Unsupervised explainable AI for molecular evolutionary study of forty thousand SARS-CoV-2 genomes

Area of Science:

• Genomics
• Artificial Intelligence
• Virology

Background:

• Unsupervised artificial intelligence (AI) discovers novel patterns in big data without predefined models.
• Characterizing SARS-CoV-2 genome evolution is crucial for public health.
• Batch-learning self-organizing maps (BLSOM) were previously developed to analyze large genomic datasets.

Purpose of the Study:

• To apply BLSOM to analyze oligonucleotide compositions of SARS-CoV-2 genomes.
• To identify genomic features responsible for viral clade clustering.
• To explore potential RNA modification sites within the SARS-CoV-2 genome.

Main Methods:

• Application of BLSOM, an unsupervised AI technique, to analyze oligonucleotide compositions.
• Processing of forty thousand SARS-CoV-2 genomes.
• Utilizing the explainable AI capabilities of BLSOM for feature identification.

Main Results:

• BLSOM successfully clustered SARS-CoV-2 genomes based on oligonucleotide composition, aligning with known clades.
• The explainable AI nature of BLSOM identified specific oligonucleotide features driving clade clustering.
• BLSOM indicated potential genomic regions undergoing RNA modifications.

Conclusions:

• BLSOM facilitates efficient knowledge discovery in viral evolution through its image display capabilities.
• This AI approach complements traditional phylogenetic methods based on sequence alignment.
• BLSOM offers a powerful tool for analyzing viral genomic data and understanding evolutionary processes.