Application of regression learning for gas chromatographic analysis and prediction of toxicity of organic molecules

An important task of physical chemistry is to predict the properties of chemical compounds from their structure. Prediction of the chromatographic retention allows one to reject false candidates in gas chromatography/mass spectrometry analysis and to elucidate structures of unknown compounds. Immediately after establishing the structure of an unknown analyte, the next task is to predict its properties, in particular, toxicity. In this work, the problem of prediction of gas chromatographic retention is considered in detail. A new method for predicting the retention indices on different stationary phases using regression learning is demonstrated in relation to flavors and fragrances. The achieved accuracy is higher than the accuracy of previously published methods. The median absolute error does not exceed 14 units. In addition, prediction of acute toxicity from the molecular structure is considered. The efficiency of various regression learning methods for predicting retention indices and acute toxicity (median lethal dose) of chemical compounds is compared.