Unsupervised Anomaly Detection Algorithms Unveil Relevant Temporal and Spatial Patterns in the SARS COV2 Codon Usage in México

Abstract

Genomes are complex biological structures that encode information that can be translated onto several levels, such as genes and proteins. Identification of relevant patterns in genomes is of paramount importance, as they may indicate states of biological or medical relevance. Among the patterns that can be detected, anomalies are especially relevant. Anomalies are instances that do not resemble, under certain metrics, the rest of the observations under study. Anomalies and their detection are relevant since their presence may indicate a systematic error in some stage of the analyzed process or structure, or may indicate that the studied system or phenomenon is undergoing a phase transition or other relevant drift in its dynamics. Here, we applied unsupervised anomaly detection algorithms to the codon usage of the genomes of thousands of SARS COV2 virus isolated in Mexico. Codon usage condenses the relative frequency of appearance of nucleotide triplets, or codons, which code for amino acids, the basic blocks of proteins. By applying several algorithms, we detected patterns that are of epidemiological relevance. The detected patterns are anomalous genomes based on their codon usage. Anomalous patterns are relevant not only because they have not been previously detected in data from Mexico, but also because they allow identification of one of the possible sources of the anomalies. Most of these anomalies were identified in two neighboring states in Mexico, namely Puebla and Tlaxcala. In addition, we identified that almost all anomalies come from subjects who were treated in the same laboratory. Based on the evidence we present here, we conclude that anomaly detection algorithms are relevant in the surveillance of epidemics.