High-order Topology for Deep Single-cell Multi-view Fuzzy Clustering

IEEE Transactions on Fuzzy Systems, Journal Year: 2024, Volume and Issue: 32(8), P. 4448 - 4459

Published: May 13, 2024

Single-cell multi-view clustering is essential for analyzing the different cell subtypes of same from views. Some attempts have been made, but most these models still struggle to handle single-cell sequencing data, primarily due their non-specific design cellular data. We observe that such data distinctively exhibits: (1) a profusion high-order topological correlations, (2) disparate distribution information across views, and (3) inherent fuzzy characteristics, indicating cell's potential associate with multiple cluster identities. Neglecting key patterns could significantly impair medical clustering. In response, we propose specialized application namely deep Multi-view Fuzzy Clustering (scMFC) method. Concretely, employ random walk technique capture relationships on graph developed cross-view aggregation mechanism adaptively assigns weights Furthermore, accurately reflect dynamic insight in development, strategy allows cells diverse clusters. Extensive experiments conducted three real-world datasets demonstrate our method's superior performance.

Language: Английский

---

Graph contrastive learning as a versatile foundation for advanced scRNA-seq data analysis

Briefings in Bioinformatics, Journal Year: 2024, Volume and Issue: 25(6)

Published: Sept. 23, 2024

Single-cell RNA sequencing (scRNA-seq) offers unprecedented insights into transcriptome-wide gene expression at the single-cell level. Cell clustering has been long established in analysis of scRNA-seq data to identify groups cells with similar profiles. However, cell is technically challenging, as raw have various analytical issues, including high dimensionality and dropout values. Existing research developed deep learning models, such graph machine models contrastive learning-based for using summarized unsupervised a human-interpretable format. While advances profound, we are no closer finding simple yet effective framework high-quality representations necessary robust clustering. In this study, propose scSimGCL, novel based on paradigm self-supervised pretraining neural networks. This facilitates generation crucial Our scSimGCL incorporates cell-cell structure enhance performance Extensive experimental results simulated real datasets suggest superiority proposed scSimGCL. Moreover, assignment confirms general applicability state-of-the-art algorithms. Further, ablation study hyperparameter efficacy our network architecture robustness decisions setting. The can serve practitioners developing tools source code publicly available https://github.com/zhangzh1328/scSimGCL.

Language: Английский

---

When bipartite graph learning meets anomaly detection in attributed networks: Understand abnormalities from each attribute

Neural Networks, Journal Year: 2025, Volume and Issue: 185, P. 107194 - 107194

Published: Jan. 22, 2025

Language: Английский

---

MSS-PAE: Saving autoencoder-based Outlier Detection from unexpected reconstruction

Pattern Recognition, Journal Year: 2025, Volume and Issue: unknown, P. 111467 - 111467

Published: Feb. 1, 2025

Language: Английский

---

scMDCL: A Deep Collaborative Contrastive Learning Framework for Matched Single-Cell Multiomics Data Clustering

Journal of Chemical Information and Modeling, Journal Year: 2025, Volume and Issue: unknown

Published: March 11, 2025

Single-cell multiomics clustering integrates multiple omics data to analyze cellular heterogeneity and is crucial for uncovering complex biological processes disease mechanisms. However, existing matched single-cell methods often neglect the full utilization of intercellular relationships interactions synergy between features from different omics, leading suboptimal performance. In this paper, we propose a deep collaborative contrastive learning framework clustering, named scMDCL. This fully leverages intercell while enhancing feature among identical cells across data, thereby facilitating efficient data. Specifically, utilize topological information cells, graph autoencoder enhancement module are designed enabling extraction augmentation cell features. Additionally, techniques employed strengthen same cell. Ultimately, modules utilized achieve clustering. Extensive experiments conducted on nine publicly available datasets demonstrate superior performance proposed in integrating tasks.

Language: Английский

---

FactVAE: a factorized variational autoencoder for single-cell multi-omics data integration analysis

Briefings in Bioinformatics, Journal Year: 2025, Volume and Issue: 26(2)

Published: March 1, 2025

Abstract Single-cell multi-omics technologies have revolutionized the study of cell states and functions by simultaneously profiling multiple molecular layers within individual cells. However, existing methods for integrating these data struggle to preserve critical feature information fail exploit known regulatory knowledge, which is essential understanding functions. This limitation hinders their ability provide comprehensive accurate insights into Here, we propose FactVAE, an innovative factorized variational autoencoder designed robust single-cell data. FactVAE integrates factorization principle framework, ensuring preservation while leveraging non-linear capture sample neural networks. Additionally, knowledge incorporated during model training, a transfer strategy employed embedding optimization augmentation. Comparative analyses datasets from different protocols spatial dataset demonstrate that not only outperforms benchmark in clustering performance but also generates augmented reveals clearest cell-type-specific motif expression. Moreover, embeddings captured enable inference potential reliable gene relationships. Overall, FactVAE’s superior strong scalability make it promising new solution analysis.

Language: Английский

---

Multiple Kernel Clustering with Adaptive Multi-scale Partition Selection

IEEE Transactions on Knowledge and Data Engineering, Journal Year: 2024, Volume and Issue: 36(11), P. 6641 - 6652

Published: May 13, 2024

Multiple kernel clustering (MKC) enhances performance by deriving a consensus partition or graph from predefined set of kernels. Despite many advanced MKC methods proposed in recent years, the prevalent approaches involve incorporating all kernels default to capture diverse information within data. However, learning may not be better than one few kernels, particularly since some exhibit higher proportion noise semantic content. Additionally, existing methods, whether based on early-fusion late-fusion approaches, predominantly rely pairwise relationships among samples cluster structures, neglecting potential correlations between these two aspects. To this end, we propose multiple with an adaptive multi-scale selection method (MPS), which exploits multiple-dimensional representations and structure for clustering. By framework, potentially harmful are dynamically excluded during fusion process, then partitions similarity graphs derived retained utilized facilitate improved generation. Finally, extensive experiments conducted demonstrate effectiveness MPS eight benchmark datasets.

Language: Английский

---

scEGG: an exogenous gene-guided clustering method for single-cell transcriptomic data

Briefings in Bioinformatics, Journal Year: 2024, Volume and Issue: 25(6)

Published: Sept. 12, 2024

In recent years, there has been significant advancement in the field of single-cell data analysis, particularly development clustering methods. Despite these advancements, most algorithms continue to focus primarily on analyzing provided matrix data. However, within medical contexts, often encompasses a wealth exogenous information, such as gene networks. Overlooking this aspect could result information loss and produce outcomes lacking clinical relevance. To address limitation, we introduce an innovative deep method for that leverages generate discriminative cell representations. Specifically, attention-enhanced graph autoencoder developed efficiently capture topological signal patterns among cells. Concurrently, random walk protein-protein interaction network enabled acquisition gene's embeddings. Ultimately, process entailed integrating reconstructing gene-cell cooperative embeddings, which yielded representation. Extensive experiments have demonstrated effectiveness proposed method. This research provides enhanced insights into characteristics cells, thus laying foundation early diagnosis treatment diseases. The datasets code can be publicly accessed repository at https://github.com/DayuHuu/scEGG.

Language: Английский

---

MOCHA’s advanced statistical modeling of scATAC-seq data enables functional genomic inference in large human cohorts

Nature Communications, Journal Year: 2024, Volume and Issue: 15(1)

Published: Aug. 9, 2024

Single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq) is being increasingly used to study gene regulation. However, major analytical gaps limit its utility in studying regulatory programs complex diseases. In response, MOCHA (Model-based single cell Open CHromatin Analysis) presents advances over existing analysis tools, including: 1) improving identification of sample-specific open chromatin, 2) statistical modeling technical drop-out with zero-inflated methods, 3) mitigation false positives analysis, 4) alternative transcription-starting-site regulation, and 5) modules inferring temporal networks from longitudinal data. These advances, addition analyses, provide a robust framework after quality control labeling human disease. We benchmark four state-of-the-art tools demonstrate advances. also construct cross-sectional networks, identifying potential mechanisms COVID-19 response. provides researchers tool functional genomic inference scATAC-seq

Language: Английский

---

Simple Yet Effective: Structure Guided Pre-trained Transformer for Multi-modal Knowledge Graph Reasoning

Published: Oct. 26, 2024

Language: Английский

---