Aims and objectives: Selection pressure, such as drug treatment or host microenvironment, drives the evolution of M. tuberculosis in vivo. The emergence of drug resistance by the accumulation of mutations leading to drug target modifications, drug transport or activation pathways inactivation, is the leading cause of treatment failure. Besides that, secondary mutations involved in bacterial adaptation, fitness-compensation, and resistance level were identified in the genome by microevolution studies. The genotype-phenotype correlation is complicated also by the partial cross-resistance between first- and second-line drugs: rifampicin and rifabutin acting on beta-subunit of RNA-polymerase in a slightly variable manner, or isoniazid and ethionamide possessing different activation pathways by InhA and EthA, respectively. Genome-wide association studies (GWAS) of a large number of sequenced clinical isolates allows for the identification of minor determinants of resistance, co-evolution markers, and its epistatic interactions. Such type of data could lead to an increase in molecular tests sensitivity and treatment schemes personalization. Various approaches are used for such analysis; most of them analyze the impact of a single mutation, omitting the mutation interactions due to calculation complexity. Methods: Whole-genome sequencing data of 2212 clinical isolates of M. tuberculosis with resistance data were downloaded from SRA. Identification of single nucleotide variations (SNVs) was performed using BWA-MEM and C-Sibelia scripts using AL123456.3 as reference. The SNVs database handling and calculations were performed with custom Python scripts. Fisher test analysis was performed as for individual mutations, also for mutation combinations. Pairs of mutations were joined using logical AND function. Thus, n^2/2 records were analyzed for correlation with resistance, where n – total number of mutations in the database. Resistance to the rifampicin, isoniazid, and ethionamide was analyzed. Results: A novel approach for GWAS was developed: we studied the correlation of simultaneous occurrence of mutations pairs and phenotypic resistance for all possible pairs. We identified several SNVs associated with resistance in combinations with other mutations, and its statistical significance was below the threshold when analyzed separately. Thus, Rv0175 coding for a membrane-associated protein with an unknown function was associated with rifampin resistance and intergenic mutation in Rv1829-Rv1830 – with isoniazid resistance. Previously described, as associated with MDR and XDR phenotype, frameshift glpK mutation was found to affect the resistance to rifampin and ethionamide, but not to isoniazid. However, in the case of rifampin, it was also identified from the individual mutation GWAS, in case of ethionamide – only in combination with other mutations. In addition to glpK, novel SNVs in rv0083 gene and promoter regions of rv1451 and whiB4 genes were identified as associated with ethionamide resistance, synergistically acting in combinations with other mutations. Conclusions: Fisher exact test for mutation combinations allows for the identification of genetic determinants missed by conventional GWAS approach. Further evaluation of novel resistance determinants using other sets of clinical isolates data from various world regions is needed. Acknowledgements This study was supported by the Russian Foundation for Basic Research (project 20-015-00463).