A k-mer Based Approach for SARS-CoV-2 Variant Identification

Abstract

With the rapid spread of the novel coronavirus (COVID-19) across the globe and its continuous mutation, it is of pivotal importance to design a system to identify different known (and unknown) variants of SARS-CoV-2. Identifying particular variants helps to understand and model their spread patterns, design effective mitigation strategies, and prevent future outbreaks. It also plays a crucial role in studying the efficacy of known vaccines against each variant and modeling the likelihood of breakthrough infections. It is well known that the spike protein contains most of the information/variation pertaining to coronavirus variants. In this paper, we use spike sequences to classify different variants of the coronavirus in humans. We show that preserving the order of the amino acids helps the underlying classifiers to achieve better performance. We also show that we can train our model to outperform the baseline algorithms using only a small number of training samples (1% of the data). Finally, we show the importance of the different amino acids which play a key role in identifying variants and how they coincide with those reported by the USA's Centers for Disease Control and Prevention (CDC).

Effective and scalable clustering of SARS-CoV-2 sequences

“…Sequence classification and clustering are popular approaches [8,25], most methods requiring the sequences to be aligned [17]. Such aligned sequences are used to design fixed-length feature vector representations that can be input to classical machine learning (ML) based algorithms.…”

“…A popular approach to preserve the order of the sequences in converting the alphabets vector into numeric vector is to use the one-hot encoding approach [27]. In theory, while the one-hot encoding preserves the order of amino acid sequences, the preserved order is of no use while computing pairwise distance (e.g., euclidean distance) [8]. To deal with this problem, each sequence is first converted into length 𝑘 substrings (called 𝑘-mers).…”

“…To deal with this problem, each sequence is first converted into length 𝑘 substrings (called 𝑘-mers). These 𝑘-mers successfully preserve the order of each sequence, that can be crucial to the performance of the classification/clustering tasks [8,18]. In this paper, we propose a method to quickly and effectively cluster the spike amino acid sequences of SARS-CoV-2.…”

AI-based search for convergently expanding, advantageous mutations in SARS-CoV-2 by focusing on oligonucleotide frequencies

“…The AI-based method has also identified another selection mechanism, vaccine breakthrough (or antibody resistance) [ 6 ]. In addition, by focusing on the amino acid sequence of the S protein, various k-mer based AI approaches (e.g., Support Vector Machine, Naive Bayes and Random Forest) have been reported [ 7 ], and the UMAP-assisted K-means clustering with the SNP profiles of SARS-CoV-2 has also been reported [ 8 ]. Unsupervised machine learning can provide new information without relying on particular models or assumptions.…”