TL;DR
This paper introduces a robust framework for assessing the quality of whole-genome cancer sequencing data from diverse sources. By developing a star-rating system based on five key quality control measures, the authors provide a standardized way to evaluate and compare sequencing data, crucial for large-scale cancer genomics projects.
What Problem Does This Paper Address?
Sequencing cancer genomes is a complex undertaking, especially when data comes from numerous centers, using different protocols, and over extended periods. This heterogeneity can lead to variations in data quality, making it challenging to compare results and draw reliable conclusions. The Pan-Cancer Analysis of Whole Genomes (PCAWG) project, involving over 2800 donors, highlighted the urgent need for a standardized, validated framework to assess and compare the quality of these diverse whole-genome cancer sequences. Without such a framework, it's difficult to identify and exclude low-quality data, potentially leading to spurious findings and wasted computational resources.
Methods
The researchers developed a quality assessment framework by selecting five non-redundant quality control (QC) measures that capture different aspects of sequencing quality. These measures were then used to create a "star rating" system. The five key measures are:
- Mean Coverage: Ensuring sufficient depth of sequencing across the genome.
- Evenness of Coverage: Assessing how uniformly the genome is covered, crucial for accurate variant calling. This was evaluated using two methods: the ratio of median to mean coverage (MoM) and the full width at half maximum (FWHM) of the normalized coverage distribution in genomic windows.
- Somatic Mutation Calling Coverage: Evaluating how well the genome is covered for reliable identification of somatic mutations.
- Paired Reads Mapping to Different Chromosomes: Identifying potential technical artifacts or rearrangements.
- Ratio of Difference in Edits Between Paired Reads: Detecting issues in sequencing cycles that might lead to errors.
A star rating is awarded for each QC measure, with specific thresholds for normal and tumor samples. The system is designed to be sensitive to quality issues while minimizing the exclusion of samples with biologically valid reasons for certain variations.
Datasets
The primary dataset used in this study is from the Pan-Cancer Analysis of Whole Genomes (PCAWG) project, which includes approximately 2959 cancer genomes (normal-tumor pairs) from 2830 donors. This dataset represents a wide range of cancer types and populations, making it ideal for developing and validating a generalizable quality assessment framework.
Key Insights and Ideas
- Standardization is Key: The PCAWG project underscored the critical need for a standardized approach to quality control in large-scale cancer genomics. Without it, comparing data from different sources is fraught with challenges.
- Multifaceted Quality Assessment: No single QC measure is sufficient. The framework's strength lies in combining five distinct measures, each capturing a different facet of sequencing quality, to provide a comprehensive picture.
- Tumor vs. Normal Differences: The study highlights that tumor samples often exhibit greater variability in certain quality metrics (like evenness of coverage) due to biological factors such as deletions and rearrangements, necessitating different thresholds for normal and tumor samples.
- Impact on Downstream Analysis: The quality of sequencing data, as captured by the star rating system, directly impacts the reliability of downstream analyses, particularly in calling somatic mutations (SNVs, indels, and SVs). Higher quality data leads to a higher proportion of calls supported by multiple callers.
- Open-Source Tools for Reproducibility: The availability of a Docker Container for the QC software promotes wider adoption and reproducibility of their framework.
Key Results
- The developed framework successfully identified and flagged samples with quality issues, leading to the exclusion of some from downstream analyses. Anecdotally, this saved significant computational time.
- The star rating system was embedded in variant call format files, making quality information readily accessible to researchers.
- Analysis of the PCAWG data revealed that sequencing quality has generally improved over the 2009-2014 period, likely due to protocol changes like the adoption of PCR-free methods.
- Samples with higher star ratings (≥ 4 stars) showed a significantly higher proportion of somatic mutation calls (SNVs, indels, and SVs) supported by multiple callers, indicating greater confidence in these calls. For example, p-values for SNVs, indels, and SVs were around 10⁻⁵, 10⁻⁵, and 10⁻¹⁸, respectively.
- The study found that sequencing quality, as measured by the star rating, has a measurable effect on downstream analyses, with different QC measures having varying levels of importance for different types of variant calling.
Limitations and Future Work
- Limited Metadata: The study notes that for many PCAWG projects, detailed metadata on sequencing protocols and timing was limited, making it harder to fully attribute quality variations to specific technical factors.
- Trade-offs in Thresholds: The chosen thresholds represent a balance between penalizing poor quality and avoiding the exclusion of samples with biologically valid reasons for certain variations. Future work could refine these thresholds.
- Adaptability: The authors believe their method can be adapted for similar projects using whole-genome sequences from various sources. They recommend using their methods to ascertain quality before downstream analyses.
- Large-Scale Meta-Analysis: The framework is intended to facilitate more robust large-scale meta-analyses by providing a standardized quality assessment.
Summary updated