Experimental

Kinetics obtained for a library of 115 oximes screened for reactivation of BChE inhibited by the NAs tabun, sarin, cyclosarin and VX were adopted from the study by Čadež et al. [9] Dataset determined for each NA contained reactivation parameters obtained with 0.1 mM oximes: the first-order observed reactivation rate constant (_k_obs / min-1), the maximal percentage of reactivation (Reactmax; %), time at which Reactmax was achieved (t / h), and BChE inhibition, % by the 0.01 mM oxime calculated against control enzyme activity (Supplementary material, Table S1). Due to a large span of _k_obs values (four orders of magnitude), normalisation was applied to narrow it, which is beneficial for the PC analysis.

ADME dataset of selected oximes

The 3D structures of a library of 111 oximes and a set of four standard oximes were minimized with the MMFF94 force field using ChemBio3D Ultra 12.0 (PerkinElmer, Inc., Waltham, MA, USA). The Discovery Studio 21.1 (BioVia, San Diego, CA, USA; https://www.3ds.com/products/biovia/discovery-studio) ADME (absorption, distribution, metabolism, and excretion) descriptor protocol was used for the calculation of the following molecular parameters: lipophilicity coefficient (Alog _P_98) and topological polar surface area (PSA 2D) [19-21]. The correlation between Alog _P_98 and PSA 2D was used to predict BBB permeability from a training set of known CNS-active compounds with good adsorption and BBB permeability properties. In addition to parameter Alog _P_98 and PSA 2D, physicochemical properties: molecular weight (MW), number of hydrogen bond donor atoms (HBD), and number of hydrogen bond acceptor atoms (HBA), dipole moment, molecular surface, and fraction of polar molecular surface were calculated [22].

Principal component analysis (PCA) of the oxime library

The reactivation dataset of the studied oxime library in combination with the ADME dataset was analysed using principal component analysis (PCA) statistical model in GraphPad Prism 9 software (Dotmatics, UK; https://www.graphpad.com/features). The first PCA model, applied to all four OP compounds and including 27 variables in total, generated 27 PCs with matching variances and eigenvectors of the covariance matrix. The second PCA model was applied to each of the four OP compounds, with 18 variables in total, generating 18 PCs with matching variances and eigenvectors. For each variable, the loading was calculated as the Pearson correlation coefficient between the original data and the principal component scores [13,14]. Loadings were calculated by scaling the eigenvectors (weights) by the square root of their corresponding eigenvalues, representing the contribution of a variable to a principal component. Plots for PCA visualization comprise PC1 vs. PC2, and PC2 vs. PC3 correlation plot, and a plot of PC1-PC3 loadings distribution. Prior to analysis, data describing first-order observed reactivation rate constant (_k_obs / min-1) were normalised, calculating a negative logarithmic value due to a very broad data range, which is four orders of magnitude. Reactivation of inactive compounds resulted in _k_obs equal to zero, and due to logarithmic normalisation, a negative logarithmic value for inactive compounds is set to 5.