Buying time: detecting VOCs in SARS-CoV-2 via co-evolutionary signals

Abstract

We present a novel framework facilitating the rapid detection of variants of interest (VOI) and concern (VOC) in a viral multiple sequence alignment (MSA). The framework is purely based on the genomic sequence data, without requiring prior established biological analysis. The frameworks building blocks are sets of co-evolving sites (motifs), identified via co-evolutionary signals within the MSA. Motifs form a weighted simplicial complex, whose vertices are sites that satisfy a certain nucleotide diversity. Higher dimensional simplices are constructed using distances quantifying the co-evolutionary coupling of pairs and in the context of our method maximal motifs manifest as clusters. The framework triggers an alert via a cluster with a significant fraction of newly emerging polymorphic sites. We apply our method to SARS-CoV-2, analyzing all alerts issued from November 2020 through August 2021 with weekly resolution for England, USA, India and South America. Within a week at most a handful of alerts, each of which involving on the order of 10 sites are triggered. Cross referencing alerts with a posteriori knowledge of VOI/VOC-designations and lineages, motif-induced alerts detect VOIs/VOCs rapidly, typically weeks earlier than current methods. We show how motifs provide insight into the organization of the characteristic mutations of a VOI/VOC, organizing them as co-evolving blocks. Finally we study the dependency of the motif reconstruction on metric and clustering method and provide the receiver operating characteristic (ROC) of our alert criterion.