Additive SMILES-based optimal descriptors in QSAR modelling bee toxicity: Using rare SMILES attributes to define the applicability domain

doi:10.1016/j.bmc.2008.03.048

Bioorganic & Medicinal Chemistry

Volume 16, Issue 9, 1 May 2008, Pages 4801-4809

https://doi.org/10.1016/j.bmc.2008.03.048 Get rights and content

Abstract

The additive SMILES-based optimal descriptors have been used for modelling the bee toxicity. The influence of relative prevalence of the SMILES attributes in a training and test sets to the models for bee toxicity has been analysed. Avoiding the use of rare attributes improves statistical characteristics of the model on the external test set. The possibility of using the probability of the presence of SMILES attributes in training and test sets for rational definition of the applicability domain is discussed.

Graphical abstract

Introduction

Quantitative structure-property/activity relationship (QSPR/QSAR) is a tool for the estimation of unavailable numerical data on endpoints of interest by means of correlations ‘descriptor-property/activity’.¹ The descriptor is a numerical index of the molecular structure that can be calculated using information on the molecular architecture, for instance, represented by molecular graphs.2, 3, 4, 5 As an alternative to the graph, the simplified molecular input line entry system (SMILES) can be used for the elucidation of the molecular structure in the QSPR/QSAR analyses.6, 7, 8

The toxicity of pesticides towards bee is an important ecologic indicator, since the bees have influence to many natural processes related to fruit trees.⁹ The experimental definition of the numerical values of the toxicity towards bee involves considerable time and resources.¹⁰ Thus there are motivations to search for robust models of the toxicity towards bee.9, 10, 11, 12

The number of SMILES-oriented databases in the internet is gradually increasing. SMILES-based optimal descriptors gave reasonable prediction for the bee toxicity.¹² The aim of the present study is an attempt to use more transparent version of the SMILES-based descriptor. In fact, each SMILES attribute is a representation of molecular fragment. The additive scheme¹³ is a well-known approach of modelling properties by the selection of special contributions for molecular fragments. The new version of the SMILES-based optimal descriptors is a SMILES-based realization of the additive scheme.

Section snippets

Results and discussion

Figure 1 shows the plot of ARP versus Lim S. One can see from Figure 1 that the ARP and Lim S are increasing. However for Lim S = 4 the ARP is unexpectedly decreased. This an interesting point probably it can be an indicator for a robust selection of the Lim S. The hypothesis is that the ARP can be used for a preliminary estimation of the split into training and test sets as well as for estimation of the Lim S.

Figure 2 shows the correlation coefficients for the training and test sets for different Lim S

Conclusion

The additive SMILES-based optimal descriptors can be used as a tool for predicting the values of the bee toxicity. The suggested Lim S index can be used as a tool for the selection of the list of the SMILES attributes for the robust SMILES-based model.

Method

As endpoint we used the decimal logarithm log(1/C), where C is the concentration of the pesticides expressed in mmol/bee, which kills 50% of the bees.¹² These 105 pesticides have been split into training set (n = 85) and test set (n = 20).

Optimal descriptors examined in the present study have been defined as $DCW = CW (b) + CW (db) + CW (tb) + CW (N) + CW (Cl) + CW (Br) + CW (F) + \sum CW ({SS}_{k})$ where CW(b) is the correlation weight of the given number of branching (i.e., number of brackets in the SMILES); CW(db) is the

Acknowledgement

The authors thank the Marie Curie fellowship for financial support (the contract ID 39036, CHEMPREDICT).

References and notes (15)

J. Devillers et al.
Comput. Electron. Agric.
(2004)
M. Vighi et al.
Sci. Total Environ.
(1991)
A.A. Toropov et al.
Comput. Biol. Chem.
(2007)
I.G. Zenkevich et al.
J. Chromatogr., A
(2004)
Q.N. Hu et al.
J. Data Sci.
(2003)
A.A. Toropov et al.
J. Chem. Inf. Comput. Sci.
(2003)
A.A. Toropov et al.
J. Chem. Inf. Comput. Sci.
(2004)

There are more references available in the full text version of this article.

Cited by (47)

Predicting the cytotoxicity of ionic liquids using QSAR model based on SMILES optimal descriptors
2015, Journal of Molecular Liquids
Citation Excerpt :
The QSPR/QSAR models have been applied widely to forecast the various properties of compounds, such as the toxicity of ILs compounds [25–32]. Simplified molecular input line entry system has been employed as an alternative for molecular graphs in the QSPR/QSAR models [33–37]. Recently, CORAL software (available at http://www.insilico.eu/coral) has been proposed as an efficient approach for the QSAR analysis.
In recent years, ionic liquids (ILs) have attracted great attention due to their remarkable physico-chemical properties. In spite of all advantages of ILs, these compounds probably cause persistent contaminations and constitute environmental risks. Cytotoxicity (Log₁₀(CE₅₀)) data 225 ILs were used in the creation of quantitative structure–activity relationship (QSAR) models by CORrelation And Logic (CORAL) software (http://www.insilico.eu/coral) in which molecular structures are indicated by the simplified molecular input line entry system (SMILES) notation. These global SMILES descriptors are illustrating the presence of some chemical elements and various types of chemical bonds (double, triple, and stereochemical). The balance of correlations (BC) of constructing a quantitative structure–property/activity relationships and the classic scheme were compared. Results from three random splits displayed reliable models for predicting the external test set with the correlation coefficient (R²) and cross validated correlation coefficient (Q²) in ranges of 0.7315–0.8760 and 0.7062–0.8490, respectively. The best predictions obtained by a classic scheme along with the global SMILES descriptors are included in the modeling process. The average statistical characteristics of the external test set are the following: n = 44, R² = 0.8760, Q² = 0.8540, standard error of estimation (s) = 0.529, mean absolute error (MAE) = 0.400 and Fischer F-ratio (F) = 297.
QSPR studies on refractive indices of structurally heterogeneous polymers
2015, Chemometrics and Intelligent Laboratory Systems
Citation Excerpt :
In previous QSPR studies, we have shown the importance of the methodology of flexible descriptors, which is able to provide models having a comparable or sometimes better quality to the ones found by searching the best descriptors in a pool containing thousands of 0D-3D descriptors [26–28]. Thereby, we investigated the most appropriate molecular structure representation for the flexible descriptor calculation, which can be done in different ways such as by using a chemical graph [29–31], using the Simplified Molecular Input Line Entry System (SMILES) [32–34], or with an hybrid representation which includes both graph and SMILES [35,36]. The high quality experimental refractive indices measured at 298 K on 234 polymer compounds were collected from two published compilations [17,37].
We developed a predictive Quantitative Structure–Property Relationship (QSPR) for the refractive indices of 234 structurally diverse polymers. The model involves a single molecular descriptor and a conformation-independent approach. The most appropriate polymer structure representation was investigated by considering 1–5 monomeric repeating units. The established equations were validated and tested through various well-known techniques, such as the use of an external test set of compounds, the Cross-Validation method, Y-Randomization and Applicability Domain, and finally a comparison was also performed to published results from the li terature. The developed QSPR could be useful for assisting the development of new polymeric materials.
Amino substituted nitrogen heterocycle ureas as kinase insert domain containing receptor (KDR) inhibitors: Performance of structure-activity relationship approaches
2015, Journal of Food and Drug Analysis
A quantitative structure–activity relationship (QSAR) study was performed on a set of amino-substituted nitrogen heterocyclic urea derivatives. Two novel approaches were applied: (1) the simplified molecular input-line entry systems (SMILES) based optimal descriptors approach; and (2) the fragment-based simplex representation of molecular structure (SiRMS) approach. Comparison with the classic scheme of building up the model and balance of correlation (BC) for optimal descriptors approach shows that the BC scheme provides more robust predictions than the classic scheme for the considered pIC₅₀ of the heterocyclic urea derivatives. Comparison of the SMILES-based optimal descriptors and SiRMS approaches has confirmed good performance of both techniques in prediction of kinase insert domain containing receptor (KDR) inhibitory activity, expressed as a logarithm of inhibitory concentration (pIC50) of studied compounds.
Conformation-independent QSAR on c-Src tyrosine kinase inhibitors
2014, Chemometrics and Intelligent Laboratory Systems
The main idea of this work was to find predictive quantitative structure–activity relationships (QSAR) for a wide set of c-Src tyrosine kinase inhibitors, by means of resorting to a conformation-independent representation of the chemical structure. In this way, our attempt was to avoid the availability of X-ray crystallographic structural information of the target. Therefore, in a set composed of 80 pyrrolo-pyrimidine derivatives, 1179 theoretical descriptors were simultaneously analyzed through linear regression models obtained with the replacement method variable subset selection technique. Alternatively, the flexible (activity dependent) descriptor approach was also applied in this study. The models were validated and tested through the use of an external test set of compounds, the leave-group-out cross validation method, Y-randomization and applicability domain analysis. Our results were compared with previously published ones based on docking analysis and 3D-QSAR. The obtained conformation-independent approach was in good agreement with experimental observations.
QSAR models for HEPT derivates as NNRTI inhibitors based on Monte Carlo method
2014, European Journal of Medicinal Chemistry
Citation Excerpt :
Recent published papers have reported the applicability of solo SMILES based descriptors in QSAR analysis [34–38] as well as SMILES based descriptors in combination with topological descriptors [39–41]. All QSAR models are based on Monte Carlo optimization method where appropriate activity is treated as random event [34–41]. The aim of this study is to build QSAR models based on SMILES and graph optimal descriptors using Monte Carlo method for HEPT derivatives as NNRTIs inhibitors and an attempt to define the molecular structure responsible for stated inhibitory effect.
A series of 107 1-[(2-hydroxyethoxy)-methyl]-6-(phenylthio) thymine (HEPT) with anti-HIV-1 activity as a non-nucleoside reverse transcriptase inhibitor (NNRTI) has been studied. Monte Carlo method has been used as a tool to build up the quantitative structure–activity relationships (QSAR) for anti-HIV-1 activity. The QSAR models were calculated with the representation of the molecular structure by simplified molecular input-line entry system and by the molecular graph. Three various splits into training and test set were examined. Statistical quality of all build models is very good. Best calculated model had following statistical parameters: for training set r² = 0.8818, q² = 0.8774 and r² = 0.9360, q² = 0.9243 for test set. Structural indicators (alerts) for increase and decrease of the IC₅₀ are defined. Using defined structural alerts computer aided design of new potential anti-HIV-1 HEPT derivates is presented.
Study on the antagonists for the orphan G protein-coupled receptor GPR55 by quantitative structure-activity relationship
2014, Chemometrics and Intelligent Laboratory Systems
The orphan G protein-coupled receptor GPR55 has been proposed as a new potential drug target for the treatment of diabetes, Parkinson's disease, neuropathic pain, and cancer. Coumarin and 8-amido-chromen-4-one-2-carboxylic acid derivatives were identified as novel antagonists for GPR55. In this study, we established reliable models and explored the valuable information by quantitative structure–activity relationship (QSAR). Firstly, we obtained a quite reliable multiple linear regression (MLR) model with correlation coefficient (R²) of 0.8204 for the training set, and R² of 0.7770 for the test set. Next, we built a better model with R² of 0.8763 for the training set, and R² of 0.8179 for the test set based on the simplified molecular input line entry system (SMILES). Lastly, we established dependable comparative molecular field analysis (CoMFA) and comparative molecular similarity indices analysis (CoMSIA) models. Then we validated these models' predictive ability using various validation methods for the test set. Besides, we designed and confirmed six new potential inhibitors with higher predicted activities and tested twenty-eight compounds without clear activities by the established models. The results obtained from molecular modeling not only provide models to predict the activities of inhibitors but also lead to a better understanding of the essential features that should be considered when designing novel inhibitors with desired activities.

View all citing articles on Scopus

View full text

Additive SMILES-based optimal descriptors in QSAR modelling bee toxicity: Using rare SMILES attributes to define the applicability domain

Abstract

Graphical abstract

Introduction

Section snippets

Results and discussion

Conclusion

Method

Acknowledgement

Comput. Electron. Agric.

Sci. Total Environ.

Comput. Biol. Chem.

J. Chromatogr., A

J. Data Sci.

J. Chem. Inf. Comput. Sci.

J. Chem. Inf. Comput. Sci.