345
11 Computational Toxicology in Drug Discovery: Opportunities and Limitations
PharmaExpert is a commercial software providing data on relationships between
drug interactions with antitargets and specific toxicity [45]. The part of known antitargets associated with specific toxicity and side effects are freely available in
Drug Adverse Reaction Target [http://bidd.nus.edu.sg/group/drt/dart.asp] and DrugInduced Toxicity-Related Protein [http://bioinf.xmu.edu.cn/databases/DITOP/] databases. Interaction with a primary target may also result in adverse reactions due to
no local distribution of a target in the body and its multiple functions.
All side effects are usually divided into five classes [46, 47]:
a. Includes dose-dependent side effects, which are often found in preclinical
studies.
b. Includes side effects, the frequency of which is not dose-dependent, they are
often detected in observation of marketed drugs.
c. Includes adaptive functional changes in the body during a long-term drug usage.
They are identified in the measurement of functional parameters during the longterm studies.
d. Includes delayed side effects such as carcinogenicity and teratogenicity.
e. Includes those side effects, which can lead to refuse of the drug.
Side effects of B, C, D and E classes are of considerable interest due to the difficulty
of timely registration.
Currently, QSAR (quantitative structure-activity relationship) and SAR (structure-activity relationship) methods are widely used for computational prediction of
different toxicity types, such as cardio-, hepato-, renal toxicity, teratogenicity, and
carcinogenicity.
In addition to traditional SAR methods used in DEREK, TOPKAT, MCASE
there are examples of using the method of molecular docking to predict side effects
of drugs. For example, Ji and co-authors described the search of targets associated
with side effects for various anti-HIV drugs available on the market [48]. For the
docking procedure, the authors used the docking program INVDOCK [49]. This
program was designed for an automated search of targets for low-weight ligands,
by attempting to integrate them into “cavities” of each protein, e.g. search of the
corresponding binding sites. Three-dimensional protein structures, data about inhibitors, activators, agonists, antagonists, and the toxic side effects data caused by
interaction with targets were obtained from the DART database [50]. As a result,
for 11 anti-HIV drugs, which are inhibitors of HIV protease and reverse transcriptase, nonviral target molecule interactions have been found which lead to the side
effects. Existence of two targets was shown for delavirdine: DNA polymerase beta
and DNA topoisomerase I. The action on these targets causes pancreatitis, nausea,
vomiting, leukopenia, peripheral neuropathy, and abdominal pain.
This method also has significant drawbacks. Firstly, it requires a lot of computational power and time. Secondly, a clear correlation between the target and side
effects is not always established, and mechanisms of adverse drug reactions are not
always known. Thirdly, three-dimensional structures are known only for a limited
number of proteins. All these drawbacks limit the application of the method.
11 Computational Toxicology in Drug Discovery: Opportunities and Limitations
PharmaExpert is a commercial software providing data on relationships between
drug interactions with antitargets and specific toxicity [45]. The part of known antitargets associated with specific toxicity and side effects are freely available in
Drug Adverse Reaction Target [http://bidd.nus.edu.sg/group/drt/dart.asp] and DrugInduced Toxicity-Related Protein [http://bioinf.xmu.edu.cn/databases/DITOP/] databases. Interaction with a primary target may also result in adverse reactions due to
no local distribution of a target in the body and its multiple functions.
All side effects are usually divided into five classes [46, 47]:
a. Includes dose-dependent side effects, which are often found in preclinical
studies.
b. Includes side effects, the frequency of which is not dose-dependent, they are
often detected in observation of marketed drugs.
c. Includes adaptive functional changes in the body during a long-term drug usage.
They are identified in the measurement of functional parameters during the longterm studies.
d. Includes delayed side effects such as carcinogenicity and teratogenicity.
e. Includes those side effects, which can lead to refuse of the drug.
Side effects of B, C, D and E classes are of considerable interest due to the difficulty
of timely registration.
Currently, QSAR (quantitative structure-activity relationship) and SAR (structure-activity relationship) methods are widely used for computational prediction of
different toxicity types, such as cardio-, hepato-, renal toxicity, teratogenicity, and
carcinogenicity.
In addition to traditional SAR methods used in DEREK, TOPKAT, MCASE
there are examples of using the method of molecular docking to predict side effects
of drugs. For example, Ji and co-authors described the search of targets associated
with side effects for various anti-HIV drugs available on the market [48]. For the
docking procedure, the authors used the docking program INVDOCK [49]. This
program was designed for an automated search of targets for low-weight ligands,
by attempting to integrate them into “cavities” of each protein, e.g. search of the
corresponding binding sites. Three-dimensional protein structures, data about inhibitors, activators, agonists, antagonists, and the toxic side effects data caused by
interaction with targets were obtained from the DART database [50]. As a result,
for 11 anti-HIV drugs, which are inhibitors of HIV protease and reverse transcriptase, nonviral target molecule interactions have been found which lead to the side
effects. Existence of two targets was shown for delavirdine: DNA polymerase beta
and DNA topoisomerase I. The action on these targets causes pancreatitis, nausea,
vomiting, leukopenia, peripheral neuropathy, and abdominal pain.
This method also has significant drawbacks. Firstly, it requires a lot of computational power and time. Secondly, a clear correlation between the target and side
effects is not always established, and mechanisms of adverse drug reactions are not
always known. Thirdly, three-dimensional structures are known only for a limited
number of proteins. All these drawbacks limit the application of the method.