DeepVISP: Deep Learning for Virus Site Integration Prediction and Motif Discovery

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Viruses contribute to approximately 15% of human cancers through genome integration.
• Viral integration can cause genomic instability and lead to carcinogenesis.
• Accurate identification of virus integration sites is crucial for cancer research.

Purpose of the Study:

• To develop a deep convolutional neural network (CNN) model, DeepVISP, for precise prediction of oncogenic virus integration sites (VISs).
• To evaluate DeepVISP's performance against conventional methods using benchmark data from hepatitis B virus (HBV), human herpesvirus (HPV), and Epstein-Barr virus (EBV).
• To identify potential cis-regulatory factors involved in virus integration and tumorigenesis.

Main Methods:

• Development of a deep convolutional neural network (CNN) with attention architecture (DeepVISP).
• Training and validation using curated integration data from HBV, HPV, and EBV.
• Comparative analysis with conventional machine learning methods using Area Under Curve (AUC).
• Clustering analysis of informative motifs to understand virus-host gene recognition.

Main Results:

• DeepVISP achieved high accuracy and robust performance in predicting VISs for HBV, HPV, and EBV.
• DeepVISP outperformed conventional machine learning methods by 8.43-34.33% in AUC enhancement.
• Identified potential cis-regulatory factors (e.g., HOXB7, IKZF1, LHX6) implicated in virus integration and tumorigenesis.
• Clustering analysis revealed representative k-mers guiding virus recognition of host genes.

Conclusions:

• DeepVISP offers a powerful and accurate method for predicting oncogenic virus integration sites.
• The model aids in uncovering regulatory mechanisms underlying virus-induced cancers.
• A user-friendly web server is available for predicting putative oncogenic VISs, facilitating further research.