We use computational models based on Direct Coupling Analysis - DCA - trained on PFAM domains of distant distant homologues to accurately predict the polymorphisms segregating in a panel of 61,157 Escherichia coli genomes. We show that the genetic context (i.e. the rest of the protein sequence) strongly constrains the tolerable amino acids in 30% to 50% of amino-acid sites. Our study also suggests the gradual build-up of genetic context over long evolutionary timescales by the accumulation of small epistatic contributions. Please refer to the README file for additional information on the structure of this dataset. Code to analyse this dataset is available at https://github.com/GiancarloCroce/DCA_polymorphism_Ecoli.

Our work was partially funded by the French Agence Nationale pour la Recherche ANR GeWiEp (ANR-18-CE35-0005-01, to L.V. and O.T.), the French Fondation pour la Recherche M��dicale (EQU201903007848, to L.V. and O.T.), the PhD program AMX of ��cole polytechnique and Min- ist��re de l'Enseignement Sup��rieur, de la Recherche et de l'Innovation (to L.V.) and EU H2020 Research and Innovation Programme MSCA-RISE-2016 (Grant Agreement No. 734439 InferNet, to M.W.).