viRNAtrap

viRNAtrap identifies and characterizes viral RNA expression in tumor RNA-seq data using an alignment-free deep learning approach to detect known and divergent viruses, including human endogenous retroviruses (HERVs), across cancer samples.

Key Features:

• Alignment-Free Pipeline: Employs an alignment-free methodology to identify viral sequences without mapping reads to existing virus databases, enabling detection of divergent viruses.
• Deep Learning Model: Uses a deep learning model trained to distinguish viral RNA sequencing reads from non-viral reads and to enable assembly of viral contigs directly from RNAseq data.
• Comprehensive Virome Characterization: Applied to RNA-seq data from 14 cancer types in The Cancer Genome Atlas (TCGA) to uncover expression of exogenous viruses and human endogenous retroviruses (HERVs), including unexpected associations.
• Clinical Relevance: Detects divergent viral sequences and assesses their expression levels, revealing associations of certain viral expressions with poor overall survival.

Scientific Applications:

• Discovery of Novel Viral Associations: Identification of viruses not previously linked to cancer to support studies of viral oncogenesis.
• Characterization of Human Endogenous Retroviruses (HERVs): Detection and expression analysis of HERVs to investigate their potential roles in cancer biology and patient survival.
• Broad Cancer Research: Enables comparative analysis of the tumor virome across multiple cancer types to study viral diversity and potential clinical impacts.

Methodology:

An alignment-free pipeline applies a deep learning classifier to RNA-seq reads to distinguish viral from non-viral sequences, followed by assembly of viral contigs from classified reads; the approach was applied to RNA-seq data from 14 cancer types in The Cancer Genome Atlas (TCGA).