A rational selection of graph-theoretical indices in the QSAR
Milan Randi, A. Sablji, Sonja Nikoli, Nenad Trinajsti · International Journal of Quantum Chemistry · 1988
In this paper we have addressed an important problem: how to develop a strategy for identification of the topological indices which are most important in establishing a model for structure–activity analysis. In particular we consider a rational selection of graph theoretical (topological) indices in QSAR. The approach is illustrated by the study of the toxicity of ethers on mice. The indices examined include the connectivity indices based on adjacency matrix, the conceptually related graph distance indices, and an information theoretic index based on partitioning of electrons in various domains. We have outlined the similarities and differences and discuss advantages and limitations among different approaches. Our emphasis is on the problem how to select in a rational way appropriate graph-theoretical descriptors and how to introduce higher “corrections” in order to improve derived correlation.