HYBRID TRANSFORMER, VARIATIONAL AUTOENCODER, AND GENETIC ALGORITHM APPROACH FOR PREDICTING CROSS-SPECIES TRANSMISSION IN INFLUENZA A VIRUS DRIVEN BY CLIMATE FACTORS.
Track: Tiny Paper Track
Keywords: Influenza virus, Cross-Species Transmission, Early Warning System, RNA sequences, Environmental Surveillance, Transformer, Variational Autoencoder, Genetic Algorithm, Reassortment, Semi-Supervised Learning, Pandemic Preparedness, Bioinformatics, Climate Data, Data-driven Decision-making
TL;DR: Predicting cross-species transmission potential of Influenza A viruses using Transformer, Variational Autoencoder, and Genetic Algorithm with RNA sequence data.
Abstract: Influenza remains one of the most significant respiratory diseases, with influenza A viruses posing a substantial risk due to their rapid mutations, genetic reassortment, and ability to infect multiple hosts. Their segmented genome facilitates reassortment, enabling genetic exchange between different viral strains, which can lead to the emergence of novel variants with pandemic potential. This increases the risk of zoonotic spillover and the emergence of highly transmissible and virulent strains [\cite{morse2012prediction}]. The highly pathogenic avian influenza (H5N1) has already caused over 860 confirmed human infections and more than 450 deaths, with outbreaks in poultry leading to billions of dollars in economic losses. Similarly, the 2009 H1N1 swine flu pandemic resulted in many fatalities worldwide, highlighting the devastating consequences of cross-species transmission. Accurate prediction of cross-species transmission is therefore critical for pandemic preparedness and outbreak prevention. Despite these risks, traditional phylogenetic-based reassortment detection methods lack predictive power, while purely machine learning-based approaches often fail to incorporate biological constraints. Additionally, the limited availability of labeled genomic data constrains fully supervised learning approaches. To address these challenges, we propose a semi-supervised learning framework that integrates Transformers, Variational Autoencoders (VAEs), and Genetic Algorithms (GAs) to enhance feature extraction and simulate novel reassortment events. Our hybrid framework incorporates Transformers for robust sequence feature extraction, VAEs for structured latent-space representations, and GAs for biologically plausible reassortment simulations. Furthermore, considering the seasonal nature of influenza and its dependence on environmental conditions, we integrate climate factors—including temperature, rainfall, and humidity—into our model to better understand climate-driven mutations [\cite{HE2023104593}].
Submission Number: 88
Loading