Quantitative Prediction of Rat Hepatotoxicity by Molecular Structure

Quantitative Prediction of Rat Hepatotoxicity by Molecular Structure

Ester Papa (University of Insubria, Varese, Italy), Alessandro Sangion (University of Insubria, Varese, Italy), Olivier Taboureau (Inserm UMR-S 973, Paris Diderot University, Paris, France) and Paola Gramatica (University of Insubria, Varese, Italy)
DOI: 10.4018/IJQSPR.2018070104

Abstract

In this article, Quantitative Structure Activity Relationships (QSAR) were generated to link the structure of over 120 heterogeneous drugs to rat hepatotoxicity. Existing studies, performed on the same data set, could not highlight relevant structure-activity relationships, and suggested models for the prediction of hepatotoxicity based on genomic data. Binary activity responses were used for the development of classification QSARs using theoretical molecular descriptors calculated with the software PaDEL-Descriptor. A statistically powerful QSAR based on six descriptors was generated by using k-Nearest Neighbour (k-NN) method and by applying the Genetic Algorithm (GA) as variable selection procedure. The new k-NN QSAR outperforms published models by providing better accuracy and less false negatives. This model is a valid alternative to approaches based on genomic descriptors, which cannot be used in virtual screening of new compounds (pre- or post-synthesis) without experimental data.
Article Preview

Introduction

In recent years, increasing concerns about the potential harmful effects of compounds present in the environment have led to the fast development of technologies, such as in silico models, which link potential adverse effects of chemicals to their molecular structure (i.e. models based on Quantitative Structure Activity Relationships (QSARs)) or to modifications of gene expression (i.e. toxicogenomic-models). Computational toxicology has been indicated as an effective support for hazard and risk assessment procedures (Afshari, Hamadeh, & Bushel, 2011; Ankley et al., 2006; Boverhof & Zacharewski, 2006; Kienhuis et al., 2011), and finds valuable application in drug design and drug development, to guarantee the safety of new pharmaceuticals (Suter, Babiss, & Wheeldon, 2004). In particular, many toxicological and toxicogenomic studies have addressed toxicity phenomena involving the liver as a target organ for chemicals and drugs (Kienhuis et al., 2011). Liver plays a major role in the metabolism and biotransformation of chemicals and xenobiotics and the early identification of potential for liver injury is a priority in the early stages of the development of pharmaceuticals. Models based on Quantitative Structure-Activity Relationships (QSARs) have been indicated in 2011 by the U.S. Food and Drug Administration (FDA) among the strategies to be applied for the identification of so called drug-induced liver injury (DILI) in the pre-marketing phase of drugs clinical evaluation (Huang, Tung, Fülöp, & Li, 2015). Several studies have highlighted the role of QSAR to predict hepatotoxicity related endpoints (Greene et al., 2010; Huang et al., 2015; Rodgers, Zhu, Fourches, Rusyn, & Tropsha, 2010). A recent study (Low et al., 2011) investigated the complementary role of structural and biological descriptors (i.e. TG-GATEs transcripts (Igarashi et al., 2015)) to model hepatotoxicity in rat induced by exposure to 127 structurally heterogeneous drugs, and derived some mechanistic insights from these observations.

Eighty-five genes were selected from an initial pool of over 31000 transcripts (measured in rat) and highlighted as relevant for rat hepatotoxicity. The modelling procedure involved the development of multiple classification models based on structural theoretical descriptors (i.e. QSAR approach), toxicogenomic information (i.e. the transcripts of the 85 genes) or hybrid information (i.e. combination of structural and toxicogenomic descriptors). Results reported in the literature study show (Low et al., 2011) that models based on toxicogenomic descriptors and hybrid models outperformed results generated by QSAR approach (i.e. based on computed structural properties only). These results are of high importance since they demonstrate the utility of toxicogenomic and hybrid models to provide valuable mechanistic information such as the identification of hepatotoxicity related pathways. However, a main drawback of models based on biological descriptors compared to the QSAR approach, is the large amount of genomic experimental information required to generate predictions for new chemicals (i.e. the 85 transcripts)

Therefore, the main objective of this study is to develop a valid QSAR based alternative to the toxicogenomic model proposed by Low (Low et al., 2011), to predict hepatotoxicity from the chemical structure. The advantage of using the QSAR approach is the possibility to identify and prioritize, at virtual screening level (i.e. pre- or post-synthesis), potential hepatotoxic chemicals possibly starting from a limited number of theoretical molecular descriptors. This is not trivial, since it allows for the a priori identification of chemicals which are “benign by structural design”, i.e. independent of the availability of experimental data.

Data Sets

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 4: 4 Issues (2019): 1 Released, 3 Forthcoming
Volume 3: 2 Issues (2018)
Volume 2: 2 Issues (2017)
Volume 1: 2 Issues (2016)
View Complete Journal Contents Listing