Existing learning-based arbitrary-scale point cloud upsampling methods are usually challenged with limited point cloud feature representation and noise-sensitive refinement of coarse point cloud. In this paper, we introduce GraphDAE-PU, a novel framework for point cloud upsampling that addresses the challenges associated with arbitrary-scale point cloud upsampling. Our approach integrates graph representation with a denoising auto-encoder architecture, harnessing the intrinsic structure of point clouds to enhance feature extraction and refine the upsampled points. Specifically, the use of a graph neural network enables our model to capture complex geometric relationships within the data, thereby improving the accuracy of the generated point clouds. Concurrently, the denoising auto-encoder refines coarse, midpoint-interpolated point clouds, resulting in surfaces that are not only accurate and uniform but also robust to noise. Experimental evaluations, both qualitative and quantitative, reveal that GraphDAE-PU consistently outperforms existing methods, demonstrating its accuracy and robustness in arbitrary-scale upsampling tasks.

Concept drift remains a critical challenge in non-stationary industrial fault diagnosis processes, where evolving operational conditions induce continuous distributional shifts and hinder model generalization. While existing continual learning approaches primarily focus on mitigating forgetting, they often lack explicit mechanisms to detect and quantify drift in a timely manner. To address this, this paper proposes a novel multi-scale feature-probability consistency-based drift detection (MFPC-DCD) framework. It comprises three core components: (1) a feature-level divergence module that leverages Maximum Mean Discrepancy (MMD) to measure latent representation shifts; (2) a probability-level divergence module utilizing Jensen–Shannon divergence to quantify predictive inconsistencies; and (3) a multi-scale consistency fusion strategy that aggregates drift signals across short-, mid-, and long-term temporal windows, thereby enhancing robustness and temporal granularity. The MFPC-DCD method can explicitly detect domain shifts by jointly capturing structural and semantic distributional variations through multi-resolution analysis and then the resulting unified drift coefficient provides precise quantification of domain shifts and can be seamlessly integrated into downstream continual learning fault diagnosis for adaptive regularization. Experiments conducted on two fault diagnosis datasets — CWRU, and our proprietary DPS — demonstrate that MFPC-DCD achieves state-of-the-art drift detection performance and adaptability under diverse drift scenarios.

Recent works try to combine clustering and contrastive learning for unsupervised out-of-distribution (OOD) detection, since these two schemes can exploit semantic information and bring in discriminative representation learning. However, most methods based on clustering and contrastive learning struggle with the problems of hard assignment and low-level clustering, i.e., they usually assign each sample to one single cluster and obtain clusters of similar low-level features, which can easily bring in numerous incorrect assignments and hinder the learning of semantic information. To address these problems, this paper proposes a novel framework for unsupervised OOD detection named Soft Cluster-aware Equivariant Contrastive Learning (SCECL). Different from previous works, SCECL devises two modules named Soft Cluster-aware Semantic Relationship Mining (SCSRM) and Contrastive Learning with Invariance and Equivariance (CLIE): SCSRM assigns each sample to multiple clusters with soft assignment weights and utilizes the soft assignment weights with semantic relationships to guide unsupervised contrastive learning for OOD detection, while CLIE introduces the equivariance principle as an additional inductive bias, encouraging the model to learn more discriminative semantic features to avoid low-level clustering. Extensive experimental results on various OOD detection benchmarks demonstrate that the proposed SCECL can effectively utilize semantic information for discriminative representation learning and achieve state-of-the-art performance.

Channel-spatial attention mechanisms have been extensively investigated in computer vision. However, it is still a difficult problem that how to efficiently utilize global and local contextual information laid in a feature tensor to generate an accurate 3D attention map. This paper proposes a novel attention module for convolutional neural networks named Dual Rank-1 Tensor Attention Module, which can reach a good balance between global and local contextual information utilization for attention map generation. In our module, given a feature tensor, we sequentially generate two rank-1 3D tensor attention maps, i.e., the initial rank-1 tensor attention map containing global contextual information, and the complement rank-1 tensor attention map containing partial local contextual information. Then, we obtain a 3D tensor attention map based on the combination of these two rank-1 tensor attention maps for feature recalibration. Experimental results on ImageNet-1K and PASCAL VOC datasets demonstrate that the proposed module can achieve competitive performance compared with other state-of-the-art attention modules. The source code will be available at https://github.com/KevinBHLin/.

Multi-agent collaborative Simultaneous Localization and Mapping (SLAM) is an effective way for large-scale mapping. However, this approach, which relies on Visual-Inertial Odometry(VIO) as input, suffers from limitations such as susceptibility to environmental influences and the difficulty in accurately constructing dense 3D maps. To address these challenges, this paper presents Multi-Estimation Robust SLAM System (MERSYS), a novel framework for three-dimensional dense mapping based on the fusion of Lidar-Inertial Odometry(LIO) and VIO. Benefiting from lower communication’s costs and dense information acquisition capability, the proposed framework aims to achieve compatibility in processing both LIO and VIO inputs, establish joint loop closure detection to enable multi-map fusion, and then create a comprehensive global 3D dense point cloud map. Furthermore, an efficient communication strategy has been proposed to enable bidirectional transmission of dense and voluminous data. Experimental evaluations conducted on the publicly available HILTI SLAM 2021 dataset [10] as well as a real world dataset. Experimental results show that MERSYS achieves better results than state-of-the-art methods. The source code is available on the GitHub1.

MOTIVATION: Single-cell DNA methylation sequencing can assay DNA methylation at single-cell resolution. However, incomplete coverage compromises related downstream analyses, outlining the importance of imputation techniques. With a rising number of cell samples in recent large datasets, scalable and efficient imputation models are critical to addressing the sparsity for genome-wide analyses. RESULTS: We proposed a novel graph-based deep learning approach to impute methylation matrices based on locus-aware neighboring subgraphs with locus-aware encoding orienting on one cell type. Merely using the CpGs methylation matrix, the obtained GraphCpG outperforms previous methods on datasets containing more than hundreds of cells and achieves competitive performance on smaller datasets, with subgraphs of predicted sites visualized by retrievable bipartite graphs. Besides better imputation performance with increasing cell number, it significantly reduces computation time and demonstrates improvement in downstream analysis. AVAILABILITY AND IMPLEMENTATION: The source code is freely available at https://github.com/yuzhong-deng/graphcpg.git.

Anomaly detection aims to separate anomalies from normal samples, and the pretrained network is promising for anomaly detection. However, adapting the pretrained features would be confronted with the risk of pattern collapse when finetuning on one-class training data. In this paper, we propose an anomaly detection framework called constrained adaptive projection with pretrained features (CAP). Combined with pretrained features, a simple linear projection head applied on a specific input and its k most similar pretrained normal representations is designed for feature adaptation, and a reformed self-attention is leveraged to mine the inner-relationship among one-class semantic features. A loss function is proposed to avoid potential pattern collapse. Concretely, it considers the similarity between a specific data and its corresponding adaptive normal representation, and incorporates a constraint term slightly aligning pretrained and adaptive spaces. Our method achieves state-of-the-art anomaly detection performance on semantic anomaly detection and sensory anomaly detection benchmarks including 96.5% AUROC on CIFAR-100 dataset, 97.0% AUROC on CIFAR-10 dataset and 89.9% AUROC on MvTec dataset.

Advances in high throughput sequencing have enabled DNA methylation profiling at single-cell resolution. The generation of single-cell methylation sequencing (scM-Seq) data provides unprecedented opportunities for a comprehensive dissection of epigenetic heterogeneity. An important step of exploring epigenetic heterogeneity is clustering cells according to their single-cell methylation profiles. However, the inherent sparsity and stochastic measurement characteristic of the data make it challenging. To this end, we introduce SINCEF, using spectral embedding fusion to reconstruct cell-to-cell pairwise distance for clustering single-cell methylation data. SIN CEF first calculates multiple basic distance matrices to capture cell-to-cell methylation dissimilarity relationships according to the global methylation status. Then it adopts spectral embedding to transform these basic distance matrices into the latent representations, pooling information from the basic distance measures. Finally, it reconstructs a novel distance matrix and implements hierarchical clustering to yield cell partitions. Assessments on several public scM-Seq datasets demonstrated that SINCEF could generate a more appropriate distance matrix to measure the methylation distance between cells, which considerably improved the clustering performance. As an additional benefit, the reconstructed novel distance matrix could help to visually assess the heterogeneity across cell populations through presenting the block structures in the hierarchical clustering heat maps. SINCEF is freely available on GitHub at https://github.com/TQBio/SINCEF.

MOTIVATION: Single-cell DNA methylation sequencing detects methylation levels with single-cell resolution, while this technology is upgrading our understanding of the regulation of gene expression through epigenetic modifications. Meanwhile, almost all current technologies suffer from the inherent problem of detecting low coverage of the number of CpGs. Therefore, addressing the inherent sparsity of raw data is essential for quantitative analysis of the whole genome. RESULTS: Here, we reported CaMelia, a CatBoost gradient boosting method for predicting the missing methylation states based on the locally paired similarity of intercellular methylation patterns. On real single-cell methylation datasets, CaMelia yielded significant imputation performance gains over previous methods. Furthermore, applying the imputed data to the downstream analysis of cell-type identification, we found that CaMelia helped to discover more intercellular differentially methylated loci that were masked by the sparsity in raw data, and the clustering results demonstrated that CaMelia could preserve cell-cell relationships and improve the identification of cell types and cell subpopulations. AVAILABILITY AND IMPLEMENTATION: Python code is available at https://github.com/JxTang-bioinformatics/CaMelia. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: DNA methylation signatures in rheumatoid arthritis (RA) have been identified in fibroblast-like synoviocytes (FLS) with Illumina HumanMethylation450 array. Since <2% of CpG sites are covered by the Illumina 450K array and whole genome bisulfite sequencing is still too expensive for many samples, computationally predicting DNA methylation levels based on 450K data would be valuable to discover more RA-related genes. RESULTS: We developed a computational model that is trained on 14 tissues with both whole genome bisulfite sequencing and 450K array data. This model integrates information derived from the similarity of local methylation pattern between tissues, the methylation information of flanking CpG sites and the methylation tendency of flanking DNA sequences. The predicted and measured methylation values were highly correlated with a Pearson correlation coefficient of 0.9 in leave-one-tissue-out cross-validations. Importantly, the majority (76%) of the top 10% differentially methylated loci among the 14 tissues was correctly detected using the predicted methylation values. Applying this model to 450K data of RA, osteoarthritis and normal FLS, we successfully expanded the coverage of CpG sites 18.5-fold and accounts for about 30% of all the CpGs in the human genome. By integrative omics study, we identified genes and pathways tightly related to RA pathogenesis, among which 12 genes were supported by triple evidences, including 6 genes already known to perform specific roles in RA and 6 genes as new potential therapeutic targets. AVAILABILITY AND IMPLEMENTATION: The source code, required data for prediction, and demo data for test are freely available at: http://wanglab.ucsd.edu/star/LR450K/ CONTACT: [email protected] or [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Over 50% of human genes contain CpG islands in their 5'-regions. Methylation patterns of CpG islands are involved in tissue-specific gene expression and regulation. Mis-epigenetic silencing associated with aberrant CpG island methylation is one mechanism leading to the loss of tumor suppressor functions in cancer cells. Large-scale experimental detection of DNA methylation is still both labor-intensive and time-consuming. Therefore, it is necessary to develop in silico approaches for predicting methylation status of CpG islands. RESULTS: Based on a recent genome-scale dataset of DNA methylation in human brain tissues, we developed a classifier called MethCGI for predicting methylation status of CpG islands using a support vector machine (SVM). Nucleotide sequence contents as well as transcription factor binding sites (TFBSs) are used as features for the classification. The method achieves specificity of 84.65% and sensitivity of 84.32% on the brain data, and can also correctly predict about two-third of the data from other tissues reported in the MethDB database. AVAILABILITY: An online predictor based on MethCGI is available at http://166.111.201.7/MethCGI.html CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data available at Bioinformatics online and http://166.111.201.7/help.html.