<html> <head> </head> <body> Nonlinear QSAR: artificial neural network for the Daphnia magna reproduction test </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> 10.10.2010 </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> 12.04.2010 </body> </html> <html> <head> </head> <body> Training, selection and test sets are available. Model algorithm is available (snn file). </body> </html> <html> <head> </head> <body> None to date. </body> </html> <html> <head> </head> <body> Daphnia magna </body> </html> <html> <head> </head> <body> see 3.6 </body> </html> <html> <head> </head> <body> mmol/L </body> </html> <html> <head> </head> <body> LogEC50 </body> </html> <html> <head> </head> <body> The reproduction toxicity to Daphnia was determined using the OECD 211 (EU C.20) test guideline [ref 1, sect 9.2]. Young female Daphnia (the parent animals), aged less than 24 hours at the start of the test, are exposed to the test substance added to water at a range of concentrations. The test duration is 21 days. At the end of the test, the total number of living offspring produced per parent animal alive at the end of the test is assessed. This means that juveniles produced by adults that die during the test are excluded from the calculations. The reproductive output of the animals exposed to the test substance is compared to that of the control(s) in order to determine the median effective concentration EC50 (LC50). This is the concentration of the test substance dissolved in water that results in a 50% reduction in reproduction of Daphnia magna within 21days. The concentrations of the substances are given in mmol per litre (mmol/L). D. magna was obtained from the National Institute for Environmental Studies (NIES), Tsukuba, Japan. The reproduction test was performed for 21 days according to the methods for survival and reproduction tests on D. magna proposed by the OECD. Females less than 24 h old were used as the founding females in each test. They were exposed to various concentrations of the test substance according to the OECD test conditions, then fed and observed daily for 21 days. Cultures were kept in an incubator at a temperature of 24±10C and a photoperiod of 14 h light/10 h dark. Six nominal concentrations of each test chemical, including a culture water control, were prepared by dilution with fresh culture water. All 21-day experiments were conducted with a dilution factor of 3 for test substances. Eight replicate glass jars (100 ml), each containing an individual D. magna female in 50 ml of media, were used for each concentration. The jars were covered with Teflon caps to prevent volatilization of the test chemicals. The water quality (pH and dissolved oxygen concentration) was measured every 2 days (right after changing of water). A suspension of 0.05 ml of Chlorella (4.3 • 108 cells ml/1) was added to each jar daily. Water hardness, pH, and dissolved oxygen concentration were 75–85 mgl/1, 7.0–7.5, and 80–99%, respectively. The medium was changed every 2 days, and neonates were removed from the jar every day and were counted by eye. The total number of neonates born over 21 days at each concentration of test chemical, as well as the total number born to the control group, were calculated and compared [ref 2 – 3, sect 9.2]. </body> </html> <html> <head> </head> <body> The data are taken from one source [ref 1, sect 9.2]. However, it is uncertain whether all experimental data points were obtained from a single laboratory. </body> </html> <html> <head> </head> <body> QSAR </body> </html> <html> <head> </head> <body> The algorithm is based on regression neural network predictor with structure 7-6-5-1 </body> </html> <html> <head> </head> <body> Initial pool of ~899 descriptors. Stepwise descriptor selection based on a set of statistical selection rules as F statistic and p. The first highest F (low p) descriptors (7) were selected from the total number of descriptors. These 7 descriptors were used as inputs to the network. 16 networks with different structures were tested in order to find the best ANN with lowest RMS (root-mean-squared error) and highest correct predictions (for training, selection and test sets). Then 555 epochs were used to train the final network with architecture depicted in 4.2. Optimization of the weights was performed with Levenberg-Marquardt algorithm encoded in the backpropagation scheme using linear and hyperbolic activation functions. </body> </html> <html> <head> </head> <body> All descriptors were generated using QSARModel on structures optimized by AM1 semiempirical quantum mechanical model. </body> </html> <html> <head> </head> <body> 28 (196 chemicals / 7 descriptors) </body> </html> <html> <head> </head> <body> Applicability domain based on training set: a)functional groups such as phenols, aldehydes, nitro, amino, alcohols, halides, aromatics, aliphatic functional groups b)The model is suitable for compounds that have descriptors values in the following range: Desc: 1 2 3 4 5 6 7 min: 0.000; 0.633; 0.077; -17.765; -8.052; -11.095; 1.000 max: 0.005; 0.925; 0.667; -8.194; -6.684; 0.000; 2.600 </body> </html> <html> <head> </head> <body> Presence of functional groups in structures. Range of descriptor values in training set with ±30% confidence. Descriptor values must fall between maximal and minimal descriptor values (see 5.1) of training set ±30%. </body> </html> <html> <head> </head> <body> See 5.1, 5.2 </body> </html> <html> <head> </head> <body> 196 data points </body> </html> <html> <head> </head> <body> Standardization and normalization of the inputs by taking into account the mean and standard deviation </body> </html> <html> <head> </head> <body> TrainingLogEC50; SelectionLogEC50; TestLogEC50 Data Mean: 4.389; 4.099; 4.214 Data SD: 2.135; 2.065; 2.165 Error Mean: 0.006; 0.203; 0.799 Error SD: 0.840; 2.545; 2.188 Abs E. Mean: 0.632; 1.384; 1.416 SD Ratio: 0.393; 1.232; 1.011 Correlation: 0.919; 0.527; 0.750 </body> </html> <html> <head> </head> <body> See 6.7 </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> RMS(Training)=0.068; RMS(Selection)=0.207; RMS(Test)=0.189 In this ANN, 2 randomly chosen sets (50) were used to test the network – selection set and test set; see also 6.7 </body> </html> <html> <head> </head> <body> The method used two validation sets: selection (50) and test (50) </body> </html> <html> <head> </head> <body> Randomly selected 50 selection and 50 test set points </body> </html> <html> <head> </head> <body> See 6.7 and 6.12 </body> </html> <html> <head> </head> <body> The descriptors for the test set are in the limit of applicability; see 6.7 and 6.12 </body> </html> <html> <head> </head> <body> Overall predictions for the selection set (used to stop the ANN training and not to overfit it) and the test set (used to test the external prediction of the net after training) are significant according to the RMS error and the standard deviation ratio (SD ratio); see 6.7 and 6.12 </body> </html> <html> <head> </head> <body> Most of the descriptors are related to the reactivity of the compounds related to the C and H atoms. A rough estimation can be made based on their values. Regarding the descriptor Avg nucleophilic reactivity index (AM1), for H atoms, it can be noted that it has slight negative correlation with the modelled property. This might suggest that with the increase of this descriptor, the property would decrease. The same holds for the descriptor Relative number of H atoms (correl -0.5). In contrast, the descriptor No. of occupied electronic levels (AM1) / # atoms leads to larger LogEC50 values (correlation 0.5). </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> </body> </html> <html> <head> </head> <body> Supporting information for: training set(s), delection set(s), test set(s) </body> </html> Q19-22-1-336 2011/12/19 Daphnia magna, reproduction, Molcode, artificial neural network To be entered by JRC