329
11 Computational Toxicology in Drug Discovery: Opportunities and Limitations
Some database may contain information about the weak activity of compounds
that are labeled like LD 50 > 300 mg/kg. This type of data should be censored. There
are different methods of using this data for QSAR modeling [13, 14], but it is better
to avoid using them. Especially, these values could not be assigned for a specific
threshold (for example, LD 50 > 300 mg/kg cannot be assigned to the LD 50 300 mg/
kg or 600 mg/kg or 900 mg/kg). Quality and reproducibility of the data used for
the toxicity modeling is one of the most important issues for creation of (Q)SAR
models. For example, can rodent’s experimental test results be used for the (Q)SAR
modeling? Although, currently, there are more than 5000 experimental results of
carcinogenicity in rodents for compounds, most of them are not public available
(e.g., private and proprietary archival research) [15]. As well as quality of data, the
transparency of standardized bioassays using specific protocols, such as NTP (National Toxicology Program: http://ntp.niehs.nih.gov/), is important. It is considered
that the usage of rodent’s carcinogenicity data is efficient for (Q)SAR modeling.
However, it was found that the data of experimental protocols are varied depending
on the sources. For some compounds contradictions in the measurement data were
found in the published reports [16], while for other compounds results were well
reproducible [17].
Since the correlation strategy of toxicity values with molecular structures of the
training set is used during in silico modeling, an inability of software to process the
mixture of compounds with a small amount of salts containing ions such as HCl- or
SO 4 or hydrated condition indicates that the wrong representation of endogenous
molecules can result in inaccurate predictions [18]. Some (Q)SAR developers ignore the small molecule ions, but they can affect the pKa of the molecule, which
depends on various physiological conditions and may affect the behavior of the
whole molecule, such as absorption. Therefore, this ignoring leads to errors in the
model as well as prediction results.
11.2.2 Databases with Experimental Toxicity Data of Compounds
Toxicity databases are widely used for developing models, prediction of undesirable drug effects, safety assessment of different xenobiotics, selection of promising
compounds and, ideally, estimation of the risk assessment of compounds. The main
aim in the design of such database is aggregation of the acceptable scientific data
from different toxicity studies for constructing an electronic resource that can be
used for search of chemicals, for model developing and for read-across strategy of
structurally similar compounds. OECD has published the guidance on the quantitative and qualitative read-across approach, which can be used to fill the data gaps
in the risk assessment of chemicals [19]. Computer toxicology databases are often
used by regulatory agencies and industry for the safety assessment and predictions
of xenobiotics side effects [20].
Definition of toxicology databases is varied as the definition of computational
toxicology. The most common definition of toxicology databases is a set of electronic information that can be related to the toxicity of compounds, which is organized
11 Computational Toxicology in Drug Discovery: Opportunities and Limitations
Some database may contain information about the weak activity of compounds
that are labeled like LD 50 > 300 mg/kg. This type of data should be censored. There
are different methods of using this data for QSAR modeling [13, 14], but it is better
to avoid using them. Especially, these values could not be assigned for a specific
threshold (for example, LD 50 > 300 mg/kg cannot be assigned to the LD 50 300 mg/
kg or 600 mg/kg or 900 mg/kg). Quality and reproducibility of the data used for
the toxicity modeling is one of the most important issues for creation of (Q)SAR
models. For example, can rodent’s experimental test results be used for the (Q)SAR
modeling? Although, currently, there are more than 5000 experimental results of
carcinogenicity in rodents for compounds, most of them are not public available
(e.g., private and proprietary archival research) [15]. As well as quality of data, the
transparency of standardized bioassays using specific protocols, such as NTP (National Toxicology Program: http://ntp.niehs.nih.gov/), is important. It is considered
that the usage of rodent’s carcinogenicity data is efficient for (Q)SAR modeling.
However, it was found that the data of experimental protocols are varied depending
on the sources. For some compounds contradictions in the measurement data were
found in the published reports [16], while for other compounds results were well
reproducible [17].
Since the correlation strategy of toxicity values with molecular structures of the
training set is used during in silico modeling, an inability of software to process the
mixture of compounds with a small amount of salts containing ions such as HCl- or
SO 4 or hydrated condition indicates that the wrong representation of endogenous
molecules can result in inaccurate predictions [18]. Some (Q)SAR developers ignore the small molecule ions, but they can affect the pKa of the molecule, which
depends on various physiological conditions and may affect the behavior of the
whole molecule, such as absorption. Therefore, this ignoring leads to errors in the
model as well as prediction results.
11.2.2 Databases with Experimental Toxicity Data of Compounds
Toxicity databases are widely used for developing models, prediction of undesirable drug effects, safety assessment of different xenobiotics, selection of promising
compounds and, ideally, estimation of the risk assessment of compounds. The main
aim in the design of such database is aggregation of the acceptable scientific data
from different toxicity studies for constructing an electronic resource that can be
used for search of chemicals, for model developing and for read-across strategy of
structurally similar compounds. OECD has published the guidance on the quantitative and qualitative read-across approach, which can be used to fill the data gaps
in the risk assessment of chemicals [19]. Computer toxicology databases are often
used by regulatory agencies and industry for the safety assessment and predictions
of xenobiotics side effects [20].
Definition of toxicology databases is varied as the definition of computational
toxicology. The most common definition of toxicology databases is a set of electronic information that can be related to the toxicity of compounds, which is organized