You are here:
USE OF A GRAPH THEORETIC SIMILARITY INDEX IN PREDICTION STUDIES OF LINEAR DISCRIMINANTS AND MODELS
Citation:
Jaeger, E. AND P. Jurs. USE OF A GRAPH THEORETIC SIMILARITY INDEX IN PREDICTION STUDIES OF LINEAR DISCRIMINANTS AND MODELS. U.S. Environmental Protection Agency, Washington, D.C., EPA/600/D-86/099 (NTIS PB86194487), 1986.
Description:
A goal of many structure-activity and structure-property studies is to develop the capability to predict the activity or property of interest for previously untested compounds. A quantitative model or a discriminant is developed from a training set of molecules with known activities or properties. The best test of the predictive ability of a model or a discriminant is to predict the activities or properties of a set of unknown compounds, so the agreement between predicted and observed values can be judged. Unknown compounds which are structurally similar to those of the training set stand the best chance of being predicted accurately. Structural similarity can be assessed with atom pairs. The metric is based on the relative frequencies of all unique atom pairs in each compound being compared.