PEIS: a novel approach of tumor purity estimation by identifying information sites through integrating signal based on DNA methylation data

Shudong Wang, Lihua Wang, Yuanyuan Zhang, Shanchen Pang, Xinzeng WangBMC Bioinformatics 2019 · DOI

TL;DR

This paper introduces PEIS, a novel method for estimating tumor purity from DNA methylation data. By integrating multi-signal information from CpG sites and considering their correlations, PEIS identifies more biologically relevant sites, leading to more accurate purity estimations compared to existing methods. This improved accuracy is crucial for downstream cancer research.

What Problem Does This Paper Address?

Tumor samples are often a mix of cancer cells and normal cells, a phenomenon known as "tumor purity." Accurately quantifying this purity is vital for understanding cancer development, identifying relevant genes, and interpreting downstream analyses like differential methylation studies. Traditional methods relied on manual pathological analysis, which is time-consuming and costly. While computational methods have emerged using gene expression or copy number variation, DNA methylation-based approaches are less common and often struggle with selecting informative sites. Existing DNA methylation methods typically focus on mean methylation differences, neglecting other signals and the spatial correlation between adjacent CpG sites, which can lead to disparate and potentially inaccurate purity estimations. This paper aims to address this gap by developing a more robust method for selecting informative sites from DNA methylation data.

Methods

The proposed method, PEIS (Purity Estimation through Integrating Signal), employs a two-stage approach:

  1. Identifying Informative CpG Sites:
    • Integrating Signal Scores: This stage calculates an "integrating signal score" for each CpG site. This score combines information from the site-level mean signal and variance signal, derived from paired T-tests (for mean difference) and one-sided F-tests (for variance difference). The scores are weighted to balance their influence.
  2. Integrating Correlation Among Adjacent Sites: To account for the biological context, the method then integrates the correlation between adjacent CpG sites. The influence of neighboring sites (within 1000 bps) is incorporated to refine the signal score. This step aims to select sites that are not only differentially methylated but also biologically relevant in their genomic neighborhood.
  3. Site Selection: After calculating these integrated scores, the top 1000 sites with the highest scores are selected as informative sites.
  4. Estimating Tumor Purity:
    • The selected informative sites are then used to estimate tumor purity. This step is based on the InfiniumPurify method, where methylation levels (beta-values) of hyper-methylated sites are used directly, and hypo-methylated sites are transformed (1-beta).
  5. Finally, a Gaussian kernel density estimation method is applied to these transformed methylation levels to estimate the tumor purity.

Datasets

The study utilized publicly available DNA methylation data from The Cancer Genome Atlas (TCGA) via the UCSC Cancer Genome Browser. The analysis included 12 different tumor types: BLCA, BRCA, COAD, HNSC, KIRC, KIRP, LIHC, LUAD, LUSC, PRAD, THCA, and UCEC. For each tumor type, paired tumor and normal-adjacent samples were used, with a minimum of twenty samples per type. The authors note that all DNA methylation CpG sites were filtered for quality control in advance.

Key Insights and Ideas

  • Beyond Mean Differences: The paper highlights the limitation of solely relying on mean methylation differences for site selection. Integrating variance and considering spatial correlations offers a more comprehensive view of methylation patterns.
  • Biological Relevance: By incorporating the correlation between adjacent sites, PEIS aims to select sites that are more likely to be biologically meaningful and less prone to noise from random variations.
  • Tumor Heterogeneity Matters: The study acknowledges that tumor heterogeneity can lead to different performance of the method across various tumor types, underscoring the complexity of cancer biology.
  • PRAD Anomaly: The PRAD dataset showed a particularly low correlation with PEIS and other methods, suggesting that specific characteristics of this tumor type (e.g., age-related factors, sex chromosome involvement) might influence methylation patterns and purity estimation.

Key Results

  • PEIS demonstrated a high degree of consistency with other established tumor purity estimation methods (ABSOLUTE, ESTIMATE, InfiniumPurify, CPE) across 12 tumor types.
  • Specifically, PEIS showed the highest Pearson's correlation (R = 0.76) with InfiniumPurify and a correlation of R = 0.55 with ESTIMATE.
  • When comparing different information site selection methods, PEIS was shown to be superior, providing more accurate results in tumor purity evaluation.
  • The study found that traditional hypothesis testing methods (like paired T-test) often yielded lower correlations with other methods, while methods considering integrated signals and correlations performed better.

Limitations and Future Work

  • Sample Size: The authors acknowledge that the relatively small sample size for some tumor types might influence the observed correlations and lead to inconsistencies.
  • PRAD Specifics: The poor performance of PEIS in PRAD is noted as a specific challenge, potentially related to age-related factors or sex chromosome involvement, which warrants further investigation.
  • CpG Islands: The authors suggest that applying their approach to CpG islands might yield better results, as some selected sites might not be differential methylation sites themselves but are influenced by surrounding sites.
  • Continuous Efforts: The paper emphasizes that selecting information sites containing more tumor information is a key direction for future research.
  • Exploring Other Data Types: The authors mention the potential of integrating CNV and SNV data for a more comprehensive tumor purity evaluation, as DNA methylation data is considered a supplement rather than a definitive standard.

Summary updated

BibTeX
@article{wang2019peis,
  author = {Shudong Wang and Lihua Wang and Yuanyuan Zhang and Shanchen Pang and Xinzeng Wang},
  title = {{PEIS: a novel approach of tumor purity estimation by identifying information sites through integrating signal based on DNA methylation data}},
  journal = {BMC Bioinformatics},
  year = {2019},
  doi = {10.1186/s12859-019-3227-1},
}
Figure