engleski

Estimation of the random correlation level of molecular descriptors in structure‐property modeling

Few years after the introduction of structure-property/activity models [1], the research community has become aware that some correlations can be obtained by chance [2]. In models relating structure of molecules described by molecular structural descriptors and their properties, the level of chance correlation present in each model is usually estimated by different randomization procedures. Such an approach is a part of validation protocols defined for structure-toxicity models used in environmental protection [3]. It came out that an appropriate model should have a smaller number of significant molecular descriptors containing maximal amount of useful and stable structural information. Stability of structural information means that the values of descriptor should not be significantly changed upon small conformational structural changes. However, many molecular descriptors calculated for most data sets are highly monotonous [4]. By performing intensive simulations of correlations between two randomized variables, we can obtain information about the distribution of chance correlation coefficients. The information content in a variable will be related to the range of random correlation with variables having the same (or similar) distribution, as well as with variables having arbitrary distributions. The level of minimal, maximal and the most probable chance correlation will be calculated and related to the distribution of descriptors. Analysis will be made on two-class classification variables, and an analogy with results for other variables will be established. Obtained results and conclusions will be used to define new steps in validation procedure for estimating the quality of structure-property models. Acknowledgement: The work of doctoral student Viktor Bojović has been fully supported by the “Young researchers' career development project – training of doctoral students” of the Croatian Science Foundation funded by the European Union from the European Social Fund. References [1] C. Hansch and T.J. Fujita, J. Am. Soc. 86 (1964) 1616-1626. [2] J.G. Topliss and R.J. Costello, J. Med. Chem. 15 (1972) 1066-1068. [3] Guidance Document on the Validation of [(Q)SAR] Models, ENV/JM/MONO(2007)2, 2017. [4] Talete srl, DRAGON program, www.talete.mi.it.

molecular descriptors ; variable distribution ; random correlation, structure-property modeling ; classification variable ; maximal and minimal correlation

nije evidentirano