Using an Unsupervised Clustering Model to Detect the Early Spread of SARS-CoV-2 Worldwide.

Area of Science:

• Genomics
• Virology
• Computational Biology

Background:

• Understanding Severe Acute Respiratory Syndrome-Coronavirus-2 (SARS-CoV-2) population structure is vital for public health and controlling viral spread.
• The increasing volume of SARS-CoV-2 genomes necessitates effective methods for classifying viral population structures.

Purpose of the Study:

• To apply unsupervised deep learning clustering to group a large dataset of SARS-CoV-2 genomes.
• To identify distinct subtypes of SARS-CoV-2 based on genetic variations.
• To analyze the geographical distribution of identified subtypes and explore contributing factors.

Main Methods:

• Utilized an unsupervised deep learning clustering algorithm on 16,873 SARS-CoV-2 genomes.
• Employed single nucleotide polymorphisms (SNPs) as input features for the clustering analysis.
• Analyzed the continental proportions and geographical distribution of the identified clusters.

Main Results:

• Identified six major subtypes of SARS-CoV-2 through deep learning-based clustering.
• Observed distinct geographical distributions for these subtypes across different continents.
• Found that both genetic variations and human migration patterns influenced the observed geographical distribution.

Conclusions:

• Deep learning clustering offers a novel approach to studying the population structure of rapidly evolving viruses like SARS-CoV-2.
• The identified subtypes and their geographical distributions provide insights into viral evolution and dissemination.
• Clustering techniques can be valuable for future investigations into localized SARS-CoV-2 population structures.