Chemical Information and Molecular Similarity

Molecular structures are determinants of molecular properties including physicochemical properties, biological activities, and toxicities. The atom types, bond types, functionalities, interatomic distances, arrangements of functionality within a molecular skeleton, branching, cyclicity, hydrogen bonding propensity, molecular size, etc. are critical information in determining the interaction of a molecule with other molecules of the same compound or different compounds using physicochemical forces. Thus, quantifying the information present within a chemical structure is critical in the predictive modeling of physicochemical properties or biological activities. This is done with the help of different fingerprints and/or chemical descriptors of different dimensions. All predictions are made based on the assumption that molecules with similar structural features will behave similarly. Thus, it becomes necessary to define similarity from a structural sense and also based on their properties, bioactivities, metabolism, and toxicities. Different measures of similarity coefficients or distance measures may be considered for the quantification of chemical similarity. In silico methodologies like Quantitative Structure–Activity Relationship (QSAR) and Read-Across are based on the concept of similarity. While QSAR is a statistical modeling technique to predict the response values of query compounds, Read-Across is a non-statistical approach that derives predictions from the close congeners of a particular query compound. Recently, a fusion of Read-Across and QSAR has led to the development of a concept called Read-Across Structure–Activity Relationship (RASAR) that has been shown to provide enhanced prediction quality using the same amount of chemical information.