Search for antibacterial activity in a number of new S-derivatives (1,2,4-triazole-3(2H)-yl)methyl)thiopyrimidines

Results. Derivatives of 1,2,4-triazole-3(2H)-thiol hybrids with a pyrimidine moiety showed high antibacterial activity against gramnegative microorganisms (E. coli, P. aeruginosa). The obtained experimental results allowed to establish not only the role of the main structural features of the compounds in the manifestation of antimicrobial properties, but also to evaluate the effectiveness of the created classification and regression QSAR models. Based on the presented parameters for individual predictive QSAR models, it is possible to conclude about the effectiveness, stability and feasibility of using these models to search for new S-derivatives (1,2,4-triazole-3(2H)-yl) methyl)thiopyrimidines as promising antimicrobial agents.

At the present stage of development of organic chemistry, many basic synthetic approaches to the synthesis of 1,2,4-triazole [1,2], which exhibits a high antibacterial [3], fungicidal [4] activity. To date, it is known that modification of azole he te rocycles leads to increased efficiency and reduced toxicity.
1,2,4-Triazole derivatives are widely known as antibacterial, fungicidal and antiprotozoal drugs, so it is interesting to select and analyze compounds with high antibacterial activity.
It is known that the most biologically active are those compounds whose molecule sizes provide them with optimal bioavailability. In this regard, the most promising salts and esters. Salts -due to the peculiarities of its pharmacokinetics (good dissociation, rapid absorption), and esters with low molecular weight alcohol residues -due to the relatively strong ester bond and good permeability to the cell.
The relevance of the study of 1,2,4-triazole derivatives with pyrimidine fragment is due to the synthesis of potential broad-spectrum antibacterial drugs, low molecular weight inducers of interferon and antitumor agents, search for mole-cular descriptors of their structure, important for establishing patterns "structure -biological activity".

Aim
The aim of the work is a computer search for the antibacterial action of new hybrids of 1,2,4-triazole-3(2H)-thiols with a pyrimidine fragment in relation to 4 test cultures, to establish the dependence of "structure -action".

Materials and methods
To create predictive QSAR models, individual samples of 1,2,4-triazole derivatives were formed, the main number of which were 1,2,4-triazole-3(2H)-thiol derivatives, and entered into the OCHEM server database [https://ochem. eu/] in Excel format [5]. Sets of experimental data included 110 structures of inhibitors P. aeruginosa, E. coli, S. aureus and C. albicans. The k-Nearest Neighbor Method (k-NN) was used to construct QSAR regression models. Classification models were built using the method of random forest (WEKA-RF, Random Forest). To calculate the molecular descriptors used 6 software packages that combine both simple descriptors for counting chemical groups, and descriptors of a wide range of possibilities for counting chemical structures, such as: ALOGPS, E-State, ADRIANA. Code, Dragon V6.0, Chemaxon, Inductive descriptors, available on the OCHEM server.
Classification quality was assessed by statistical indicators -total accuracy as a percentage of correctly classified compounds (total accuracy), prediction accuracy for active and inactive compounds (precision), and class efficiency rates (class hit rate). The predictive power of QSAR regression models was estimated using a crossestimation factor q 2 .
When creating QSAR classification models, MIC values were divided into two conditional groups in a ratio of approximately 1:1 for bacteria, where 50 % of all compounds were considered active and 50 % inactive and in a ratio of 1.0:1.5 for the fungus C. albicans where 40 % of all compounds considered active and 60 % inactive.
The percentage of correctly classified compounds in relation to all microbial cultures (total accuracy) was 80-88%, which indicates the high predictive power of the constructed QSAR classification models.
Statistical coefficients of the developed regression consensus QSAR models for predicting the activity of new S-derivatives (1,2,4-triazole-3(2H)-yl)methyl)thiopyrimidines against the studied microorganisms are presented in Table 2 and graphically shown in Fig. 1-4.
Antimicrobial activity, represented as MIC, was transformed into log (1/MIC) and used as a dependent variable to build QSAR models.
This conclusion is confirmed by the graphical results ( Fig. 1-4)  Prediction of antimicrobial activity of synthesized S-derivatives (1,2,4-triazole-3(2H)-yl)methyl)thiopyrimidines according to QSAR classification models. The created QSAR classification models were used to predict the "class" of antimicrobial activity of the synthesized compounds by the criterion of "active" and "inactive". The results of the prediction by classification models are given in Table 3.
According to table 3 almost 90 % of compounds according to the classification models of activity are provided as active.

Discussion
Based on the presented parameters for individual predictive QSAR models ( Table 2), we can conclude about the efficiency, stability and feasibility of using these models to search for new S-derivatives (1,2,4-triazole-3(2H)-yl)methyl) thiopyrimidines as promising antimicrobial agents. This is evidenced by the high value of the crossestimation coefficient -q 2 , determined for all consensus models in the range of 0.84-0.91 and the optimal range of values of the standard error of the forecast -RMSE 0.26-0.48. The use of regression QSAR models to predict the antimicrobial activity of the compounds allowed to divide them into 4 conditional groups by activity value (MIC) in the range of  10 μmol, 100 μmol, 1000 μmol and 10000 μmol. Moreover, for each type of microorganism there was a different level of predicted activity of S-derivatives (1,2,4-triazole-3(2H)-yl) methyl)thiopyrimidines. The obtained experimental results allowed to establish not only the role of the main structural features of the compounds in the manifestation of antimicrobial properties, but also to evaluate the effectiveness of the created classification and regression QSAR models.
Developed QSAR classification models based on the percentage of correctly predicted compounds (70 %) are the most effective compared to regression (50 %) for the search for new antimicrobial agents in a number of S-derivatives (1,2,4-triazole-3(2H)-yl)methyl)thiopyrimidines.

Conclusions
1. It was found that the studied derivatives of hybrids of 1,2,4-triazole-3(2H)-thiol with a pyrimidine moiety showed high antibacterial activity against gram-negative microorganisms.
2. The developed QSAR classification models based on the percentage of correctly predicted compounds (70 %) are the most effective in comparison with regression (50 %) for the search for new antimicrobial agents in a number of derivatives of hybrids 1,2,4 triazole-3(2H)-thiol with pyrimidine fragment .