ReviewThe computational prediction of toxicity
Introduction
The individual properties of a chemical are all derived from, and related to, the unique molecular structure of that chemical. These properties include physical properties, chemical properties and toxicological properties. Because these properties are all derived from the molecular structure of the chemical, it follows that relationships also exist between the different properties of the chemical. These principles form the underlying basis for the prediction of toxicity from chemical structure.
There are essentially two basic approaches to the prediction of toxicity from chemical structure. These are the mechanistic approach (the one favoured by the authors) and the statistical approach. In the mechanistic approach, a hypothesis is proposed that links a group of related chemicals with a particular toxicological endpoint. The hypothesis is then used to select physical, chemical or reactivity parameters to establish a structure/activity relationship (SAR). The resulting relationship can then be tested and the hypothesis and parameters refined, until an adequate prediction model is obtained.
The biological activity of a chemical depends on two factors: first, it must be transported from its site of administration to its site of action; second, it must bind to or react with the receptor or target (i.e. biological activity is a function of partition and reactivity) [1]. The chemical may also undergo metabolic transformation during these processes. If a structure/activity model is deficient in modelling either partition or reactivity, only a partial correlation with the biological response is likely to be observed. For a model to be reliable, the dependent property (in this case, toxicity) for all of the chemicals covered by the relationship has to be elicited by a mechanism that is both common to all of the chemicals in the set, as well as relevant to that dependent property. Careful selection of the chemicals to be used in a particular model is implicit in the mechanistic approach. The mechanistic approach to the prediction of toxicity from chemical structure has been reviewed recently [2•].
In the statistical approach, a structure/activity association or correlation is generated between structural fragments or from (often large) numbers of computed parameters from these fragments, and toxicity from collections of toxicological data. These systems use little or no expert judgement in organising or selecting the data to be processed either on the basis of chemical class or by putative mechanism.
In this review some of the recent developments in the computational prediction of toxicity are discussed.
Section snippets
Skin sensitization
The skin sensitization potential of a chemical can be considered to depend upon reactivity (the ability to react with skin protein either directly or after appropriate metabolism) and partition (the ability of the chemical to reach its site of action).
SAR studies of skin sensitization and respiratory sensitization have been reviewed by Karol et al. [3]. In this review, some key criteria for the selection of chemicals for SAR modelling of skin sensitization were set out:
- 1.
There are a number of
Eye irritation
Worth and Cronin [17] described the application of embedded cluster modelling to the ECETOC (European Chemical Industry, Ecology and Toxicology Centre) eye irritation data set [18]. The irritant chemicals were found to cluster in a two-dimensional scatter plot of logP (log [octanol/water partition coefficient]) and the first order difference valence connectivity index; the latter appears to be a measure of the degree of branching and/or cyclicity of the structure of the chemical (Fig. 1). The
Mutagenicity/carcinogenicity
Benigni et al. [23•] reviewed the value of QSAR models in predicting carcinogenicity of well-defined classes of chemicals such as aromatic amines. A statistically reliable and informative model of aromatic amine carcinogenic potential was constructed using in vivo data. It was determined that aromatic amines require metabolic activation to yield the ultimate carcinogen or mutagen. The principle pathway was identified as formation of a hydroxylamine that decomposes to a reactive nitrenium ion
Oestrogenic activity
The US Environmental Protection Agency initiative to screen and test over 80 000 chemicals for potential to disrupt the endocrine systems of humans and other vertebrates [29] has stimulated the development of QSAR prediction models to this end [30].
A recent study compares the analysis of relative binding affinity values for the displacement of tritiated 17β-oestradiol from the oestrogen receptor of calf uterine cytosol, for a large number of chemicals [31]. Three QSAR methods were
Decision support
The German Federal Institute for Health Protection of Consumers and Veterinary Medicine (BgVV) has developed a rule-based SAR system using the data submitted within the notification procedure for new chemicals of the EU. The system was developed to support regulatory decision making in the assessment of local (skin and eye) irritation/corrosivity hazards of new and existing chemicals. The BgVV database contains proprietary data that cannot be published directly. The database has been evaluated
Conclusions
In reviewing recent developments in the prediction of toxicity from chemical structure, we have attempted to draw attention to some of the problems that can be encountered when working in this area. The prediction of toxicity from chemical structure requires a multi-disciplinary approach covering detailed knowledge of chemistry, toxicology and statistical methods. Given the complexity of most toxicological endpoints, it is not sufficient just to collect data and then attempt to analyse it with
Update
Development of QSAR models for the prediction of oestrogenic activity has continued [43]. Further refinement of the models has resulted in the generation of a ‘flowchart’ approach to filter out chemicals that are unlikely to be active as oestrogen receptor ligands [44•]. The flow chart uses five key structural features (e.g. if a chemical does not contain a ring structure, then it is unlikely to be an oestrogen receptor ligand, etc). The QSAR models are then used to make a quantitative estimate
References and recommended reading
Papers of particular interest, published within the annual period of review,have been highlighted as:
•of special interest
••of outstanding interest
References (44)
- et al.
Computer prediction of possible toxic action from chemical structure; an update on the DEREK system
Toxicology
(1996) - et al.
An expert system rule-base for identifying contact allergens
Toxicol Vitro
(1994) - et al.
Development of an expert system rulebase for the prospective identification of photoallergens
J Photochem Photobiol
(2000) A quantitative structure-activity relationship for the eye irritation potential of neutral organic chemicals
Toxicol Lett
(1995)- et al.
A QSAR model for the eye irritation of cationic surfactants
Toxicol Vitro
(2000) - et al.
QSARs of mutagens and carcinogens: two case studies illustrating problems in the construction of models for noncongeneric chemicals
Mut Res
(1996) - et al.
A new highly specific method for predicting the carcinogenic potential of pharmaceuticals in rodents using enhanced MCASE QSAR-ES software
Regul Tox Pharmacol
(1998) Predicting metabolic pathways by logic programming
J Mol Graph
(1988)Physicochemical descriptors
Prediction of toxicity from chemical structure
Cell Biol Toxicol
(2000)
Structure-activity relationship studies of hypersensitivity
Comments Toxicol
Computer prediction of possible toxic action from chemical structure; the DEREK system
Hum Toxicol
Chemical aspects of contact hypersensitivity
Chemikalien und Kontaktallergie — Eine bewertende Zusammenstellung
Chemikalien und Kontaktallergie — Eine Bewertende Zusammenstellung
Validation and subsequent development of the DEREK skin sensitization rulebase by analysis of the BgVV list of contact allergens
J Chem Inf Comput Sci
Validation and development of the DEREK skin sensitisation rulebase by analysis of the BgVV List of contact allergens
A neural network SAR model for allergic contact dermatitis
Toxicology Methods
Multivariate QSAR analysis of a skin sensitization database
SAR QSAR Environ Res
Exploring the potential for allergic contact dermatitis via computed heats of reaction of haptens with protein end-groups — heats of reaction of haptens with protein end-groups by computation
Quant Struc-Act Relatsh
Correlations between chemical reactivity and mutagenic activity against S. typhimurium TA100 for alpha-dicarbonyl compounds as a proof of the mutagenic mechanism
Mut Res
Embedded cluster modelling: a novel quantitative structure-activity relationship method for generating elliptic models of biological activity
Cited by (47)
ToxDP2 Database: Toxicity prediction of dietary polyphenols
2022, Food ChemistryCitation Excerpt :Q)SARs, are used to predict the physio-chemical, biological, and toxicological properties of compounds from a knowledge of their chemical structure. The underlying principle for the in silico toxicity predictions using these QSARs approaches lies in the relationship of the unique chemical structure and its activity (Barratt & Rodford, 2001). Most of these (Q)SAR models/databases identify toxicity endpoints, including mutagenicity, carcinogenicity, developmental toxicity, skin sensitization, hepatotoxicity, and eye irritation (Bitsch et al., 2006).
In silico toxicity profiling of natural product compound libraries from African flora with anti-malarial and anti-HIV properties
2018, Computational Biology and ChemistryCitation Excerpt :Since these properties are all derived from the latter, it follows that a relationship also exists between the various properties of the compound. This forms the underlying basis for the prediction of toxicity from the chemical structure of a compound (Barratt and Rosemary, 2001; Barratt, 2000; Cronin, 2001). Traditional methods of drug discovery through the evaluation of the biological activities of a large number of compounds have proven to be financially costly, time-consuming and above all, the probability of the evaluated compounds to finally succeed is low (Bala et al., 2010).
Large-scale prediction of drug-target interactions using protein sequences and drug topological structures
2012, Analytica Chimica ActaCitation Excerpt :This is partially caused by the unacceptable toxicity for those drug candidates with the satisfactory activity [5]. Toxicity is usually responsible for many compounds failing to reach the market and for the withdrawal of a significant number of compounds from the market once they have been approved [6–8]. Such failures are very expensive and time-consuming, and have given rise to the saying: “Fail early, fail fast, fail cheap”.
Assessing chronic liver toxicity based on relative gene expression data
2008, Journal of Theoretical Biology