Study On Mixture Toxicity Prediction Based On New Index Of Mode Of Action Similarity

The long-term mixed exposure of pollutants is a common phenomenon in the natural environment.Due to the limitations of the long cycle and the high cost of toxicity tests,the mixture toxicity prediction method in silicon has attracted more and more attention and has become a research hotspot in the field of environmental science.At present,the toxicity prediction methods of mixtures are mainly the concentration addition（CA）model based on the similar mode of action（MOA）and the independent action（IA）model based on dissimilar MOA.However,the application is often limited because the MOA of various chemical components in the mixture is not always similar or dissimilar.There are two ways to solve this problem.The first idea is to use models or algorithms,such as neural network algorithms,to directly"learn"and use the toxicity information of the whole mixture to construct a mixture toxicity prediction method independent of MOA.Although this idea can establish a model with a good prediction effect,it needs enough mixture toxicity data to build the model,which is contrary to the reality of insufficient mixture toxicity data.Therefore,this paper focused on the second idea:To establish a toxicity prediction method for mixtures composed of different modes of action similarity（MOASim）chemicals by exploring the quantitative characterization method of MOASim between chemicals（components）.In the process of establishing MOASim quantitative characterization and prediction method,firstly,the whole genome sequence with functional annotation and the three-dimensional structure of luminescence related proteins and membrane proteins of Vibrio qinghaiensis sp.-Q67（Q67）were obtained by using complete genome sequencing technology and homology modeling technology;Then,the chemical protein interaction binding energy network was constructed by molecular docking technology,and the characterization method of MOASim was established;Finally,through the mixture toxicity experiment,the quantitative relationship between MOASim and toxicology was studied,and the toxicity prediction method for mixtures composed of different MOASim chemicals was established;In addition,molecular dynamics was used to investigate the MOA between representative chemicals and proteins.The main results are summarized in detail as follows:（1）Based on back propagation neural network（BPNN）,taking the concentration（logarithm）of each component of the mixture as the input,BPNN with two hidden layers as the core and the toxicity of the mixture as the output,a new mixture toxicity prediction method（BNNmix）independent of MOA was proposed.The method has been successfully applied to predict the toxicity of a seven-component mixture system composed of two substituted phenols,two pesticides,two ionic liquids,and one heavy metal to Caenorhabditis elegans.This seven-component mixture system had time-dependent synergistic toxic interaction.Classical CA or IA could not accurately predict the toxicity of the mixture with toxic interaction,but BNNmix could accurately predict it,and the root mean square error was as low as 0.060.（2）The complete genome of Q67 was obtained by complete genome sequencing technology.It was found that Q67 had two chromosomes.2661 and 1421 protein-coding genes on the two chromosomes were annotated based on COG,GO,and KEGG databases,respectively,and the genome maps of the two chromosomes were obtained.Combined with the luminescent mechanism of other Vibrionaceae,the related genes or proteins in Q67 that may be involved in the luminescent mechanism and its regulatory pathway were determined.Based on the proteins in the quorum sensing（QS）signal pathway and all proteins on cell membrane annotated by complete genome sequencing,135 protein homology models were established by using MODELLER and SWISS-MODEL software.（3）The binding patterns between 207 chemicals and 135 proteins were explored by molecular docking technology.It was found that all molecules were more likely to bind to the largest and most representative pocket or the same active site.According to the binding free energy of the optimal conformational complex,a chemical protein interaction binding energy network representing the difficulty of binding between chemicals and proteins was constructed.In order to ensure the rationality of the proteins constructing the network,the roles of all 135proteins in the binding energy network were analyzed from the perspectives of optimal protein receptor statistics,principal component analysis（PCA）and cluster analysis（agglomerative hierarchical clustering,AHC）.A total of 28 receptor proteins were optimally bound to 207chemicals,and 111 proteins with a load value greater than 0.8 were obtained by PCA,both of which were distributed in 7 functional partitions.AHC found that the information contained in proteins showed great differences as a whole.In general,almost all proteins played an important role in the binding energy network,which also showed that the screened proteins are reasonable.（4）Five parameters(S_ECFP,S_PFP,S_MACCSFP,S_pearson,and S_grade)were proposed to characterize the MOASim from the perspective of molecular similarity and combination difficulty.The p-value of the correlation analysis of the five parameters is 0,the minimum correlation coefficient is 0.198 and the maximum correlation coefficient is 0.815,indicating that there are different degrees of correlation between the parameters.Because the parameters S_grade based on the grade of combination difficulty not only considered the interaction between chemicals and macromolecules but also avoided the influence of the maximum（positive）which has little effect on the binding free energy,it is considered that it can better characterize MOASim.Based on this,five representative chemical pairs with equal gradients were screened,which were CAR-MOL(S_grad value of 0.9495),CAR-PRO（0.8392）,CAR-DIM（0.7398）,CAR-GLY（0.5986）,and CAR-KAS（0.4915）respectively.（5）The photoinhibition toxicity of five mixture rays with different concentration ratios and six chemical components of representative chemicals to Q67 at five exposure time points was determined by time toxicity microplate analysis.Taking EC50 as the toxicity index,the more similar the structure or MOA of chemical components,the closer the toxicity was.Different binary mixture systems had concentration ratio-dependent toxicity and different time-dependent toxicity.There was a clear linear relationship between the toxicological interaction（model deviation ratio（MDR）and its logarithm）and S_grade characterizing MOASim of different mixture systems,and the value of the interaction increased with the increase of S_grade and gradually approached 1,indicating that the more similar the mixture of MOA was,the easier it was to produce additive action.Based on this linear relationship,the improved additive reference model which can predict the toxicity of mixtures composed of different MOASim chemicals was proposed.（6）Through the molecular docking,it was found that the binding pockets of six small molecules（CAR,MOL,PRO,DIM,GLY,and KAS）bound to Lux R protein were consistent.The binding modes of these six chemicals with Lux R were analyzed by molecular dynamics from the aspects of hydrogen bond,binding site,free energy composition,and residue contribution.The results showed that these six small molecules had the same binding pocket when binding with Lux R.The enthalpy and entropy changes of all systems were less than 0,indicating that the main binding forces were van der Waals force and hydrogen bond,but the types and lifetime of existing hydrogen bond,pocket shape,and contributing residues are different in different systems.