327
11 Computational Toxicology in Drug Discovery: Opportunities and Limitations
The advantage of computational prediction methods in comparison with the
experimental biological toxicity is their lower cost and time, high efficiency and
reproducibility using the same model. (Q)SAR models have no restrictions related
to chemical synthesis, they can be continuously improved (allow adding important
properties, descriptors, and expansion of the chemical space), and can also help to
reduce the number of experimental animals. It is important to understand that, in
spite of all possibilities for computation prediction methods, they cannot be used
separately from experimental studies and cannot fully replace biological experiments designed to determine the toxicity of compounds. In most cases the computational prediction of toxicity is an effective tool to support the decision about experimental testing of compounds. In addition, an ability of computational methods
to predict ADME (Adsorption, Distribution, Metabolism and Excretion) properties
for virtual structures allows investigating the chemical space without chemical synthesis and experimental testing of compounds.
11.2 General Principles of the Computational Prediction
of Toxicity
Toxicity prediction is based on the assumption that an activity of chemicals depends
on their structures. This statement is valid both for the creation of (Q)SAR models
and for calculation of risk assessment using expert rules, which allow detecting the
toxic compounds based on the so-called alerts, simple structural components associated with the manifestation of toxicity [8, 9, 10].
For the construction of any toxicological model the three key components are
used:
1. Data of chemical compounds (structure, physicochemical and biological properties) and biological experimental systems (species, strain, sex, clinical characteristics, gene expression and protein synthesis);
2. Descriptors are used for description of chemical structures (constitutional
descriptors, topological, electro-topological, quantum-chemical, structural fragments, fingerprints, physicochemical descriptors);
3. Mathematical methods are used to identify the relationship between descriptors
and the biological effects (multiple linear regressions, neural networks, nearest
neighbors, support vector machine, random forest, etc.).
In 2002, the Organization for Economic Cooperation and Development (OECD)
had developed and introduced the guidelines for creation of predictive (Q)SAR
models [11]. The guidelines are that a valid QSAR/QSPR should have:
1. A defined endpoint. An independent variable, which is used for modeling has to
be well defined;
2. An unambiguous algorithm. It is a necessary to use an unambiguous algorithm
for the model building;
11 Computational Toxicology in Drug Discovery: Opportunities and Limitations
The advantage of computational prediction methods in comparison with the
experimental biological toxicity is their lower cost and time, high efficiency and
reproducibility using the same model. (Q)SAR models have no restrictions related
to chemical synthesis, they can be continuously improved (allow adding important
properties, descriptors, and expansion of the chemical space), and can also help to
reduce the number of experimental animals. It is important to understand that, in
spite of all possibilities for computation prediction methods, they cannot be used
separately from experimental studies and cannot fully replace biological experiments designed to determine the toxicity of compounds. In most cases the computational prediction of toxicity is an effective tool to support the decision about experimental testing of compounds. In addition, an ability of computational methods
to predict ADME (Adsorption, Distribution, Metabolism and Excretion) properties
for virtual structures allows investigating the chemical space without chemical synthesis and experimental testing of compounds.
11.2 General Principles of the Computational Prediction
of Toxicity
Toxicity prediction is based on the assumption that an activity of chemicals depends
on their structures. This statement is valid both for the creation of (Q)SAR models
and for calculation of risk assessment using expert rules, which allow detecting the
toxic compounds based on the so-called alerts, simple structural components associated with the manifestation of toxicity [8, 9, 10].
For the construction of any toxicological model the three key components are
used:
1. Data of chemical compounds (structure, physicochemical and biological properties) and biological experimental systems (species, strain, sex, clinical characteristics, gene expression and protein synthesis);
2. Descriptors are used for description of chemical structures (constitutional
descriptors, topological, electro-topological, quantum-chemical, structural fragments, fingerprints, physicochemical descriptors);
3. Mathematical methods are used to identify the relationship between descriptors
and the biological effects (multiple linear regressions, neural networks, nearest
neighbors, support vector machine, random forest, etc.).
In 2002, the Organization for Economic Cooperation and Development (OECD)
had developed and introduced the guidelines for creation of predictive (Q)SAR
models [11]. The guidelines are that a valid QSAR/QSPR should have:
1. A defined endpoint. An independent variable, which is used for modeling has to
be well defined;
2. An unambiguous algorithm. It is a necessary to use an unambiguous algorithm
for the model building;