Applicability Domain (APD) based on the Euclidean distances.

Domain of model applicability must be defined to flag compounds in the test set for which predictions may be unreliable. In this node similarity measurements are used to define the domain of applicability of the model based on the Euclidean distances among all training compounds and the test or virtual screening compounds. The distance of a test compound to its nearest neighbor in the training set is compared to the predefined applicability domain threshold (APD). If the similarity is beyond this threshold, the prediction is considered unreliable (S. Zhang, A. Golbraikh, S. Oloff, H. Kohn, A. Tropsha J. Chem. Inf. Model., 46 (2006), pp. 1984–1995).

APD is calculated as follows:

APD = 'd' + Zσ

Calculation of 'd' and σ is performed as follows: First, the average of Euclidean distances between all pairs of training compounds is calculated. Next, the set of distances that were lower than the average is formulated. 'd' and s are finally calculated as the average and standard deviation of all distances included in this set. Z is an empirical cutoff value and the default value is 0.5. More details and examples can be found here:
www.novamechanics.com/knime.php

If this node is useful to you, please cite the following papers:

G. Melagraki, Α. Afantitis, H. Sarimveis, P.A. Koutentis, O. Igglessi – Markopoulou, G. Kollias "In Silico Exploration for Identifying Structure–Activity Relationship of MEK Inhibition and Oral Bioavailability for Isothiazole Derivatives" Chemical Biology and Drug Design 2010; 76: 397–406

A. Afantitis, G. Melagraki, P.A. Koutentis, H. Sarimveis, G. Kollias. Ligand - based virtual screening procedure for the prediction and the identification of novel β-amyloid aggregation inhibitors using Kohonen Maps and Counterpropagation Artificial Neural Networks” European Journal of Medicinal Chemistry 46 (2011) 497-508