Dimension related distance and its application in QSAR/QSPR model error estimation
Yongjin Xu, Hua Gao · QSAR & Combinatorial Science · 2003
Abstract In this paper, the concept of dimension related distance (DRD) is introduced to measure the position of a test compound in the descriptor space of training set compounds in a QSAR/QSPR model. DRD can be used to perform similarity comparison between one compound and a group of compounds. In a QSAR/QSPR model, we find that prediction error is related to the similarity between a test compound and the model training set compounds. DRD seems able to capture this similarity and make it possible to estimate prediction error range of a test compound from calculated DRD.