In Silico Modelling of Drug Toxicity - Reducing Drug Attrition

In Silico Modelling of Drug Toxicity: Reducing Drug
Attrition

Content
s
 Abbreviations:...................................................................................5
 Abstract:............................................................................................6
 Introduction:......................................................................................8
 Role of In Silico Technique in Reducing Drug
Attrition……………………………….12
 Figure 1: In Silico Technique Helps in Predicting Toxicites............................13
 Figure 2: Steps Invoved in the Generation of Prediction
Models……………....16
 Table 1:Example of Descriptors used In Silico Methods................................21
 Methods:..........................................................................................23
 Results:.............................................................................................25
 Figure 3: Graph Shows Percetange of Agreement using DEREK Software....25
 Table 2: Percentage of Predicted Drugs Adverse Reaction...........................25
 Discussion: ......................................................................................27
 Figure 4: Metabolic Oxidation of Nefazodone
Drug...................................27
 Conculsion:......................................................................................32
 References:……………………………………………………………………………….…
33
 Appendices: ....................................................................................37

Abbreviations:
ADME……Absorption, Distribution, Metabolism, and Excretion
MOE…..……………………Molecular Operating Environment
PK............................... Pharmacokinetics
QSAR………………Quantitative Structure-Activity Relationship
QSPR…………Quantity Structure Toxicity/Property Relationship
SMILES……….. Simplified Molecular Input Line Entry System

Abstract:
There is a greater need for most pharmaceutical industries to apply different
mechanism to address the issues of drug failure. Both pharmacokinetic and
toxicokinetic profiles play imperative roles in the drug discovery and
development. In silico methods is one of the alternatives to animal testing
that complement in vitro and in vivo toxicity assessments to potentially
minimise animal testing, reduce the cost and time of drug discovery, and
improve toxicity prediction and safety assessment.
This project is looking to see if modern software such as DEREK Nexus is
able to predict toxicities of drugs which have historically been withdrawn
from the market. The project also determines if the ability to use DEREK
software for toxicity prediction would have made a difference if they could
be available at the drugs discovery stage. SMILES notation for example has
been considered by most researchers as the most favourable method of
coding for chemical structures. The study employed the use of Molecular
Operating Environment (MOE) to generate 3D structures and the SDF
(Structure Data File). DEREK nexus software was then used to read the 3D

structures and the SDF generated by theme to predict toxicities. Based on
the data obtained from toxicity predictions by the DEREK Nexus software,
most of its toxicity predictions agree with the previously predicted toxicity
that leads to their withdrawal from the market. Few predictions were in
disagreement with the DEREK Nexus predictions. The use DEREK software
is effective for toxicity prediction. The use of structural alerts in Derek
nexus software in the identification of the reasons for drug withdrawal show
how effective the use expert systems are.
Introduction:
The discovery of various drugs requires a significant investment in
terms of time and resources. It has been considered very significant to
address the high attrition rates in drug development (Agamah et al, 2019).
Drug attrition is the failure of drugs to meet the market entry requirements
due to its adverse effects to the target. This is caused by high level of
toxicity during drug development stages. Toxicity can be defined as the
processing of measuring the side effects of drugs before released into the
clinics and further into the market (Berg 2019). The pharmaceutical
industries, therefore, remain under huge pressure to provide solutions to the
high rates of attrition in drug development.

It is imperative to note that failure of drugs during early stages of
development attrition drug candidates remain a problem. This is despite the
continued application of various software such as TOPKAT, CASE,
ToxTree, and DEREK to predict toxicity from the structure of chemical/drug
compounds. Moreover, for drug development to meet both financial and
economic sustainability there is a great need to reduce animal testing for
toxicological risk assessment. The use of animal in testing for toxicity is
uneconomical in terms finance and time due to high chances of drug failing
to meet the requirements (Sakti, Nishimura, and Nakai, 2018). In silico can
be defined as the process of integration of modern computing and
information technology to improve the process of assessing drugs (Baur et
al., 2020). In silico methods, therefore, offer an alternative that can be used
to address the challenges associated with the high attrition rates in drug
development.
Poor pharmacokinetic and toxicokinetic profiles in the field of medicine
have significantly increased the number of drug failing at the late stage of
development. Toxicokinetic or biokinetics in toxicology refers to processes
such as absorption, distribution, metabolism, and excretion. The
pharmacokinetic profile of drugs that interact with living organisms
significantly relates to the above processes (Clippinger et al, 2018).The fate

of any drug inside the body of an organism is determined by the four
processes.
The overall profiling properties of the four processes and toxic effects
of substances is expressed by the term ADME (T). In silico methods
significantly rely on the ADME properties of compounds to test and assess
the side effects of any drug before they are released for use. The ADME (T)
properties of compounds include absorption, distribution, metabolism and
excretion properties used especially by the in silico methods to test for drug
toxicity level. In silico methods give a better interpretation of in silico
toxicity data for their relevance in terms of a toxic dose by incorporating the
use of physiologically based biokinetics modelling. The concentration of a
given dose in a tissue is usually determined by this model (Naven, & Louise-
May, 2015). It also estimates the effective concentration of an external dose
in the target tissue. In integrated testing work in unison within in silico
methods to make predictions on chemical compounds.
The modelling of various physiological properties of drugs such as
toxicity, metabolism, drug-drug interactions, and carcinogenesis have been
facilitated by the use of Quantitative Structure-Activity Relationship
(QSAR) techniques. The application and use of QSAR in modelling various
physiological properties of drugs require the application of SMILES

(Simplified Molecular Input Line Entry System). SMILES is a scientific
language usually developed to process information on modern chemistry
(Rognan, 2017). Moreover, this technique applies the principles of
molecular graph theory and very small and natural grammar to allow
rigorous structure specification. Molecular graph theory is the use of graph
to represent the structural formula of a chemical compound.
Most of the highly efficient chemical computer applications are
designed to generate unique chemical notation to allow constant-speed
database retrieval, flexible substructure searching, and property prediction
models through their ease of usage by the chemist and machine
compatibility. SMILES is a language used in software to code for structure.
The use and application of SMILES in predicting toxicities are due to
several advantages. Due to its ease of use and accessibility by chemists,
interpretation and generation of chemical notation that are independent of
the specific computer system in use is significantly simplified (Przekwas and
Somayaji, 2018). The ability of SMILES to represent molecular structure
using a linear string of symbols similar to natural language makes is use
very popular by most computational chemist in predicting drug toxicity.
SMILES notation simplifies the chemical structure of chemical compounds
through a two-dimensional valence-oriented picture graph. Hydrogen can

either be included or excluded in the representation of molecular structure
using this language (Worth, 2018). The use of SMILES follows certain rules
and specifications. Letters are always used to represent the atoms within the
structure as shown below.
CC Ethane (C2H6)
N ammine (NH2)
For toxicity of various drugs to be predictd by Derek software, structure
must first be encoded in the SMILES language.
is for example encoded in the SMILE language as
O=C-O-C-C-C
SMILES notation has been considered by most researchers as the most
favourable method for developing a large database for molecular structures
(Effinger et al, 2018. QSAR is a technique that analyses the equation
connecting the structures of molecules to their respective measured activity

and property to predict the activity, reactivity, and properties of an unknown
set of molecules. As earlier mentioned, there is a great need to provide
solutions to the problem drug attrition in both early and late stages of drug
development as well as the need to reduce animal testing for toxicological
risk assessment. This project, therefore, investigates the capabilities of the
modern software in predicting toxicities of drugs which have historically
been withdrawn from the market.
Role of in silico technique in reducing drug attrition:
In silico technique has been extensively applied to help in reducing
the rate at which new drugs are failing in pharmaceutical research and
development. This is due to its ability to identify the gaps that exist in the
current mechanistic and chemical toxicology used in the testing of drug
toxicity levels (Atienzar et al 2016). Computational toxicology has,
therefore, engaged in understanding these gaps to provide the solution to
factors that lead to high drug attrition rate. This is through the application of
modern software such DEREK nexus in enhancing toxicity prediction. In
silico models tend to provide solutions to drug failure by identifying the
gaps that exist in the current safety screening cascades through

understanding the structural, mechanistic, chemical, and the
pharmacological space of the drug molecules (Ekins et al, 2019).
The model contributes significantly in minimizing the rate of drug
failure by identifying the toxicological knowledge gaps that exist in the
already exposed miss-predicted compounds through validated analytical
processes that provide the means for understanding the applicability domain
of each assay or model. The applicability domain (AD) of a QSAR model is
the physico-chemical, structural or biological space, knowledge or
information on which the training set of the model has been developed, and
for which it is applicable to make predictions for new compounds. Naven &
Louise-May conducted a study using regression analysis to analyse
preclinical drugs that were annotated concerning their level of toxicity. Their
findings indicate that in vivo are not very effective in predicting toxicities of
compounds (Worth, 2018). The use of regression analysis yielded an
increased number of drug candidates in the preclinical screening that could
be rendered toxic if the in vivo technique could be used. Therefore, in vivo
technique is ineffective predicting drug toxicity. This was due to the ability
of this method to predict its toxicity level based on the drug physicochemical
properties, conclusively, in silico modeling method is essential in reducing
drug attrition during drug development.

The high probability of drug attrition in the late stages of drug
development is due to inaccurate estimation of drug toxicity level. In silico
techniques through the application and use of QSAR has greatly contributed
in reducing drug failure.
Figure 1: The use of QSAR in predicting toxicity of compounds (Hodos et
al, 2016).
Different categories of assays that describe the safety liabilities of
drugs can be effectively identified using the above regression analysis.
Based on the results from the cytotoxicity, in silico method is a chief player
in determining the level of toxicity in acidic drugs. Figure1 shows the results
obtained from several acidic drugs screened in a cytotoxicity assay using
QSAR. The apparent cytotoxicity of acidic compounds is enhanced by the
reduced level of fatal bovine serum from the downward shift (Richardson et
al., 2017). The result obtained from the above regression analysis help in

identification of compounds that cause significant in silico toxicity hence
reducing drug attrition rate.
In silico differ significantly with other traditional toxicology methods. This
is due the large number of chemicals that are studied, breadth of endpoints
and pathways covered levels of biological organization examined and the
range of exposure conditions considered during toxicity (Parthasarathi, and
Dhawan, 2018).
In vivo toxicology is used to establish the relationship between a chemical
agent and the toxic effects it elicits by testing on living animals. In vitro
approaches apply a different methodology; cultured cells are exposed to
potentially harmful substances and then the toxicity towards these cells
(either of human or animal origin) is measured. In vitro toxicity assays are
conducted primarily with cells or cell lines, ideally from humans or
transfected prokaryotic cells for mechanistic investigations. In silico
toxicology, on the other hand, relies on computerized techniques such as
methods, algorithm, software, and data for toxicity assessment (Patel et al,
2019). Toxicity testing in silico toxicology involves drug analysis,
modelling, and prediction of toxicity.

In silico pharmacology is increasingly becoming relevant in the field
of drug discovery due to its ability to use SMILES in capturing, analysing,
and integrating biological and medical data from different sources (Prior et
al, 2019). The method plays a major role in the reduction of drug attrition
during drug development. It can also be defined as the level of potential
effects drugs or chemicals cause on their final users either through single-
exposure or multiple-exposure (Troth et al., 2019). Toxicity endpoints are
the specific categories of side effects caused by drug testing. Toxicity tests,
on the other hand, are the identification of harmful effects caused by
substances on plants, animals, and human beings. Time and gain
significantly play a critical role in any successful industry. Drug discovery
firms are such firms with huge and complex information to handle and
interpret and therefore need to save on time for a gain in the discovery
process. For the newly discovered drugs to proceed quickly into the clinic
and the market, there is a great need for incorporation of modern prediction
software such as DEREK Nexus software (Parthasarathi, and Dhawan,
2018). In silico method is promising candidates in ensuring problems
associated with late discovery are minimised.
The high rate at which the discovered drugs fail to reach the clinic and
into the market is as a result of the high cost involved and a lot of time

consumed in the process assessing the risks associated with such drugs.
This, therefore, calls for the use and application of in silico computational
methods. A computational method has the ability to provide real time
analysis of drugs pre- synthesis. This again gives it a unique advantage over
in vitro and in vivo. In silico toxicology employs the use of computational
tools in the reduction of attrition in a drug. They include toxicity database of
compounds, various simulation tools, and modelling tools such as statistical
packages (Lucas et al., 2019). Prediction models are generated through
various steps as illustrated in figure 2 below.
Figure 2: Showing all the steps involved in the generation of prediction
models (Kenna, and Ram, 2019).

In silico toxicology also employs several methods or models in its role
in reducing drug attrition. They include, pharmacokinetic models,
pharmacodynamics model, and Quantitative Structure-Activity Relationship,
and Quantity Structure Toxicity/Property Relationship (Przekwas and
Somayaji, 2018). The project heavily relies on QASR/QSPR models in
illustrating the role played by in silico toxicology in reducing drug attrition.
There are two main broad categories of in silico testing methods. They
include the expert system and the predictive systems. Global or expert
systems predict toxicity in compounds by mimicking human reasoning and
formalizing the existing knowledge while predictive expert systems predict
toxicity in compounds through the application of various models
(Richardson et al., 2017). Quantitative Structure-Activity Relationship
(QSAR) is an example of such models that predicts chemicals’ toxicity
through the use of molecular descriptors. It can be defined as the process of
correlating a quantitative measure of chemical structure to either a physical
property or biological effects such as toxic outcomes using a mathematical
model (Hasselgren and Myatt, 2018)
QSAR makes predictions about the toxicity level of a substance using
various parameters also known as molecular descriptors. Molecular
descriptors are the numerical representation of the molecular structure. Apart

from descriptors, QSAR also employs the use of linear regression analysis as
well as other multivariate statistical techniques to make the prediction
(Loiodice, Nogueira da Costa, and Atienzar, 2019). Moreover, molecular
descriptors can be used to make prediction on the level of toxicity during
drug development by calculating the ligands and their target macromolecules
atomic interactions (Papas, and De Leon, 2017). These atomic interactions
form the basis through which the level of toxicity in drugs is determined by
the QSAR.
QSAR also relies on the differences in the chemical reactivity of
molecules to make its prediction in the assumption that, the analogue and
target molecules show toxicology significant metabolic by either converging
or diverging. Moreover, it considers the effects on the bioavailability and
consequently biological responses by identifying the differences in the
physicochemical properties. In silico toxicology, therefore, applies this
technique by first studying the changes in the molecular framework of
known drug samples by monitoring the trend in the variation in their
activity/property and their toxicity (Agamah et al., 2019). Through this
mechanism, in silico toxicology minimizes the unethical practices in the
testing of drug toxicity such as the use of animals in toxicity testing.
Moreover, different molecular properties of drugs can be quantified through

the application and utilization of the mathematical models in the developing
of QSAR equations.
The application and use of QSAR has significantly simplified the
prediction of chemical activity and toxicity levels of drug molecules. This is
due to its importance of the model in building models that correlate
structure, activity, and toxicity (Alex, Harris, and Smith, 2016). The
application of QSAR in toxicology prediction in the drug development,
therefore, makes in silico a good candidate for complementing in vitro and
in vivo toxicology techniques.
QSAR model is preferred due to its simplicity and accuracy in
providing predictions. Its use in in silico methods depends on its ease of
interpretation as a result of its well-defined descriptors and the relationship
between them with fewer instances of overfitting (Holmes et al., 2017).
Overfitting is where a model cannot effectively describe the external
data despite its extreme fit in the training data (Przekwas and Somayaji,
2018). The use of excess or incorrect as well as developing models that
cover all cases in the training set results in overfitting of models. To increase
the effectiveness of toxicity prediction in in silico methods, the QSAR
model applies an appropriate chemical per descriptors.

The suitability of models used for making predictions can be
determined using either the R2 (coefficient of multiple determination) or (s)
standard error of estimate (Waring et al, 2015). The success of the regression
line in measuring the biological data variations such as the measured
responses is estimated by using the coefficient of multiple determination.
The deviation on the predicted response from the regression line, on the
other hand, is estimated using the standard error of estimate. It is imperative
to note that, the effectiveness of the predictive models is directly
proportional to the coefficient of multiple determination and inversely
proportional to the standard error of an estimate. The affectivity of any
predictive model increases when (R2 ) is extremely high and (s) is extremely
low.
In silico toxicology as mentioned earlier, heavily rely on the use of
descriptors in the prediction of toxicity of various drugs. In the selection of
descriptors for use in in silico methods for making toxicity prediction,
several factors should be put into consideration. Such factors include, the
number of descriptors should be made very small. The selected descriptors
for use in in silico methods should be independent, orthogonal, uncorrelated,
and non-redundant (Kalyaanamoorthy and Barakat, 2018). Feature selection,

therefore, ensures such factors are taken into account when applying the use
of descriptors in making predictions in QSAR.
The probability of the QSAR model being closer to the optimal during
the process of making predictions can be increased by ensuring there is no
any correlation between descriptors. Moreover, model interpretation is
simplified when the descriptors used in the QSAR models are made
meaningful (Holmes et al., 2017). Descriptors are, therefore, useful in
predicting toxicology in drugs. Besides, the application of popular
qualitative chemical concepts such as density functional theory based
electronegativity and hardness in understanding of various aspects of
chemical reactivity has significantly simplified the accuracy in predicting
toxicity in the computational method. Several types of descriptors are
applied by in silico toxicology to predict drug toxicity levels.
Table 1: shows summarized examples of the commonly used in silico for
reducing drug attrition (Parthasarathi and Dhawan, 2018).
Type of descriptor Role in In silico Toxicology
EHOMO Represent the highest occupied molecular orbital in the energy
level
ELUMO Represent the lowest occupied molecular orbital in the energy

level
Μ Represent chemical potential where μ= (ELUMO + EHOMO)/2
Δ Represent charge distribution in the drug molecule
Lipole Show lipophilicity distribution within the drug molecule
Hydrophobic/hydrophilic
descriptors
Used to indicate the level of solubility of the drug molecules in
either aqueous or organic medium
The above descriptors, therefore, plays a significant role in supporting
in silico models in predicting potentially toxic molecules. Through the
identification of key molecular descriptors of compounds that are very
crucial in the development of predictive models for testing drug toxicity in
silico methods, therefore, contribute imperatively in reducing drug attrition
(Waring et al, 2015). An appropriate statistical approach such as linear
multiple regression, discriminant analysis, recursive portioning or artificial
neural networks is used to correlate the descriptors with a toxicological
endpoint once the predictive models are developed.
Databases and web tools also contribute a significantly in supporting
the development and use of in silico methods in reducing drug attrition
during drug development. This is made possible through proper organization
of chemical and toxicity data on a particular drug molecule in a consistent

way (Kalyaanamoorthy & Barakat, 2018). In silico toxicology improves
several applications in the testing of drug toxicity and estimation of potential
toxic effects associated with such drug by using the data on individual
chemicals and their relationship with other series of drugs.
In silico methods gives a better interpretation of toxicity data for their
relevance in terms of a toxic dose by incorporating the use of
physiologically based biokinetics modelling. In silico method uses
structural alerts and kinetic models to assess the probability of toxicity of
any drug compound before it is released into the clinic and finally into the
market.
This project is looking to see if modern software such as DEREK Nexus is
able to predict toxicities of drugs which have historically been withdrawn
from the market. The project also determines if the ability to use DEREK
software for toxicity prediction would have made a difference if they could
be available at the drugs discovery stage.
Method:
The data below were obtained from Onakpoya, Heneghan and Aronson,
2019. Toxicity for over 400 drugs were predicted using the DEREK Nexus
software. The 400 compounds analyzed were drugs previously withdrawn

from the market due to their level of toxicity. The toxic endpoints obtained
from the software was used to determine if the previous predictions agrees
with the DEREK predictions on toxicities.
Dataset:
Table2: Summary of the Dataset
Field 1 Compound
Name
Alert
Count
Reason for
Withdrawal
Does
it agree
Medifoxamine Structure-
235
3 Liver Yes
Mefenorex Structure-
236
5 Drug abuse,
psychiatric
Yes
Megestrol acetate Structure-
237
5 Tumorigenicity Yes
Mepacrine
(quinacrine)
Structure-
238
5 Carcinogenicity;
ectopic
pregnancy;
possibly political
Yes
Mepazine Structure-
239
8 Cardiotoxicity Yes
Mephenesin Structure-
240
1 Drug abuse Yes
Norpseudoephedrine
(Phenylpropanolamine)
Structure-
284
0 Hemorrhagic
stroke
No
Nikethamide Structure-
277
0 Nervous system No
The above data set were obtained from Derek predictions. Molecular
Operating Environment (MOE) was used to import the SMILES and convert
them to 3D structures. The 3D structures were then saved as structure 1-400
and opened by DEREK nexus. The toxicities of the over 400 drugs already

withdrawn from the market were then predicted using DERK Nexus
software. The above compounds are drugs recently remove from the market
due to toxicity reasons. Structure-235 is a drug compound from the field of
Medifoxamine as decoded by SMILES as
O(C(Oc1ccccc1)CN(C)C)c1ccccc1. Compound -235 can be represented
structurally as;
Use of software:
Dataset was stored in the SMILES software which contained the following
information: Structures, SMILES string, chemical name.
SMILES strings were obtained from Chemspider online chemical database
(www.chemspider.com). Molecular Operating Environment (MOE) was
used to import the SMILES and convert them to 3D structures. The 3D
structures were then saved (as what?) and opened by DEREK nexus. The

toxicities of the over 400 drugs already withdrawn from the market were
then predicted using DERK Nexus software.
Result:
Graph of Predicted Toxicities against Percentages of Agreement to DEREK
Nexus Toxicity Prediction:
Yes No
0
50
100
150
200
250
300
350
400
11%
89%
Number of compounds
89% of toxicity prediction among the 400 drug compounds analyzed using
DEREK Nexus Software are in agreement with DEREK toxicity predictions.
However, 11% does not agree with Derek toxicity predictions.

Figure 3: Percentages of Agreement using DEREK Nexus software
Table 3: Percentages of Predicted Drug Adverse Reaction
Table 3: Summary of dissagreeing structural alerts
Compounds Reasons for withdrawal
Vitamin B complex (injectable) Cardiovascular, hematologic
Xenazoic acid (xenalamine) Death, hematologic, liver, urinary tract.
Dithiazanine iodideç Cardiovascular, metabolism
Thalidomide Nervous system, teratogenicity
The number structural alert was the basis of determining the degree of
agreement between DEREK Nexus predictions and the previous predictions.
Cardiotoxicity due to adverse drug reaction is the leading cause of drug
attrition by 7%. Infections and infestations adverse drug effects contribute
2% to level of drug attrition. Cardiotoxicity and Hepatotoxicity is the
leading adverse reaction type by 52% among the number drugs allowed.

Based on the frequencies of each toxicity endpoints obtained from Derek
predictions, cardiotoxicity with its structural alerts forms the main reasons
for the removal of drug compounds in the Mepacrine (quinacrine) field. The
data obtained from toxicity prediction by the DEREK Nexus software shows
that most of its toxicity predictions agrees with the previously predicted
toxicity that lead to their withdrawal from the market. 65% of the drugs
withdrawn from the market due their association with endocrine disorders
agree with DEREK’s prediction. Withdrawal of drugs due to skin and
subcutaneous disorders was found to be in 100% agreement with DEREK’s
predictions. Few predictions were in disagreement with the DEREK Nexus
predictions. Drug compounds from the field of Norpseudoephedrine
(Phenylpropanolamine) withdrawn due its Haemorrhagic stroke effects was
in disagreement with DEREK Nexus predictions. The drug was rendered
toxic due to its cardiotoxicity and livertoxicity effects leading to its
withdrawal from the market.
Table 4: Fired DEREK structural alerts and their respective toxicities
Compound Structural Alert Toxicity

Thalidomide Nervous system,
teratogenicity
Aliskiren Angioedema
Dithiazanine iodideç Cardiovascular,
metabolism
Timonacic acid
(thioproline)
Liver, nervous system
Vitamin B complex
(injectable)
Cardiovascular,
hematologic
Xenazoic acid
(xenalamine)
Death, hematologic,
liver, urinary tract.
Discussion:
Most drugs fail or get terminated due to the adverse effects caused by
the drug to the patients. Based on the data obtained from toxicity prediction

by the DEREK Nexus software, most of its toxicity predictions agree with
the previously predicted toxicity that leads to their withdrawal from the
market. Cardiotoxicity with its high frequency toxicity endpoints of 8 alerts
forms the main reason for drug withdrawal. About 11% percent of drugs fail
to be accepted due to failure to meet clinical safety criteria. Non-clinical
toxicology reasons contribute the highest to drug attrition.
The highest rate of drugs failing at their development stage due non-
clinical toxicology is due to poor pharmacokinetics. Failure to determine the
amount of drug in the body and how long it will take in the body contributes
to drug failure in the non-clinical toxicology. Drug attrition due to adverse
drug effects such as cardiac failure, hepatobiliary disorders, endocrine
disorder, renal and urinary disorder, and nervous system disorder results
from incompatibility between the drug and the predicted target. Other
adverse drug effects such as infestations and affections that are toxicology
related results from failure to apply in vitro toxicology screening. According
to Waring et al. (2015), most adverse drug effects results from the difficulty
in interpretation of in vitro data. In silico methods can be applied to give a
better interpretation of in silico toxicity data for their relevance in terms of a
toxic dose by incorporating the use of physiologically based biokinetics
modelling. Besides, failure to effectively predict human pharmacokinetics

from preclinical parameters also plays a significant role in the high rate of
drug attrition. Skin and subcutaneous tissue disorders adverse drug effect is
associated with the difference in the lipophilicity between the patients and
the drugs. The inhibition of the primary target often contributes to a
reasonable proportion of toxicology and safety failures. It is imperative to
note that, the identification of compounds drug quality is attributed to the
effective control of physicochemical properties during compound
optimizations. The use of DEREK software helps in creating structural alerts
to test the level of toxicity of any drug compound before it is released into
the clinic and finally into the market. MOE is, therefore, an important tool in
reducing drug attrition level through its application of DEREK Nexus
software in predicting toxicity.
Despite the promising solutions provided by in silico approach in an
attempt to complement the in vitro and in vivo system, the computational
methods still face few challenges that need to be addressed for its
effectiveness in the drug discovery. In silico method is faced with what is
referred to as the black box dilemma due to poor understanding of the model
constructions (Naven, & Louise-May, 2015). Moreover, the some model
cannot still present multiple mechanisms of compound toxicity.
Additionally, the descriptors used by in silico tools are sometimes confusing,

there is also a lack of transparency in the program as well as clarity on what
is being modelled (Will et al, 2016). The training set of the experimental
data used in this method also lacks quality and transparency. Finally, the
method can also be used to predict the carcinogenicity of non-genotoxic
compounds. The efficiency of costly drug discovery can be improved by
predicting the toxicity of any new drug compound. Many drugs such as
nefazodone have been historically withdrawn from the due to safety reasons.
The UK markets recently withdraw nefazodone from the market owing to
severe, albeit rare, hepatotoxicity (Ivanov, Lagunin, and Poroikov, 2016).
There are several safety liabilities associated with nefazodone, therefore,
made it not reach the market. Moreover, chloroanaline moiety is also
currently undergoing extensive tests to determine whether it meets the
market entry requirements in terms of toxicity.
Figure 4: The metabolic oxidation of nefazodone to form a quinone-imide as a
reactive intermediate as drafted from (Effinger et al, 2018).

The serious organ toxicities caused by the chloroanaline moiety,
therefore, calls for the quick implementation of in silico tools to address the
high level of organ toxicity caused by these drugs (Rezaee and Abdollahi,
2017). Moreover, the inability of the in vitro system to effectively assess
drug safety before released into the market calls for the use and application
of in silico in the testing of toxicity in drugs.
The high level of off-target toxicity has further led to the failure of
other drugs such as ximelgatran, an antithrombotic and anticoagulant. Off-
target toxicity refers to effects caused by drugs binding to a wrong target.
Efficacy of such drugs are reduced by such unintended binding. Such drugs
were recently removed from the UK markets owing to their high liver
toxicity (Richardson et al., 2017). The failure of these drugs and final
removal from the market is due to a lack of signals on their toxicity liability
during the preclinical or clinical screening as a result of lack of assays and
clarity on the underlying mechanisms at the time of their development
(Ivanov, Lagunin, and Poroikov, 2016). The failure of such drugs, therefore,
significantly triggered the implantation of in silico methods to reduce drug
attrition during the early stages of drug development.
Despite the above limitations, the method is promising in providing an
alternative for in vitro techniques in several ways. In silico methods is very

cost-effective to acquire and use, it is less time consuming compared to in
vivo methods hence no delay in drug discovery especially in the late stage of
drug development. Moreover, the method allows for constant optimization
of its models. The use of the same model results in higher reproducibility. In
silico method has limited requirements for compound synthesis and most
importantly, the method reduces the unethical practices in testing the level of
toxicity in drugs.
Reasons for Suitability of Derek in Predicting Toxicities
Derek works best in predicting toxicities leading drug withdrawal due to its
ability to employ the use of structural alerts. Besides, its results are accurate
compared to in vivo and in silico methods.
Conclusion:
DEREK Nexus software is effective in identifying possible drug
toxicities. There is a great need by the pharmaceutical industries to intensify
the use of such software to improve on their methods of toxicity predictions.
The selection of in silico methods to replace other methods of toxicological
prediction is also triggered by its cost-effectiveness to acquire and use. In
silico method also has limited requirements for compound synthesis and

most importantly, the method reduces the unethical practices in testing the
level of toxicity in drugs. Conclusively, ability of the in silico techniques to
employ QSAR models in conjunction with various software to test the rate
of drug toxicity makes it be the best method to offer an alternative that can
be used to address the challenges associated with the high attrition rates in
drug development.
References:
Agamah, F.E., Mazandu, G.K., Hassan, R., Bope, C.D., Thomford, N.E.,
Ghansah, A. and Chimusa, E.R., 2019. Computational/in silico methods in
drug target and lead prediction. Briefings in bioinformatics.
Alex, A., Harris, C.J. and Smith, D.A., 2016. Attrition in the pharmaceutical
industry...
Atienzar, F.A., Blomme, E.A., Chen, M., Hewitt, P., Kenna, J.G., Labbe, G.,
Moulin, F., Pognan, F., Roth, A.B., Suter-Dick, L. and Ukairo, O., 2016.
Key challenges and opportunities associated with the use of in vitro models
to detect human DILI: integrated risk assessment and mitigation plans.
BioMed research international, 2016.
Barton, P. and Riley, R.J., 2016. A new paradigm for navigating compound
property related drug attrition. Drug discovery today, 21(1), pp.72-81.

Berg, E.L., 2019. Human Cell-Based In Vitro Phenotypic Profiling for Drug
Safety-Related Attrition. Frontiers in Big Data, 2, p.47.
Bauer, U. and Breeze, A.L., 2016. “Ligandability” of Drug Targets:
Assessment of Chemical Tractability via Experimental and In Silico
Approaches. Lead Generation: Methods and Strategies, pp.37-63.
Clippinger, A.J., Allen, D., Jarabek, A.M., Corvaro, M., Gaca, M., Gehen,
S., Hotchkiss, J.A., Patlewicz, G., Melbourne, J., Hinderliter, P. and Yoon,
M., 2018. Alternative approaches for acute inhalation toxicity testing to
address global regulatory and non-regulatory data requirements: an
international workshop report. Toxicology in Vitro, 48, pp.53-70.
Effinger, A., O’Driscoll, C.M., McAllister, M. and Fotaki, N., 2018. In Vitro
and In Silico ADME Prediction. In ADME Processes in Pharmaceutical
Sciences (pp. 301-330). Springer, Cham.
Ekins, S., Puhl, A.C., Zorn, K.M., Lane, T.R., Russo, D.P., Klein, J.J.,
Hickey, A.J. and Clark, A.M., 2019. Exploiting machine learning for end-to-
end drug discovery and development. Nature materials, 18(5), p.435.
Fisk, L., Guesne, S., Williams, R., Hewitt, P., Müller, I., Mason, C., Tetley,
T., Ebbrell, D.J., Madden, J.C., Cronin, M.T. and Enoch, S.J., 2018.
Abstracts from the In Vitro Toxicology Society Annual Meeting Senate
House London, United Kingdom November 23–24, 2017. Applied In Vitro
Toxicology, 4(1), pp.69-79.

Gintant, G., Sager, P.T. and Stockbridge, N., 2016. Evolution of strategies to
improve preclinical cardiac safety testing. Nature reviews Drug discovery,
15(7), p.457.
Holmes, A.M., Charlton, A., Derby, B., Ewart, L., Scott, A. and Shu, W.,
2017. Rising to the challenge: applying biofabrication approaches for better
drug and chemical product development. Biofabrication, 9(3), p.033001.
Hasselgren, C. and Myatt, G.J., 2018. Computational Toxicology and Drug
Discovery. In Computational Toxicology (pp. 233-244). Humana Press, New
York, NY.
Hodos, R.A., Kidd, B.A., Shameer, K., Readhead, B.P. and Dudley, J.T.,
2016. In silico methods for drug repurposing and pharmacology. Wiley
Interdisciplinary Reviews: Systems Biology and Medicine, 8(3), pp.186-210.
Ivanov, S.M., Lagunin, A.A. and Poroikov, V.V., 2016. In silico assessment
of adverse drug reactions and associated mechanisms. Drug Discovery
Today, 21(1), pp.58-71.
Kalyaanamoorthy, S. and Barakat, K.H., 2018. Binding modes of hERG
blockers: an unsolved mystery in the drug design arena.
Kenna, J.G. and Ram, R., 2019. Safety Assessment of Pharmaceuticals. In
The History of Alternative Test Methods in Toxicology (pp. 167-176).
Academic Press.

Loiodice, S., Nogueira da Costa, A. and Atienzar, F., 2019. Current trends in
in silico, in vitro toxicology, and safety biomarkers in early drug
development. Drug and chemical toxicology, 42(2), pp.113-121.
Lucas, A.J., Sproston, J.L., Barton, P. and Riley, R.J., 2019. Estimating
human ADME properties, pharmacokinetic parameters and likely clinical
dose in drug discovery. Expert opinion on drug discovery, 14(12), pp.1313-
1327.
Naven, R.T., and Louise-May, S., 2015. Computational toxicology: Its
essential role in reducing drug attrition. Human & experimental toxicology,
34(12), pp.1304-1309
Onakpoya, I., Heneghan, C. and Aronson, J., 2019. Correction to: Post-
marketing withdrawal of 462 medicinal products because of adverse drug
reactions: a systematic review of the world literature. BMC Medicine,
[online] 17(1).
Przekwas, A. and Somayaji, M.R., 2018. Computational pharmacokinetic
modeling of organ-on-chip devices and microphysiological systems. In
Organ-on-a-chip (pp. 311-361). Academic Press.
Papas, K.K. and De Leon, H., 2017. Pancreas and islet preservation. In
Transplantation, Bioengineering, and Regeneration of the Endocrine
Pancreas (pp. 503-527). Academic Press.

Parthasarathi, R. and Dhawan, A., 2018. In silico approaches for predictive
toxicology. In In Vitro Toxicology (pp. 91-109). Academic Press.
Paparella, M., Colacci, A. and Jacobs, M.N., 2017. Uncertainties of testing
methods: What do we (want to) know about carcinogenicity? ALTEX-
Alternatives to animal experimentation, 34(2), pp.235-252.
Polak, S., Tylutki, Z., Holbrook, M. and Wiśniowska, B., 2019. Better
prediction of the local concentration–effect relationship: the role of
physiologically based pharmacokinetics and quantitative systems
pharmacology and toxicology in the evolution of model-informed drug
discovery and development. Drug discovery today.
Patel, C.N., Kumar, S.P., Rawal, R.M., Patel, D.P., Gonzalez, F.J. and
Pandya, H.A., 2019. A multiparametric organ toxicity predictor for drug
discovery. Toxicology Mechanisms and Methods, pp.1-8.
Prior, H., Monticello, T., Boulifard, V., Brennan, F.R. and Kimber, I., 2019.
Integration of Consortia Recommendations for Justification of Animal Use
within Current and Future Drug Development Paradigms. International
Journal of Toxicology, p.1091581819852922.
Richardson, C., Matthews, F., Robinson, L. and Stephan, B.C., 2017. The
Changing Face of Mild Cognitive Impairment (MCI) and Mild Dementia.
Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 13(7),
pp.P544-P545.

Rognan, D., 2017. The impact of in silico screening in the discovery of
novel and safer drug candidates. Pharmacology & therapeutics, 175, pp.47-
66.
Rezaee, R. and Abdollahi, M., 2017. The importance of translatability in
drug discoveryTaylor, K. and Alvarez, L.R., 2019. Regulatory drivers in the
last 20 years towards the use of in silico techniques as replacements to
animal testing for cosmetic-related substances. Computational Toxicology,
p.100112.
Sakti, A.W., Nishimura, Y. and Nakai, H., 2018. Cover Image, Volume 10,
Issue 1. Wiley Interdisciplinary Reviews: Computational Molecular Science,
10(1).
Troth, S.P., Simutis, F., Friedman, G.S., Todd, S. and Sistare, F.D., 2019,
March. Kidney Safety Assessment: Current Practices in Drug Development.
In Seminars in nephrology (Vol. 39, No. 2, pp. 120-131). WB Saunders.
Vian, M., Raitano, G., Roncaglioni, A. and Benfenati, E., 2019. In silico
model for mutagenicity (Ames test), taking into account metabolism.
Mutagenesis, 34(1), pp.41-48.
Worth, A.P., 2018. The future of in silico chemical safety… and beyond.
Will, Y., McDuffie, J.E., Olaharski, A.J. and Jeffy, B.D. eds., 2016. Drug
Discovery Toxicology: From Target Assessment to Translational
Biomarkers. John Wiley & Sons.

Waring, M. J., Arrowsmith, J., Leach, A. R., Leeson, P. D., Mandrell, S.,
Owen, R. M., ... & Wallace, O. (2015). An analysis of the attrition of drug
candidates from four major pharmaceutical companies. Nature reviews Drug
discovery, 14(7), 475-486.
Yamagata, Y., Yamada, H. and Horii, I., 2019. Current status and future
perspective of computational toxicology in drug safety assessment under
ontological intellection. The Journal of toxicological sciences, 44(11),
pp.721-735
Appendices:
The list of appendices includes:
 Appendix I:
 Graph of drug withdrawal due to adverse drug reactions to
the target:

Blood and lymphatic system disorders
Cardiac disorders
Congenital, familial and genetic disorders
Endocrine disorders
Gastrointestinal disorders
Hepatobiliary disorders
Immune system disorders
Injury, poisoning and procedural complications
Neoplasms benign, malignant and unspecified (incl cysts
and polyps)
Nervous system disorders
Psychiatric disorders
Renal and urinary disorders
Respiratory, thoracic and mediastinal disorders
Skin and subcutaneous tissue disorders
Fatilites
Infections and infestations
Other
Total
0% 20% 40% 60% 80% 100%
Adverse drug reaction type
Number of drugs:Yes Number of drugs:No
 Appendix II :
 Dataset of DEREK toxicity predictions
Field 1 Compound
Name
Alert
Count
Reason for
Withdrawal
Does
it agree
Metrizamide Structure-
235
3 Liver Yes
Mefenorex Structure-
236
5 Drug abuse,
psychiatric
Yes
Megestrol acetate Structure-
237
5 Tumorigenicity Yes
Mepacrine
(quinacrine)
Structure-
238
5 Carcinogenicity;
ectopic
pregnancy;
Yes

possibly political
Mepazine Structure-
239
8 Cardiotoxicity Yes
Mephenesin Structure-
240
1 Drug abuse Yes
Norpseudoephedrine
(Phenylpropanolamine)
Structure-
284
0 Hemorrhagic
stroke
No
Nikethamide Structure-
277
0 Nervous system No