Molecular graph fingerprint: a new molecular structural characterization method for the modelling and prediction of chromatographic retention behaviour of several persistent organic pollutants

How to extract and characterize information on molecular microstructures is deemed to be the key task to accurately simulate and predict molecular properties. In terms of atomic attributes, atoms in a molecule are divided into three levels. Based upon that, inter-atomic correlations are mapped to certain reasonable spatial coordinates in virtue of radial distribution function, generating the novel molecular graph fingerprint (MoGF), which essentially provides insight into molecular inner structures. MoGF, committing itself to transformation of molecular structures into characteristic graph curves, shows valuable advantages such as easy calculation, experimental parameters-free, rich information content, and structural significance and intuitive expressions. QSRR studies were performed for 115 polychlorinated dibenzofurans (PCDFs), 41 polychlorinated dibenzo-p-dioxins (PCDDs), 62 polychlorinated naphthalenes (PCNs), and 210 polychlorinated biphenyls (PCBs including the biphenyl)) tested for their retention behaviours on gas chromatographic column DB-5. The resulting PLS models showed good performances with correlation coefficients for both training and test sets above 0.97.