Scholarlink

<p>Single-cell clustering on single-cell RNA sequencing (scRNA-seq) data enables precise and detailed discrimination of cell populations and their tissue functions, garnering significant interest. A category of outstanding methods performs node clustering on cell networks constructed from preprocessed or dimensionally reduced data, capturing the complex relationships between cells to accomplish the cell clustering task. However, these methods typically employ a <i>k</i> nearest neighbors (KNN) approach to construct the cell network, using the same number <i>k</i> of highly similar cells for each cell. This approach overlooks the distribution characteristics of the similarity values, which can degrade subsequent clustering performance. In this paper, we analyze the similarity distribution of cells and propose determining the <i>k</i> value for each cell based on its similarity distribution. Consequently, we develop an adaptive network construction algorithm that selects a varying number of highly similar cells for each cell, with the number automatically determined according to the cell’s similarity distribution. Our experiments on several datasets demonstrate that this proposed adaptive network construction algorithm outperforms the KNN-based approach and enhances clustering performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An adaptive network construction for single-cell clustering

  • 期刊论文

  • 发表日期：

    2025年4月30日

查看全文
  • Tianyu Yang,
  • Yanmei Hu,
  • Yihang Wu,
  • Yingxi Zhang,
  • Bin Duo,
  • Xiaochuan Tang,
  • Xiangtao Li

摘要

Single-cell clustering on single-cell RNA sequencing (scRNA-seq) data enables precise and detailed discrimination of cell populations and their tissue functions, garnering significant interest. A category of outstanding methods performs node clustering on cell networks constructed from preprocessed or dimensionally reduced data, capturing the complex relationships between cells to accomplish the cell clustering task. However, these methods typically employ a k nearest neighbors (KNN) approach to construct the cell network, using the same number k of highly similar cells for each cell. This approach overlooks the distribution characteristics of the similarity values, which can degrade subsequent clustering performance. In this paper, we analyze the similarity distribution of cells and propose determining the k value for each cell based on its similarity distribution. Consequently, we develop an adaptive network construction algorithm that selects a varying number of highly similar cells for each cell, with the number automatically determined according to the cell’s similarity distribution. Our experiments on several datasets demonstrate that this proposed adaptive network construction algorithm outperforms the KNN-based approach and enhances clustering performance.