Significance The COVID-19 pandemic is a worldwide public health emergency caused by the β -coronavirus SARS-CoV-2. A very large and continuously increasing number of high-quality whole-genome sequences are available. We have investigated whether these sequences show effects of epistatic contributions to fitness. In a population evolving under a high rate of recombination, such effects of natural selection can be detected by direct coupling analysis, a global model learning technique. The paper opens up the prospect to leverage very large collections of genome sequences to find combinatorial weaknesses of highly recombinant pathogens.