Risk stratification of cervical lesions using capture sequencing and machine learning method based on HPV and human integrated genomic profiles

Abstract

From initial human papillomavirus (HPV) infection and precursor stages, the development of cervical cancer takes decades. High-sensitivity HPV DNA testing is currently recommended as primary screening method for cervical cancer, whereas better triage methodologies are encouraged to provide accurate risk management for HPV-positive women. Given that virus-driven genomic variation accumulates during cervical carcinogenesis, we designed a 39 Mb custom capture panel targeting 17 HPV types and 522 mutant genes related to cervical cancer. Using capture-based next-generation sequencing, HPV integration status, somatic mutation and copy number variation were analyzed on 34 paired samples, including 10 cases of HPV infection (HPV+), 10 cases of cervical intraepithelial neoplasia (CIN) grade and 14 cases of CIN2+ (CIN2: n = 1; CIN2-3: n = 3; CIN3: n = 9; squamous cell carcinoma: n = 1). Finally, the machine learning algorithm (Random Forest) was applied to build the risk stratification model for cervical precursor lesions based on CIN2+ enriched biomarkers. Generally, HPV integration events (11 in HPV+, 25 in CIN1 and 56 in CIN2+), non-synonymous mutations (2 in CIN1, 12 in CIN2+) and copy number variations (19.1 in HPV+, 29.4 in CIN1 and 127 in CIN2+) increased from HPV+ to CIN2+. Interestingly, ‘common’ deletion of mitochondrial chromosome was significantly observed in CIN2+ (P = 0.009). Together, CIN2+ enriched biomarkers, classified as HPV information, mutation, amplification, deletion and mitochondrial change, successfully predicted CIN2+ with average accuracy probability score of 0.814, and amplification and deletion ranked as the most important features. Our custom capture sequencing combined with machine learning method effectively stratified the risk of cervical lesions and provided valuable integrated triage strategies.

Association of human papillomavirus integration with better patient outcomes in oropharyngeal squamous cell carcinoma

“…Of the integration sites detected, only eight (23%) were within cellular genes. Previous studies have proposed that integration is a directed process that occurs preferentially in/near genes or other genomic features, such as miRNAs or lncRNAs, 6,13,40,42 but our results show more of a stochastic pattern given the wide range of chromosomes affected and low percentage involving genes. However, the number of events we detected is relatively small, and therefore it is challenging to detect predilections for a specific type of location or chromosomal hotspots.…”

“…One such process that has been investigated as a potential driver of HPV + HNSCC is the process of viral integration. Viral integration has been well characterized in cervical cancer as a marker of disease progression 40 . Studies in cervical cancer and HNSCC have shown that integration into the genome can have a variety of effects on both the cellular and viral genomes, including large‐scale rearrangements, amplifications, deletions, alterations in gene expression and generation of viral‐cellular fusion transcripts 6‐8,11‐13,19 .…”

“…Dashed black line labeled SCAFF, genomic scaffold regionintegration. Viral integration has been well characterized in cervical cancer as a marker of disease progression 40. Studies in cervical cancer and HNSCC have shown that integration into the genome can have a variety of effects on both the cellular and viral genomes, including largescale rearrangements, amplifications, deletions, alterationsF I G U R E 5 Fusion transcript amplification.…”

SearcHPV: a novel approach to identify and assemble human papillomavirus-host genomic integration events in cancer

“…To overcome this issue, others have begun to use HPV targeted capture sequencing. 5,[15][16][17][18] This strategy allows for better coverage of integration sites than an untargeted approach like WGS but requires sensitive and accurate viral-human fusion detection bioinformatic tools, of which the field has been lacking. In our lab, we have found the previously available viral integration callers to have a relatively low validation rate and limitations on the structural information surrounding the fusion sites, which impairs mechanistic studies.…”

Hub gene identification and prognostic model construction for isocitrate dehydrogenase mutation in glioma

“…Machine learning, an integral part of computational biology, is being increasingly adopted for analyzing health-care and biomedical data to address some of these challenges. In the last decade, machine learning algorithms have gained popularity in medicine and scientific research [ 35 , 58 , 75 ]. A study conducted by Zhang and co-workers successfully discriminated hepatocellular carcinoma tissues from cirrhosis tissues by establishing a computational model based on machine learning method [75] .…”