A Straightforward HPV16 Lineage Classification Based on Machine Learning.

Area of Science:

• Genomics
• Machine Learning
• Oncology

Background:

• Human Papillomavirus (HPV) causes 5% of global cancers, notably cervical, oropharyngeal, and anogenital cancers.
• HPV type 16, responsible for over 60% of cervical cancers, is classified into lineages (A-D) linked to disease progression.
• Current lineage assessment via Maximum Likelihood Tree (MLT) is slow, manual, and struggles with poor sequencing.

Purpose of the Study:

• To develop a novel, machine learning-based model for accurate HPV type 16 lineage assessment.
• To create a faster and more efficient alternative to the existing MLT method for HPV lineage classification.

Main Methods:

• Genome-Wide Association Study (GWAS) on 645 HPV16 genomes identified 56 lineage-specific Single Nucleotide Polymorphisms (SNPs).
• Machine learning models (Random Forest, SVM, KNN) were trained using identified SNPs.
• The Random Forest (RF) model was validated on 1,028 HPV16 sequences with pre-determined lineages via MLT.

Main Results:

• The RF-based model achieved 99.5% accuracy in assigning HPV16 lineages.
• The RF model successfully determined lineages for 273 samples that MLT could not analyze.
• The RF model demonstrated a ~40-fold increase in speed compared to MLT.

Conclusions:

• The developed RF model offers a precise, rapid, and efficient method for HPV16 lineage classification.
• This approach can overcome limitations of current methods, including handling poorly sequenced samples.
• Implementation of this model could facilitate lineage classification as a clinical triage or prognostic marker.