For three random splits, one-variable models of oximes reactivation of sarin inhibited acetylcholinesterase (logarithm of the AChE reactivation percentage by oximes with concentration of 0.001 M) have been calculated with CORAL software. The total number of considered oximes was 42. Simplified molecular input line entry system (SMILES) and hydrogen-suppressed graph (HSG) are used to represent the molecular structure. Using CORAL software by means of the calculation with Monte Carlo optimization of the so called correlation weights for the molecular fragments, optimal SMILES-based descriptors were defined, which are correlated with an endpoint for the training set. The predictability of these descriptors for an external test are estimated. In this study hybrid representation HSG together with SMILES was used. The "classic" scheme (i.e. split data into the training set and test set) of building up quantitative structure-activity relationships was employed. Computational experiments indicated that this approach can satisfactorily predict the desired endpoint. Best model had following statistical characteristics n=32, r2= 0.6012, s= 0.279, F= 45 for training and n=10, r2= 0.9301, s= 0.076, Rm2=0.9206 for test set.