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

Area of Science:

• Genomics
• Oncology
• Biotechnology

Background:

• Cervical cancer development is a lengthy process from human papillomavirus (HPV) infection.
• Current HPV DNA testing requires improved triage methods for accurate risk management in HPV-positive women.
• Genomic variations accumulate during cervical carcinogenesis, necessitating advanced detection strategies.

Purpose of the Study:

• To design a custom capture panel for next-generation sequencing (NGS) targeting HPV and cervical cancer-related genes.
• To analyze HPV integration, somatic mutations, and copy number variations in cervical lesions.
• To develop a machine learning-based risk stratification model for cervical precursor lesions.

Main Methods:

• A 39 Mb custom capture panel was designed targeting 17 HPV types and 522 mutant genes.
• Capture-based NGS was performed on 34 paired samples (HPV infection, CIN1, CIN2+).
• A Random Forest machine learning algorithm was used to build a risk stratification model.

Main Results:

• HPV integration events, non-synonymous mutations, and copy number variations increased from HPV+ to CIN2+ stages.
• A common deletion of the mitochondrial chromosome was significantly observed in CIN2+ cases (P = 0.009).
• The model accurately predicted CIN2+ with an average probability score of 0.814, with amplification and deletion as key features.

Conclusions:

• Custom capture sequencing combined with machine learning effectively stratifies cervical lesion risk.
• The developed method provides valuable integrated triage strategies for HPV-positive women.
• This approach enhances risk management for cervical cancer precursor lesions.