Correction to: LightCpG: a multi-view CpG sites detection on single-cell whole genome sequence data

Limin Jiang, Chongqing Wang, Jijun Tang, Fei GuoBMC 2019 · DOI · OpenAlex

Overview

This article serves as a formal correction to the original publication of "LightCpG: a multi-view CpG sites detection on single-cell whole genome sequence data" in BMC Genomics. The correction addresses a typesetting error where the illustrations for Figure 2 and Figure 3 were inadvertently swapped in the initial release.

The Challenge of Single-Cell Epigenetics

Single-cell whole-genome bisulfite sequencing (scWGBS) is a transformative technology for mapping epigenetic diversity. However, it is plagued by extreme data sparsity and high noise levels. Low sequencing coverage often leads to missing or incorrectly identified CpG sites—regions where a cytosine is followed by a guanine, which are critical for understanding gene regulation.

To overcome these limitations, the authors developed LightCpG, a computational framework designed to accurately predict methylation states. Instead of analyzing sites in isolation, LightCpG leverages the surrounding genomic context and patterns observed across multiple cells.

Methodology: Multi-View Feature Extraction

Based on the corrected figures, the LightCpG method relies on two primary feature extraction strategies to fill in the gaps of sparse data:

1. The Skip-k Method

This technique focuses on the genomic neighborhood of a target CpG site. By calculating the distance ($d$) and methylation state ($s$) of neighboring sites at specific intervals (the $k$-th neighbors), the model captures both immediate local context and longer-range dependencies along the DNA strand.

2. Multi-view Feature Extraction

The model employs a "multi-view" perspective by integrating two types of features to gain a more holistic understanding of the dataset:

  • Feature $\mathcal{F}$: Focuses on the spatial relationship and physical distances between CpG sites within and across different cells.
  • Feature $\mathcal{D}$: Incorporates methylation states across the entire dataset, allowing the model to learn from a "consensus" of patterns found in other cells to infer missing data in a specific single cell.
The sketch map of feature F and feature D, illustrating how spatial relationships and methylation states across cells are integrated.

Key Insights

  • Structured Context: The "skip-k" approach demonstrates that methylation states are influenced by structured patterns of neighbors at varying distances, not just the immediate adjacent site.
  • Collective Intelligence: By treating detection as a multi-view problem, the sparsity of an individual cell's data is mitigated by the collective behavior of the cell population.
  • Bio-Geometric Integration: The framework successfully integrates physical genomic structure (base pair distance) with biological state (methylation), reflecting the complex nature of epigenetic regulation.

Note: This summary focuses on the methodological corrections provided in the formal notice. For full experimental results and datasets, readers are referred to the original study.

Summary updated

Abstract

Following the publication of this article [1], the authors reported that the images of Fig. 2 and Fig. 3 were switched during typesetting.
BibTeX
@inproceedings{jiang2019correction,
  author = {Limin Jiang and Chongqing Wang and Jijun Tang and Fei Guo},
  title = {{Correction to: LightCpG: a multi-view CpG sites detection on single-cell whole genome sequence data}},
  booktitle = {BMC},
  year = {2019},
  doi = {10.1186/s12864-019-5742-x},
}
Figure