VIDHOP, viral host prediction with deep learning.

Area of Science:

• Virology
• Genomics
• Machine Learning

Background:

• Zoonotic diseases pose a significant global threat, with recent outbreaks highlighting the need for rapid identification of virus origins.
• Current methods for predicting viral hosts from genomic data are often limited in accuracy and the number of hosts they can identify.

Purpose of the Study:

• To develop a fast and accurate deep learning approach for predicting viral hosts based solely on viral genome sequences.
• To create a user-friendly tool, VIDHOP, that can predict viral hosts and be adapted for new viruses.

Main Methods:

• A deep neural network (DNN) was trained using viral genome sequences.
• The DNN was tested on influenza A virus, rabies lyssavirus, and rotavirus A.
• The model utilizes short sequence fractions (100-400 bp) and handles unbalanced data.

Main Results:

• The DNN achieved high prediction accuracy, with an Area Under the Curve (AUC) ranging from 0.93 to 0.98 for the tested virus species.
• The approach demonstrates the effectiveness of deep learning for viral host prediction even with limited and imbalanced genomic data.

Conclusions:

• Deep neural networks are a powerful tool for accurate viral host prediction.
• The VIDHOP tool provides a robust and adaptable solution for identifying virus origins, aiding in the control of zoonotic disease spread.