Theoretical Descriptor for the Correlation of the Fish Embryo Toxicity test (FET) By QSAR Model of LC50

Quantitative structure–toxicity relationships were developed for the prediction of the fish embryo toxicity test using the CODESSA treatment, based on the linear heuristic method (HM) and support vector machines (SVM). Each kind of compound was represented by several calculated structural descriptors, derived for a diverse set of 97 compounds using DFTB3LYP/6-31+G (d) level of the theory. A six-parameter correlation was found for the 97 compounds. In the HM method, the value of square of the correlation coefficient r2 is 0.8142, s2 is 0.0380, in the SVM method, the value of the r2 is 0.7105 and s2 is 0.0604. The HM model may be used for prediction of toxicity, safety and risk assessment of chemicals to achieve better ecotoxicological management. environmental, and pharmaceutical areas [4,5], a potential or useful technique to estimate the toxicity especially for the compounds whose toxicity are not easy to test. The advantage of this approach over the experimental method lies in the fact that the descriptors used can be calculated from the structures alone and are not dependent on any experimental properties. Once the structure of a compound is known, a limited set of molecular properties, said descriptors, which can be obtained observing the molecular structure or semi-empirical quantum-mechanical calculations [6]. Therefore, once a reliable model is established, the model can be applied to help in identifying and describe important structural features of the molecules that are relevant to variations in molecular properties. In this work we proposed a QSAR model to evaluate fish embryo toxicity data. 97 structures are calculated by the CODESSA software package, developed by Katritzky et al. [7,8] based on descriptors constitutional, topological, geometrical, geometrical, electrostatic and quantum chemical. Materials and Method In order to develop the QSAR, a database of LC50, reported in terms of micromolar (μM) concentration for correlation purposes, reported LC50 values were converted to their molar units and subsequently to free energy related negative logarithmic state, i.e., log (1/ LC50). These LC50 values have been obtained from experiments and literature [9]. The selected series consists of total 97 numbers of compounds, divided in to two data sets, one was training set having 83 compounds and other was test set with the remaining 14 compounds based on random selection criteria. The biological activities were expressed in terms of median lethal concentration (EC50). Organic Chemistry Current Research O r g a ni c Ch em istry: Crent Rese a r c h


Introduction
The fish acute toxicity is a mandatory component in the base set of data requirements for eco-toxicity testing but fish suffer distress and perhaps pain [1]. Animal alternative considerations have also been incorporated into new REACH regulations through strong advocacy for the reduction of testing with live animals. REACH, as a regulatory tool, was envisaged to revise the existing chemical policy and to harmonize chemical legislations in Europe and the major aim of REACH is to systematically evaluate the risk to human health and the environment of approximately 30,000 chemical substances [2]. The fish embryo toxicity (FET) testan alternative approach to classical acute fish toxicity testing with live fish, has been a mandatory component in routine whole effluent testing in Germany since 2005 and has already been standardized at the international level. A modified version has been submitted by the German Federal Environment Agency (UBA) as a draft guideline for an alternative to chemical testing with intact fish [1]. The fish embryo test is a surrogate for the OECD 203 acute fish toxicity test (or other guideline equivalent acute fish assays). Recently, the regulations of the People's Republic of China on Control over Safety of Hazard Chemicals revised and passed, therefore, FET is useful for the acute fish toxicity testing. In order to analyze the applicability of the FET also in chemical testing, a comparative re-evaluation was carried out for a total of 97 substances, and QSAR was developed to evaluate fish embryo toxicity data. QSAR to evaluate fish embryo toxicity data is described to be useful for future validation exercises.
The quantitative structure-property/activity relationship (QSPR/ QSAR) method is based on the assumption that the variation of the behavior of the compounds, as expressed by many measured physicochemical properties, can be correlated with changes in molecular features of the compounds termed descriptors [3]. This method can be used for the prediction of the properties of new compounds. It can also be applied to identify and describe important structural features of the molecules that are relevant to variations in molecular properties. Computational models are useful because they rationalize a large number of experimental observations. The QSAR model is widely used for the prediction of physicochemical properties and biological activities in chemical, environmental, and pharmaceutical areas [4,5], a potential or useful technique to estimate the toxicity especially for the compounds whose toxicity are not easy to test. The advantage of this approach over the experimental method lies in the fact that the descriptors used can be calculated from the structures alone and are not dependent on any experimental properties. Once the structure of a compound is known, a limited set of molecular properties, said descriptors, which can be obtained observing the molecular structure or semi-empirical quantum-mechanical calculations [6]. Therefore, once a reliable model is established, the model can be applied to help in identifying and describe important structural features of the molecules that are relevant to variations in molecular properties.
In this work we proposed a QSAR model to evaluate fish embryo toxicity data. 97 structures are calculated by the CODESSA software package, developed by Katritzky et al. [7,8] based on descriptors constitutional, topological, geometrical, geometrical, electrostatic and quantum chemical.

Materials and Method
In order to develop the QSAR, a database of LC 50 , reported in terms of micromolar (μM) concentration for correlation purposes, reported LC 50 values were converted to their molar units and subsequently to free energy related negative logarithmic state, i.e., log (1/ LC 50 ). These LC 50 values have been obtained from experiments and literature [9]. The selected series consists of total 97 numbers of compounds, divided in to two data sets, one was training set having 83 compounds and other was test set with the remaining 14 compounds based on random selection criteria. The biological activities were expressed in terms of median lethal concentration (EC 50 ).

Calculation of the descriptors
The chemicals in the data set (

Development of linear model by HM [13]
The heuristic procedure was used to obtain quantitative structure property relationship. Once the molecular descriptors are generated, the HM in CODESSA is used to pre-select the descriptors and build the near model. The advantages of the HM are the high speed, usually produces correlations 2-5 times faster than other methods with comparable quality and no software restrictions on the size of the data set [14]. The HM can either quickly give a good estimation about what quality of correlation to expect from the data, or derive several best regression models. HM of the descriptors selection proceeds with a pre-selection of the descriptors to eliminate: (1) those descriptors that are not available for each structure;(2) descriptors having a small variation in magnitude for all structures;(3)descriptors that provide a F-test value below 1.0 on the one-parameter correlation and (4) and the descriptors whose t-values are less than the user-specified value, etc. New descriptors were added one by one until the suitable number descriptors without losing any important information in the model is achieved [15]. The final result is a list of the 10 best models according to the values of the F-test and correlation coefficient. The statistical quality of the generated models was gauged by the F-test (F), squared correlation co-efficient (R 2 ), square of cross-validate coefficient regression (R CV 2 ), t-test and the standard deviation (s 2 ). Fischer's value (F), which represents F-ratio between the variance of calculated and observed activity, and chance statistics assuring that the results are not merely based on chance correlations and t-test reflects significance of the parameter within the model. From the above processes, seven descriptors are selected from descriptors pool and the linear model is produced by the HM [16]. The experimental values of all compounds are summarized in Table 1.

Development of linear model by SVM
The support vector machine algorithm was developed by Vapnik [17]. Support vector machine (SVM), as a novel type of learning machine is gaining rapid popularity due to its remarkable generalization performance [18], has gained much interest in pattern recognition and function approximation applications recently. In bioinformatics, SVMs have been successfully used to solve classification and correlation problems, with the basic idea of SVM is to map the original data into a higher dimensional feature space by a kernel function and then to do classification in this space by constructing an optimal separating hyperplane. Compared with traditional regression and neural networks methods, SVMs have some advantages, including the absence of local minima, good generalization ability, simple implementation, few free parameters, dimensional independence and its ability to condense information contained in the training set [19]. The flexibility in classification and ability to approximate continuous function make SVMs very suitable for QSAR and QSPR studies. Because it is difficult to predict in advance which descriptors are most relevant to the problem at hand, however, SVM is well known to tolerate irrelevant features [20][21][22]. In some cases, the best average performance was achieved when all the features were given to SVM [23].

Results of HM
The heuristic method is not made a prior choice of descriptors, but it performs an analysis on several intrinsic molecular properties available from the Gaussian output. We performed correlations for a growing number of descriptors in the range from 1 to 10. Figure 1 shows the plots of R 2 , cross-validated coefficient (R CV 2 ) and the standard deviation (s 2 ) for the training set as a function of the number of descriptors models. R 2 and R CV 2 are increased with the increasing number of descriptors. However, the values of s 2 decreased with the increasing number of descriptors [24,25]. When adding another descriptor did not improve significantly the statistics of a model, it was determined that the optimum subset size had been achieved. From Figure 1, it can be seen that six descriptors appear to be sufficient for a successful regression model. Then the corresponding descriptors are applied as inputs for the non-linear model.Best models were selected on the basis of their statistical significance. High r 2 and high F and low s 2 values in each model reflect their good predictive potential. Pair wise correlations between the descriptors used in the QSAR models obtained for series is summarized in Table 2. The molecular descriptors used in the selected QSAR models are defined in Table 3. The predict results are shown in Figure 2.

Results of SVM
SVM is a machine-learning approach that allows one to learn from experimental data within a given set, and build a computer model to make predictions on new data sets. The associated parameters are c, p and g, -t kernel function is the RBF kernel function: exp(gamma*|uv|^2), the calculated optimal correlation parameters are c:3.9192, g:168.9664, p: 0.9892. The predict results are shown in Figure  3.
Unsaturated double bond, based on positive and negative regression coefficients, can be judged according to their contribution to the toxic compounds, The coefficients of number of double bonds is positive, the LC 50 increase with the double bond value increases. As for the electron donation of the elide, it is noted that the HOMO energy can be used a new method for showing the basicity of elides. The descriptors that matter most is on the Highest Occupied Molecular Orbital (HOMO) calculated from quantum chemical calculation, which describes the effect of HOMO energies on the toxicity. The results also show that the contribution from the HOMO-1 becomes less important than HOMO. Molecular volume is important to toxicity, the information contained in high-resolution structural determination of macromolecules. The HOMO-1, which is lower in energy than the HOMO, features a large contribution for toxicity.
The HOMO of 4-Aminophenol, 2-Bromophenol, Butyldiglycol and 2-Propanol, were calculated by B3LYP/6-31G method. 2-Propanol was the highest toxicity with the highest HOMO, as well as 4-Aminophenol was the lowest toxicity with the lowest HOMO ( Figure 4). The structure of 4-Aminophenol was similar to 2-Bromophenol. The toxicity 2-Bromophenol, with higher HOMO, was higher than 4-Aminophenol. The Highest Occupied Molecular Orbital (HOMO) describes the effect of HOMO energies on the toxicity.

Conclusions
A QSAR study for fish embryo toxicity test (FET) was performed using HM and SVM, based on electrostatic, CPSA, constitutional, and quantum-chemical descriptors and satisfactory results were obtained. Through analyzing the obtained results, the present study gives rise to QSAR with good statistical significance and predictive capacity. The 6-parameter model can be considered as the best linear modelthe correlations are satisfying. Additionally, HM produced good model with vigorous predictive ability. Furthermore, the proposed approach can also be extended in other QSAR investigations. Fish embryo testing of chemicals has matured to the point that international standardization, method validation, and broadening of chemical coverage is rapidly occurring. Fish embryo tests offer a reasonable alternative to increased use of fish in the future.