Quality testing of distance-based molecular descriptors for benzenoid hydrocarbons

In this paper, we investigate the prediction power of all the well-known distance-based molecular structure descriptors in the literature for 22 lower polycyclic aromatic hydrocarbons (PAHs). The standard enthalpy/heat of formation is chosen to be the representative of thermal properties, whereas, the normal boiling point is selected to represent intermolecular and van-der-Waals type interactions. First, we present a computational technique to compute distance-based, eccentricity-based, degree-distance-based and degree based topological descriptors. The proposed method extends certain existing techniques in the literature. The proposed method is used to compute various distance-based descriptors for the 22 PAHs and then we develop the regression models with their normal boiling point and standard enthalpy of formation. Experimental results reveal some unexpected outcomes as the correlation ability of some well-known distance-based descriptors such as the Wiener and the Szeged indices is rather weak. Unlike their reputation among researchers, the second atomic-bond connectivity index and the fourth geometric-arithmetic index yield the best performance with correlation coefficients greater than 0.95. The results warrant further usage of the second atom-bond connectivity index in the structure-activity and structure-property models. Our study contributes towards slowing down the proliferation of distance-based topological descriptors.