DemQSAR: predicting human volume of distribution and clearance of drugs

Abstract

In silico methods characterizing molecular compounds with respect to pharmacologically relevant properties can accelerate the identification of new drugs and reduce their development costs. Quantitative structure-activity/-property relationship (QSAR/QSPR) correlate structure and physico-chemical properties of molecular compounds with a specific functional activity/property under study. Typically a large number of molecular features are generated for the compounds. In many cases the number of generated features exceeds the number of molecular compounds with known property values that are available for learning. Machine learning methods tend to overfit the training data in such situations, i.e. the method adjusts to very specific features of the training data, which are not characteristic for the considered property. This problem can be alleviated by diminishing the influence of unimportant, redundant or even misleading features. A better strategy is to eliminate such features completely. Ideally, a molecular property can be described by a small number of features that are chemically interpretable. The purpose of the present contribution is to provide a predictive modeling approach, which combines feature generation, feature selection, model building and control of overtraining into a single application called DemQSAR. DemQSAR is used to predict human volume of distribution (VD(ss)) and human clearance (CL). To control overtraining, quadratic and linear regularization terms were employed. A recursive feature selection approach is used to reduce the number of descriptors. The prediction performance is as good as the best predictions reported in the recent literature. The example presented here demonstrates that DemQSAR can generate a model that uses very few features while maintaining high predictive power. A standalone DemQSAR Java application for model building of any user defined property as well as a web interface for the prediction of human VD(ss) and CL is available on the webpage of DemPRED: http://agknapp.chemie.fu-berlin.de/dempred/ .

Applying Linear and Non-Linear Methods for Parallel Prediction of Volume of Distribution and Fraction of Unbound Drug

“…The early attempts for predicting volume of distribution were based on small data sets and did not specify the type of volume of distribution that was used as the endpoint or in some cases used several types of volume of distribution for the model building [3]-[8], [11], [13], [14], [17]. In 2008, a major advance was the publication of a clean, manually curated dataset of V ss [18] that subsequently has been used successfully to build predictive models for V ss [12], [16].…”

Predicting volume of distribution with decision tree-based regression methods using predicted tissue:plasma partition coefficients

“…The Geometric Mean Fold Errors (GMFEs) shown in that table were calculated as: GMFE = antilog 10 (MAE) [3]. The GMFE measure has the advantage of being less affected by extreme outliers, by comparison with the coefficient of determination (which measures the quadratic error) [35]. In order to interpret the GMFE measure, note that a model with a GMFE of 2 makes Vss predictions which are on average twofold off – i.e., 100% above or 50% below the true Vss value.…”

“…Concerning dataset variations, there are several other studies based on Obach et al’s dataset [14], including [5,22,35,45]. However, unlike our work, most of those studies tend to use a substantially smaller version of Obach et al’s dataset, focusing on a single type of compounds or removing compounds that are more difficult to predict for some reason.…”

“…In the work of Demir-Kavuk et al [35], in order to avoid missing values for some descriptors, many compounds for which some descriptors could not be calculated were removed from the original Obach et al’s dataset. More precisely, compounds containing phosphorous, boron and metal atoms, all macrocycles, and some fragment-like compounds (like Metformin) were removed, which reduced the size of the dataset to 584 compounds.…”

“…By contrast, the studies performed by Demir-Kavuk et al [35], Louis and Agrawal [5], and Zhivkova and Doytchinova [22] focused mainly on using variations of multiple linear regression, without building decision tree models. Hence, in the following we initially focus on discussing the decision tree-based regression methods used by del Amo et al [45], contrasting them with the decision tree-based methods used in this work.…”