A novel approach to analyzing the evolution of SARS-CoV-2 based on visualization and clustering of large genetic data compactly represented in operative memory.

Abstract

SARS-CoV-2 is a virus for which an outstanding number of genome variants were collected, sequenced and stored from sources all around the world. Raw data in FASTA format include 16.8 million genomes, each ≈29,900 nt (nucleotides), with a total size of ≈500 ∙ 109 nt, or 465 Gb. We suggest an approach to data representation and organization, with which all this can be stored losslessly in the operative memory (RAM) of a common PC. Moreover, just ≈330 Mb will be enough. Aligning all genomes versus the initial Wuhan-Hu-1 reference sequence allows each to be represented as a data structure containing lists of point mutations, deletions and insertions. Our implementation of such data representation resulted in a 1:1500 compression ratio (for comparison, compression of the same data with the popular WinRAR archiver gives only 1:62) and fast access to genomes (and their metadata) and comparisons between different genome variants. With this approach implemented as a C++ program, we performed an analysis of various properties of the set of SARS-CoV-2 genomes available in NCBI Genbank (within a period from 24.12.2019 to 24.06.2024). We calculated the distribution of the number of genomes with undetermined nucleotides, 'N's, vs the number of such nucleotides in them, the number of unique genomes and clusters of identical genomes, and the distribution of clusters by size (the number of identical genomes) and duration (the time interval between each cluster's first and last genome). Finally, the evolution of distributions of the number of changes (editing distance between each genome and reference sequence) caused by substitutions, deletions and insertions was visualized as 3D surfaces, which clearly show the process of viral evolution over 4.5 years, with a time step = 1 week. It is in good correspondence with phylogenetic trees (usually based on 3-4 thousand of genome variant representatives), but is built over millions of genomes, shows more details and is independent of the type of lineage/clade classification.