Robust Cross-Validation of Linear Regression QSAR Models

Abstract

A quantitative structure-activity relationship (QSAR) model is typically developed to predict the biochemical activity of untested compounds from the compounds' molecular structures. "The gold standard" of model validation is the blindfold prediction when the model's predictive power is assessed from how well the model predicts the activity values of compounds that were not considered in any way during the model development/calibration. However, during the development of a QSAR model, it is necessary to obtain some indication of the model's predictive power. This is often done by some form of cross-validation (CV). In this study, the concepts of the predictive power and fitting ability of a multiple linear regression (MLR) QSAR model were examined in the CV context allowing for the presence of outliers. Commonly used predictive power and fitting ability statistics were assessed via Monte Carlo cross-validation when applied to percent human intestinal absorption, blood-brain partition coefficient, and toxicity values of saxitoxin QSAR data sets, as well as three known benchmark data sets with known outlier contamination. It was found that (1) a robust version of MLR should always be preferred over the ordinary-least-squares MLR, regardless of the degree of outlier contamination and that (2) the model's predictive power should only be assessed via robust statistics. The Matlab and java source code used in this study is freely available from the QSAR-BENCH section of www.dmitrykonovalov.org for academic use. The Web site also contains the java-based QSAR-BENCH program, which could be run online via java's Web Start technology (supporting Windows, Mac OSX, Linux/Unix) to reproduce most of the reported results or apply the reported procedures to other data sets.

Structural Determinants of Tau Aggregation Inhibitor Potency

“…ODS detergent was used as aggregation inducer in these experiments because of its efficacy with full-length Tau isoforms (47) and because of the reported ability of micellar detergents to depress small-molecule aggregation associated with promiscuous activity (48). Overall, the nine cyanines were found to vary in IC 50 value over nearly 2 orders of magnitude Յ10 M (Table 1). When these values were compared against polarizability (␣) calculated for the ground state using ab initio quantum methods (Table 1), a strong log-linear correlation was observed (R 2 ϭ 0.72; Fig.…”

“…On the basis of a training set of 39 molecules, the best PLR model rationalized pIC 50 (i.e. -log IC 50 ) in terms of 35 molecular descriptors (x variables) collapsed into six linear combinations (latent t variables) ( Table 4).…”

Relating methanogen community structure and anaerobic digester function

“…Cross-validation was performed using the leave-group-out (LGO) procedure (Konovalov et al, 2008). Others have jackknifed data in this way to assess the predictability of QSARs (Nirmalakhandan and Speece, 1988a, Eriksson et al, 2003and Konovalov et al, 2008.…”

3-Dimensional QSAR and molecular docking studies of a series of indole analogues as inhibitors of human non-pancreatic secretory phospholipase A<sub>2</sub>

“…16 QSAR are statistically derived models that can be used to predict the biological activity of untested compounds from their molecular structures. 17,18 This concept helps to understand the role of physicochemical descriptors of compounds in determining the biological activity and in estimating the characteristics of the new and potent compounds, without the chemical synthesis of the compounds. 16 Docking various ligands to the protein of interest followed by scoring to determine the affinity of binding and to reveal the strength of interaction has also become increasingly important in the context of drug discovery.…”