A linear regression method for predicting the potential of atmospheric organic amines to promote the formation of new particles driven by methanesulfonic acid

Abstract

The present invention provides a linear regression method for predicting the potential of atmospheric organic amines to promote the formation of new methanesulfonic acid-driven particles. The construction process is as follows: collection of atmospheric organic amine species and search for their gas-phase basicity GBs; optimization of the organic amine structures; calculation of the free energy of formation of a 1:1 cluster between the organic amine and methanesulfonic acid; calculation of the average first ionization potential (MI) and radial distribution function (RDF040s) of the organic amine molecular descriptors; training of a multivariate linear regression model; evaluation of model performance using metrics such as the calculated-to-predicted determination coefficient; and characterization of the model's application domain. The prediction model established by the present invention has excellent fitting ability, robustness, and predictive power, and can effectively predict the free energy of formation of a 1:1 cluster between organic amines and methanesulfonic acid within the application domain, namely, the potential of organic amines to promote the formation of new methanesulfonic acid particles. This method is simple, efficient, and low-cost, saving resources for experimental testing. It is of great significance for screening atmospheric particulate matter precursors and for air quality prediction and management.

Description

Technical Field

[0001] The present invention belongs to the technical field of the sources and causes of atmospheric particulate matter, and relates to a linear regression method for predicting the potential of organic amines in the atmosphere to promote the formation of new particles driven by methanesulfonic acid. Specifically, the present invention is a method for predicting the potential of organic amines in the atmosphere to promote the formation of new particles driven by methanesulfonic acid based on a quantitative structure-activity relationship (QSAR) model. Background Art

[0002] Secondary particulate matter (PSM) formed during the formation of new particles (NPs) is a significant source of atmospheric particulate matter, contributing over half of the total particle count and a core area of ​​research on atmospheric particulate matter pollution. The NPs-driven formation of methanesulfonic acid (MSA) significantly contributes to the formation of atmospheric particulate matter. Organic amines, as important organic bases in the atmosphere, may promote MSA-driven NPs through acid-base reactions. Methanesulfonic acid has a wide range of sources, primarily from the oxidation of organic sulfur compounds (OSCs) in the atmosphere, with dimethyl sulfide (DMS) emitted from marine biological processes being the primary precursor. In addition to marine sources, recent studies have revealed that similar OSCs can be emitted from inland agriculture, industry, and aquaculture. Organic amines are a significant class of organic bases in the atmosphere and have a wide range of sources. Anthropogenic sources include livestock farming, food processing, chemical and leather manufacturing, composting operations, and engine operation, all of which contribute significant amounts of amines to the atmosphere. Other anthropogenic sources of amines include sewage, cooking, and pesticides. Natural sources of amines include marine biological activity, biomass burning, and vegetation emissions. Numerous studies have confirmed that in the presence of organic amines, the nucleation rate driven by key nucleation species of organic acids such as SA (gaseous sulfuric acid) or MSA (methanesulfonic acid) is significantly increased. Due to their high gas-phase alkalinity, organic amines are much more capable of promoting nucleation than NH3. Studies by Place et al. have shown that even in the presence of large amounts of NH3, organic amines still play an important role in nucleation. Currently, research is further clarifying and refining the atmospheric multiphase transformation of organic amines, especially the mechanisms and kinetics of nucleation and growth involved in the formation of new particles.

[0003] However, there are many different types of organic amines. Currently, more than 160 have been found in the gas phase, and the number of organic amines reaches nearly 200 if amino acids in the condensed phase are included. However, only five organic amines have been studied for their effects on the formation of new methanesulfonic acid particles. Studying the effects of organic amines in the atmosphere on the formation of new methanesulfonic acid particles, whether through computational simulation or experimental research, is inefficient, time-consuming, and expensive. With the use of various volatile chemicals, the variety of organic amines in the atmosphere is expected to increase. For organic amines, the ability to form initial 1:1 clusters with methanesulfonic acid is crucial to the new particle formation process. Therefore, the development of efficient (high-throughput, low-cost) simulation and prediction technologies is needed. Summary of the Invention

[0004] To solve the above technical problems, based on the computational simulation technology of quantitative structure-activity relationship (QSAR), by establishing the correlation between the formation free energy ΔG of 1:1 clusters of organic amines and methanesulfonic acid and the molecular characteristic parameters of each organic amine, the potential of organic amines to promote the formation of new methanesulfonic acid particles can be effectively predicted. The present invention provides a linear regression method for predicting the potential of organic amines in the atmosphere to promote the formation of new particles driven by methanesulfonic acid. By collecting literature, a comprehensive collection of 195 organic amines currently present in the atmosphere (including 31 amino acids present in the atmospheric condensation phase) was obtained. 50 organic amines were selected, covering aliphatic amines, aromatic amines, amides, piperazines, amino acids and other organic amine types. The formation free energy ΔG of these 50 organic amines with MSA 1:1 clusters was calculated, and a linear regression model between the formation free energy and the molecular characteristic parameters of the organic amines was constructed using the multivariate linear regression method. The application domain of the model was characterized, and the scope of application of the model was clarified.

[0005] Specifically, the present invention constructs a simple and efficient multivariate linear regression model method for predicting the formation free energy of 1:1 clusters generated by organic amines and methanesulfonic acid. This method can predict the formation free energy of 1:1 clusters of organic amines and methanesulfonic acid based on the gas phase basicity GB, average first ionization potential Mi, and radial distribution function RDF040s of the organic amine, providing a basic tool for screening organic amines that efficiently promote methanesulfonic acid nucleation. During the modeling process, internal and external validation was carried out with reference to the Organization for Economic Cooperation and Development (OECD) guidelines for the construction and use of QSAR models to examine the robustness and predictive ability of the model.

[0006] The technical solutions of the present invention are as follows:

[0007] A multivariate linear regression method for predicting the potential of organic amines to promote the formation of new methanesulfonic acid particles is developed, with the following steps:

[0008] (1) Data collection and calculation

[0009] We searched the literature for 195 organic amines currently present in the atmosphere (including 31 amino acids), selected 50 of them, optimized their molecular structures, and obtained their most stable configurations. We then used the thermodynamic parameters in the output file to calculate the formation free energies ΔG of 1:1 clusters with methanesulfonic acid. We also collected the gas-phase basicity GB of the organic amines.

[0010] (2) Calculation of molecular descriptors of organic amines

[0011] According to the optimized molecular structure of the organic amine, the .log file of the most stable configuration of the organic amine was converted into a .mol ​​file, and the molecular descriptors average first ionization potential Mi and radial distribution function RDF040s of 50 organic amines were calculated based on the .mol ​​file;

[0012] (3) Model training

[0013] The formation free energy ΔG, gas-phase basic GB, average first ionization potential Mi, and radial distribution function RDF040s of the 1:1 cluster of organic amines and methanesulfonic acid were combined; the dataset was randomly split into a training set and a validation set at a ratio of 3:1. The formation free energy ΔG (unit: kcal mol) of the 1:1 cluster of organic amines and methanesulfonic acid was used to calculate the formation free energy ΔG (unit: kcal mol -1 ) as the dependent variable, organic amine gas phase basicity GB, average first ionization potential Mi, radial distribution function RDF040s as the independent variables, the stepwise multiple linear regression method was used to train the linear regression model, the model is shown in Equation 1;

[0014] ΔG=-0.0500GB-123.780Mi-0.189RDF040s+182.222 (1)

[0015] (4) Model evaluation

[0016] Calculate the coefficient of determination R between the training set and the predicted value 2 , root mean square error RMSE, characterizes the goodness of model fitting; R 2 ,RMSE,Q 2 ext Characterize the model's predictive ability; use the elimination method to characterize the robustness of the training set and use the validation coefficient Q 2 LOO To express;

[0017] The final model’s prediction effect is:

[0018] The formation free energy ΔG of the 1:1 cluster of organic amine and methanesulfonic acid (unit: kcal mol -1 )’s prediction effect:

[0019] R 2 train =0.823, RMSE train =0.809, Q 2 LOO =0.822, R 2 test =0.823, RMSE test =0.321, Q 2 ext =0.757;

[0020] (5) Application domain representation

[0021] The Williams diagram is used to characterize the application domain of the model. That is, the leverage value h of the selected 50 organic amines is plotted against the standard residual δ. The calculation methods of h, δ and the warning value h* are as follows:

[0022]

[0023]

[0024]

[0025] where y i and are the calculated value and the model predicted value of the i-th data point respectively, n is the number of data in the dataset, A is the number of descriptors involved in the model, h i and x i represent the leverage value and the descriptor vector of the i-th data respectively, X is the descriptor matrix, X T is the transpose of the descriptor matrix, n tra is the number of data in the training set. If |δ| > 3, it is regarded as an outlier. If |δ| < 3, h i <h* indicates that the data is within the application domain. For the organic amines in the training set, h i > h* indicates that the structure of this molecule appears less frequently and has an impact on the establishment of the model; for the organic amines in the validation set, h i > h* and |δ| < 3 indicate that the prediction result of this substance is an extrapolation of the model, indicating that the model is also applicable to this substance.

[0026] Furthermore, the 50 kinds of organic amines at least include aliphatic amines, aromatic amines, amides, piperazines and amino acids.

[0027] Furthermore, in step (1), the molecular structures of the organic amines are optimized by using the density functional DFT method and the ab initio method in the GAUSSIAN 09 and ORCA4.0 programs respectively; in step (2), the.mol file is obtained by converting the output.log file of GAUSSIAN 09 into a.mol file through the OpenBabel 2.3.2.2 software.

[0028] Furthermore, in step (2), calculations are performed by inputting the.mol file into the Dragon 6.0.0 software.

[0029] Furthermore, the h* = 0.3243.

[0030] The present invention has the following beneficial effects: the constructed model can effectively predict the formation free energy ΔG of a 1:1 cluster of an organic amine and methanesulfonic acid, reflecting the potential of organic amines to promote the formation of new methanesulfonic acid particles. It has good fitting, robustness, and predictive capabilities, and has a clearly characterized application domain. This method is simple, efficient, and low-cost, and is expected to enrich the particle formation mechanisms in macroscopic atmospheric models, play a role in air quality simulations, and provide a basic tool for atmospheric environmental quality prediction, serving the national needs of atmospheric environmental governance and the Blue Sky Defense Campaign. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A process for constructing a linear regression method to predict the potential of atmospheric organic amines to promote the formation of new particles driven by methanesulfonic acid;

[0032] Figure 2 Figure 3 (a) shows the linear fitting plot of the calculated and predicted values ​​of the formation free energy ΔG of the 1:1 cluster of organic amines and methanesulfonic acid, and the Williams plot (b) showing the application domain of the model. The types of organic amines in the training set and validation set are 37 and 13, respectively. DETAILED DESCRIPTION

[0033] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and technical solutions.

[0034] Example 1

[0035] like Figure 1 As shown in FIG, a multiple linear regression method for predicting the potential of organic amines to promote the formation of new methanesulfonic acid particles is described. The steps are as follows:

[0036] (1) Data collection and calculation

[0037] A literature search revealed 195 organic amines currently present in the atmosphere (including 31 amino acids). Fifty of these amines, including aliphatic amines, aromatic amines, amides, piperazines, and amino acids, were selected. Density functional theory (DFT) and ab initio optimization were performed in the GAUSSIAN09 and ORCA4.0 programs, respectively, to obtain their most stable configurations. Thermodynamic parameters in the output files were used to calculate the formation free energies, ΔG, of 1:1 methanesulfonic acid clusters. The gas-phase basicity (GB) of the organic amines was also collected.

[0038] (2) Calculation of molecular descriptors of organic amines

[0039] According to the optimized molecular structure of the organic amine, the output .log file of GAUSSIAN09 was converted into a .mol ​​file using OpenBabel 2.3.2.2 software. The .mol ​​file was then input into Dragon 6.0.0 software to calculate the molecular descriptors of the organic amine, namely the average first ionization potential Mi and the radial distribution function RDF040s.

[0040] (3) Model training

[0041] The formation free energy ΔG, gas-phase basic GB, average first ionization potential Mi, and radial distribution function RDF040s of the 1:1 cluster of organic amines and methanesulfonic acid were combined; the dataset was randomly split into a training set and a validation set at a ratio of 3:1. The formation free energy ΔG (unit: kcal mol) of the 1:1 cluster of organic amines and methanesulfonic acid was used to calculate the formation free energy ΔG (unit: kcal mol -1 ) as the dependent variable, organic amine gas phase basicity GB, average first ionization potential Mi, radial distribution function RDF040s as the independent variables, the stepwise multiple linear regression method was used to train the linear regression model, the model is shown in Equation 1;

[0042] ΔG=-0.0500GB-123.780Mi-0.189RDF040s+182.222 (1)

[0043] (4) Model evaluation

[0044] The coefficient of determination R of the calculated value-predicted value adjusted using the training set 2 , root mean square error RMSE, characterizes the goodness of model fitting; R 2 ,RMSE,Q 2 ext Characterize the model's predictive ability; use the elimination method to characterize the robustness of the training set and use the validation coefficient Q 2 LOO To express;

[0045] The final model’s prediction results are as follows:

[0046] The formation free energy ΔG of the 1:1 cluster of organic amine and methanesulfonic acid (unit: kcal mol -1 )’s prediction effect:

[0047] R 2 train =0.823, RMSE train =0.809, Q 2 LOO =0.822, R 2 test =0.823, RMSE test =0.321, Q 2ext = 0.757;

[0048] The fitting performance of the model is as Figure 2 (a) shown, R 2 train 、R 2 test Greater than 0.8 indicates good fitting performance of the model; RMSE train 、RMSE test With small values, it indicates high prediction accuracy of the model, and R 2 train And Q 2 LOO The difference less than 0.3 indicates that there is no overfitting phenomenon in the model; Q 2 LOO Greater than 0.8 indicates high robustness of the model.

[0049] (5) Characterize the application domain according to OECD guidelines

[0050] Use the Williams plot to characterize the application domain of the model, that is, plot the leverage value (h) of the selected 50 organic amines against the standard residual (δ). The calculation methods of h, δ and the warning value (h*) are as follows:

[0051]

[0052]

[0053]

[0054] Among them, y i And Are the calculated value and the model prediction value of the i-th data point respectively, n is the number of data in the data set, A is the number of descriptors involved in the model, h i 、x i Represent the leverage value and the descriptor vector of the i-th data respectively, X is the descriptor matrix, X T Is the transpose of the descriptor matrix, n tra Is the number of data in the training set. |δ| > 3 is regarded as an outlier. |δ| < 3, h i < h* = 0.3243, indicating that the data is within the application domain. For the organic amines in the training set, h i > h* indicates that the structure of this molecule appears less frequently and has an impact on the establishment of the model; for the organic amines in the validation set, h i > h*, |δ| < 3, indicating that the prediction result of this substance is an extrapolation of the model, which means the model is also applicable to this substance. The Williams plot characterizing the application domain of the model is as Figure 2As shown in (b), the |δ| of all data points in the training set is less than 3, and all data points in the validation set are within the application domain.

[0055] Example 2

[0056] Given an organic amine, C3H9N (CAS number: 107-10-8), we need to predict its potential to promote the formation of new methanesulfonic acid particles, that is, to predict the formation free energy ΔG of a 1:1 cluster with methanesulfonic acid. First, we find its gas-phase basicity GB. Then, we use GAUSSIAN 09 and ORCA4.0 to find the most stable configuration of the organic amine. This configuration is then input into Dragon 6.0.0 software to obtain its molecular descriptor, the average first ionization potential Mi, and the radial distribution function RDF040s. The leverage value (h = 0.036) and standard residual (δ = -0.430) are calculated to determine its application domain. Finally, we use the linear regression model constructed in this invention to perform predictions, obtaining the following results:

[0057] GB=883.9kJ mol -1 , Mi=1.166, RDF040s=2.231, ΔG 预测 = -6.70 kcal mol -1

[0058] The corresponding calculated value is: ΔG 计算 = -6.00 kcal mol -1 , the predicted values ​​are in good agreement with the calculated values.

[0059] Example 3

[0060] Given an organic amine C6H 15 N (CAS No. 918-02-5) was used to predict its potential to promote the formation of new methanesulfonic acid particles, that is, to predict the formation free energy ΔG of a 1:1 cluster with methanesulfonic acid. First, its gas-phase basicity GB was determined. Then, GAUSSIAN 09 and ORCA 4.0 were used to find the most stable configuration of the organic amine. This configuration was then input into Dragon 6.0.0 software to obtain its molecular descriptor, average first ionization potential Mi, and radial distribution function RDF040s. The leverage value (h = 0.070) and standard residual (δ = 0.354) were calculated to determine its applicability within the model. Finally, the linear regression model constructed in the present invention was used for prediction, yielding the following results:

[0061] GB=948.6kJ mol -1 , Mi=1.155, RDF040s=16.686, ΔG 预测 = -11.31 kcal mol -1

[0062] The corresponding calculated value is: ΔG计算 = -11.88 kcal mol -1 , the predicted values ​​are in good agreement with the calculated values.

[0063] Example 4

[0064] Given an organic amine C9H 21 To predict the potential of N (CAS No. 102-69-2) to promote the formation of new methanesulfonic acid particles, that is, to predict the formation free energy ΔG of a 1:1 cluster with methanesulfonic acid, we first searched for its gas-phase basicity GB. Then, we used GAUSSIAN 09 and ORCA 4.0 to find the most stable configuration of the organic amine. This configuration was then input into Dragon 6.0.0 software to obtain its molecular descriptor, average first ionization potential Mi, radial distribution function RDF040s, and leverage value (h = 0.102) and standard residual (δ = -0.040) to determine its application domain within the model. Finally, we used the linear regression model constructed in this invention to perform predictions, yielding the following results:

[0065] GB=960.1kJ mol -1 , Mi=1.150, RDF040s=20.408, ΔG 预测 = -11.97 kcal mol -1

[0066] The corresponding calculated value is: ΔG 计算 = -11.90 kcal mol -1 , the predicted values ​​are in good agreement with the calculated values.

Claims

1. A linear regression method for predicting the potential of organic amines in the atmosphere to promote the formation of new particles driven by methanesulfonic acid, characterized in that: Here are the steps: (1) Data collection and calculation 50 organic amines and their gas-phase basic GBs were collected from the literature, their structures were optimized, and the most stable configurations of the organic amines were obtained. The formation free energies ΔG of the 1:1 clusters with methanesulfonic acid were calculated using the thermodynamic parameters in the output file. (2) Calculation of descriptors of organic amine molecules Convert the .log file of the most stable configuration of the organic amine into a .mol ​​file; Calculate the molecular descriptors of 50 organic amines, including the average first ionization potential Mi and radial distribution function RDF040s, based on the .mol ​​file; (3) Model training The formation free energy ΔG, gas-phase basicity GB, average first ionization potential Mi, and radial distribution function RDF040s of the 1:1 cluster of organic amines and methanesulfonic acid were merged; the data set was randomly split into a training set and a validation set at a ratio of 3:

1. The formation free energy ΔG of the 1:1 cluster of organic amines and methanesulfonic acid was used as the dependent variable, and the gas-phase basicity GB, average first ionization potential Mi, and radial distribution function RDF040s were used as independent variables. The linear regression model was trained using the stepwise multiple linear regression method. The model is shown in Equation 1. (1) (4) Model performance evaluation Calculate the coefficient of determination R between the training set and the predicted value 2 , root mean square error RMSE, characterizes the goodness of model fitting; R 2 , RMSE characterizes the prediction ability of the model; the internal cross-validation coefficient Q of the training set is used 2 LOO Characterize model robustness; (5) Application domain representation The Williams plot was used to characterize the application domain of the model. That is, the leverage value h of 50 organic amines was plotted against the standard residual δ. The calculation methods of h, δ and the warning value h* are as follows: (2) (3) (4) in, and are the calculated value and model predicted value of the i-th data point, n is the number of data in the dataset, A is the number of descriptors involved in the model, and h i 、x i Represent the leverage value and descriptor vector of the i-th data respectively, X is the descriptor matrix, is the transpose of the descriptor matrix, n tra is the number of data in the training set, |δ| > 3, it is considered an outlier, |δ| < 3, h i < h* indicates that the data is within the application domain. In the training set, organic amine h i > h* indicates that the structure of the molecule appears less frequently, which has an impact on the establishment of the model; h of the organic amines in the validation set i > h*, |δ| < 3, indicating that the prediction result of the substance is an extrapolation of the model, indicating that the model is also applicable to this substance.

2. The method according to claim 1, characterized in that The 50 types of organic amines include at least aliphatic amines, aromatic amines, amides, piperazines and amino acids.

3. The method according to claim 1, characterized in that The .mol ​​file in step (2) was obtained by converting the GAUSSIAN 09 output .log file into a .mol ​​file using Open Babel 2.3.2.2 software.

4. The method according to claim 1, wherein In step (2), the .mol ​​file is input into Dragon 6.0.0 software for calculation.

5. The method according to claim 1, wherein The h* = 0.3243.

Images