Scholarlink

With a significant breakthrough in profiling chromatin accessibility at single-cell resolution (scATAC-seq), it is promising to apply deep learning (DL) to cell classification tailored for scATAC-seq data. However, scATAC-seq data possess ambiguous feature spaces and sparse expression levels, and existing cell-typing methods typically either align modalities in the latent space or perform transfer learning based on scRNA-seq data. In this study, we propose SCBC for scATAC-seq data, a two-stage computational method based on batch correction to ensure data consistency. First, we present best-performing entropy from samples with distribution differences to guide the model to quickly learn features and patterns of high-quality samples in limited labeled data. Then, a batch correction strategy is proposed to address the influence of noisy labels and the inconsistency of data distribution. Experimental results demonstrate that SCBC shows excellent discriminative performance and embedding quality.

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

SCBC: A Supervised Single-Cell Classification Method Based on Batch Correction for ATAC-Seq Data

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jiawei Hai,
  • Zhen Xie,
  • Na Liu,
  • Ye Yuan

摘要

With a significant breakthrough in profiling chromatin accessibility at single-cell resolution (scATAC-seq), it is promising to apply deep learning (DL) to cell classification tailored for scATAC-seq data. However, scATAC-seq data possess ambiguous feature spaces and sparse expression levels, and existing cell-typing methods typically either align modalities in the latent space or perform transfer learning based on scRNA-seq data. In this study, we propose SCBC for scATAC-seq data, a two-stage computational method based on batch correction to ensure data consistency. First, we present best-performing entropy from samples with distribution differences to guide the model to quickly learn features and patterns of high-quality samples in limited labeled data. Then, a batch correction strategy is proposed to address the influence of noisy labels and the inconsistency of data distribution. Experimental results demonstrate that SCBC shows excellent discriminative performance and embedding quality.