Simplified molecular input line entry system (SMILES) as an alternative for constructing quantitative structure-property relationships (QSPR)
Abstract
Flexible descriptors calculated with correlation weights of fragments in the SMILES notation of molecular sy stems have been used as a tool for modeling normal boiling points of acyclic carbonyl compounds. Four variants of th e Optimizati on of Correlation Weights o f SMILES Fragments (OCWSF) have been examined. The difference between th em is in the number o f symbols in th e SMILES fragments. T hus. fragments involving o ne-, two-, three-. and four-symbols have been examined. Correlation weights for three calculable features of SMILES are used in the OCWSF sc he me: number of oxygen atoms (No), number of double bonds (NDn) , and (No - NDn +10). In order to take into account the hydrogen bo nd interacti ons, correlation weights of these three feature s have bee n included in th e OCWSF sch~me. The best OCWSF model is based on three-symbol fra gments together with the menti oned three features of the SMILES notati o n. Its stati stic al characteristics are: 1/=100, ? =0.9795, 5=5.35 0c, F=4673 (training set); 1/= 100, ,.2=0.9764, s=5 .38 °C , F=4055 (te st set) .