Complementary PLS and KNN algorithms for improved 3D-QSDAR consensus modeling of AhR binding

Abstract

Multiple validation techniques (Y-scrambling, complete training/test set randomization, determination of the dependence of R2test on the number of randomization cycles, etc.) aimed to improve the reliability of the modeling process were utilized and their effect on the statistical parameters of the models was evaluated. A consensus partial least squares (PLS)-similarity based k-nearest neighbors (KNN) model utilizing 3D-SDAR (three dimensional spectral data-activity relationship) fingerprint descriptors for prediction of the log(1/EC50) values of a dataset of 94 aryl hydrocarbon receptor binders was developed. This consensus model was constructed from a PLS model utilizing 10 ppm x 10 ppm x 0.5 Å bins and 7 latent variables (R2test of 0.617), and a KNN model using 2 ppm x 2 ppm x 0.5 Å bins and 6 neighbors (R2test of 0.622). Compared to individual models, improvement in predictive performance of approximately 10.5% (R2test of 0.685) was observed. Further experiments indicated that this improvement is likely an outcome of the complementarity of the information contained in 3D-SDAR matrices of different granularity. For similarly sized data sets of Aryl hydrocarbon (AhR) binders the consensus KNN and PLS models compare favorably to earlier reports. The ability of 3D-QSDAR (three dimensional quantitative spectral data-activity relationship) to provide structural interpretation was illustrated by a projection of the most frequently occurring bins on the standard coordinate space, thus allowing identification of structural features related to toxicity.

Partial least square and k‐nearest neighbor algorithms for improved 3D quantitative spectral data–activity relationship consensus modeling of acute toxicity

“…Supplemental Data, Table S1 (see also Figure ) indicates that the PLS and KNN algorithms achieve their best predictive performance at a different granularity of the 3D‐SDAR space. Similar behavior has been observed previously and was attributed in part to the effect of 13 C simulation error and the way PLS and KNN process input data. Although the parametric space surfaces shown in Figure for both types of model have multiple local maxima, in general PLS is more effective at bin sizes (0.5 Å × 14 ppm × 14 ppm, average R 2 test = 0.743) larger than those for KNN (0.5 Å × 8 ppm × 8 ppm, average R 2 test = 0.764).…”

Rigorous 3‐dimensional spectral data activity relationship approach modeling strategy for ToxCast estrogen receptor data classification, validation, and feature extraction

“…As described in the Modeling approach section, 3D-SDAR performs multiple randomizations on the modeling set and reports averages of the aggregated predicted values [33,39]. Because it is not based on topology or structural fragments, 3D-SDAR can determine a set of substructural units, as well as the distances between them that are related to the compounds' biological activity.…”

“…This generalized structural pattern is commonly referred to as either a pharmacophore or toxicophore. As described in the Modeling approach section, 3D-SDAR performs multiple randomizations on the modeling set and reports averages of the aggregated predicted values [33,39]. Simultaneously, the PLS algorithm records the first N (N is a user-defined parameter) heavily weighted bins for each latent variable.…”

Alignment-independent technique for 3D QSAR analysis

“…In this project, information about the presence of atoms other than carbon was not explicitly included. We have found in several of our previous SDAR modeling projects that the high sensitivity of 13 C δs to their environment is often sufficient for useful reflection of chemical structure in the vicinity, including the presence of nearby heteroatoms [ 1 , 3 , 4 ]. A tessellation of the 3D-SDAR space into regular grids (“binning”) is further used to convert the information contained in a fingerprint into a set of 3D-SDAR descriptors.…”

“…Our earlier studies indicated that 3D-QSDAR models based on lowest energy conformations perform well [ 1 , 3 , 4 ]. However, we hypothesized that use of substrates internally aligned with respect to molecular template molecules rather than energy-minimized conformations might prove beneficial.…”