Predicting the Risk of Avian Influenza Zoonosis using Viral Genome Sequencing Data

Avian Influenza viruses (AIVs) have the ability to infect a wide range of hosts, despite originating from wild aquatic birds. Most strains remain confined within their host species, but some undergo genetic changes that allow them to jump the species barrier. Zoonotic cases, where humans become infected, are of particular concern due to their high mortality rates and potential for human-to-human transmission, which could lead to a pandemic.

Researchers have used genetic features from 8 AIV proteins, obtained from viral sequence data, to develop machine-learning algorithms capable of classifying AIV cases as zoonotic or not. These genetic features include host signatures such as dipeptide composition and amino acid physiochemical properties. The XGBoost algorithm consistently outperformed other classification models.

The study employed ten models, with the best prediction for AIV zoonosis achieved through a multi-model approach. All 8 proteins were found to play a role in predicting zoonotic transmission, with the PB2 and HA proteins being particularly important. Specific amino acid physiochemical properties, such as charge, secondary structure, and hydrophobicity, were identified as the most indicative features in the combined models.

An alignment-free computational study was conducted to identify AIV cases that are still circulating within avian hosts but are genetically closest to zoonotic cases, highlighting those most likely to cross the species barrier. In resource-limited settings, the model can be utilized to quickly pinpoint high-priority cases for further investigation.