Stelmakh S. Analysis and prediction of properties of molecular nanoobjects by chemoinformatics methods

The thesis is devoted to the study of influence of the peculiarities of nature and dimensional characteristics on cytotoxicity and other properties of a row of oxide nanoparticles by nanoQSAR/QSPR methods within simplex representation of molecular structure approach. The Combined Database of Nanooxides (CDN) has been constructed, which contains information on 188 oxide nanoparticles including such characteristics as structure and size parameters as well as some activities/properties: zeta potential, energy Eg (band gaps - an energy range in a solid where no electronic states can exist), data on cytotoxicity to Escherichia coli cells and HaCaT cells. A system of 1D descriptors of oxide nanoparticles is constructed with the application of simplex, integral, "liquid drop", individual and cross-descriptors. The obtained nanoQSAR models of cytotoxicity to Escherichia coli cells (R2 = 0.93, R2test = 0.97) and HaCaT cells (R2 = 0.83, R2test = 0.91) were validated and interpreted. Cross-validation (Q2LOO = 0.90, Q2LOO = 0.71 respectively) was performed by the Leave-One-Out procedure. The cluster analysis and interpretation by descriptors relative influence showed that one of the main factors determining the cytotoxicity of oxide nanoparticles is the magnitude of the metal ion charge. A consensus nanoQSPR model for the zeta potential was constructed (R2 = 0.89, R2test = 0.81). Model validation (Q2cv = 0.81) was performed according to the five-fold cross-validation procedure. An additional assessment of the predictive ability of the model was performed using external testing (R2ext.test = 0.83). The performed interpretation showed a significant influence of the interaction of structural factors - total relative contribution of cross-descriptors was ~ 81%. NanoQSPR modeling of Eg energy was performed. Statistical indicators of the constructed consensus model (R2 = 0.83, Q2cv = 0.74, R2test = 0.73) indicate the ability to adequately predict the studied property. A structural interpretation of the obtained model was performed, it was found that the most influential factors are electrostatic and Van der Waals interactions. The obtained nanoQSAR/QSPR models for cytotoxicity to Escherichia coli and HaCaT cells, zeta potential, Eg energy were combined into the “nanoExpert” expert system. The expert system was integrated into the Methods of Data Analysis software (© Artemenko A.G.) and the corresponding user interface. It is shown that the expert system is software aimed at a wide audience and suitable for modeling of a row of activities/properties by users without special skills in chemoinformatics and nanoQSAR/QSPR modeling.