NextVir: Enabling classification of tumor-causing viruses with genomic foundation models

Area of Science:

• Genomics
• Bioinformatics
• Oncology

Background:

• Oncoviruses are pathogens linked to cancer development.
• Detecting oncoviruses in tumor samples is crucial for cancer research.
• Current methods struggle with multiple oncovirus species in a single sample.

Purpose of the Study:

• To develop a computational framework for precise oncovirus classification.
• To adapt genomic foundation models for oncovirus detection.
• To address the challenge of identifying multiple oncoviruses in tumor DNA.

Main Methods:

• Introduced NextVir, a multi-class viral classification framework.
• Adapted genomic foundation models (DNABERT-S, Nucleotide Transformer, HyenaDNA).
• Fine-tuned models for accurate sequencing read origin identification.

Main Results:

• NextVir demonstrated superior performance compared to existing deep learning methods.
• The framework accurately identifies sequencing reads from various oncoviruses.
• Successfully classified multiple oncovirus species within simulated tumor samples.

Conclusions:

• Genomic foundation models show significant potential in genomics and cancer research.
• NextVir provides a robust solution for complex oncovirus detection.
• This framework can advance the understanding of virus-associated cancers.