A new quantitative structure–property relationship approach using dissimilarity measurements based on topological distances of non-isomorphic subgraphs

A Quantitative Structure-Property Relationship model has been developed using a new method proposed in this paper, which is aimed at overcoming disadvantages related to the use of similarity calculations in quantitative approaches. The method uses the concept of topological descriptor but applied to non-isomorphic subgraphs. A symmetrical matrix comprising Euclidean distances according to differences between the non-isomorphic subgraphs is built. This symmetrical matrix is used as input of Partial Least Squares Regression processes for predicting sublimation enthalpies of Polychlorinated Biphenyls. Statistical results (R² in full cross validation, Standard Error in Cross Validation, slope and bias) of our model were obtained and compared with those from the use of similarity values, univariate topological descriptors and literature approaches.