HPViewer: Sensitive and specific genotyping of human papillomavirus in metagenomic DNA

Yuhan Hao, Liying Yang, Antonio Galvao Neto, Milan R. Amin, Dervla Kelly, Stuart M. Brown, Ryan C. Branski, Zhiheng Pei

TL;DR

This paper introduces HPViewer, a novel computational tool for accurately identifying and genotyping Human Papillomavirus (HPV) from metagenomic DNA. By masking repetitive and homologous sequences, HPViewer overcomes limitations of existing methods, offering improved sensitivity and specificity for HPV detection across diverse human samples.

What Problem Does This Paper Address?

Detecting and genotyping HPV from metagenomic data is challenging due to shared DNA sequences between HPV, human, and microbial genomes, leading to false positives. Conversely, methods relying on assembling short reads into longer contigs can miss HPV types present at low abundance, resulting in false negatives. Existing tools struggle to balance sensitivity and specificity, particularly for the vast number of known HPV types (over 210 identified). This paper aims to develop a more robust and accurate method for comprehensive HPV detection in complex biological samples.

Methods

HPViewer employs a novel "masking" approach. It utilizes two custom HPV reference databases:

  • Repeat-mask database: This database masks low-complexity and simple repeat regions within HPV genomes, which are often shared with human and prokaryotic DNA, thus reducing false positives.
  • Homology-mask database: This database masks homologous regions shared between different HPV types, aiming to improve genotyping accuracy and reduce misclassification of closely related HPV types.

HPViewer directly analyzes short DNA reads rather than relying on de novo assembly of contigs. A key innovation is the use of a homology distance matrix to combine the strengths of the repeat-mask and homology-mask approaches, creating a "hybrid-mask" mode. This hybrid mode is designed to optimize the trade-off between sensitivity and specificity, particularly for samples with varying HPV read depths.

Datasets

The study utilized several datasets for evaluation:

  • Simulated HPV shotgun sequencing data: 100,100 simulated samples were generated, each containing one of 143 HPV types, at seven different sequencing depths (2 to 1000 reads per sample). This dataset was used to evaluate the sensitivity and specificity of HPViewer's different modes and compare it to HPVDetector.
  • Recurrent respiratory papillomatosis (RRP) patient samples: 12 samples (6 tumor tissues and 6 matched oral washes) from patients with known HPV6/11 infections were used to compare HPViewer's performance against HPVDetector, VirusTAP, and Vipie in a real-world clinical context.
  • Human Microbiome Project (HMP) datasets: 1,573 metagenomic datasets from 18 different body sites of healthy individuals were analyzed to assess HPViewer's ability to identify HPV prevalence and co-occurrence patterns in a broad range of human samples.

Key Insights and Ideas

  • HPV's "noisy" genome: HPV genomes are not just unique; they share a significant amount of simple repeats with human and prokaryotic genomes, and homologous sequences with other HPV types. This genomic "noise" is a primary source of error in current detection methods.
  • Contigs are not always better: While de novo assembly into contigs can reduce false positives, it comes at the cost of significantly reduced sensitivity, especially for low-abundance HPV types. HPViewer's direct read-based approach bypasses this limitation.
  • Masking is key: The paper demonstrates that strategically masking repetitive and homologous sequences is a highly effective strategy for improving both the specificity and sensitivity of HPV detection.
  • Hybrid approach optimizes performance: Combining repeat-masking and homology-masking strategies through a homology distance matrix (the hybrid-mask mode) provides the best balance of sensitivity and specificity, making it robust across different HPV read depths.

Key Results

  • High Sensitivity and Specificity: HPViewer, particularly in its hybrid-mask mode, achieved high sensitivity (up to 98.9% at 50 reads) and specificity (up to 100% at 2-10 reads, and 98.7% at 1000 reads) in simulated data.
  • Improved performance over HPVDetector: In RRP samples, HPViewer was more specific and sensitive than HPVDetector, which misidentified human repeats as HPV. HPViewer also showed better read counts for HPV6 and HPV11.
  • Effective for low read counts: HPViewer could detect HPV with as few as two reads per sample, a capability that VirusTAP and Vipie lacked, leading them to miss HPV infections in some RRP samples.
  • Broad detection spectrum: In the HMP dataset, HPViewer identified 104 HPV types across 16 body sites, with 81.73% of these types not detectable by standard cervical HPV detection kits.
  • Computational efficiency: HPViewer was significantly faster than contig-based methods like VirusTAP and Vipie, taking approximately two minutes for analysis compared to 12-32 minutes for others. It also used less data storage.

Limitations and Future Work

  • Limitations: While HPViewer demonstrated excellent performance, the paper notes that the specificity of the repeat-mask mode decreased at higher read depths due to errors in genotyping closely related HPV types. The homology-mask mode, while specific, had lower sensitivity at very low read depths.
  • Future Work: The authors suggest that HPViewer is a broad-range detection tool suitable for evaluating HPV presence beyond the female genital system. The tool is available as an open-source python package, encouraging further development and application in various research and clinical settings. The ability to process large numbers of samples efficiently makes it well-suited for the increasing use of metagenomic sequencing.

Summary updated

Abstract

Abstract Background Shotgun DNA sequencing provides sensitive detection of all 182 HPV types in tissue and body fluid. However, existing computational methods either produce false positives misidentifying HPV types due to shared sequences among HPV, human, and prokaryotes, or produce false negative since they identify HPV by assembled contigs requiring large abundant of HPV reads. Results We show that HPV shares extensive simple repeats with human and prokaryotes and homologous sequences among different HPV types. The shared sequences caused errors in HPV genotyping and the repeats of human origin caused false positives in HPVDetector. Programs, such as VirusTAP and Vipie, which require de novo assembly of shotgun reads into contigs, eliminated false positives at a cost of substantial reduction in sensitivity. Here, we designed HPViewer with two custom HPV reference databases masking simple repeats and homology sequences respectively and one homology distance matrix to hybridize these two databases. It directly identified HPV from short DNA reads rather than assembled contigs. Using 100,100 simulated samples, we revealed that HPViewer was robust for samples containing either high or low number of HPV reads. Using 12 shotgun sequencing samples from respiratory papillomatosis, HPViewer was equal to VirusTAP, and Vipie and better than HPVDetector with the respect to specificity and was the most sensitive method in the detection of HPV types 6 and 11. We demonstrated that contigs-based approaches had disadvantages of detection of HPV. In 1,573 sets of metagenomic data from 18 human body sites, HPViewer identified 104 types of HPV in a body-site associated pattern and 89 types of HPV co-occurring in one sample with other types of HPV at least once. Conclusions We demonstrated HPViewer was sensitive and specific for HPV detection in metagenomic data. It was also suggested that masking shared sequences is an effective approach to avoid false positive detection and identifying HPV from short metagenomic reads is more sensitive than assembled contigs. The innovative homology distance matrix connecting two HPV databases, repeat-mask and homology-mask, optimized the balance of sensitivity and specificity. HPViewer can be accessed at https://github.com/yuhanH/HPViewer/ .
BibTeX
@article{hao2018hpviewer,
  author = {Yuhan Hao and Liying Yang and Antonio Galvao Neto and Milan R. Amin and Dervla Kelly and Stuart M. Brown and Ryan C. Branski and Zhiheng Pei},
  title = {{HPViewer: Sensitive and specific genotyping of human papillomavirus in metagenomic DNA}},
  journal = {Bioinformatics},
  year = {2018},
  doi = {10.1093/bioinformatics/bty037},
}

Loading related content...

Figure