In infected individuals, viruses are present as a population consisting of dominant and minor variant genomes. Most databases contain information on the dominant genome sequence. Since the emergence of SARS-CoV-2 in late 2019, variants have been selected that are more transmissible and capable of partial immune escape. Currently, models for projecting the evolution of SARS-CoV-2 are based on using dominant genome sequences to forecast whether a known mutation will be prevalent in the future. However, novel variants of SARS-CoV-2 (and other viruses) are driven by evolutionary pressure acting on minor variant genomes, which then become dominant and form a potential next wave of infection. In this study, sequencing data from 96 209 patients, sampled over a 3-year period, were used to analyse patterns of minor variant genomes. These data were used to develop unsupervised machine learning clusters to identify amino acids that had a greater potential for mutation than others in the Spike protein. Being able to identify amino acids that may be present in future variants would better inform the design of longer-lived medical countermeasures and allow a risk-based evaluation of viral properties, including assessment of transmissibility and immune escape, thus providing candidates with early warning signals for when a new variant of SARS-CoV-2 emerges.