Spatial transcriptomics provides a direct view of tissue architecture, but ligand-receptor (LR) communication is often inferred with spatial neighborhoods or distance kernels that are shared across many LR pairs. This is limiting because contact-dependent, paracrine and chemokine-like niche programs can require distinct effective spatial supports. We present ScaleComm, a graph-wavelet framework for event-level spatial cell-cell communication inference. ScaleComm uses a tissue graph as a multi-scale scaffold, learns an LR-specific scale mixture and reports a dataset-specific scale profile for each LR pair, while scoring candidate sender-receiver-LR events with joint sender-receiver context features. In a mechanism-aware semi-synthetic benchmark, ScaleComm improved the balance between LR recovery, event localization and suppression of mechanism-mismatched decoys relative to representative spatial and cell-type-level baselines. Across tumor, inflammatory and lymphoid spatial datasets, ScaleComm produced localized candidate events whose receiver maps were supported by downstream response signatures and tissue-region enrichment, including TGFB1-TGFBR2 remodeling in colorectal cancer and CXCL13-CXCR5 enrichment in germinal-center regions. ScaleComm provides inspectable scale profiles, spatially localized candidate events and response-based validation readouts for studying LR communication in tissue context.

Tabular anomaly detection (TAD), which focuses on identifying abnormal samples that deviate from the majority in tabular data, has received growing attention. Recently, there has been an emerging trend towards unified TAD, which seeks to detect anomalies across different datasets using a single generalizable model. In unified TAD, aligning heterogeneous data remains challenging. While existing methods often rely on distance-based unified feature construction, they may obscure the semantics of the original features. Moreover, existing approaches typically formulate anomaly detection as a binary classification task, which may overlook diverse anomaly patterns from various datasets and be misled by unrepresentative synthetic anomalies. To address these challenges, we propose an in-COntext REconstruction approach for unified TAD (CORE for short). It introduces a decorrelated feature alignment module to directly align heterogeneous features into a unified representation space, which retains their semantic information. Meanwhile, CORE formulates unified TAD as an in-context reconstruction problem, eliminating the need for labeled or synthesized anomalies. Specifically, the in-context reconstruction module reconstructs each sample by leveraging contextual normal samples to capture dataset-specific distributions, such that reconstruction errors reflect its deviation from normality, facilitating unified TAD on arbitrary unseen datasets.

Graph anomaly detection (GAD) is critical for identifying abnormal nodes in graph-structured data from diverse domains, including cybersecurity and social networks. The existing GAD methods often focus on the learning paradigms of "one-model-for-one-dataset", requiring dataset-specific training for each dataset to achieve optimal performance. However, this paradigm suffers from limitations, such as high computational and data costs, limited generalization and transferability to new datasets, and challenges in privacy-sensitive scenarios where access to full datasets or sufficient labels is restricted. To address these limitations, we propose a novel generalist GAD paradigm that aims to develop a unified model capable of detecting anomalies on multiple unseen datasets without retraining/fine-tuning or customization. To this end, we propose a few-shot generalist GAD method with three key designs, namely feature Alignment, a Residual encoder, and in-Context learning, abbreviated as ARC. As a generalist approach, ARC only requires a few labeled normal samples during prediction on any unseen graphs. Specifically, ARC consists of three modules: a feature Alignment module to unify and align features across datasets, a Residual graph encoder to capture dataset-agnostic anomaly representations, and a cross-attentive in-Context learning module to score anomalies using few-shot normal context. Building on ARC, we further introduce ARC(zero ) for the zero-shot generalist GAD setting, which selects representative pseudo-normal nodes via a pseudo-context mechanism and thus enables fully label-free inference on unseen datasets. Experiments on 17 real-world datasets demonstrate that ARC and ARC(zero) effectively detect anomalies, exhibit strong generalization ability, and perform efficiently under few-shot and zero-shot settings.

Relational databases are widely used for managing structured data in real-world systems. Detecting anomalies from such relational data is crucial for identifying fraud, risks, and abnormal behaviors, yet remains under-explored. The key challenges lie in the intrinsic complexity of relational data: multi-table attributes are high-dimensional and heterogeneous, making sparse abnormal clues easy to overwhelm by normal or irrelevant information; and anomalies may further manifest as abnormal connection patterns across different foreign-key relations, which existing tabular and graph anomaly detection methods are ill-suited to capture. To address them, we propose RelAD, a reconstruction-based framework that captures anomalies from both attribute and relational edge reconstruction. RelAD contains two core modules: conditional sparse-gated attribute reconstruction, which suppresses redundant multi-table attributes and emphasizes abnormal semantic blocks, and dual-view multi-relational edge reconstruction, which detects relation-specific abnormal connections from both intrinsic and behavioral entity profiles. The resulting attribute and relational signals are integrated through a lightweight fusion module to produce the final anomaly score. We further construct 6 benchmark datasets with systematic anomalies, on which extensive experiments show that RelAD consistently outperforms other baselines while achieving competitive efficiency.

The development of single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) has significantly advanced the study of cell heterogeneity in the epigenetic landscape. Numerous studies have leveraged scATAC-seq data to explore deeper gene regulatory relationships. However, scATAC-seq usually faces dropout events which may result in data sparsity and noise. In this work, we propose a method (scMCG) for analyzing scATAC-seq data that employs contrastive learning and a generative adversarial network (GAN). First, the scMCG method uses two distinct encoders for contrastive learning to solve the issues of feature redundancy and data sparsity in scATAC-seq data. Subsequently, a generator is used to reconstruct the latent embedding. Finally, a decoder is used to generate binary accessibility. We conduct experiments on multiple scATAC-seq datasets. The results demonstrate that the scMCG method achieves excellent performance in multiple tasks such as cell clustering and transcription factor activity influence.

Breast cancer is a major global health threat to women. Its molecular heterogeneity leads to distinct subtypes with different treatments. Traditional diagnostic methods using either single-omics data or whole-slide imaging (WSI) cannot fully capture tumor complexity. Although multimodal models have emerged, most rely on simple feature concatenation, ignoring differences between omics and imaging data. Moreover, missing values in omics data further limit clinical applicability. In this work, we introduce a Deep Incomplete Multi-view Information Recovery framework (DIMIR) for breast cancer subtype classification. DIMIR utilizes cross-attention and latent self-attention mechanisms to model relationships between multi-omics views and reconstruct missing omics views with a context-aware way. Then, it reconstructs missing omics views by learning high-level semantic representations across multiple views. Further, the reconstructed multi-omics views are used to train a subtype classification network that fully exploits multimodal discriminative information. Experiments on the Cancer Genome Atlas (TCGA) breast cancer demonstrate that DIMIR significantly outperforms state-of-the-art methods in cancer subtypes prediction and identifies biomarkers closely associated with cancer subtypes.

Recent contrastive multi-view clustering methods have achieved remarkable performance by using two-branch contrastive learning. However, most existing studies focus on the optimization of the false negatives (FNs) identification strategy, ignoring the critical issue of cluster center alignment between fused view and single views. To address this limitation, we present a robust multi-view clustering method (CAFE) based on cross-view adaptive fusion and cluster center enhancement. Specifically, we first design a cross-view adaptive fusion module that incorporates dual weights at both the view level and the sample level, enabling effective coordination of consistent and complementary information across views. Subsequently, we propose a dual-driven cluster center enhancement framework to refine cluster structures. It introduces a dual alignment mechanism between single-view cluster centers and fused-view cluster centers to systematically coordinate view-specific discriminative patterns and cross-view consensus representations. Furthermore, we develop a second-order proximity graph embedding method to more effectively rectify FNs by computing neighborhood similarity. It constructs second-order proximity to identify structurally related samples that may be spatially distant in feature space. Extensive experiments on six widely used multi-view benchmark datasets demonstrate that CAFE achieves state-of-the-art performance under both complete and incomplete multi-view scenarios.

Graph Neural Networks (GNNs) are widely used in real-world applications for their ability to process graph-structured data. However, they can amplify biases in training data, leading to unfair predictions against groups defined by sensitive attributes (e.g., race or gender). To ensure fairness in GNNs, most existing methods focus on removing information related to sensitive attributes from nodes. However, these approaches often lead to the unintended removal of useful information, resulting in a significant utility loss. To address this limitation, we propose FairNCI, a novel fair graph representation learning method that leverages Neighbor Counterfactual Information (NCI). It employs labels and sensitive attributes to directly identify two types of counterfactual neighbors for each center node: sensitive attribute counterfactual neighbors and label counterfactual neighbors. By leveraging adversarial learning to guide the learning of both types of NCI and integrating them into the center node's representation, FairNCI effectively mitigates bias while preserving valuable information. Experiments on multiple real-world datasets demonstrate that FairNCI achieves a favorable balance between predictive utility and fairness, offering a promising approach to fair graph representation learning.

Graph-level anomaly detection (GLAD) is crucial for ensuring the reliability of graph-driven applications by identifying abnormal graphs that deviate from the majority. Considering the privacy concerns in distributed scenarios, federated graph-level anomaly detection (FedGLAD) has emerged as a promising solution to enable collaborative detection without sharing raw data. However, existing methods suffer from poor generalization due to the reliance on unrealistic synthetic anomalies and insufficient personalization capabilities under data heterogeneity. To address these challenges, we propose a novel Federated graph-level anomaly detection approach with Cluster-adaptIve GAted Reconstruction (FedCIGAR). Specifically, we design a reconstruction-based paradigm trained on normal graphs to avoid synthetic data. Furthermore, we introduce a client-side node contribution gating mechanism and a server-side sliding window-based clustering strategy to tackle data heterogeneity. Extensive experiments demonstrate that FedCIGAR achieves superior performance and robustness compared to state-of-the-art methods.

Accurate prediction of drug-target affinities (DTA) is critical for drug discovery. However, this task remains a significant challenge due to the complexity of modeling interactions between small ligands and large targets. In this study, we propose a multi-modal deep learning framework (CrossSG-DTA) to predict drug-target affinity by integrating sequence semantics with graph structural information. We leverage ChemBERTa and ESM-2 to extract rich semantic features for drugs and targets, respectively. In addition, a modified Graph Convolutional Network (GCN) is utilized to simultaneously capture structural data. To effectively fuse these heterogeneous features, we design a new symmetric dual cross-attention fusion mechanism for drugs and targets. This mechanism enables the model to capture complex dependencies between global sequence representations and local topological structures. Subsequently, the fused drug and target features are concatenated and fed into a three-layer Multi-Layer Perceptron (MLP) to obtain the final binding affinity. Experimental results on the Davis and KIBA datasets demonstrate that CrossSG-DTA significantly outperforms state-of-the-art methods. Finally, a case study on a glaucoma-related target highlights the practical utility of our model as a powerful in silico tool for DTA tasks.

Background:Hepatocellular carcinoma (HCC) is a major global health burden, with persistently low 5-year survival rates. The major clinical challenge in HCC lies in the lack of accurate biomarkers for early detection, prognostic evaluation, and prediction of tumor aggressiveness. This study aimed to identify reliable biomarkers for HCC diagnosis and outcome prediction, and to elucidate their functional relevance in tumor progression. Methods:Transcriptomic profiles of HCC and adjacent normal tissues, along with single-cell RNA sequencing (scRNA-seq) data from six HCC patients, were obtained from public databases. The differentially expressed genes (DEGs) were identified and functionally annotated. HCC tumor cell-specific genes were determined using the "FindAllMarkers" function. Genes that were upregulated in both HCC tissues and tumor cells were considered candidate biomarkers. Western blotting and quantitative reverse transcription polymerase chain reaction (qRT-PCR) were used to assess spondin-2 (SPON2) and WNT-related gene expression. Cell counting kit-8 (CCK-8), 5-ethynyl-2'-deoxyuridine (EdU), as well as colony formation assays evaluated cell proliferation. Additionally, wound healing and transwell assays assessed migration and invasion capacities. Results:Integration of bulk transcriptomic and scRNA-seq datasets identified SPON2 as an HCC-specific biomarker. SPON2 showed favorable diagnostic accuracy in distinguishing HCC patients from healthy controls across four independent cohorts, and its elevated expression was linked to poor overall survival (OS) and increased metastatic potential. At the single-cell level, SPON2+ tumor cells showed higher invasion and migration scores. Functional assays confirmed that SPON2 promoted the proliferative, migrative, and invasive capacities of HCC cells in vitro. Moreover, SPON2 expression exhibited a robust positive correlation with the activation of WNT signaling pathway; while SPON2 knockdown suppressed WNT-related protein expression. Conclusions:Through integrative transcriptomic and single-cell analyses, SPON2 was identified as a robust HCC biomarker, aberrantly upregulated in HCC tissues and tumor cells, and predictive of shorter OS and enhanced metastatic potential. By activating the WNT signaling pathway, SPON2 enhances the proliferative, migrative, and invasive capacities of HCC cells, underscoring its pivotal role in tumor aggressiveness and providing insights into novel diagnostic and prognostic targets for HCC management.

Recommender systems aim to predict user feedback on unseen items, but confounding bias, particularly from latent confounders, presents a major challenge. Existing debiasing methods in recommender systems often overlook the complex interplay among multiple features and the subtleties of user preferences. To address this, we propose a novel framework called Multi-View-based Identifiable Debiased Learning (MViDL) for recommendations, even in the presence of latent confounders. Specifically, MViDL first employs a multi-view framework to discern interactions between user and item features, unearth user interests in specific items, and capture fundamental user and item ID information. To mitigate the effects of latent confounders, MViDL incorporates the identifiable Variational Auto-Encoder (iVAE) to efficiently infer the latent representation from a set of proxy variables and adjusts for the learned latent representation to mitigate confounding bias. We further provide a theoretical analysis of the identifiability of the latent representations. Extensive evaluations on three real-world datasets highlight the superiority of MViDL. Specifically, our approach achieves average improvements of approximately 6.03% and 5.70% in NDCG@K and Recall@K over the state-of-the-art (SOTA) baselines on Coat, 3.54% and 2.29% on Yahoo!R3, and 1.49% and 2.47% on KuaiRand.

Text anomaly detection (TAD) plays a critical role in various language-driven real-world applications, including harmful content moderation, phishing detection, and spam review filtering. While two-step “embedding–detector” TAD methods have shown state-of-the-art performance, their effectiveness is often limited by the use of a single embedding model and the lack of adaptability across diverse datasets and anomaly types. To address these limitations, we propose to exploit the embeddings from multiple pretrained language models and integrate them into MCA ^2 , a multi-view TAD framework. MCA ^2 adopts a multi-view reconstruction model to effectively extract normal textual patterns from multiple embedding perspectives. To exploit inter-view complementarity, a contrastive collaboration module is designed to leverage and strengthen the interactions across different views. Moreover, an adaptive allocation module is developed to automatically assign the contribution weight of each view, thereby improving the adaptability to diverse datasets. Extensive experiments on 10 benchmark datasets verify the effectiveness of MCA ^2 against strong baselines. The source code of MCA ^2 is available at https://github.com/yankehan/MCA2 .

With the rapid proliferation of graph-structured data in critical domains, unsupervised graph-level anomaly detection (GLAD) has emerged as a key task in graph analysis. The goal of GLAD is to identify graph structures that deviate significantly from normal patterns. However, existing graph neural network (GNN) models often fail to account for data biases, such as feature-level and structural-level data biases, which can obscure true anomalies and severely degrade model generalization. To address this issue, we present DeGlad, a novel Debiased representation learning framework that explicitly tackles bias effects on GLAD tasks. Specifically, DeGlad decomposes the input graph into core subgraphs and environmental subgraphs to disentangle relevant structures from bias influences. To further mitigate feature-level bias, we introduce counterfactual sample generation, effectively eliminating spurious correlations and improving model reliability. In particular, DeGlad employs a structural-level de-biasing mechanism that mixes environmental subgraphs to generate counterfactual samples, enhancing generalization across diverse distributions. Experiments on eight real-world datasets demonstrate that DeGlad significantly outperforms existing GLAD methods, showcasing superior anomaly detection performance.

Graph Neural Networks (GNNs) are widely applied to graph-structured data, but they often suffer from fairness concerns, as inherent biases in node attributes and graph topology can result in discriminatory predictions. Existing approaches typically attempt to mitigate all sources of bias within a single, entangled representation, thereby limiting the effectiveness of debiasing. To overcome this limitation, we propose FairGID, a novel framework for fair graph representation learning that enhances fairness by separating topology from node attributes and disentangling node representations. Specifically, FairGID first learns attribute-only and structure-only representations independently. It then further disentangles the attribute representations into multiple latent factors and applies sensitive attribute masking to suppress bias-related information. Finally, an adversarial fusion module integrates the attribute and structural representations into a unified embedding that is both informative and fair. Extensive experiments on five real-world datasets demonstrate that FairGID achieves a superior accuracy-fairness trade-off compared with state-of-the-art baselines, highlighting its potential as an effective solution for fair graph representation learning.

tRNA-derived small RNAs (tsRNAs) have emerged as a novel class of regulatory molecules implicated in the pathogenesis of numerous human diseases, positioning them as promising biomarkers and therapeutic targets. Existing computational methods provide a cost-effective alternative to experimental method, but they tend to ignore biological attributes and complex feature interactions. To overcome these limitations, we propose ERFMTDA, an enhanced rotative factorization machine framework for predicting potential tsRNA-disease associations. ERFMTDA explicitly models complex interactions among heterogeneous biological features while integrating latent structural representations derived from the global association matrix. In addition, a biologically informed negative sampling strategy based on motif-level sequence similarity is introduced to improve the reliability of negative samples. Extensive experiments demonstrate that ERFMTDA consistently surpasses the other eleven state-of-the-art methods. Two case studies on diabetic retinopathy and hepatocellular carcinoma further corroborate the model's ability to prioritize biologically meaningful tsRNA-disease associations.

Graph neural networks (GNNs) are essential for graph classification. Recent studies have shown that graph transformer (GT) outperforms vanilla GNNs in learning global graph representations. GT effectively filters out irrelevant neighbours by applying attention mechanisms to identify highly relevant nodes, thereby enhancing its performance. However, their high complexity and substantial memory requirements make them impractical for large-scale applications. Although linear GT and its variants attempt to mitigate these costs through kernel tricks, their performance is inferior to that of vanilla GT. Some variants aim to enhance accuracy by incorporating parameters; however, this can increase computational overhead and compromise linear efficiency. This article proposes LiSA, a linear GT framework with sample aggregation. Its core innovation is replacing traditional node-neighbor attention with node-neighbor-cluster attention via sample aggregation. Further, mathematical refinements reduce the number of parameters without compromising accuracy, resulting in an average running time reduction of 16.4%. Extensive experiments on eight benchmark datasets demonstrated that LiSA is an efficient framework that accurately approximates the attention distribution with high robustness. The results demonstrate that our method outperforms 14 baseline methods and matches or exceeds the state-of-the-art performance. Code is available at: https://github.com/cirila9/LiSA.

Graph Neural Networks (GNNs) excel at modeling graph data but often amplify biases tied to sensitive attributes like gender and race. Existing causality-based methods use isolated interventions on graph topology or features but struggle to produce representations that balance predictive power with fairness. This leads to two issues: (1) weak predictive power, where representations miss critical task-relevant features, and (2) bias amplification, where representations encode sensitive attributes, causing unfair outcomes. To address these issues, we introduce the Probability of Necessity and Sufficiency (PNS), where necessity ensures representations capture only essential features for predictions, and sufficiency guarantees these features are adequate without relying on sensitive attributes. We propose FairSNR, a fairness-aware graph representation learning framework that introduces constraints based on the PNS. This leverages PNS to guide the learning of fair representations from graph data. In particular, FairSNR employs an encoder to learn node representations with high PNS for downstream tasks. To compute and optimize PNS, FairSNR introduces an intervenor to generate the most challenging counterfactual interventions on the representations, thereby enhancing the model’s causal stability even under worst-case scenarios. Further, a discriminator is trained to detect and mitigate sensitive information leakage in the learned representations, effectively disentangling sensitive biases from task-relevant features. Experiments on real-world graph datasets demonstrate that FairSNR outperforms existing state-of-the-art (SOTA) methods in both fairness and utility.

By generating massive gene transcriptome data and analyzing transcriptomic variations at the cell level, single-cell RNA-sequencing (scRNA-seq) technology has provided new way to explore cellular heterogeneity and functionality. Clustering scRNA-seq data could discover the hidden diversity and complexity of cell populations, which can aid to the identification of the disease mechanisms and biomarkers. In this paper, a novel method (DSINMF) is presented for clustering single cell RNA sequencing data by using deep matrix factorization. Our proposed method comprises four steps: first, the feature selection is utilized to remove irrelevant features. Then, the dropout imputation is used to handle missing value problem. Further, the dimension reduction is employed to preserve data characteristics and reduce noise effects. Finally, the deep matrix factorization with bi-stochastic graph regularization is used to obtain cluster results from scRNA-seq data. We compare DSINMF with other state-of-the-art algorithms on nine datasets and the results show our method outperformances than other methods. The code can be downloaded from https://github.com/lanbiolab/DSINMF.