Prediction method for composite biological effect of carbon nanomaterial

Abstract

The invention discloses a method for predicting a composite biological effect of a carbon nanomaterial. The method comprises the following steps: measuring the growth inhibition toxicity of different types of carbon-based nanomaterials to scenedesmus obliquus in the presence of singly; the growth inhibition toxicity is calculated by taking the effect multiple change generated by a treatment group compared with a control group as a toxicity index; based on the obtained growth inhibition toxicity data, constructing an SEML prediction model through a support vector machine regression algorithm or an artificial neural network regression algorithm; and predicting the composite biological effect of the carbon nanomaterial according to the SEML prediction model. According to the method, the effect multiple change FC is adopted as a toxicity index, biotoxicity evaluation is more accurate, meanwhile, the prediction model is directly constructed based on growth inhibition toxicity data, material characteristic parameters of the carbon-based nano material do not need to be measured, and the modeling process is more convenient.

Description

Technical Field

[0001] The invention relates to the field of environmental science, and in particular to a prediction of the composite biological effect of carbon-based nanomaterials. Background Art

[0002] Chemical pollutants (CPs) released into the environment often do not exist in isolation, but may act together with other CPs on the ecosystem, showing complex compound biological effects. In recent years, a series of experimental studies have confirmed that when multiple CPs are exposed together, the toxicological effects on multiple species in the ecosystem are more significant and serious than when they are exposed alone. However, current toxicological studies on CPs mainly focus on the toxic effects of single substances, while relevant data on the toxic effects of CPs mixtures are particularly scarce. Biological testing of CPs (especially CPs mixtures) is inefficient, has a long cycle, and is costly, making it difficult to meet the actual needs of a large number of existing CPs and new chemicals that urgently need risk assessment. With the increasing maturity and widespread use of computer technology, the establishment of effective non-biological (reducing biological exposure) alternative methods (i.e., predictive toxicology methods) that can be used for risk assessment has become a consensus and focus of attention in academia, industry, and chemical regulatory agencies.

[0003] With the help of advanced computer-aided tools such as big data analysis, artificial intelligence and machine learning, predictive toxicology methods can be used to analyze, simulate, visualize or predict chemical toxicity to reduce experimental and computational costs and ultimately achieve risk assessment of toxic chemicals. In addition, predictive toxicology methods based on machine learning better associate CPs characteristics and toxic effects through linear or nonlinear relationships, enhancing the understanding of CPs complex systems. Existing predictive toxicology models lack the development of multivariate CPs mixed toxicity prediction models. The limitations of this study may be related to the lack of data related to CPs mixed toxicity and the singleness of mixed toxicity assessment indicators. In addition, the regression method selected in the modeling process is still insufficient in its prediction and interpretation capabilities when dealing with such highly nonlinear data. These challenges together lead to the limitations of the current construction of CPs predictive toxicology models.

[0004] With the continuous increase in market demand for carbon-based nanomaterials (such as fullerenes, carbon nanotubes, etc.), coupled with the continuous emergence of advanced carbon nanomaterials (graphene, etc.), an environmental issue that cannot be ignored is that massive amounts of carbon-based nanoparticles will inevitably penetrate into the atmosphere, water bodies and soil environments, thereby transforming into potential environmental pollutants. In today's rapidly changing nanotechnology, compared with the rapid development of its production and application fields, human awareness of the toxicological effects and potential risks of carbon-based nanomaterials lags behind. Assessing the environmental risks brought about by carbon-based nanomaterials is no longer just a challenge faced by the scientific community, but a national strategic need for sustainable development in the field of nanotechnology in various countries.

[0005] At the same time, carbon-based nanomaterials are relatively less toxic, such as based on traditional toxicity indicators EC 50 ,EC 10 Therefore, up to now, no research has successfully developed and fully verified a predictive toxicology model that can predict the combined biological effects between carbon-based nanomaterials and coexisting pollutants. Summary of the invention

[0006] The purpose of the present invention is to solve the defects in the prior art and provide a method for effectively predicting the composite biological effects of carbon-based nanomaterials.

[0007] In order to achieve the above object, the technical solution provided by the present invention is as follows:

[0008] A method for predicting the composite biological effect of carbon nanomaterials comprises the following steps:

[0009] (1) Determine the growth inhibition toxicity of different types of carbon-based nanomaterials on Scenedesmus obliquus in their sole presence; the growth inhibition toxicity is calculated using the effect fold change produced by the treatment group compared with the control group as a toxicity index.

[0010] (2) constructing a SEML prediction model based on the growth inhibition toxicity data obtained in step (1) by using a support vector machine regression algorithm or an artificial neural network regression algorithm;

[0011] (3) Predict the biological effects of carbon nanomaterial composites based on the SEML prediction model obtained in step (2).

[0012] The present invention adopts the effect fold change FC as a toxicity index, which makes the biological toxicity evaluation more accurate. At the same time, a prediction model is directly constructed based on the growth inhibition toxicity data, without measuring the material characteristic parameters of the carbon-based nanomaterials, and the modeling process is more convenient.

[0013] Furthermore, the treatment group was set as Scenedesmus obliquus exposed to different types of carbon-based nanomaterials at different concentrations for 96 hours, and the control group was set as Scenedesmus obliquus not exposed to carbon-based nanomaterials, and the cell numbers of the control group and the treatment group were determined. The fold change (FC) of the effect (based on the dynamic change of cell number) produced by the treatment group compared with the control group was obtained by formula 1:

[0014]

[0015] here, is the average level of the number of algal cells in the treatment group, and Nc is the number of algal cells in the control group.

[0016] From formula 2, we get:

[0017]

[0018] Here, x n Represents the number of algal cells in the nth group in the treatment groups with different exposure concentrations.

[0019] Among them, the exposure concentrations of carbon-based nanomaterials were set at 0.1 mg / L, 1 mg / L and 10 mg / L respectively.

[0020] The method for predicting the composite biological effects of carbon nanomaterials of the present invention also includes the following steps: judging the goodness of fit, robustness and predictive ability of the constructed SEML prediction model through the fitting coefficient; and using the Y randomization test method to comprehensively evaluate and verify the non-randomness or robustness of the constructed SEML prediction model.

[0021] The fitting coefficients include the coefficient of determination R 2 , root mean square error RMSE, mean absolute error MAE.

[0022] Furthermore, the different types of carbon-based nanomaterials in the above step (1) include fullerenes, single-walled carbon nanotubes, multi-walled carbon nanotubes, graphene nanosheets and graphene oxide.

[0023] Furthermore, when the support vector machine regression algorithm is used to construct the SEML prediction model in the above step (2), the basis function RBF is selected as the kernel function, and the expression is as follows:

[0024] K(x i , x j )=exp(-γ|||x i -x j || 2 )

[0025] In the formula, x i ,x jis a data point and γ is a hyperparameter.

[0026] Furthermore, in the above step (2), an artificial neural network regression algorithm is preferably used to construct a SEML prediction model.

[0027] Compared with the prior art, the present invention has the following advantages:

[0028] The present invention uses the effect fold change FC as the toxicity index, which is more accurate and avoids the traditional toxicity index EC. 50 ,EC 10 The impact is due to the low toxicity of the material and its difficulty in obtaining.

[0029] The present invention directly constructs a prediction model through toxicity prediction index data, and the modeling process is more concise and transparent.

[0030] The composite biological effect prediction model between carbon nanomaterials constructed by the present invention has higher accuracy and robustness in predicting the toxicity of carbon nanomaterial mixtures than the traditional independent action (IA) model, and can accurately and efficiently evaluate the potential risks of these complex interactions in environmental systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The modeling flow chart of the SEML model in the prediction method of the composite biological effect of carbon nanomaterials of the present invention;

[0032] Figure 2 The SEML model constructed based on the support vector machine (SVM) algorithm in Example 1 of the present invention is used to compare FC 生长抑制 The fitting diagram of the predicted value and the experimental value;

[0033] Figure 3 The SEML model constructed based on the artificial neural network (ANN) algorithm in Example 2 of the present invention is used to compare FC 生长抑制 The fitting diagram of the predicted value and the experimental value;

[0034] Figure 4 Comparison of the linear relationship between the pre-test and measured values ​​of the three models in Example 3 of the present invention;

[0035] Figure 4 In the figure, a is the SVM-SEML model, b is the ANN-SEML model, and c is the IA model. DETAILED DESCRIPTION

[0036] like Figure 1 As shown, five carbon-based nanomaterials (C 60Taking the construction of SEML model of the composite biological effect potential of binary mixture composed of SWCNTs, MWCNTs, GN, and GO on Scenedesmus obliquus as an example, the prediction method of the composite biological effect of carbon nanomaterials of the present invention comprises the following specific steps:

[0037] (1) The toxic effects of these five carbon-based nanomaterials on S. obliquus under single and binary mixed exposure were investigated through algae growth inhibition toxicity tests. After 96 hours of exposure, the cell numbers of the control group and the treatment group were determined. The fold change (FC) of the effect produced by the treatment group compared with the control group (based on the dynamic change of cell number) was obtained by formula 1:

[0038]

[0039] here, is the average level of the number of algal cells in the treatment group, and Nc is the number of algal cells in the control group.

[0040] From formula 2, we get:

[0041]

[0042] Here, x n Represents the number of algal cells in the nth group in the treatment groups exposed to different concentrations.

[0043] Among them, for the single treatment groups of the five carbon-based nanomaterials, the exposure concentrations were set at 0.1 mg / L, 1 mg / L and 10 mg / L respectively; and for the binary mixed treatment groups, the exposure concentrations were also 0.1 mg / L, 1 mg / L and 10 mg / L, and the two components in these mixed treatment groups were prepared at a concentration ratio of 1:1. In addition, the control group was set to be Scenedesmus obliquus that was not exposed to carbon-based nanomaterials.

[0044] The obtained data set is as follows:

[0045] Table 1. Composition of the dataset

[0046]

[0047]

[0048] (2) Based on the toxicity indicators of single carbon-based nanomaterials, a SEML prediction model was constructed using a machine learning regression algorithm.

[0049] (3) In order to accurately evaluate the robustness and predictive effectiveness of the SEML prediction model, we used the leave-one-out cross-validation method (CV-LOO) to conduct rigorous testing on the training set.

[0050] (4) The coefficient of determination (R2 ), root mean square error (RMSE), and mean absolute error (MAE) were used to judge the goodness of fit and predictive ability of the SEML model. 2 The calculation formulas of RMSE and MAE are shown in Formulas 3, 4 and 5.

[0051]

[0052] (5) For the carbon nanomaterial mixture whose combined biological effect potential is to be determined, the toxic effect of each material on Scenedesmus obliquus under single exposure is measured, and the SEML prediction model is used to calculate its combined biological effect.

[0053] Example 1

[0054] This embodiment adopts the above method to construct the SEML prediction model, wherein in step (2), the support vector machine (SVM) regression algorithm is used to construct the SEML prediction model, and the radial basis function (RBF) is selected as the kernel function (as shown in Formula 6).

[0055] K(x i , x j ) = exp(-y||x i -x j || 2 ) [6]

[0056] In the formula, x i ,x j is a data point and γ is a hyperparameter.

[0057] The fit and prediction ability of the constructed SEML prediction model (SVM-SEML) are judged, such as Figure 2 As shown, R 2 =0.680, RMSE=0.045, MAE=0.039. The prediction model has good goodness of fit and prediction ability.

[0058] Example 2

[0059] This embodiment adopts the above method to construct the SEML prediction model, wherein in step (2), an artificial neural network (ANN) regression algorithm is used to construct the SEML prediction model, and the parameters are set as: solver = "lbfgs", activation = "relu", hidden_layer_sizes = 100, max_iter = 200.

[0060] The goodness of fit and predictive ability of the constructed SEML prediction model (ANN-SEML) are judged, such as Figure 3 As shown, R 2=0.698, RMSE=0.043, MAE=0.035. Compared with the SVM-SEML prediction model constructed in Example 2, this prediction model has better goodness of fit and prediction ability.

[0061] Example 3

[0062] The Y randomization test method was used to comprehensively evaluate and verify the non-randomness or robustness of the constructed SEML prediction model.

[0063] Among them, the modeling data of the random model is generated by the random function RAND().

[0064] The linear correlation coefficient R and the mean absolute difference between the predicted value and the measured value were used to compare the prediction ability of the SEML model and the IA model. The expression of the IA model is:

[0065]

[0066] In the formula, E(C mix ) is the effect value of the mixed system; E(C i ) is the effect value of the mixture component i.

[0067] The SEML prediction models constructed in Examples 1 and 2 were used to predict the composite biological effects of binary combinations of carbon-based nanomaterials. The results are shown in Table 2 below:

[0068] Table 2. Composite biological effect prediction dataset

[0069]

[0070]

[0071] As shown in Table 3 below, the performance of the SEML prediction model constructed by the present invention is much higher than that of the Y randomization model.

[0072] Table 3. Performance of the Y-randomization model developed based on the composite biological effect prediction dataset

[0073]

[0074] As shown in Table 4 and Figure 4 As shown, the average absolute difference between the predicted value and the measured value of the SVM-SEML prediction model and the ANN-SEML prediction model constructed by the present invention is 3.88 and 3.53% respectively, which is lower than the average absolute difference (12.53%) between the predicted value and the measured value of the traditional and commonly used IA model. This shows that the SEML model constructed by the present invention has a high accuracy for the toxicity of the carbon nanomaterial mixture.

[0075] Table 4. Absolute differences between the predicted values ​​and the measured values ​​of the three models

[0076]

[0077]

Claims

1. A method for predicting the composite biological effects of carbon nanomaterials, characterized in that: The following steps are involved: (1) Determining the growth inhibition toxicity of different types of carbon-based nanomaterials on Scenedesmus obliquus when present alone; the growth inhibition toxicity is calculated using the effect fold change produced by the treatment group compared with the control group as a toxicity index; (2) constructing a SEML prediction model based on the growth inhibition toxicity data obtained in step (1) by using a support vector machine regression algorithm or an artificial neural network regression algorithm; (3) Predict the biological effects of carbon nanomaterial composites based on the SEML prediction model obtained in step (2).

2. The prediction method according to claim 1, characterized in that: The effect fold change FC of the treatment group compared with the control group in step (1) is obtained by the following method: The oblique Scenedesmus was exposed to different types of carbon-based nanomaterials at different concentrations for 96 hours, which were set as the treatment group; at the same time, the control group was set as the oblique Scenedesmus that was not exposed to the carbon-based nanomaterials. The cell numbers of the control group and the treatment group were determined and calculated using the following formula (1): In the formula, is the average level of the number of algal cells in the treatment group, and Nc is the number of algal cells in the control group. From formula (2), we get: In the formula, x n Represents the number of algal cells in the nth group in the treatment groups with different exposure concentrations.

3. The prediction method according to claim 2, characterized in that: The different exposure concentrations set in the control group were 0.1 mg / L, 1 mg / L and 10 mg / L respectively.

4. The prediction method according to claim 3, characterized in that: The different types of carbon-based nanomaterials in step (1) include fullerene, single-walled carbon nanotube, multi-walled carbon nanotube, graphene nanosheet and graphene oxide.

5. The prediction method according to claim 4, characterized in that: When the support vector machine regression algorithm is used to construct the SEML prediction model in step (2), the basis function RBF is selected as the kernel function, and the expression is as follows: K(x i ,x j )=exp(-γ|||x i -x j || 2 ) In the formula, x i ,x j is a data point and γ is a hyperparameter.

6. The prediction method according to claim 4, characterized in that: In the step (2), an artificial neural network regression algorithm is used to construct a SEML prediction model.

Images