Applicability Domain for QSAR Models: Where Theory Meets Reality

Applicability Domain for QSAR Models: Where Theory Meets Reality

Domenico Gadaleta (University of Bari, Bari, Italy), Giuseppe Felice Mangiatordi (University of Bari, Bari, Italy), Marco Catto (University of Bari, Bari, Italy), Angelo Carotti (University of Bari, Bari, Italy) and Orazio Nicolotti (University of Bari, Bari, Italy)
DOI: 10.4018/IJQSPR.2016010102
OnDemand PDF Download:
No Current Special Offers


Quantitative Structure-Activity Relationships are widely acknowledged predictive methods employed, for years, in organic and medicinal chemistry. More recently, they have assumed a central role also in the context of the explorative toxicology for the protection of environment and human health. However, their real-life application has not been always enthusiastically welcomed, being often retrospectively used and, thus, of limited importance for prospective goals. The need of making more trustable predictions has thus addressed studies on the so-called Applicability Domain, which represents the chemical space from which a model is derived and where a prediction is considered to be reliable. In the present study, the authors survey a number of approaches used to build the Applicability Domain. In particular, they will focus on strategies based on: a) physico-chemical, b) structural and c) response domains. Moreover, some examples integrating different strategies will be also discussed to meet the needs of both model developers and downstream users.
Article Preview


In the last years, great concerns were raised by numerous troubles in designing and developing new drugs. Failures can occur at different stages of the drug pipeline, from lead/hit discovery to pre-clinical testing to clinical trials (Hay, Thomas, Craighead, Economides, &Rosenthal, 2014). However, it was observed that a great percentage of these flops (about the 60% of all the drugs fail) is due to the occurrence of inadequate ADMET (Absorption, Distribution, Metabolism, Excretion and Toxicity) properties, not thoroughly evaluated at the very early phases of drug discovery (Weaver, 2008). Needless to say that the need of making trustable predictions is still extremely important and, thus, the role of in silico methods has never been in question although greater attention should be paid in their application and in the interpretation of results (Kubinyi, 1997a; Kubinyi, 1997b; Pisani et al., 2014).

As well-known, Quantitative Structure-Activity Relationships (QSARs) are in silico approaches developed to predict a certain response (i.e., endpoint: pharmacological or toxic effect or physico-chemical property) of chemicals. Normally, a QSAR model consists in a mathematical equation relating an endpoint to a series of meaningful molecular descriptors, numerical attributes featuring the chemicals, which are object of investigation (Nicolotti & Carotti, 2006).

Besides rational drug design (Kubinyi, 1997b), QSARs have more recently proved to be also a valuable tool in the regulatory context for the evaluation and prediction of risks posed by chemicals on the environment and on human health. In December 2006, the European Community (EC) introduced a new regulation concerning the Registration, Evaluation, Authorization and Restriction of Chemicals (REACH) (EC, 2006). The aim of REACH is to ensure a high level of protection of environment and human health by gathering and analyzing a great number of toxicological data for thousands of chemical substances that are on the European market (Nicolotti et al., 2014). Importantly, REACH Article 1 explicitly encourages the use of alternative methods, such as QSARs, for obtaining such information.

The usage of QSAR for regulatory purposes must fulfill a number of important requisites itemized by the Organization for Economic Co-operation and Development (OECD) as follows: 1) a well-defined endpoint; 2) an unambiguous algorithm for model derivation; 3) a clearly defined Applicability Domain (AD); 4) appropriate measures of goodness-of-fit, robustness, and predictivity; and 5) a mechanistic interpretation, if possible. (OECD, 2007).

The present review focuses on the third point, which explicitly claims the need to report, along with a QSAR model, its AD or, in other terms, the areas of the physico-chemical, structural, or biological space where the model is expected to be exploitable and the predictions are assumed to be trustable. Indeed, a QSAR model cannot reliably predict any chemical but only those having physico-chemical, structural and biological features similar to those of Training Set (TS) compounds, used to create a causative relationship with a given endpoint. In this respect, TS compounds define the boundaries of the chemical space where predictions could be considered as the result of data interpolation. On the other hand, predictions made outside such space are the result of data extrapolation and, thus, meaningless, chancy and likely wrong.

Complete Article List

Search this Journal:
Open Access Articles
Volume 7: 4 Issues (2022): Forthcoming, Available for Pre-Order
Volume 6: 4 Issues (2021): 3 Released, 1 Forthcoming
Volume 5: 4 Issues (2020)
Volume 4: 4 Issues (2019)
Volume 3: 2 Issues (2018)
Volume 2: 2 Issues (2017)
Volume 1: 2 Issues (2016)
View Complete Journal Contents Listing