Deciphering Genetic Overlaps: A Comprehensive Study On Viral Host Determination Using Machine Learning And Deep Learning Models

The study uses machine learning and deep learning models to study the intricate relationship between viral genetic DNA sequences and host organisms. It uses a comprehensive dataset from databases like ExPASy and NCBI, which encodes crucial genetic information for viral replication. The study aimed to create a viral DNA dataset and develop robust machine learning and deep learning models to classify viruses into eight host categories. Despite extensive experimentation using various models, performance improvement was elusive due to genetic overlaps. Viral genomes from different classes had significant shared genetic sequences, making it difficult for these models to identify unique class-specific features, blurring the lines of differentiation. The study reduced the number of classes from eight to three, focusing on plants, animals, and microorganisms. This resulted in improved evaluation metrics, with the Random Forest Machine learning model reaching a maximum accuracy of 70% and the LSTM deep learning model surpassing 85%, overcoming earlier challenges. The discovery that viral genomes from different classes share significant genetic overlaps challenges conventional molecular distinctions, emphasizing the complexity of molecular differentiation in viral genomes. This pragmatic approach aligns molecular understanding with genetic data in viral host determination.