346
A. Zakharov and A. Lagunin
Current methods based on “structure–activity” relationships, have also some disadvantages. Firstly, they are not applicable to predict biological activities of polymers, especially proteins, and inorganic compounds that can act as drugs. Secondly,
many experimental data is obtained from animals. Therefore predictions based on
these end-points cannot be always extrapolated to humans. Thirdly, predictions
were made for the substance itself, but toxic or side effects might be provided by its
metabolites. Fourthly, most of the methods allow predicting only one type of toxicity or one mechanism of side effects.
Thus, the desirable method is to predict the full range of side effects, without a
three-dimensional structure of the target, without spending a lot of time and computing resources. These capabilities are presented in the program PASS—Prediction of Activity Spectra for Substances (www.way2drug.ru/PASS). Its algorithm is
based on MNA descriptors for describing the structure of compounds and modified
Bayesian approach for prediction of biological activities [31]. The program allows
predicting both the mechanisms of toxic effects (interaction with antitargets) and
main side effects of compounds [51]. Table 11.11–11.12 shows the main toxic side
effects and interaction with antitargets predicted by PASS 2012, the number of active compounds in the training set and the prediction accuracy. An average accuracy
of prediction calculated by leave-one-out cross-validation is 87.7 % for side and
toxic effects and 96.7 % for interaction with antitargets.
GUSAR software is another method corresponded to above mentioned requirements. It is based on Multilevel and Quantitative Neighbourhoods of Atoms (MNA,
QNA) descriptors [52, 53] and the self-consistent regression (SCR) algorithm [52,
54]. It was shown that GUSAR may successfully be applied for multiple QSAR
tasks [35, 52, 54, 55, 56].
A freely available on-line service for the simultaneous prediction of thirty two
antitarget end-points (IC 50 , K i and K act ) has been developed on the basis of GUSAR
[http://www.way2drug.com/GUSAR/Antitargets/]. These antitarget end-points are
related to 18 proteins: 13 receptors, 2 enzymes and 3 transporters. The relationships
between predicted drug interactions with antitargets and adverse side effects are
represented in Table 11.10. The accuracy of end-point predictions, calculated for
the appropriate external test sets was typically in the range of R
2
test
= 0.6–0.9. This
service provides a reasonable computational speed (about 2 compounds per second
for the simultaneous prediction of 32 antitarget end-points).
In addition, the web service allows calculating the total number of targets for
which the input compound has been predicted to be active. This can be useful for
selection and prioritization of compounds during the drug discovery process. A particular compound can be considered as a potential source of adverse drug reactions
(ADRs) if interactions with three or more antitargets have been predicted and exceed the cut-off value (1 µM). Compounds for which antitargets are not predicted
can be selected for further development as potential drugs. The service can also help
medical chemists to determine targets (molecular mechanism of toxicity) on which
a particular compound should be tested experimentally, to avoid ADRs.
Fourteen known drugs, which had been withdrawn from the market, were analyzed by Zakharov with co-authors using GUSAR Online web-service [57]. In addi-
A. Zakharov and A. Lagunin
Current methods based on “structure–activity” relationships, have also some disadvantages. Firstly, they are not applicable to predict biological activities of polymers, especially proteins, and inorganic compounds that can act as drugs. Secondly,
many experimental data is obtained from animals. Therefore predictions based on
these end-points cannot be always extrapolated to humans. Thirdly, predictions
were made for the substance itself, but toxic or side effects might be provided by its
metabolites. Fourthly, most of the methods allow predicting only one type of toxicity or one mechanism of side effects.
Thus, the desirable method is to predict the full range of side effects, without a
three-dimensional structure of the target, without spending a lot of time and computing resources. These capabilities are presented in the program PASS—Prediction of Activity Spectra for Substances (www.way2drug.ru/PASS). Its algorithm is
based on MNA descriptors for describing the structure of compounds and modified
Bayesian approach for prediction of biological activities [31]. The program allows
predicting both the mechanisms of toxic effects (interaction with antitargets) and
main side effects of compounds [51]. Table 11.11–11.12 shows the main toxic side
effects and interaction with antitargets predicted by PASS 2012, the number of active compounds in the training set and the prediction accuracy. An average accuracy
of prediction calculated by leave-one-out cross-validation is 87.7 % for side and
toxic effects and 96.7 % for interaction with antitargets.
GUSAR software is another method corresponded to above mentioned requirements. It is based on Multilevel and Quantitative Neighbourhoods of Atoms (MNA,
QNA) descriptors [52, 53] and the self-consistent regression (SCR) algorithm [52,
54]. It was shown that GUSAR may successfully be applied for multiple QSAR
tasks [35, 52, 54, 55, 56].
A freely available on-line service for the simultaneous prediction of thirty two
antitarget end-points (IC 50 , K i and K act ) has been developed on the basis of GUSAR
[http://www.way2drug.com/GUSAR/Antitargets/]. These antitarget end-points are
related to 18 proteins: 13 receptors, 2 enzymes and 3 transporters. The relationships
between predicted drug interactions with antitargets and adverse side effects are
represented in Table 11.10. The accuracy of end-point predictions, calculated for
the appropriate external test sets was typically in the range of R
2
test
= 0.6–0.9. This
service provides a reasonable computational speed (about 2 compounds per second
for the simultaneous prediction of 32 antitarget end-points).
In addition, the web service allows calculating the total number of targets for
which the input compound has been predicted to be active. This can be useful for
selection and prioritization of compounds during the drug discovery process. A particular compound can be considered as a potential source of adverse drug reactions
(ADRs) if interactions with three or more antitargets have been predicted and exceed the cut-off value (1 µM). Compounds for which antitargets are not predicted
can be selected for further development as potential drugs. The service can also help
medical chemists to determine targets (molecular mechanism of toxicity) on which
a particular compound should be tested experimentally, to avoid ADRs.
Fourteen known drugs, which had been withdrawn from the market, were analyzed by Zakharov with co-authors using GUSAR Online web-service [57]. In addi-