Bo Yang

Accurate forecasting of citywide electric vehicle (EV) charging load is critical for alleviating station-level congestion, improving energy dispatching, and supporting the stability of intelligent transportation systems. However, large-scale EV charging networks exhibit complex and heterogeneous spatio-temporal dependencies, and existing approaches often struggle to scale with increasing station density or long forecasting horizons. To address these challenges, we develop a modular spatio-temporal prediction framework that decouples temporal sequence modeling from spatial dependency learning under an encoder–decoder paradigm. For temporal representation, we introduce a global aggregation mechanism that compresses multi-station time-series signals into a shared latent context, enabling efficient modeling of long-range interactions while mitigating the computational burden of cross-channel correlation learning. For spatial representation, we design a dynamic multi-scale attention module that integrates graph topology with data-driven neighbor selection, allowing the model to adaptively capture both localized charging dynamics and broader regional propagation patterns. In addition, a cross-step transition bridge and a gated fusion unit are incorporated to improve stability in multi-horizon forecasting. The cross-step transition bridge maps historical information to future time steps, reducing error propagation. The gated fusion unit adaptively merges the temporal and spatial features, dynamically adjusting their contributions based on the forecast horizon, ensuring effective balance between the two and enhancing prediction accuracy across multiple time steps. Extensive experiments on a real-world dataset of 18,061 charging piles in Shenzhen demonstrate that the proposed framework achieves superior performance over state-of-the-art baselines in terms of MAE, RMSE, and MAPE. Ablation and sensitivity analyses verify the effectiveness of each module, while efficiency evaluations indicate significantly reduced computational overhead compared with existing attention-based spatio-temporal models.

Spatial transcriptomics enables the measurement of gene expression in intact tissues. Despite this, reconstructing anatomically accurate spatial domains remains challenging, primarily due to expression sparsity, complex tissue architecture that is characterized by sharp boundaries and long-range continuity, and weak spatial signals. Traditional pipelines typically rely on expression-driven clustering and spatial smoothing, which underperform at boundaries and in sparse regions while neglecting morphological information. To address these challenges, AugGCL is proposed, an augmented graph-convolutional learning framework that enhances spatial structure decoding and gene expression reconstruction through targeted augmentation of both gene and image data. A key component of AugGCL is neighborhood information aggregation mechanism, which integrates expression similarity and spatial proximity to construct a weighted graph and an enhanced expression matrix, addressing sparsity without sacrificing boundary clarity. Additionally, a two stream weighted graph convolutional network jointly models refined gene features and image-derived morphological information, with image-aware auxiliary reconstructions enhancing weak spatial signals and sharpening boundaries. On datasets from the human dorsolateral prefrontal cortex, breast cancer, and mouse embryo, AugGCL outperforms baseline methods across multiple metrics, showing robustness and generalization across a range of datasets. Downstream analysis validated the reliability of the method, confirming its effectiveness in cell annotation, functional enrichment, and mechanistic studies. AugGCL generates clearer spatial domains and significantly advances the application of spatial transcriptomics in tissue structure and disease research.

Cancer, with its inherent heterogeneity, is commonly categorized into distinct subtypes based on unique traits, cellular origins, and molecular markers specific to each type. However, current studies primarily rely on complete multi-omics datasets for predicting cancer subtypes, often overlooking predictive performance in cases where some omics data may be missing and neglecting implicit relationships across multiple layers of omics data integration. This paper introduces Multi-Layer Matrix Factorization (MLMF), a novel approach for cancer subtyping that employs multi-omics data clustering. MLMF initially processes multi-omics feature matrices by performing multi-layer linear or nonlinear factorization, decomposing the original data into latent feature representations unique to each omics type. These latent representations are subsequently fused into a consensus form, on which spectral clustering is performed to determine subtypes. Additionally, MLMF incorporates a class indicator matrix to handle missing omics data, creating a unified framework that can manage both complete and incomplete multi-omics data. Extensive experiments conducted on 12 multi-omics cancer datasets, both complete and with missing values, demonstrate that MLMF achieves results that are comparable to or surpass the performance of several state-of-the-art approaches. MLMF is open source and available at (https://github.com/renyingxuan/MLMF.git).

Cancer is a highly diverse group of diseases, and each type of cancer can be further divided into various subtypes according to specific characteristics, cellular origins, and molecular markers. Subtyping helps in tailoring treatment and prognosis accuracy. However, the existing studies are more concerned with integrating different omics data to discover potential connections, but ignoring the relationships between consensus information and individual information within each omics level during the integration process. To this end, we propose a novel fusion-free method called multi-view multi-level contrastive graph convolutional network (M$^{2}$CGCN) for cancer subtyping. M$^{2}$CGCN learns multi-level features, i.e. high-level and low-level features, respectively. The low-level features from each view capture the intrinsic information in each omics by reconstruction of node attribute and graph structures. The high-level features achieve cancer subtyping via contrastive learning. Comprehensive experiments were performed on 34 multi-omics cancer datasets. The findings indicate that M$^{2}$CGCN achieves results comparable to or surpassing many state-of-the-art methods.

The frequent turnover of college graduates is a key factor leading to the frictional unemployment and structural unemployment of youth, which are important research fields concerned with pedagogy, sociology, and management; however, there is little research on the prediction of college graduates’ turnover. Therefore, this study investigated the turnover status of 17,268 college graduates from 52 universities in China, constructed and optimized a random forest model for predicting the turnover of college graduates, and analyzed the influencing mechanism of college graduates’ turnover and the importance of influencing factors. The enhanced random forest model could deal with the unbalanced data and has a higher prediction accuracy as well as stronger generalization ability in predicting the turnover of college graduates. Individual background variables, job characteristic variables, and work environment variables are all important factors influencing whether college graduates resign or not. The top five factors that affect the turnover of college graduates by more than 10% are income level, job satisfaction degree, job opportunities, and job matching degree. The conclusion of this study is conducive to improving the accuracy of turnover prediction, systematically exploring the influencing factors of college graduates’ turnover, and effectively guaranteeing the overall stability of youth employment.