Modeling of the Inhibition Constant (K_i) of Some Cruzain Ketone‐Based Inhibitors Using 2D Spatial Autocorrelation Vectors and Data‐Diverse Ensembles of Bayesian‐Regularized Genetic Neural Networks

Abstract

The inhibition constant (K i ) of a set of 46 ketone-based cruzain inhibitors against cysteine protease cruzain was successfully modeled by means of data-diverse ensembles of Bayesian-regularized genetic neural networks. 2D spatial autocorrelation vectors were used for encoding structural information yielding a nonlinear model describing about 90 and 75% of ensemble training and test set variances, respectively. From the results of a ranking analysis of the neural network inputs, it was derived that atomic van der Waals volume distributions at topological lags 3, 5, and 6 in the 2D topological structure of the inhibitors have a high nonlinear influence on the inhibition constants. Furthermore, optimum subset of autocorrelation vectors well mapped the studied compounds according to their inhibition constant values in a Kohonen self-organizing map.

Design and Fabrication of Open-Tubular Array Gas Sensors Based on Conducting Polypyrrole Modified With Crown Ethers for Simultaneous Determination of Alkylamines

“…The BRANN models are a class of multilayer perceptron networks with high prediction power and reproducibility of results [30]. More details about this class of networks can be found in literature [31,32]. The number of neurons in hidden layer has been optimized using cross-validation methodology and the optimum number was 2.…”

QSAR Modeling of 1‐(3,3‐Diphenylpropyl)‐Piperidinyl Amides as CCR5 Modulators Using Multivariate Adaptive Regression Spline and Bayesian Regularized Genetic Neural Networks

“…Bayesian regularized artificial neural networks (BRANNs) have the potential to give models which are relatively independent of neural network architecture above a minimum architecture [26,27]. The Bayesian regularization estimates the number of effective parameters which are lower than the number of weights.…”

Protein radial distribution function (P-RDF) and Bayesian-Regularized Genetic Neural Networks for modeling protein conformational stability: Chymotrypsin inhibitor 2 mutants

“…Differently to other GA-based approach, the objective of our algorithm is not to obtain a sole optimum model but a reduced population of well-fitted models, with MSE lower than a threshold MSE value, at which the Bayesian regularization guaranties network to posses good generalization abilities [47]. This is because we used MSE of data training fitting instead of cross-validation or test set MSE values as cost function and therefore the optimum model cannot be directly derived from the best-fitted model yielded by the genetic search.…”

“…This process also assures to avoid chance correlations. This approach have shown to be highly efficient in comparison with cross-validation-based GA approach since only optimum models, according to the Bayesian regularization, are cross-validated at the end of the routine and not all the model generated throughout all the search process [47].…”

“…In this work, we validated our regression model using a reasonable method recently employed by our group that consists into a robust validation process by means of NNE [47]. Recently Baumann [51] demonstrated that ensemble averaging significantly improve prediction accuracy by averaging the predictions of several models that are obtained in parallel with bootstrapped training sets and provide a more realistic meaning of the predictive capacity of any regression model.…”