0192. SHAP analysis indicates that a more carbon-intensive energy structure is associated with higher predicted emissions, whereas stronger technological innovation is associated with lower predicted emissions, and feature-importance patterns vary across development-stage groups. Forecast trajectories diverge during 2025–2035 and show a convergence tendency by 2050 under the model assumptions, informing differentiated near-term mitigation pathways and longer-term cross-regional coordination and technology diffusion. The results provide an interpretable evidence base for sustainability-oriented provincial decarbonization policies.

Traditional bulk RNA sequencing often masks cell‐to‐cell variability, leading to a loss of individual heterogeneity information. Single‐cell RNA sequencing (scRNA‐seq) preserves cellular heterogeneity by reverse‐transcribing, amplifying, as well as sequencing mRNA molecules from individual cells, enabling in‐depth studies of cell development, differentiation, and disease mechanisms. However, scRNA‐seq data are inherently high‐dimensional and noisy with prevalent dropout events, posing challenges for accurate clustering and subtype identification. To address these issues, this study proposes an Adaptive Fractional‐Order Total Variation Regularised Low‐Rank Representation (AFTV‐LRR) model that integrates adaptive fractional‐order total variation into the low‐rank representation framework. The proposed method reconstructs low‐rank subspace structures to learn cell similarities while preserving fine‐grained cellular features through fractional‐order gradient learning. The optimisation problem is efficiently solved using the Alternating Direction Method of Multipliers (ADMM), and spectral clustering is applied to the learnt similarity matrix for accurate cell type identification. Extensive experiments on 11 publicly available scRNA‐seq datasets demonstrate that AFTV‐LRR achieves competitive and often superior performance compared with eight representative single‐cell clustering algorithms in terms of Adjusted Rand Index (ARI) and Normalised Mutual Information (NMI). Visualisation with t‐SNE further confirms that the proposed model yields clearer inter‐cluster separations and higher intra‐cluster compactness. Moreover, marker gene analysis using the mouse embryo dataset supports the biological interpretability and robustness of the clustering results. Overall, this work provides an adaptive computational framework for improving the accuracy and reliability of single‐cell clustering analysis.

Medical image segmentation is crucial for clinical diagnosis and treatment planning. Although methods based on CNN, particularly U‐Net and its variants, have achieved remarkable success in automated segmentation tasks, they still face challenges in effectively capturing long‐range dependencies, refining multi‐level features, and efficiently integrating cross‐level information. To address these issues, we propose a novel U‐Net architecture incorporating a multi‐scale feature refinement mechanism (MFR‐UNet). This network enhances segmentation accuracy and robustness by integrating three innovative modules. First, we designed a wavelet transform convolution (WtConv) module. By decomposing, processing, and reconstructing features in the frequency domain, this module enables the model to learn high‐frequency details and low‐frequency contours with greater precision. Second, we introduce a large receptive field attention (LRFA) module in the encoder. Combining deep separable convolutions with multi‐head attention, LRFA efficiently captures global contextual information at low computational cost. Finally, in the skip connections and decoding path, our weighted contextual fusion module (WCF) module dynamically generates channel attention weights for one feature stream to another, achieving efficient adaptive feature fusion. Simulation experiments on multiple public medical image segmentation datasets demonstrate that our MFR‐UNet outperforms several existing mainstream methods in key metrics such as Dice coefficient and IoU, proving its effectiveness in enhancing segmentation accuracy and boundary clarity.

Magnetic resonance imaging (MRI) has a pivotal role in both pretreatment staging and post‐treatment evaluation of rectal cancer. This study presents an innovative deep learning model, CAAFE‐ResNet18*, based on the residual neural network ResNet18*. The model features an ingeniously designed feature extraction and complementation module (i.e., CAAFE), which leverages a multiscale dilated convolution parallel architecture combined with a channel attention mechanism (CAM) to achieve multilevel information fusion, spatial feature enhancement and channel feature optimisation. This enables in‐depth exploration and augmentation of multilevel downsampled features, significantly improving feature representation capability and overall performance. Testing on rectal cancer MRI data demonstrates that the CAAFE‐ResNet18* model significantly outperforms convolutional neural network (CNN) backbone networks and recent state‐of‐the‐art (SOTA) models. This result indicates that the CAAFE model, by complementing and extracting MR images of patients with locally advanced rectal cancer (LARC) features at different scales from ResNet18*, may help to identify patients who will show complete response (CR) at the end of treatment and those who will not respond to therapy (NR) at an early stage of the treatment.

Single-cell RNA sequencing is a valuable technique for identifying diverse cell subtypes. A key challenge in this process is that the detection of rare cells is often missed by conventional methods due to low abundance and subtle features of these cells. To overcome this, we developed SCLCNF (Local Connectivity Network Feature Sharing in Single-Cell RNA sequencing), a novel approach that identifies rare cells by analyzing features uniquely expressed in these cells. SCLCNF creates a cellular connectivity network, considering how each cell relates to its neighbors. This network helps to pinpoint coexpression patterns unique to rare cells, utilizing a rarity score to confirm their presence. Our method performs better in detecting rare cells than existing techniques, offering enhanced robustness. It has proven to be effective in human gastrula data sets for accurately pinpointing rare cells, and in sepsis data sets where it uncovers previously unidentified rare cell populations.

Single-cell RNA sequencing (scRNA-seq) is currently an important technology for identifying cell types and studying diseases at the genetic level. Identifying rare cell types is biologically important as one of the downstream data analyses of single-cell RNA sequencing. Although rare cell identification methods have been developed, most of these suffer from insufficient mining of intercellular similarities, low scalability, and being time-consuming. In this paper, we propose a single-cell similarity division algorithm (scSID) for identifying rare cells. It takes cell-to-cell similarity into consideration by analyzing both inter-cluster and intra-cluster similarities, and discovers rare cell types based on the similarity differences. We show that scSID outperforms other existing methods by benchmarking it on different experimental datasets. Application of scSID to multiple datasets, including 68K PBMC and intestine, highlights its exceptional scalability and remarkable ability to identify rare cell populations.

Dysregulation of microRNAs (miRNAs) is closely associated with refractory human diseases, and the identification of potential associations between small molecule (SM) drugs and miRNAs can provide valuable insights for clinical treatment. Existing computational techniques for inferring potential associations suffer from limitations in terms of accuracy and efficiency. To address these challenges, we devise a novel predictive model called RPCA$\Gamma $NR, in which we propose a new Robust principal component analysis (PCA) framework based on $\gamma $-norm and $l_{2,1}$-norm regularization and design an Augmented Lagrange Multiplier method to optimize it, thereby deriving the association scores. The Gaussian Interaction Profile Kernel Similarity is calculated to capture the similarity information of SMs and miRNAs in known associations. Through extensive evaluation, including Cross Validation Experiments, Independent Validation Experiment, Efficiency Analysis, Ablation Experiment, Matrix Sparsity Analysis, and Case Studies, RPCA$\Gamma $NR outperforms state-of-the-art models concerning accuracy, efficiency and robustness. In conclusion, RPCA$\Gamma $NR can significantly streamline the process of determining SM–miRNA associations, thus contributing to advancements in drug development and disease treatment.