Antibacterial Activity of Imidazolium-Based Ionic Liquids Investigated by QSAR Modeling and Experimental Studies

Article abstract of DOI:10.1111/cbdd.12770

Predictive QSAR models for the inhibitors of B. subtilis and Ps. aeruginosa among imidazolium-based ionic liquids were developed using literary data. The regression QSAR models were created through Artificial Neural Network and k-nearest neighbor procedures. The classification QSAR models were constructed using WEKA-RF (random forest) method. The predictive ability of the models was tested by fivefold cross-validation; giving q² = 0.77–0.92 for regression models and accuracy 83–88% for classification models. Twenty synthesized samples of 1,3-dialkylimidazolium ionic liquids with predictive value of activity level of antimicrobial potential were evaluated. For all asymmetric 1,3-dialkylimidazolium ionic liquids, only compounds containing at least one radical with alkyl chain length of 12 carbon atoms showed high antibacterial activity. However, the activity of symmetric 1,3-dialkylimidazolium salts was found to have opposite relationship with the length of aliphatic radical being maximum for compounds based on 1,3-dioctylimidazolium cation. The obtained experimental results suggested that the application of classification QSAR models is more accurate for the prediction of activity of new imidazolium-based ILs as potential antibacterials.

Full text of DOI:10.1111/cbdd.12770

Chem Biol Drug Des 2016
Research Article
Antibacterial Activity of Imidazolium-Based Ionic
Liquids Investigated by QSAR Modeling and
Experimental Studies
Diana Hodyna1,*, Vasyl Kovalishyn1, Sergiy
Rogalsky2, Volodymyr Blagodatnyi3, Kirill Petko4
and Larisa Metelytsia1
Received 1 December 2015, revised 11 March 2016 and
accepted for publication 28 March 2016
The development of antibiotic resistant bacteria, especially
biofilm-forming variants, is becoming a serious threat in
human society since they cause numerous health care-
associated infections (1,2). In some cases the resistance
of bacterial biofilms against antibiotics can be increased
up to 1000-fold compared to isolated colonies (2).
Research in alternative biocides with decreased potential
for resistance development is therefore topical nowadays.
1Department of Medical and Biological Researches,
Institute of Bioorganic Chemistry and Petrochemistry,
NAS of Ukraine, 1, Murmanska Street, Kyiv 02160,
Ukraine
2Laboratory of Modification of Polymers, Institute of
Bioorganic Chemistry and Petrochemistry, NAS of
Ukraine, 50, Kharkivske Schose, Kyiv 02160,
Ukraine
3Department of Microbiology and Epidemiology, Shupyk
National Medical Academy of Postgraduate Education, 9,
Dorogozhytska Street, Kyiv 04112, Ukraine
4Organofluorine Compounds Chemistry Department,
Institute of Organic Chemistry, NAS of Ukraine, 5,
Murmanska Street, Kyiv 02660, Ukraine
*Corresponding author: Diana Hodyna,
Cationic antimicrobials have been in general use within
clinical and domestic settings for over half a century.
Amongst the most useful antiseptics and disinfectants are
the quaternary ammonium compounds (cetrimide, benza-
lkonium chloride), cetylpyridinium chloride, bisbiguanides
(chlorhexidine), and polymeric biguanides (3). However, the
use of such antimicrobial agents has been questioned in
many application areas due to the narrowed spectrum of
their activity, as well as growing problems associated with
the development and spread of bacterial resistance (3–5).
Thus, conventional quaternary ammonium-based antimi-
crobials, as well as chlorhexidine showed low efficacy
against Gram-negative bacteria (3,5–7).
dianahodyna@gmail.com
Predictive QSAR models for the inhibitors of B. sub-
tilis and Ps. aeruginosa among imidazolium-based
ionic liquids were developed using literary data. The
regression QSAR models were created through Artifi-
cial Neural Network and k-nearest neighbor proce-
dures. The classification QSAR models were
constructed using WEKA-RF (random forest) method.
The predictive ability of the models was tested by
fivefold cross-validation; giving q2 = 0.77–0.92 for
regression models and accuracy 83–88% for classifi-
cation models. Twenty synthesized samples of 1,3-
dialkylimidazolium ionic liquids with predictive value
of activity level of antimicrobial potential were evalu-
ated. For all asymmetric 1,3-dialkylimidazolium ionic
liquids, only compounds containing at least one radi-
cal with alkyl chain length of 12 carbon atoms
showed high antibacterial activity. However, the activ-
ity of symmetric 1,3-dialkylimidazolium salts was
found to have opposite relationship with the length of
aliphatic radical being maximum for compounds
based on 1,3-dioctylimidazolium cation. The obtained
experimental results suggested that the application of
classification QSAR models is more accurate for the
prediction of activity of new imidazolium-based ILs as
potential antibacterials.
Over recent years ionic liquids (ILs), which are low-tem-
perature molten salts entirely consisting of discrete cations
and anions, have attracted growing interest as extremely
promising new class of ‘green’ chemicals (8). Due to the
unique combination of negligible vapor pressure, non-
flammability, and low toxicity, as well as high solvency for
a broad range of organic and inorganic compounds, ILs
have found numerous applications such as alternative
‘green’ solvents in organic synthesis and liquid–liquid
extractions (8), polymer chemistry (9), enzyme catalysis
(10,11), drug delivery systems (12) etc. Overall, the ability
to ‘tune’ the physical, chemical, and biological properties
of ILs by modification of the properties of the constituent
of cations and anions create enormous potential for their
industrial use (13).
Many researchers have examined the antimicrobial
potency of different ILs, comprising imidazolium, pyri-
dinium, pyrrolidinium cations and various anions against
Key words: antibacterial activity, imidazolium ionic liquids,
QSAR
ª 2016 John Wiley & Sons A/S. doi: 10.1111/cbdd.12770
1

Hodyna et al.
the gram-positive and gram-negative bacteria, including
biofilm-forming bacteria, and fungi (13–22). Most widely
studied ILs comprise asymmetric 1-alkyl-3-methylimidazo-
lium cations and small inorganic anions (14–16,20–23).
ILs with long alkyl chain substituents have structural simi-
larity with conventional quaternary ammonium based sur-
factants comprising positively charged hydrophilic head
group and hydrophobic tail which therefore gives them
an inherent amphiphilic nature (3,22,23). It has been
found that biological activity of imidazolium based ILs is
closely related to their surfactant properties and therefore
greatly affected by the cation alkyl side chain length,
whereas the type of anion has little effect on their activity
(14,22,23). Among the homologous investigated, ILs with
an alkyl chain length of 12 and 14 carbon atoms showed
the highest efficiency as antimicrobial agents (22,23). The
mechanism of antimicrobial action of ILs is considered
similar to those reported for cationic surfactants which
interact with phospholipid components in the cytoplasmic
membrane resulting in a deformation of membrane per-
meability and lethal leakage of cytoplasmic materials
(3,24). However, in comparison with conventional quater-
nary ammonium biocides, long-chain imidazolium based
ILs were found to possess much broader range of
antimicrobial activity against Gram-negative and Gram-
positive bacteria and fungi (15,23). Moreover, 1-alkyl-3-
methylimidazolium ILs showed strong antibiofilm activity
against a panel of pathogen micro-organisms (13,21).
Thus, ILs are promising candidates for further develop-
ment of new antimicrobial agents with much broader
application area.
Materials and Methods
Data set
The databases for our analysis were retrieved from the litera-
ture and stored in the Online Chemical Modeling environ-
ment (OCHEM)a in Excel format. The biological data
obtained as minimum inhibitory concentration (MIC) were
converted into log (1/MIC) values and used as dependent
variable in the following QSAR studies. Structural formulas
of ILs were drawn using MarvinSketch and imported in
SMILES format. The data set consisted of 47 B. subtilis inhi-
bitors (dataset 1) and 83 Ps. aeruginosa inhibitors (dataset
2). The range of MIC values of the 83 ILs was from 1.1 to
8600 mkg/mL and of 47 ILs from 0.3 to 1443 mkg/mL. The
OCHEM allows using conditions of experiments in the mod-
eling process as descriptors, that’s why the basic character-
istic of ILs such as a type of anion was used as obligatory
condition for properties in OCHEM.
Machine learning methods
In this study, we used the OCHEM to develop high accu-
racy models for predicting antibacterial activity of imida-
zolium-based ILs. Several machine-learning methods were
used to build QSAR models - Associative Neural Network
(ASNN) and k-nearest neighbor Method (k-NN) for creating
of the regression models and WEKA-RF – for classification
models.
Associative neural network
An associative neural network (ASNN) is an ensemble-
based method inspired by the function and structure of
neural network correlations in brain. The method operates
by simulating the short- and long-term memory of neural
networks. The long-term memory is represented by
ensembling of neural network weights, while the short-
term memory is stored as a pool of internal neural network
representations of the input pattern. It allows the ASNN to
incorporate new data cases in short-term memory and
provides high generalization ability without the need to
retrain the neural network weights (25).
It should be noted that ILs based on symmetric 1,3-dialky-
limidazolium cations are practically not been studied for
biological activity. However, these compounds also seem
very promising primarily due to their lower cost and sim-
plicity of the synthesis.
Overall, the increasing quantity of ILs with a huge number
of a combination of different cations and anions increases
formed the problem of creation of computer expert sys-
tems for a prediction of the properties of new ILs, in espe-
cially the bactericidal properties.
This method uses the correlation between ensemble
responses (each molecule is represented in the space of
neural network models as a vector of predictions of neural
network models) as a measure of distance amid the ana-
lyzed cases for the nearest neighbor technique. Thus,
ASNN performs k-NN in the space of ensemble predic-
tions. It provides an improved prediction by the bias cor-
rection of the neural network ensemble. The k-NN method
was used for correction of predicted values averaged over
an ensemble of neural networks based on errors in predic-
tion of k-nearest neighbors in chemical space or in space
of an ensemble of Back Propagation Neural Network
(BPNN) models. This process of correcting predicted
values based on a set of nearby patterns is targeted to
diminish the systematic error for a subset of chemical
Now a number of QSAR models for a prediction of the
ILs physical and chemical properties is known. Predict-
ing the biological properties of ILs is not a standard
procedure due to a limited number of research results,
their unification and due to the difficulties of representa-
tion of their structural features as objects for the pre-
dicting.
In the present work we offer effective QSAR models for
the prediction of antibacterial activity of a number of imi-
dazolium based ILs, comprising both symmetric and
asymmetric 1,3-dialkylimidazolium cations, and experi-
mental studies of their activity as potential antibacterial
agents.
2
Chem Biol Drug Des 2016

QSAR Modeling and Antibacterial Studies of ILs
space and known as Local Correction (LC) or Associative
Memory approach (26).
property-weighted autocorrelation descriptors (charge,
electronegativity, and others)
Dragon V6.0 descriptors (3D) include more than 4885
descriptors organized into 29 different logical blocks.
Among them, we have selected the following types of
descriptors: topological indices, information indices, drug-
like indices, ring descriptors, functional group counts, walk
and path counts, atom-centered fragments, molecular
properties.c
k-Nearest neighbor method
K-Nearest Neighbors predicts the property (activity) using
the average property value of those k compounds from
the training set that are the nearest (in the descriptor
space) to the target compound. The configurable options
are: metrics type (Euclidean distance) and the number of
neighbors. The optimal value of k in the range of 1–100
was automatically detected by OCHEM (27).
Chemaxon descriptors, including elemental analysis,
charge, geometry, and others were used to calculate a
range of physical, chemical and life-science related prop-
erties from chemical structures.d
WEKA-RF (random forest)
The random forest machine learner meaning consisting of
many individual learners (trees). The random forest uses mul-
tiple random trees classifications to votes on an overall classi-
fication for the given set of inputs. In general, in each
individual machine learner vote is given equal weight (28). RF
is a combination of tree predictors such that each tree
depends on the values of a random vector sampled indepen-
dently and with the same distribution for all trees in the forest.
Random inputs and random features produce good results in
classification. This is a high-dimensional non-parametric
method that works well on large numbers of variables (29).
Inductive descriptors (3D) were computed for bound
atoms, groups and molecules using intramolecular dis-
tances, atomic electro-negativities, charge and covalent
radii (32).
Descriptors selection
Unsupervised filtering of descriptors was applied to each
descriptor set before using it as a machine learning input.
Before the development of QSAR models, descriptors,
which contained of two or fewer non-zero values for the
whole training set were eliminated. Besides, descriptors
which were inter-correlated with a linear correlation coeffi-
cient of R2 > 0.95 were batched together and only one
descriptor from the group was taken for development of
QSAR model. This unsupervised filtering does not use any
information about the biological activity and thus does not
introduce selection bias (33), which could provide chance
correlations.
Descriptors
The descriptors were calculated using six descriptor pack-
ages, which cover different representations of chemical
structures from simple type of descriptors and a count of
chemical groups, to packages offering a wide variety of
descriptors types, such as E-State indices, ALogPS,
ADRIANA.Code, Dragon V6.0, Chemaxon, Inductive
descriptors available in the OCHEM.
ALogPS (2D) descriptors were used as parameters of
lipophilicity (logP) and water solubility of chemical com-
pounds. The AlogP estimates are provided only for com-
pounds having C, H, O, N, S, Se, P, B, Si, and halogens (30).
Validation of QSAR models
The accuracy of all individual models was evaluated using
the fivefold cross-validation (CV) procedure. In the fivefold
cross-validation the initial data set was divided into five
subsets of approximately equal size. Each QSAR model
was built using 4/5 of the compounds from the initial train-
ing set. The remaining 20% of compounds were predicted
and were used to estimate the model accuracy. This pro-
cedure was sequentially repeated five times producing five
different external validation data sets and corresponding
training set molecules.e These predictions were used to
estimate the CV validity of the model. Then the average
statistical coefficients for all five-test sets were computed
by OCHEM.
E-State indices (2D) are separated on atom/bond type,
which were attracted atom indices, atom and bonds counts.
E-State indices combine the electronic character and the
topological environment of each skeletal atom (31).
ADRIANA.Code (3D) comprises a unique combination of
methods for calculating molecular descriptors on a sound
geometric and physicochemical basis. Thus, they are all
prone to an interpretation and allow the understanding of
the influence of various structural and physicochemical
effects on the property under investigation.b ADRIANA.-
Code descriptors includes global molecular descriptors
(molecular weight, topological polar surface area, molecu-
lar dipole moment, number of atoms, number of rotatable
bonds, molecular and ring complexity, number of hydro-
gen bonding acceptors and donors) and spatial or 3D
Statistical coefficients
Classification models
The OCHEM server uses the average correct classification
rate (in percents) as a measure of classification models
Chem Biol Drug Des 2016
3

Hodyna et al.
quality. The correct classification rate is complemented
with a confusion matrix that shows a number of com-
pounds classified correctly for every class as well as
details of misclassified compounds, e.g. how many com-
pounds from the class A are classified to belong to the
class B. We used the most common case of classification
models - binary classification, where the instances belong
to either active or inactive class. To assess the classifica-
tion ability the following statistical measures which are
applicable to binary classification models: accuracy (per-
centage of correctly classified samples), class hit rate –
sensitivity for active tasks class and specificity for inactive
class, precision – active and inactive predictive value were
calculated.
forming units (CFU) in 1 mL. About 0.02 mL of the tested
compounds applied on standard paper discs. All com-
pounds were tested at identical concentrations 1% and
0.1% and were presented on a disc in moles respectively.
The compounds which formed zones of growth inhibition
of microbes ≥15 mm were accepted as active.
Results and Discussion
Statistical results of QSAR modeling
Classification and regression QSAR models for predicting
antimicrobial activity of imidazolium-based ILs was created
in the search for potential new Ps. aeruginosa and B. sub-
tilis inhibitors. In preprocessing steps using ChemAxon
Standardizerd all structures were standardized and then
optimized with Corina.h
Regression models
Regression models were evaluated with the cross-valida-
tion coefficient, q2. The prediction performance of the
methods was compared using the root mean squared
error (RMSE), mean absolute error (MAE), and squared
correlation coefficient, R2.f
Classification models
Two QSAR models were developed by the WEKA-RF
method using the ALogPS, E-State indices, ADRIANA.-
Code, Dragon V6.0, Inductive descriptors and Type of
anion. Before creating QSAR models, the numerical values
(MIC) were discretized so that 50% of the compounds
were considered active and 50% inactive. The results on
classification models of antibacterial activity of ILs against
B. subtilis and Ps. aeruginosa are summarized in Table 1.
QSAR models are considered effective at the value of
R2 > 0.6 and q2 > 0.5 and can be used to assess the
activity of new compounds (34).
Applicability domain
Each QSAR model should have an applicability domain
(AD) since the model could only cover a limited range of
the entire chemical space. AD was determined for each
model to avoid incorrect predictions. A unique feature of
the OCHEM is the automatic assessment of the prediction
accuracy. The estimation of the accuracy is based on the
concept of ‘distance to a model’, (DM) (35) i.e. some
numeric value is estimated solely from molecular structures
and experimental conditions, which correlates with the
average model performance. In the current study, we used
the standard deviation of predictions of the ensemble of
the regression models in the bagging approach (BAG-
GING-STD) or in a consensus model (CONSENSUS-STD)
as a measure to distinguish reliable and non-reliable pre-
dictions under the OCHEM.g
The percentage of correctly classified samples (Accuracy)
for classification QSAR models of antibacterial imida-
zolium-based ILs with activity against B. subtilis and
Ps. aeruginosa was 83% and 88% respectively (Table 1).
Regression models
To improve the accuracy of predicted results of classifica-
tion QSAR models four regression QSAR models were
created using ASNN method including ALogPS, E-State
indices, ADRIANA.Code, Dragon V6.0, Chemaxon, Induc-
tive descriptors, Type of anion and one model using k-NN
method including ALogPS, E-State, Type of anion descrip-
tors. The predicted activity for tests molecules using a
consensus model was calculated as a usual conditionally
weighted average from predictions of all five models.
As result, OCHEM automatically detects AD for each new
molecule (i.e. whether a molecule inside or outside of AD)
for current QSAR model.
Table 1: Comparison of parameters of classification QSAR mod-
els of antibacterial imidazolium ILs with activity against B. subtilis
(dataset 1) and Ps. aeruginosa (dataset 2) by WEKA-RF method
Antimicrobial activity of 1,3-dialkylimidazolium
ionic liquid
Parameters
dataset 1
dataset 2
Number of descriptors
Precision (active)
Precision (inactive)
Sensitivity
Specificity
Accuracy
191
0.81
0.86
0.88
0.79
83% ꢀ 5.0
146
0.86
0.90
0.90
0.86
88% ꢀ 4.0
The method of disc diffusion in Mueller-Hinton agar (36)
was used for testing antibacterial activity of imidazolium-
based ionic liquids against gram-positive B. subtilis ATCC
6633 and gram-negative Ps. aeruginosa ATCC 27853.
Zones of inhibition were formed by testing compounds
under condition of microbial loading 1 9 105 colony
4
Chem Biol Drug Des 2016

QSAR Modeling and Antibacterial Studies of ILs
Table 2: Molecular descriptors of QSAR models of antibacterial
activity
7
6
5
4
3
2
Model Method Types of descriptors
Number of descriptors
1
ASNN
ALogPS
14–18
E-State indices
Type of anion
ADRIANA.Code
Type of anion
Chemaxon
Type of anion
Dragon V6.0
Inductive descriptors
Type of anion
ALogPS
2
3
4
ASNN
ASNN
ASNN
63
37
114
5
k-NN
18
E-State indices
Type of anion
2
3
4
5
6
7
The methods, types and number of descriptors by each
studied micro-organism are presented in the Table 2.
Observed log (1/MIC)
Figure 1: Plots of experimental values and predicted values of
the consensus QSAR model for dataset 1.
Results of QSAR modeling for B. subtilis inhibitor
(dataset 1)
statistical parameters presented in Table 4, the QSAR
models of activity of imidazolium ILs as inhibitors of
Ps. aeruginosa (dataset 2) are stable and predictive both
internally, verified by the statistical parameters (high value
of cross-validation parameters q2 = 0.81–0.92 and low
MAE = 0.18–0.24). The results of the correlation between
the observed and predicted values log(MIC) of the consen-
sus QSAR model for dataset 2 are presented graphically
in Figure 2.
Table 3 summarizes the statistical parameters of the
regression QSAR models for dataset 1. According to the
statistical significance presented in Table 3, the cross-vali-
dation coefficient q2 for all models ranged from 0.77 to
0.86, confirms their high predictive ability. The correlation
between the observed and predicted values log(MIC) of
the consensus QSAR model for dataset 1 is given graphi-
cally in Figure 1.
The graphically results reported in Figure 1 show that the
difference between experimental and predicted values of
the activity of imidazolium ILs against B. subtilis (dataset 1)
does not exceed 0.9 log units for all compounds with a
high cross-validation coefficient q2 = 0.86 and low MAE
(MAE = 0.34).
After analyzing the prediction results for all compounds of
this dataset, we found that all compounds except one
compound are well-predicted with smaller residues lower
than 1 log unit (Figure 2). The total accuracy q2 of the
consensus QSAR model of antibacterial activity of ILs
against Ps. aeruginosa (dataset 2) was about 0.91,
MAE = 0.18.
Results of QSAR modeling for Ps. aeruginosa
inhibitor (dataset 2)
Statistical parameters of the regression QSAR models for
dataset 2 are presented in Table 4. According to the
As it can be seen from the Tables 1, 3 and 4, the predic-
tive ability of classification models and regression models
is high. Twenty structures of 1,3-dialkylimidazolium ionic
Table 3: Statistical coefficients of QSAR models for dataset 1
Statistical coefficients
QSAR Models
q2
R2
RMSE
MAE
Model 1
Model 2
Model 3
Model 4
Model 5
Consensus model
0.86 ꢀ 0.04
0.84 ꢀ 0.05
0.85 ꢀ 0.04
0.84 ꢀ 0.04
0.77 ꢀ 0.07
0.86 ꢀ 0.04
0.86 ꢀ 0.04
0.85 ꢀ 0.04
0.86 ꢀ 0.04
0.84 ꢀ 0.04
0.79 ꢀ 0.06
0.86 ꢀ 0.04
0.43 ꢀ 0.04
0.46 ꢀ 0.05
0.44 ꢀ 0.04
0.47 ꢀ 0.04
0.55 ꢀ 0.06
0.43 ꢀ 0.04
0.34 ꢀ 0.04
0.35 ꢀ 0.04
0.35 ꢀ 0.04
0.38 ꢀ 0.04
0.41 ꢀ 0.05
0.34 ꢀ 0.04
q2, cross-validation coefficient; R2, squared correlation coefficient; RMSE, root mean squared error; MAE, mean absolute error.
Chem Biol Drug Des 2016
5

Hodyna et al.
Table 4: Statistical coefficients of QSAR models for dataset 2
Statistical coefficients
QSAR Models
q2
R2
RMSE
MAE
Model 1
Model 2
Model 3
Model 4
Model 5
Consensus model
0.89 ꢀ 0.04
0.92 ꢀ 0.02
0.91 ꢀ 0.03
0.91 ꢀ 0.02
0.81 ꢀ 0.08
0.91 ꢀ 0.03
0.89 ꢀ 0.03
0.92 ꢀ 0.02
0.91 ꢀ 0.02
0.91 ꢀ 0.02
0.85 ꢀ 0.05
0.91 ꢀ 0.03
0.3 ꢀ 0.05
0.25 ꢀ 0.02
0.27 ꢀ 0.03
0.27 ꢀ 0.02
0.39 ꢀ 0.08
0.26 ꢀ 0.04
0.19 ꢀ 0.02
0.18 ꢀ 0.02
0.2 ꢀ 0.02
0.2 ꢀ 0.02
0.24 ꢀ 0.03
0.18 ꢀ 0.02
q2, cross-validation coefficient; R2, squared correlation coefficient; RMSE, root mean squared error; MAE, mean absolute error.
evacuated and, the atmosphere of tetrafluoroethylene was
6
done. The reaction mixture was vigorously stirred in the
atmosphere of tetrafluoroethylene at 25–30 °C while
adsorption of the gas stopped (about 30–35 h). The sol-
5
vent was removed at reduced pressure. The residue was
taken with 200 mL of ether and washed with water
(50 mL) to remove unreacted imidazole. The ether solution
4
3
2
1
was dried with MgSO4. The solvent was distilled off, and
the product was distilled in vacuum. Yield 78%, b. p.
50 °C/15 mmHg.
ILs comprising asymmetric 1.3-dialkylimidazolium cations
were synthesized according to the methods described in
Refs 37,38 (Figure 4).
To obtain water soluble chloride or bromide 1,3-dialkylimi-
dazolium ionic liquids, equimolar mixture of N-substituted
imidazole and corresponding halogenoalkane was stirred
at 120–140 °C from 2 to 20 h under an argon atmo-
sphere. The obtained viscous liquids were cooled to room
temperature and washed three times with hexane-ethyl
acetate mixture or neat hexane in order to remove unre-
acted initial compounds. Solid compounds were recrystal-
lized from hexane-ethyl acetate mixture or from neat
hexane. Residual solvent was removed in vacuum
10 mbar at 70–80 °C for 24 h.
1
2
3
4
5
6
Observed log (1/MIC)
Figure 2: Plots of experimental values and predicted values of
the consensus QSAR model for dataset 2.
liquid (Figure 3) with different activity levels of antimicrobial
potential were identified for the synthesis and biological
testing.
Symmetric 1,3-dialkylimidazoliun ILs with chloride or bro-
mide anion were synthesized by following method (Fig-
ure 5).
Synthesis of 1,3-dialkylimidazolium ionic liquid
The following chemicals were used for the synthesis of
ionic liquids: imidazole (Shanghai Synnad, China),
1-methylimidazole, 1-butylimidazole, 1-benzylimidazole,
1-allylimidazole, 1-(2-hydroxyethyl)imidazole, N,N-dimethyl-
formamide, methylene chloride, ethyl acetate, hexane
(Fluka), 1-bromohexane, 1-bromooctane, 1-bromononane,
1-bromodecane, 1-bromododecane, 1-chlorododecane,
1-bromohexadecane, tetrafluoroboric acid (48% in H2O;
Sigma-Aldrich, St. Louis, MO, USA).
Imidazole (5 g, 0.073 mol) was dissolved in 40 mL of dry N,
N-dimethylformamide. Potassium carbonate (11.5 g.
0.08 mol) and corresponding halogenoalkane (0.15 mol)
were added to the solution and then the mixture was stirred
at 80–120 °C for 20 h. After cooling to room temperature,
the mixture was poured into 100 mL of water. The top water
immiscible layer of ionic liquid was extracted with methylene
chloride (2 9 50 mL) and dried overnight with sodium sul-
fate. The solvent was removed by distillation, the obtained
viscous liquid was purified by washing with ethyl acetate-
hexane mixture (3 9 50 mL). If products were solid, they
were purified by recrystallization from ethyl acetate-hexane
mixture or from neat hexane. Residual solvents were
removed in vacuum 10 mbar at 70–80 °C for 24 h.
1-tetrafluoroethylimidazole was synthesized by following
method. To a solution of 13.6 g (0.2 mol) of imidazole in
150 mL of dry THF in 500 mL round-bottom flask,
100 mg of potassium suspended in 2 mL of hexane, was
added. After finishing of gas (H2) evolution, the flask was
6
Chem Biol Drug Des 2016

QSAR Modeling and Antibacterial Studies of ILs
CH3
CH3
CF2CF2H
N
N
N
N
_
BF4
_
BF4
+
N
_
BF4
+
N
+
N
+
N
_
BF4
CH3
CH2(CH2)4CH3
CH2(CH2)6CH3
CH2(CH2)6CH3
[CH2PhC1IM]BF4 (1 a) [C6C1IM]BF4 (1 b) [C8C1IM]BF4 (1 c) [C8CF2CF2HIM]BF4 (1 d)
CH2(CH2)6CH3
CH2(CH2)2CH3
CH2CH2OH
CH3
N
N
N
N
_
BF4
_
BF4
_
BF4
_
BF4
+
N
+
N
+
N
+
N
CH2(CH2)10CH3
CH2(CH2)10CH3
CH2(CH2)10CH3
CH2(CH2)6CH3
[C12C1IM]BF4 (1 e) [C12C4IM]BF4 (1 f) [C12CH2CH2OHIM]BF4 (1 g) [C8C8IM]BF4 (1 h)
CH2(CH2)10CH3
CH2(CH2)10CH3
CH3
Cl
CH3
N
N
N
_
BF4
N
_
_
_
+
+
N
+
N
+
N
BF4
Cl
N
CH2(CH2)10CH3
CH2(CH2)10CH3
CH2(CH2)14CH3
CH2(CH2)10CH3
[C12C12IM]BF4 (1 i) [C16C1IM]BF4 (1 j)
[C12C1IM]Cl (2 a)
[C12C12IM]Cl (2 b)
CH2CH=CH2
CH3
CH2(CH2)6CH3
CH2(CH2)7CH3
N
N
N
N
_
Br
_
_
Br
_
+
N
+
N
+
N
+
N
Br
Br
CH2(CH2)10CH3
CH2(CH2)6CH3
CH2(CH2)6CH3
CH2(CH2)7CH3
[C8C1IM]Br (3 a) [C12CH2CH=CH2IM]Br (3 b) [C8C8IM]Br (3 c)
[C9C9IM]Br (3 d)
CH2(CH2)10CH3
CH3
CH2(CH2)8CH3
CH3
N
N
N
N
+
_
_
_
Br
_
+
N
+
N
+
N
Br
Br
(CF3SO2)2N
N
CH2(CH2)8CH3
CH2(CH2)10CH3
CH2(CH2)14CH3
CH2(CH2)10CH3
Figure 3: Chemical structures of 1,3-
dialkylimidazolium ILs.
[C10C10IM]Br (3 e) [C12C12IM]Br (3 f)
[C16C1IM]Br (3 g)
[C12C1IM](CF3SO2)2N (4)
R
N
R
N
N
120–140 oC
HBF
4 , H2O
+
N
+
N
RHal
+
Hal
BF4
– HHal
N
R'
R'
R'
R = C6H13, C8H17, C12H25, C16H33, CH2Ph Hal = Cl, Br
R' = CH3, C4H9, CH2CH2OH, CH2CH=CH2, CF2CF2H
Figure 4: Synthesis of asymmetric 1,3-
dialkylimidazolium ionic liquids.
R
R
N
N
N
DMF, K2CO3
HBF
4 , H2O
+
N
+
N
2 RHal
+
Hal
BF4
80–120 o
– HHal
C
N
H
R
R
R = C8H17, C9H19, C10H21, C12H25
Hal = Cl, Br
Figure 5: Synthesis of symmetric 1,3-
dialkylimidazolium ionic liquids.
Chem Biol Drug Des 2016
7

Hodyna et al.
To obtain water-immiscible ILs, the corresponding chloride
or bromide 1,3-dialkylimidazolium IL was dissolved in
water (10 wt%) followed by the addition of tetrafluoroboric
acid (~10% molar excess over IL concentration). The
formed water-immiscible layer (or semi-solid residue) was
extracted with methylene chloride (2 9 50 mL) and dried
overnight under sodium sulfate. The solvent was distilled
off and the product was dried in vacuum 1 mbar at
70–80 °C for 12 h.
19F NMR (188 MHz, DMSO-d6): d = ꢁ148.2 (s, 4F, BF4).
1-dodecyl-3-butylimidazolium tetrafluoroborate (1f)
1H NMR (300 MHz, CDCl3): d = 0.87–0.95 (m, 6H, CH3),
1.24 (m, 20H, CH3(CH2)10), 1.87 (m, 4H, NCH2CH2), 4.21
(m, 4H, NCH2), 7.39 (br s, 1H, C4–H), 7.43 (br s, 1H, C5–
H), 8.87 (s, 1H, C2–H).
19F NMR (188 MHz, CDCl3): d = ꢁ151.7 (s, 4F, BF4).
1-benzyl-3-methylimidazolium tetrafluoroborate (1a)
m. p. 73–74 °C.
1-dodecyl-3-(2-hydroxyethyl)imidazolium tetrafluoroborate
(1g)
1H NMR (300 MHz, DMSO-D6): d = 3.86 (s, 3H, NCH3),
5.42 (s, 2H, NCH2), 7.37–7.42 (m, 5H, Haryl), 7.69 (br s,
1H, C4–H), 7.76 (br s, 1H, C5–H), 9.17 (s, 1H, C2–H).
1H NMR (300 MHz, DMSO-D6): d = 0.83 (t, 3H, CH3), 1.23
(m, 18H, CH3(CH2)9), 1.79 (m, 2H, NCH2CH2), 3.49 (br s, 1
H, OH), 3.75 (m, 2H, NCH2CH2OH), 4.17-4.22 (m, 4 H,
NCH2), 7.69–7.71 (m, 2H, C4–H, C5–H), 9.07 (s, 1H, C2–H).
19F NMR (188 MHz, DMSO-D6): d = ꢁ148.4 (s, 4F, BF4).
1-hexyl-3-methylimidazolium tetrafluoroborate (1b)
19F NMR (188 MHz, DMSO-D6): d = ꢁ148.6 (s, 4F, BF4).
1,3-dioctylimidazolium tetrafluoroborate (1h)
1H NMR (400 MHz, DMSO-D6): d = 0.86 (t, 3H, CH3),
1.27 (m, 6H, (CH2)3), 1.78 (m, 2H, CH2), 3.85 (s, 3H,
NCH3), 4.15 (t, 2H, NCH2), 7.68 (br s, 1H, C4–H), 7.75 (br
s, 1H, C5–H), 9.07 (s, 1H, C2–H).
1H NMR (300 MHz, CDCl3): d = 0.84–0.86 (m, 6H, CH3),
1.25–1.31 (m, 20H, CH3(CH2)5), 1.87 (m, 4H, NCH2CH2),
4.18–4.22 (m, 4H, NCH2), 7.4 (m, 2H, C4–H, C5–H), 8.86
(s, 1H, C2–H).
19F NMR (188 MHz, DMSO-D6): d = ꢁ147.3 (s, 4F, BF4).
1-octyl-3-methylimidazolium tetrafluoroborate (1c)
19F NMR (188 MHz, CDCl3): d = ꢁ151.6 (s, 4F, BF4).
1,3-didodecylimidazolium tetrafluoroborate (1i)
m. p. 52–53 °C.
1H NMR (300 MHz, DMSO-D6): d = 0.85 (t, 3H, CH3),
1.25 (m, 10H, CH3(CH2)5), 1.78 (m, 2H, NCH2CH2), 3.85
(s, 3H, NCH3), 4.16 (t, 2H, NCH2), 7.67 (br s, 1H, C4-H),
7.74 (br s, 1H, C5–H), 9.06 (s, 1H, C2–H).
1H NMR (400 MHz, DMSO-D6): d = 0.85 (t, 6H, CH3),
1.23–1.27 (m, 36H, CH3(CH2)9), 1.79 (m, 4H, NCH2CH2),
4.18 (t, 4H, NCH2), 7.84 (m, 2H, C4–H, C5–H), 9.42 (s,
1H, C2–H).
19F NMR (188 MHz, DMSO-D6): d = ꢁ148.8) (s, 4F, BF4).
1-octyl-3-tetrafluoroethylimidazolium tetrafluoroborate (1d)
m. p. 49–50 °C.
19F NMR (188 MHz, DMSO-D6): d = –148.4 (s, 4F, BF4).
1-hexadecyl-3-methylimidazolium tetrafluoroborate (1j)
m. p. 54–55 °C.
1H NMR (300 MHz, DMSO-D6): d = 0.88 (t, 3H, CH3),
1.29 (m, 10H, CH3(CH2)5), 2.52 (m, 2H, NCH2CH2), 4.30
(t, 2H, NCH2), 7.15–7.32 (m, 1H, CF2H), 8.17 (br s, 1H,
C4–H), 8.32 (br s, 1H, C5–H), 10.04 (s, 1H, C2–H).
1H NMR (300 MHz, DMSO-D6): d = 0.85 (t, 3H, CH3),
1.24 (m, 26H, CH3(CH2)13), 1.78 (m, 2H, NCH2CH2), 3.84
(s, 3H, NCH3), 4.14 (t, 2H, NCH2), 7.69 (br s, 1H, C4-H,
7.76 (br s, 1H, C5-H), 9.08 (s, 1H, C2-H).
19F NMR (188 MHz, DMSO-D6): d = ꢁ148.7 (s, 4F, BF4),
ꢁ137.31 to ꢁ137.53 (d t, 2F, CF2H), ꢁ99.51 (t, 2F, CF2).
1-dodecyl-3-methylimidazolium tetrafluoroborate (1e)
m. p. 27–28 °C.
19F NMR (188 MHz, DMSO-D6): d = -148.2 (s, 4F, BF4).
1-dodecyl-3-methylimidazolium chloride (2a)
m. p. 46-47 °C.
1H NMR (300 MHz, DMSO-d6): d = 0.84 (t, 3H, CH3), 1.25–
1.31 (m, 18H, CH3(CH2)9), 1.85 (m, 2H, NCH2CH2), 3.84 (s,
3H, NCH3), 4.15 (t, 2H, NCH2, J = 7.2 Hz), 7.68 (br s, 1H,
C4–H), 7.75 (br s, 1H, C5–H), 9.09 (s, 1H, C2–H).
1H NMR (300 MHz, DMSO-D6): d
= 0.85 (t, 3H, CH3),
1.24 (m, 18H, CH3(CH2)9), 1.79 (m, 2H, NCH2CH2), 3.91
8
Chem Biol Drug Des 2016

QSAR Modeling and Antibacterial Studies of ILs
Table 5: Growth inhibition of B. subtilis and Ps. aeruginosa by an
imidazolium-based ILs. The diameters of inhibition zones are given
in millimeters.
(s, 3H, NCH3), 4.22 (t, 2H, NCH2), 7.84 (br s, 1H, C4-H),
7.92 (br s, 1H, C5-H), 9.57 (s, 1H, C2-H).
1,3-didodecylimidazolium chloride (2b)
m. p. 49-50 °C.
Content on a disc
1∙10-7
Ps.
aeruginosa
B.
subtilis
ILs
mole
1H NMR (300 MHz, DMSO-D6): d
1.24-1.27 (m, 36H, CH3(CH2)9), 1.79 (m, 4H, NCH2CH2),
4.18 (t, 4H, NCH2), 7.84 (m, 2H, C4-H, C5-H), 9.32 (s, 1H,
C2-H).
= 0.85 (t, 6H, CH3),
[CH2PhC1IM]BF4
1 a
7.7a
0.8b
7.9
0.8
7.1
0.7
5.4
0.5
5.9
0.6
5.3
0.5
5.4
0.5
5.4
0.5
4.1
0.4
5.1
0.5
7.0
0.7
4.5
0.5
7.3
0.7
5.6
0.6
5.4
0.5
5.0
0.5
4.7
0.5
4.1
0.4
5.2
0.5
3.8
0.4
na
na
na
na
na
na
na
na
29
17
20
15
25
11
38
23
15
na
8
na
31
24
10
na
12
na
27
19
32
27
27
20
12
na
11
na
8
10
na
na
na
10
na
na
na
33
20
26
18
26
12
38
25
13
na
10
na
28
15
10
Na
na
na
28
20
34
24
19
14
13
na
8
[C6C1IM]BF4
1 b
[C8C1IM]BF4
1 c
1-octyl-3-methylimidazolium bromide (3a)
[C8CF2CF2HIM]BF4
1 d
1H NMR (300 MHz, DMSO-D6): d
1.24 (m, 10H, CH3(CH2)5), 1.79 (m, 2H, NCH2CH2), 3.90
(s, 3H, NCH3), 4.20 (t, 2H, NCH2), 7.85 (br s, 1H, C4-H),
7.93 (br s, 1H, C5-H), 9.59 (s, 1H, C2-H).
= 0.86 (t, 3H, CH3),
[C12C1IM]BF4
1 e
[C12C4IM]BF4
1 f
[C12CH2CH2OHIM]BF4
1 g
1-dodecyl-3-allylimidazolium bromide (3b)
[C8C8IM]BF4
1 h
1H NMR (300 MHz, CDCl3): d = 0.88 (t, 3H, CH3), 1.25
(m, 18H, CH3(CH2)9), 1.94 (m, 2H, NCH2CH2), 4.36 (t, 2H,
NCH2), 5.08 (d, 2H, NCH2CH=CH2), 5.46 (m, 2H,
NCH2CH=CH2), 6.08 (m, 1H, NCH2CH=CH2), 7.64 (br s,
1H, C4-H), 7.68 (br s, 1H, C5-H), 10.35 (s, 1H, C2-H).
[C12C12IM]BF4
1 i
[C16C1IM]BF4
1 g
[C12C1IM]Cl
2 a
[C12C12IM]Cl
2 b
1,3-dioctylimidazolium bromide (3c)
1H NMR (400 MHz, DMSO-D6): d
1.25 (m, 20H, CH3(CH2)5), 1.79 (m, 4H, NCH2CH2), 4.17
(t, 4H, NCH2), 7.83 (m, 2H, C4-H, C5-H), 9.26 (s, 1H,
C2-H).
= 0.85 (m, 6H, CH3),
[C8C1IM]Br
3 a
[C12CH2CH=CH2IM]Br
3 b
[C8C8IM]Br
3 c
1,3-dinonylimidazolium bromide (3d)
[C9C9IM]Br
3 d
1H NMR (300 MHz, CDCl3): d = 0.86 (t, 6H, CH3), 1.24-
1.3 (m, 24H, CH3(CH2)6), 1.85 (m, 4H, NCH2CH2), 4.17-
4.21 (t, 4H, NCH2), 7.42 (m, C4-H, C5-H), 8.9 (s, 1H, C2-
H).
[C10C10IM]Br
3 e
[C12C12IM]Br
3 f
na
12
na
23
11
[C16C1IM]Br
3 g
na
20
10
1,3-didecylimidazolium bromide (3e)
[C12C1IM](CF3SO2)2N
4
1H NMR (300 MHz, CDCl3): d = 0.84 (t, 6H, CH3), 1.26
(m, 28H, CH3(CH2)7), 1.85 (m, 4H, NCH2CH2), 3.93(t, 4H,
NCH2), 7.35 (m, C4-H, C5-H), 8.74 (s, 1H, C2-H).
na, not active.
aThe content of moles on the disc corresponds to concentration
of compounds 1%.
bThe content of moles on the disc corresponds to concentration
of compounds 0.1%.
1,3-didodecylimidazolium bromide (3f)
m. p. 40-41 °C.
1H NMR (300 MHz, DMSO-D6): d
1.24 (m, 26H, CH3(CH2)13), 1.79 (m, 2H, NCH2CH2), 3.90
(s, 3H, NCH3), 4.21 (t, 2H, NCH2), 7.85 (br s, 1H, C4-H,),
7.92 (br s, 1H, C5-H), 9.59 (s, 1H, C2-H).
= 0.86 (t, 3H, CH3),
1H NMR (300 MHz, DMSO-D6): d = 0.85 (t, 6H, CH3), 1.26
(m, 36H, CH3(CH2)9), 1.79 (m, 4H, NCH2CH2), 4.18 (t, 4H,
NCH2), 7.83 (m, 2H, C4-H, C5-H), 9.31 (s, 1H, C2-H).
1-hexadecyl-3-methylimidazolium bromide (3 g)
m. p. 63-64 °C.
1-dodecyl-3-methylimidazolium bis(trifluoromethylsulfonyl)
imide (4)
Chem Biol Drug Des 2016
9

Hodyna et al.
40
35
30
16
13
25
20
15
Correctly
predicted
10
7
9
Number of
compounds
3
20
20
10
11
10
10
5
9
0
Figure 6: Antibacterial activity prediction of
1,3-dialkylimidazolium ILs for dataset 1 and
dataset 2 using classification QSAR models.
All
Active
Inactive
All
Active
Inactive
compounds compounds compounds
Dataset 1
compounds compounds compounds
Dataset 2
Table 6: ILs Predicted activity via regression QSAR models for dataset 1 and dataset 2
Predicted activity for dataset 1. log(1/MIC) Predicted activity for dataset 2. log(1/MIC)
ILs
M1
M2
M3
M4
M5
Mean
M1
M2
M3
M4
M5
Mean
1a
1b
1c
1d
1e
1f
1g
1h
1i
2.6
2.8
3.0
5.0
4.1
4.7
4.5
4.7
5.2
4.9
4.1
5.2
3.0
4.4
4.7
4.9
5.0
5.2
4.9
4.1
2.7
2.8
3.0
3.1
4.4
4.8
4.4
4.9
5.4
4.9
4.4
5.4
3.0
4.6
4.9
5.2
5.3
5.4
4.9
4.4
2.5
2.7
3.0
3.0
4.5
5.1
3.8
5.2
5.4
5.2
4.5
5.4
3.0
4.9
5.2
5.4
5.7
5.4
5.2
4.5
3.1
3.0
3.1
3.0
4.2
5.3
3.9
5.4
5.7
4.9
4.2
5.7
3.1
4.5
5.4
5.4
5.5
5.7
4.9
4.2
3.6
3.6
3.6
5.2
4.3
4.3
4.3
4.3
4.3
4.3
4.3
4.3
3.6
4.3
4.3
4.3
4.3
4.3
4.3
4.3
2.9 ꢀ 0.4
3.4 ꢀ 0.32
3.2 ꢀ 0.25
3.9 ꢀ 1.0
2.7
2.3
2.9
4.8
3.0
3.1
3.1
3.1
3.2
3.1
3.0
3.2
2.8
3.3
3.1
3.2
3.2
3.2
3.1
3.0
2.1
2.2
2.7
2.7
3.1
3.2
4.7
3.3
3.5
3.2
3.1
3.5
2.7
3.2
3.3
3.5
3.4
3.5
3.2
3.1
2.5
2.4
2.8
2.8
3.1
3.4
4.0
3.4
3.3
3.3
3.1
3.3
2.8
3.3
3.4
3.4
3.4
3.3
3.3
3.1
2.1
2.5
2.9
2.2
3.0
3.9
4.6
3.9
4.2
3.2
3.0
4.2
2.8
3.3
3.9
4.0
4.1
4.2
3.2
3.0
2.6
2.6
3.0
5.2
3.0
3.0
3.0
3.0
3.0
3.0
3.0
3.0
2.6
3.0
3.0
3.0
3.0
3.0
3.0
3.0
2.4 ꢀ 0.25
2.4 ꢀ 0.26
2.9 ꢀ 0.1
3.5 ꢀ 1.21
3.1 ꢀ 0.05
3.3 ꢀ 0.31
3.9 ꢀ 0.40
3.4 ꢀ 0.31
3.5 ꢀ 0.41
3.2 ꢀ 0.1
3.1 ꢀ 0.05
3.5 ꢀ 0.40
2.8 ꢀ 0.1
3.2 ꢀ 0.1
3.4 ꢀ 0.3
3.4 ꢀ 0.33
3.5 ꢀ 0.37
3.5 ꢀ 0.41
3.2 ꢀ 0.1
3.0 ꢀ 0.05
4.3 ꢀ 0.14
4.9 ꢀ 0.34
4.2 ꢀ 0.28
4.9 ꢀ 0.38
5.2 ꢀ 0.48
4.9 ꢀ 0.29
4.3 ꢀ 0.14
5.2 ꢀ 0.48
3.2 ꢀ 0.23
4.56 ꢀ 0.2
4.9 ꢀ 0.38
5.0 ꢀ 0.41
5.17 ꢀ 0.5
5.2 ꢀ 0.48
4.88 ꢀ 0.3
4.3 ꢀ 0.14
1j
2a
2b
3a
3b
3c
3d
3e
3f
3g
4
M1, QSAR model number 1.
1H NMR (300 MHz, DMSO-D6): d
=
0.87 (t, 3H, CH3),
liquids, comprising shorter alkyl chains of C8 and C9
1.25 (m, 18H, CH3(CH2)9), 1.85 (m, 2H, NCH2CH2), 3.93
(s, 3H, NCH3), 4.15 (t, 2H, NCH2), 7.31 (br s, 1H, C4-H),
7.34 (br s, 1H, C5-H), 8.72 (s, 1H, C2-H).
carbon atoms (1h, 3c, 3d).
Experimental results and results of prediction by means of
the classification and regression QSAR models were com-
pared.
19F NMR (188 MHz, DMSO-D6): d = -79.92 (s, 6F, CF3).
Figure 6 presents the data about coincidence between the
predicted activity of classification QSAR models and
the real activity of the studied compounds.
Biological testing
The antibacterial activities of the 20 imidazolium-based ILs
(Figure 3) against B. subtilis and Ps. aeruginosa are shown
in Table 5.
Classification QSAR models correctly predicted the
antibacterial activity of 80% of compounds for dataset 1.
The 9 ILs 1e, 1f, 1g, 1h, 2a, 3b, 3c, 3d, and 4 were pre-
dicted as active against B. subtilis, which coincides with
the results of biological testing. 11 compounds were pre-
dicted as inactive, but seven of them (64%) were really
inactive.
The data presented in Table 5 indicate high antibacterial
activity of both water soluble and water immiscible asym-
metric 1,3-dialkylimidazolium ionic liquids, comprising at
least one C12-alkyl radical (compounds 1e, 1f, 1g, 2a,
3b, 4), as well as symmetric 1,3-dialkylimidazolium ionic
10
Chem Biol Drug Des 2016

QSAR Modeling and Antibacterial Studies of ILs
For dataset 2, about 65% of compounds were predicted
correctly by classification QSAR models. The 10 ILs 1e,
1f, 1g, 1h, 1i, 2a, 3b, 3c, 3d, and 4 were predicted as
active against Ps. aeruginosa that coincides with the
results of biological testing. 10 compounds were predicted
as inactive, but 3 (34%) were really inactive.
countries: systematic review and meta-analysis.
Lancet;377:228–241.
€
2. Kaali P., Stromberg E., Karlsson S. (2011) Prevention
of biofilm associated infections and degradation of
polymeric materials used in biomedical applications. In:
Laskovski A.N., editor. Biomedical Engineering, Trends
in Material Science, Chapter 22. Rijeka: In Tech, p.
513–541.
3. Gilbert P., Moore L.E. (2005) Cationic antiseptics:
diversity of action under a common epithet. J Appl
Microbiol;99:703–715.
The data presented in Table 6 show the predicted level of
ILs activity via regression QSAR models for dataset 1 and
dataset 2.
Compounds 1f, 1h, 1i, 1j, 2b, 3c, 3d, 3e, and 3f were
predicted as the most active against B. subtilis. The
results of biological screening of the imidazolium ILs
confirmed the QSAR predictions for ILs 1f, 1h, 3c, 3d,
3e (55%).
4. Russel A.D. (2003) Biocide use and antibiotic resistance:
the relevance of laboratory findings to clinical and envi-
ronmental situations. Lancet Infect Dis;3:794–803.
5. Russel A.D. (1986) Chlorhexidine: antibacterial action
and bacterial resistance. Infection;14:212–215.
6. Fazlara A., Ekhtelat M. (2012) The disinfectant effects
of benzalkonium chloride on some important foodborne
pathogens. J Agric Environ Sci;12:23–29.
7. Mc Cay P.H., Ocampo-Sosa A.A., Fleming G.T. (2010)
Effect of subinhibitory concentrations of benzalkonium
chloride on the competitiveness of Pseudomonas
aeruginosa grown in continuous culture. Microbiol-
ogy;156:30–38.
Compounds 1d, 1g, 1h, 1i, 2b, 3c, 3d, 3e, and 3f were
with high predicted activity against Ps. aeruginosa. The
real biological activity was demonstrated by compounds
ILs 1g, 1h, 3c, 3d (45%).
Conclusions
8. Welton T. (1999) Room-Temperature Ionic Liquids. Sol-
vents for Synthesis and Catalysis. Chem Rev;99:2071–
2084.
Classification and regression predictive QSAR models of
imidazolium-based ionic liquids antibacterial activity were
created. The developed classification QSAR models were
a good predictive quality with accuracy 83% and 88% for
dataset 1 and dataset 2 respectively. All the created
regression QSAR models have a good predictive quality
too with q2 ranged from 0.77 to 0.86 for dataset 1 and
ranged from 0.81 to 0.92 for dataset 2. It means that the
developed QSAR models can be used to search for a new
active antibacterial imidazolium-based ionic liquids as
potential antimicrobial agents.
9. Hong K., Zhang H., Mays J., Visser A., Brazel C., Hol-
brey J., Reichert W., Rogers R. (2002) Conventional
free radical polymerization in room temperature ionic
liquids: a green approach to commodity polymers with
practical advantages. Chem Commun;13:1368–1369.
10. Naushad M., Alothman Z., Khan A., Ali M. (2012)
Effect of ionic liquid on activity, stability, and structure
of enzymes: a review. Int J Biol Macromol;51:555–560.
11. Arning J., Stolte S., Boschen A., Stock F., Pitner W.-
R., Welz-Biermann U., Jastorff B., Ranke J. (2008)
Qualitative and quantitative structure activity relation-
ships for the inhibitory effects of cationic head groups,
functionalized side chains and anions of ionic liquids
on acetylcholinesterase. Green Chem;10:47–58.
12. Shamshina J., Barber P., Rogers R. (2013) Ionic liquids
in drug delivery. Expert Opin Drug Deliv;10:1367–1381.
13. Gilmore B.F., Earle M.J. (2011) Development of ionic
liquid biocides against microbial biofilm. Chim
Oggi;29:50–53.
The antibacterial activity of 20 synthesized compounds of
1,3-dialkylimidazolium ionic liquid with different level of pre-
dictive activity was evaluated.
The predictive ability of classification models was better
than ability of regression models. The coincidence
between the predicted activity of classification QSAR mod-
els and the real activity of the studied ILs as 80% for data-
set 1 and 65% for dataset 2 was determined. The
coincidence between the predicted activity of regression
QSAR models and the real activity of the studied ILs was
lower than 55%. Therefore, it can be better to use classifi-
cation models of evaluation of activity of imidazolium-
based ionic liquids as potential antibacterial agents.
14. Pernak J., Sobaszkiewicz K., Mirska I. (2003) Anti-micro-
bial activities of ionic liquids. Green Chem;5:52–56.
15. Demberelnyamba D., Kim K., Choi S., Park S., Lee H.,
Kim C., Yoo I. (2004) Synthesis and antimicrobial prop-
erties of imidazolium and pyrrolidinium salts. Bioorg
Med Chem;12:853–857.
16. Docherty K., Kulpa C. (2005) Toxicity and antimicrobial
activity of imidazolium and pyridinium ionic liquids.
Green Chem;7:185–189.
References
1. Allegranzi B., Nejad S.B., Combescure C., Graafmans
W., Attar H., Donaldson L., Pittet D. (2011) Burden of
endemic health-care-associated infection in developing
17. Iwai N., Nakauama K., Kitazume T. (2011) Antibacterial
activities of imidazolium, pyrrolidinium and piperidinium
salts. Bioorg Med Chem Lett;21:1728–1729.
Chem Biol Drug Des 2016
11

Hodyna et al.
18. Yu Y., Nie Y. (2011) Toxicity and Antimicrobial Activi-
ties of Ionic Liquids with Halogen Anion. J Environ
Prot;2:298–303.
32. Sushko I., Pandey A.K., Novotarskyi S. (2011) Online
Chemical Modeling Environment (OCHEM): Web Plat-
form for Data Storage. Model Development and Pub-
lishing of Chemical Information. J Cheminform;3:20.
33. Tetko I.V., Baskin I.I., Varnek A.. (2008) Tutorial on
machine learning. Part 2. Descriptor selection bias. In:
Strasbourg summer school on chemoinformatics:
cheminfoS3. Obernai. https://www.researchgate.net/
publication/236651951_Tutorial_on_Machine_Learning_
Part_2_Descriptor_Selection_Bias (accessed 5 August
2015)
19. Borowiecki P., Milner-Krawczyk M., Brzezin0ska D.,
Wielechowska M., Plenkiewicz J. (2013) Synthesis and
Antimicrobial Activity of Imidazolium and Triazolium
Chiral Ionic Liquids. Eur J Org Chem;2013:712–720.
20. Mester P., Wagner M., Rossmanith P. (2015) Antimi-
crobial effects of short chained imidazolium based
ionic liquids – influence of anion chaotropicity. Ecotoxi-
col Environ Saf;111:96–101.
21. Pendleton J.N., Gilmore B.F. (2015) The antimicrobial
potential of ionic liquids: A source of chemical diversity
for infection and biofilm control. Int J Antimicrob
Agents;46:131–139.
34. Mahobia N.K., Patel R.D., Sheikh N.W., Singh S.K.,
Mishra A., Dhardubey R. (2010) Validation Method
Used In Quantitative Structure Activity Relationship.
Der Pharm Chem;2:260–271.
22. Luczak J., Jungnickel C., Laska I., Stolte S., Hupka J.
(2010) Antimicrobial and surface activity of 1-alkyl-3-
methyl imidazolium derivatives. Green Chem;12:593–
601.
35. Tetko I.V., Sushko I., Pandey A.K., Zhu H., Tropsha A.,
Papa E., Oberg T., Todeschini R., Fourches D., Varnek
A. (2008) Critical assessment of QSAR models of envi-
ronmental toxicity against Tetrahymena pyriformis:
focusing on applicability domain and overfitting by vari-
able selection. J Chem Inf Model;48:1733–1746.
36. Bauer A., Kirby W., Sherris J., Turck M. (1966) Antibi-
otic susceptibility testing by a standardized single disk
method. Am J Clin Pathol;45:493–496.
37. Dzyuba S.V., Bartsch R.A. (2001) Efficient synthesis of
1-alkyl(aralkyl)-3-methyl(ethyl)imidazolium halides: Pre-
cursors for room-temperature ionic liquids. J Hetero-
cycl Chem;38:265–268.
23. Cornellas A., Perez L., Comelles F., Ribosa I., Manresa
A. (2011) Self-aggregation and antimicrobial activity of
imidazolium and pyridinium based ionic liquids in water
solutions. J Colloid Interface Sci;355:164–171.
24. Venkata Nancharaiah Y., Kiran Kumar Reddy G.,
Lalithamanasa P., Venugopalan V.P. (2012) The ionic
liquid 1-alkyl-3-methylimidazolium demonstrates com-
parable antimicrobial and antibiofilm behavior to a
cationic surfactant. Biofouling;28:1141–1149.
25. Tetko I.V. (2008) Associative neural network. Methods
Mol Biol;458:185–202.
26. Tetko I.V. (2002) Associative neural network. Neural
Process Lett;16:187–199.
27. Vorberg S., Tetko I.V. (2014) Modeling the Biodegrad-
ability of Chemical Compounds Using the Online
Chemical Modeling Environment (OCHEM). Molecular
Informat;33:73–85.
38. Rogalsky S., Fatyeyeva K., Lyoshina L., Tarasiuk O.,
Bulko O., Lobok S. (2014) Antimicrobial properties and
thermal stability of polycarbonate modified with 1-alkyl-
3-methylimidazolium tetrafluoroborate ionic liquids.
J Appl Polym Sci;131:40050.
Notes
28. Livingston F. (2005) Implementation of Breiman’s ran-
dom forest machine learning algorithm. ECE591Q
Mach Learn J Pap;1–13.
29. Breiman L. (2001) Random forests. Mach Learn;45:5–
32.
ahttps://ochem.eu/
bAdriana.Code web-page. http://www.molecularnetworks.
com/products/adrianacode (accessed Dec4. 2010).
chttp://www.talete.mi.it/products/dragon_description.htm/
(Accessed June 1. 2015).
30. Tetko I.V., Tanchuk V.Yu. (2002) Application of asso-
ciative neural networks for prediction of lipophilicity in
dChemAxon. https://www.chemaxon.com/ (Accessed May
1. 2015).
ALOGPS 2.1 program.
Sci;42:1136–1145.
31. Hall L.H., Kier L.B., Brown B.B. (1995) Molecular
Similarity Based on Novel Atom-Type Electrotopologi-
cal State Indices. J Chem Inf Comput Sci;35:1074–
1080.
J
Chem Inf Comput
ehttp://docs.ochem.eu/display/MAN/OCHEM+Introduction.
fhttp://docs.ochem.eu/display/MAN/Statistical+parameters
ghttp://docs.ochem.eu/display/MAN/Applicability+domain+
assessment
hhttps://www.molecular
(Accessed May 1. 2015)
networks.com/products/corina
12
Chem Biol Drug Des 2016