QSPR method and system for predicting hole mobility and stability using interpretable GBRT regression model
Through the GBRT regression model based on SHAP value prediction of hole mobility and stability of organic small molecule hole transport materials in perovskite solar cells, the problems of cumbersome and high cost in the existing technology are solved, and the rapid Low cost efficient prediction and high performance material discovery.
Patent Information
- Application Number
- CN202210398430.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-15
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2042-04-15
AI Technical Summary
The prior art is difficult to quickly and effectively predict the hole mobility and stability of organic small molecule hole transport materials in perovskite solar cells, resulting in cumbersome and high cost.
An interpretable GBRT regression model is constructed based on SHAP values, and structural descriptors are generated through Dragon software. Features are screened and modeled using SHAP nested GBRT method to predict the hole mobility and stability of organic SM-HTMs, and through Qualcomm Quickly detect high-performance materials by volume screening.
It achieves low-cost, simple and efficient prediction of hole mobility and stability of organic SM-HTMs, shortens R&D time, reduces R&D costs, and avoids experiments and pollution, which is in line with the concept of green environmental protection.
Smart Images

Figure CN114974456B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method and system for predicting the hole mobility (μ) and stability (η) of organic small molecule hole transport materials (SM-HTMs) in perovskite solar cells (PSCs), and in particular to a quantitative structural relationship (QSPR) method for predicting hole mobility and stability by constructing an interpretable GBR regression model based on SHAP values to discover high-performance organic small molecule hole transport materials (SM-HTMs). Background Art
[0002] With the rapid development of science and society, it is urgent to alleviate energy and pollution problems. As the third generation of solar cells, PSCs have exceeded 25.2% in photoelectric conversion efficiency (PCE) in just 13 years since they were first reported by Miyasaka et al. in 2009, and are the most promising candidate materials for silicon-based solar cells. As the main component of PSCs, the hole transport layer plays a leading role in preventing the perovskite layer from external stimulation to improve the stability of the device, and reducing the charge recombination at the hole collection electrode, reducing the loss of charge and thus improving the PCE of the device. Among them, organic SM-HTMs have monodispersity and a certain molecular weight. They provide an infinite variety of chemical structures, and the band gap, energy level and electron mobility can be adjusted by modifying the chemical structure. Therefore, new organic SM-HTMs have great prospects in the field of photovoltaics. The discovery of high-performance new organic SM-HTMs will effectively improve the PCE of PSCs and the stability of the device. However, the chemical space of possible organic SM-HTMs is very large, and the trial and error process of the experiment is cumbersome and expensive. In order to accelerate the research and development of organic SM-HTMs, it is necessary to design efficient potential organic SM-HTMs using computational chemistry methods before experiments.
[0003] With the continuous development of computer software and artificial intelligence, the application of machine learning has become more and more extensive, and a series of high-performance machine learning models have been developed, such as support vector machines, random forests, and deep learning. GBRT is the abbreviation of Gradient Boosting Regression Tree, which is an algorithm model that uses tree models for regression. However, the internal principles of these machine learning models are difficult to understand. The Shapley Additive Model (SHAP) method regards all features as "contributors". For each predicted sample, the model generates a predicted value, and the SHAP value is the value assigned to each feature in the sample. The biggest advantage of the SHAP value is that it reflects the contribution of each feature to the target feature, and at the same time shows the positive and negative nature of its contribution. Therefore, using SHAP to nest GBRT to select features and explain the model is very meaningful for the development of high-performance organic SM-HTMs.
[0004] The quantitative structure-activity relationship (QSPR) model established by SHAP nested GBRT technology is used to analyze the relationship between molecular structure and certain properties of the molecule. The candidate descriptors used in the model can quantitatively or qualitatively describe parameters such as molecular structure, molecular composition (such as the number of hydrogen bond donors, the number of chemical bonds), physicochemical properties (such as ester-water distribution coefficient) and molecular shape. In recent years, the use of machine learning for material design has achieved remarkable success. There is no doubt that integrating machine learning into the current material development workflow has great potential. However, how to improve the hole transport rate and stability of organic SM-HTMs still needs further exploration and research. Summary of the invention
[0005] In order to solve the problems of the prior art, the purpose of the present invention is to overcome the shortcomings of the prior art, and provide a QSPR method and system for constructing an interpretable GBRT regression model hole mobility and stability based on SHAP values, which can accelerate the discovery of organic SM-HTMs with high hole transfer rate and stability, obtain structural descriptors through Dragon calculation, use SHAP to nest GBRT screening features, and use GBRT algorithm modeling to predict the hole mobility and stability of organic SM-HTMs, and high-throughput screening of candidate organic SM-HTMs with high hole mobility and stability. The method is low-cost, simple and efficient, with complete and accurate data, no experiments, no complex calculation process, and no pollution.
[0006] In order to achieve the above invention purpose, the present invention adopts the following technical solution:
[0007] A QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values comprises the following steps:
[0008] 1) Use computer systems to search for literature, and search for the structure, theoretically calculated hole mobility and stability values of organic small molecule hole transport materials in perovskite solar cells;
[0009] 2) Draw the molecular structure of organic SM-HTMs using ChemDraw and generate descriptors using Dragon software;
[0010] 3) For the initially screened data set, randomly divide it into training set and test set, and the proportion of the test set accounts for 15% of the entire data set;
[0011] 4) The SHAP-nested GBRT method was used for feature screening, and the optimal variable subset selection for GBRT modeling was obtained by comparative leave-one-out cross-validation;
[0012] 5) Establish a rapid prediction model for hole mobility and stability of organic SM-HTMs using GBRT regression;
[0013] 6) Based on the established rapid prediction model of hole mobility and stability of organic SM-HTMs, predict the hole mobility and stability of the molecules to be detected;
[0014] 7) Further obtain the relationship between the variable and the target variable based on the positive and negative impact of the SHAP value;
[0015] 8) Based on the influence of the positive and negative SHAP values of the variables on the target variables in step 7), consult the literature, use the explanations of the features in the references, construct a QSPR model, and design new core groups and side groups;
[0016] 9) The newly designed core groups and side groups are merged into the original data set to form a core group library and a side group library. The groups in the two libraries are randomly combined to generate a large number of virtual samples. These virtual samples are put into the constructed QSPR model for high-throughput screening to quickly predict the hole mobility and stability values of organic SM-HTMs.
[0017] Preferably, in step 1), after searching the literature, the obtained data samples are preprocessed, including unifying the calculation method of sample molecules, arranging the molecular structure of the samples, arranging the target variables hole mobility and stability, and determining the number of sample data.
[0018] Preferably, in step 4), the SHAP nested GBRT method is used to screen variables according to the "contribution value" of the target variable, and the optimal variable subset selection for GBRT modeling is obtained by comparative leave-one-out cross-validation.
[0019] Preferably, in said step 9), the method for high throughput screening comprises the following steps:
[0020] The design conditions are as follows: select organic SM-HTMs molecules studied theoretically, divide the core groups and side groups into fragments, randomly combine the core groups and side groups, and draw a large number of virtual samples through ChemDraw;
[0021] Then the large number of virtual samples are put into the constructed prediction model for forecasting.
[0022] A QSPR system for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values, and executing the QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values of the present invention, comprising:
[0023] Input module: Search the molecular structure of organic SM-HTMs in perovskite solar cells, the hole mobility and stability calculated theoretically from the published literature, and use them as input data;
[0024] Data analysis module: Using the data obtained from the input module, the QSPR method based on the SHAP value is constructed to construct an interpretable GBRT regression model to predict the hole mobility and stability of organic SM-HTMs.
[0025] Output module: Output the hole mobility and stability information of organic SM-HTMs.
[0026] Compared with the prior art, the present invention has the following obvious outstanding substantial features and significant advantages:
[0027] 1. The process of predicting the hole mobility and stability of organic SM-HTMs in the present invention is simple, fast and efficient; the Dragon software generates descriptors for organic SM-HTMs, and the atomic properties and structural parameters of all samples can be obtained in just a few seconds; through SHAP screening features and GBRT regression modeling, the hole mobility and stability of organic SM-HTMs can be predicted in advance, and it can also provide a reference for R&D personnel of perovskite solar cells, shortening R&D time and reducing R&D costs;
[0028] 2. The present invention generates a prediction value for each prediction sample. The SHAP value is the value assigned to each feature in the sample. The positive and negative relationship between the SHAP value and the target feature value is combined with domain knowledge to explain the features and construct a QSPR model. This provides a deeper understanding of the relationship between the molecular structure and the hole mobility and stability, helping to solve the shortcomings of most methods that the features cannot be explained, thereby helping modelers to deeply understand and apply the model.
[0029] 3. The high-throughput screening results of the present invention do not involve experiments and chemicals for the random combination process of fragments, do not cause environmental pollution, comply with the concept of green environmental protection, and obtain high-performance new candidate dye molecules quickly, efficiently and economically. BRIEF DESCRIPTION OF THE DRAWINGS
[0030] Figure 1 Flow chart for implementing the present invention.
[0031] Figure 2 This is the sample training set of hole mobility of organic SM-HTMs molecules of the present invention, and the result diagram of SHAP nested GBRT feature screening.
[0032] Figure 3 This is the organic SM-HTMs molecular stability sample training set of the present invention, and the SHAP nested GBRT feature screening result diagram.
[0033] Figure 4The leave-one-out cross-validation results and independent test set result diagram of the GBRT regression model for predicting the hole mobility of organic SM-HTMs molecules in the present invention.
[0034] Figure 5 These are the leave-one-out cross-validation results and independent test set result diagram of the GBRT regression model for predicting the molecular stability of organic SM-HTMs in the present invention.
[0035] Figure 6 A diagram showing the influence of the features screened by the SHAP nested GBRT of the present invention on the output value (hole mobility).
[0036] Figure 7 The influence diagram of the features screened by the SHAP nested GBRT of the present invention on the output value (stability).
[0037] Figure 8 Figure 1 is an example diagram of the fragmentation of some organic SM-HTMs molecules (CBP and FDT) of the present invention. DETAILED DESCRIPTION
[0038] The above scheme is further described below in conjunction with specific implementation examples. The preferred embodiments of the present invention are described in detail as follows:
[0039] Embodiment 1:
[0040] In this embodiment, see Figure 1 A QSPR method for predicting the hole mobility and stability of organic SM-HTMs molecules by constructing an interpretable GBRT regression model based on SHAP values includes the following steps:
[0041] 1) Use computer systems to search for literature on the structure, theoretically calculated hole mobility and stability of organic SM-HTMs in perovskite solar cells;
[0042] 2) Draw the molecular structure of organic SM-HTMs using ChemDraw and generate descriptors using Dragon software;
[0043] 3) For the initially screened data set, randomly divide it into training set and test set, and the proportion of the test set accounts for 15% of the entire data set;
[0044] 4) The SHAP-nested GBRT method was used for feature screening, and the optimal variable subset selection for GBRT modeling was obtained by comparative leave-one-out cross-validation;
[0045] 5) Use GBRT regression to establish a rapid prediction model for the hole mobility and stability of organic SM-HTMs molecules;
[0046] 6) Based on the established rapid prediction model of hole mobility and stability of organic SM-HTMs molecules, predict the hole mobility and stability of the molecules to be detected;
[0047] 7) The relationship between the variable and the target variable is further obtained based on the positive and negative impact of the SHAP value;
[0048] 8) Based on the influence of the positive and negative SHAP values of the variables on the target variables in step 7), consult the literature, use the explanations of the features in the references, construct a QSPR model, and design new core groups and side groups;
[0049] 9) The newly designed core groups and side groups are merged into the original data set to form a core group library and a side group library. The groups in the two libraries are randomly combined to generate a large number of virtual samples. These virtual samples are put into the constructed QSPR model for high-throughput screening to quickly predict the hole mobility and stability values of organic SM-HTMs.
[0050] This embodiment constructs an interpretable GBRT regression model based on SHAP value to predict the QSPR method and system for hole mobility and stability, which can accelerate the discovery of organic SM-HTMs with high hole mobility and stability, obtain descriptors of organic SM-HTMs through Dragon calculation, use SHAP to nest GBRT screening features, and use GBRT algorithm modeling to predict the hole mobility and stability of organic SM-HTMs, and high-throughput screening of candidate organic SM-HTMs with high hole mobility and stability. The method is low-cost, simple and efficient, with complete and accurate data, no experiments, no complex calculation process, and no pollution.
[0051] Embodiment 2:
[0052] This embodiment is basically the same as the first embodiment, except that:
[0053] In this embodiment, in step 1), after searching the literature, the obtained data samples are preprocessed, including unifying the calculation method, sorting out the molecular structure and hole mobility value of the samples, and determining the number of sample data.
[0054] In this embodiment, the SHAP method is combined with GBRT to screen features, and the optimal feature subset for modeling is selected, from the original 3740 features to 5 features; the GBRT algorithm is used to establish a model for rapid prediction of the hole mobility of organic SM-HTMs; based on the 5 features obtained by screening, the literature is consulted to explain the positive and negative effects of SHAP on the "contribution" of the target feature in conjunction with the molecular structure, and a QSPR model is constructed;
[0055] In this embodiment, in step 9), the high-throughput screening method includes the following steps:
[0056] The design conditions are as follows: select organic SM-HTMs molecules studied theoretically, divide the core groups and side groups into fragments, randomly combine the core groups and side groups, and draw a large number of virtual samples through ChemDraw;
[0057] Then the large number of virtual samples are put into the constructed prediction model for forecasting.
[0058] The method of this embodiment calculates the descriptors of the molecular structure of organic SM-HTMs, uses SHAP combined with the GBRT method to screen features and model, and predicts the hole mobility of organic SM-HTMs. These methods can get results in a few minutes, which is convenient and fast, without experiments and complicated calculations. At the same time, high-throughput screening avoids the high-performance molecular fragment combination that is difficult to obtain by blind trial and error based on experience. This embodiment builds an interpretable GBRT regression model based on SHAP values to accelerate the discovery of a QSPR method for organic SM-HTMs with high hole mobility, establishes a data set sample; Chemdraw draws molecules; generates descriptors; randomly divides training sets and test sets; uses SHAP to nest GBRT to screen variables and select the optimal variable subset for GBRT modeling; uses GBRT regression to establish a rapid prediction model for organic SM-HTMs; according to the established model, the hole mobility of the organic SM-HTMs in the test set is rapidly predicted. According to the impact on the target features fed back by SHAP and the descriptors explained in the reference literature, the structure of the corresponding molecular fragments is used to build a QSPR model and design new molecular fragments; molecular fragments are randomly combined, a large number of virtual samples are manually drawn, and the built GBRT model is used for prediction. Based on reliable literature real values and modeling methods, the GBRT prediction model of organic SM-HTMs built in this embodiment has the advantages of being convenient and fast without chemical pollution.
[0059] Embodiment three:
[0060] This embodiment is basically the same as the second embodiment, except that:
[0061] In this embodiment, in step 1), after searching the literature, the obtained data samples are preprocessed, including unifying the calculation method, sorting out the molecular structure and stability value of the samples, and determining the number of sample data.
[0062] In this embodiment, the SHAP method is combined with GBRT to screen features, and the optimal feature subset for modeling is selected, from the original 3740 features to 6 features; the GBRT algorithm is used to establish a model for rapid prediction of the stability of organic SM-HTMs; based on the 6 features obtained by screening, the literature is consulted to explain the positive and negative effects of SHAP on the "contribution" of the target feature and the molecular structure, and a QSPR model is constructed;
[0063] In this embodiment, in step 9), the high-throughput screening method includes the following steps:
[0064] The design conditions are as follows: select organic SM-HTMs molecules studied theoretically, divide the core groups and side groups into fragments, randomly combine the core groups and side groups, and draw a large number of virtual samples through ChemDraw;
[0065] Then the large number of virtual samples are put into the constructed prediction model for forecasting.
[0066] The method of this embodiment calculates the descriptors of the molecular structure of organic SM-HTMs, uses SHAP combined with GBRT method to screen features and model, and predicts the stability of organic SM-HTMs. These methods can get results in a few minutes, which is convenient and fast, without experiments and complicated calculations. At the same time, high-throughput screening avoids the high-performance molecular fragment combination that is difficult to obtain by blind trial and error based on experience. This embodiment builds an interpretable GBRT regression model based on SHAP value to accelerate the discovery of a QSPR method for highly stable organic SM-HTMs, establishes a data set sample; Chemdraw draws molecules; generates descriptors; randomly divides training sets and test sets; uses SHAP to nest GBRT to screen variables and select the optimal variable subset for GBRT modeling; uses GBRT regression to establish a rapid prediction model for organic SM-HTMs; according to the established model, quickly predict the stability of organic SM-HTMs in the test set. According to the impact of SHAP feedback on the target features and the reference literature interpretation descriptors, the structure of the corresponding molecular fragments, constructs a QSPR model, and designs new molecular fragments; randomly combines molecular fragments, manually draws to generate a large number of virtual samples, and uses the built GBRT model for prediction. This embodiment is based on reliable literature real values and modeling methods, and the GBRT prediction model of organic SM-HTMs established has the advantages of being convenient, fast and free of chemical pollution.
[0067] Embodiment 4:
[0068] This embodiment is basically the same as the second embodiment, except that:
[0069] In this embodiment, the GBRT nested in SHAP in step 4) reduces the dimension of the features, first screens the features from 3740 to 1-10, and uses the GBRT algorithm to build a model. The results are shown in Figure 3 The optimal feature subset was obtained by comparing the evaluation criteria of the leave-one-out cross-validation method. The evaluation criteria of the leave-one-out cross-validation method included the determination coefficient R 2 and root mean square error RMSE, Figure 3 When there are 5 features in R 2 The highest, RMSE the lowest.
[0070] In this embodiment, in step 8), the five features obtained by screening are interpreted by consulting literature and combining the positive and negative effects of SHAP on the "contribution" of the target feature with the molecular structure to construct a QSPR model.
[0071] The method of this embodiment preprocesses the data, calculates and sorts the data, and is a low-cost, simple and efficient method with complete and accurate data, no need for experiments, and no need for complex calculation processes.
[0072] Embodiment five:
[0073] This embodiment is basically the same as the above embodiment, with the following features:
[0074] In this embodiment, the GBRT nested in SHAP in step 4) reduces the dimension of the features, first screens the features from 3740 to 3-8, and uses the GBRT algorithm to build a model. The results are shown in Figure 4 The optimal feature subset was obtained by comparing the evaluation criteria of the leave-one-out cross-validation method. The evaluation criteria of the leave-one-out cross-validation method included the determination coefficient R 2 and root mean square error RMSE, Figure 4 When there are 6 features in R 2 The highest, RMSE the lowest.
[0075] In this embodiment, in step 8), the six features obtained by screening are interpreted by consulting literature and combining the positive and negative effects of SHAP on the "contribution" of the target feature with the molecular structure to construct a QSPR model.
[0076] The method of this embodiment preprocesses the data, calculates and sorts the data, and is a low-cost, simple and efficient method with complete and accurate data, no need for experiments, and no need for complex calculation processes.
[0077] Embodiment six:
[0078] This embodiment is basically the same as the above embodiment, with the following features:
[0079] In this embodiment,
[0080] A QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values comprises the following steps:
[0081] 1) Through literature research, we searched for theoretically calculated hole mobility and stability of organic SM-HTMs in perovskite solar cells as data set samples for subsequent modeling. The hole mobility and stability of some SM-HTMs are shown in Table 1;
[0082] Table 1. Partial data sample set of hole mobility and stability of SM-HTMs
[0083] SM-HTMs Name <![CDATA[μ / (cm 2 V -1 s -1 )]]> η CBP 0.05 3.03 X25 0.05 2.7 H7 0.22 2.86 H2,5 0.48 1.39 Para-PyB <![CDATA[5.73×10 -5 ]]> 2.23 Py-H <![CDATA[6.47×10 -4 ]]> 2.35
[0084] 2) Using Dragon software to generate descriptors for the organic SM-HTMs molecules drawn by Chemdraw, 5271 descriptors were generated for each sample, some of which are shown in Table 2;
[0085] Table 2. Molecular descriptors of some organic SM-HTMs
[0086]
[0087]
[0088] 3) Descriptors displayed as "na" were removed, and 3740 descriptors were obtained by preliminary screening of the data set; the training set and test set were randomly divided into training set and test set in a ratio of 5:1. The sample sizes of the training set and test set in the hole mobility data sample were 37 and 7 respectively; the sample sizes of the training set and test set in the stability data sample were 41 and 8 respectively.
[0089] 4) Use SHAP nested GBRT method to filter descriptors and use SHAP nested GBRT method to filter features Figure 2 and Figure 3 , the training set organic SM-HTMs molecular data were modeled according to the leave-one-out cross-validation evaluation criteria, which included R 2 and RMSE, and the optimal feature subset is obtained. The optimal feature subset of the hole mobility regression model is 5 features, and the data of the 5 optimal feature descriptors of some samples are shown in Table 3; the optimal subset of the stability regression model is 6 features. The data of the 6 optimal descriptors of some samples are shown in Table 4:
[0090] Table 3. Data of the five best descriptors for some organic SM-HTMs (μ / (cm 2 V -1 s -1 ))
[0091]
[0092] Table 4. Data of the six optimal descriptors of partially organic SM-HTMs (η)
[0093]
[0094]
[0095] 5) The GBRT algorithm was used to perform regression modeling on the hole mobility data samples of 37 organic SM-HTMs in the training set, and a GBRT rapid prediction model for organic SM-HTMs was established. The GBRT algorithm was used to perform regression modeling on the stability data samples of 41 organic SM-HTMs in the training set, and a GBRT rapid prediction model for organic SM-HTMs was established.
[0096] In this embodiment, the leave-one-out cross-validation results of the GBRT regression model for rapid hole mobility prediction of SM-HTMs established based on 37 hole transfer rate samples are as follows: Figure 4 In the leave-one-out cross-validation results of the modeling with 5 features, the actual value of the literature of organic SM-HTMs is closest to the model prediction value, and its R 2 The average value of the regression line is 0.79 and the RMSE is 0.68.
[0097] In this example, the prediction results of the independent test set of GBRT regression for the rapid prediction of hole mobility of organic SM-HTMs based on SHAP screening features are as follows: Figure 4 As shown in the figure, the prediction results of the 7 samples in the test set are good, and the R 2 The average value of the regression line is 0.79 and the RMSE is 0.58.
[0098] In this embodiment, the leave-one-out cross-validation results of the stability rapid prediction GBRT regression model of organic SM-HTMs established based on 41 stability value samples are as follows: Figure 5 In the leave-one-out cross-validation results of the modeling with 6 features, the actual value of the literature of organic SM-HTMs is closest to the model prediction value, and its R 2 The average value of the proposed method is 0.8 and the RMSE is 0.21.
[0099] In this example, the stability of organic SM-HTMs established based on SHAP screening features is rapidly predicted by GBRT regression and the prediction results of the independent test set are as follows: Figure 7 As shown in the figure, the prediction results of the 8 samples in the test set are good, and the R 2 The average value of the regression line is 0.79 and the RMSE is 0.16.
[0100] The method in this embodiment uses sample data from literature to establish a SHAP screening feature GBRT rapid prediction model for organic SM-HTMs. It has the advantages of low cost and green environmental protection, and evaluates the stability and reliability of the data modeling method with good results. It can also guide the actual operation of the experiment and avoid blindness.
[0101] Embodiment seven:
[0102] This embodiment is basically the same as the above embodiment, with the following features:
[0103] In this embodiment, see Figure 6 , using the SHAP nested GBRT method, the positive and negative effects of each eigenvalue on the target feature (hole mobility) are calculated to further explain the relationship between the feature and the target feature. Red indicates a strong eigenvalue effect, and blue indicates a weak feature effect. Red points appear more in positive effects, and blue points appear more in negative effects, indicating that this feature is positively correlated with the target feature; red points appear more in negative effects, and blue points appear more in positive effects, indicating that this feature is negatively correlated with the target feature. It can be seen that the CATS2D_09_AL feature is negatively correlated with the target feature, the MATS5m feature is negatively correlated, the Eta_betaS_A feature is negatively correlated, the ATSC5e feature is negatively correlated, and the GATS8i feature is negatively correlated.
[0104] In this example, literature related to the five characteristics was consulted, the relationship between their structure and hole mobility was explained, and a QSPR model was established. The CATS2D_09_AL feature indicates the number of pairs of hydrogen bond acceptors (O atoms and N atoms not connected to H) and lipophilic atoms (C atoms connected only to C and H and S atoms in CSC) with a topological distance of 9 in the molecular structure, indicating that the number of AL atom pairs between 4 and 16 is beneficial to hole mobility; the MATS5m feature is the Moran autocorrelation with a lag of 5 weighted by atomic mass, indicating that the relative atomic mass of the atom pairs with a topological distance of 5 in the molecular structure is one greater than and one less than the average atomic relative mass, which is beneficial to the target value; Eta_betaS_A represents the ratio of the sum of all sigma electrons to non-hydrogen atoms; ATSC5e is the product of the difference between the electronegativity of the atom pairs with a topological distance of 5 and the average electronegativity, indicating that the closer the electronegativity of the atom pairs with a topological distance of 5 is to the average electronegativity, the more advantageous it is for hole mobility; GATS8i is the Moran autocorrelation with a lag of 8 weighted by the first ionization potential, indicating that the closer the first potential of the atom pairs with a topological distance of 8 is, the more favorable it is to the target value.
[0105] Embodiment eight:
[0106] This embodiment is basically the same as the above embodiment, with the following features:
[0107] In this embodiment, see Figure 7 , using the SHAP nested GBRT method, the positive and negative effects of each eigenvalue on the target feature (stability) are calculated to further explain the relationship between the feature and the target feature. Red indicates a strong eigenvalue effect, and blue indicates a weak feature effect. Red points appear more in positive effects, and blue points appear more in negative effects, indicating that this feature is positively correlated with the target feature; red points appear more in negative effects, and blue points appear more in positive effects, indicating that this feature is negatively correlated with the target feature. It can be seen that the TPSA (Tot) feature is negatively correlated with the target feature, the RBF feature is negatively correlated, the T (O..S) feature is negatively correlated, the F04 [CS] feature is negatively correlated, and the RBN feature and the C-025 feature have no obvious linear relationship with the target value.
[0108] In this embodiment, the literature on 6 features is consulted, the relationship between their structure and stability is explained, and the QSPR model is established. The TPSA (Tot) feature represents the sum of the surface area contributions of polar atom types (N\O\S\P) in the molecular structure, indicating that the fewer polar molecules, the more favorable the target value; the RBF feature represents the number of rotatable bonds (RBN) in the molecule divided by the number of bonds (nBT), and its characteristic value is proportional to the characteristic RBN, so reducing the number of rotatable bonds between molecules can reduce the RBF characteristic value, which is beneficial to the target value; the T (O..S) feature is the sum of the topological distances between O and S in the molecule, indicating that the shorter the distance between O and S in the molecule, the more favorable the target value; F04 [CS] represents the frequency of CS atom pairs with a topological distance of 4 in the molecular structure; the C-025 feature is defined as the number of specific atom types in the molecule.
[0109] Embodiment nine:
[0110] This embodiment is basically the same as the above embodiment, with the following features:
[0111] In this example, an interpretable GBRT regression model was constructed based on SHAP values to accelerate the QSPR method of discovering organic SM-HTMs with high hole mobility and stability, and new core groups and side groups were designed, such as Figure 8 The newly designed core groups and side groups were merged into the original data set to form a core group library and a side group library. The groups in the two libraries were randomly combined to generate a large number of virtual samples. These virtual samples were put into the constructed QSPR model for high-throughput screening. The screening results of some samples are shown in Table 5.
[0112] Table 5. Some high-throughput screening characteristic values and hole mobility (μ / cm 2 V -1 s -1 )Prediction results table
[0113]
[0114]
[0115] Table 6. Prediction results of some high-throughput screening characteristic values and stability (η)
[0116]
[0117] In this embodiment, the generated 848 virtual samples are put into the established model to predict their corresponding hole mobility and stability, and the hole mobility>2.00×10 -1 (cm 2 V -1 s -1 ), while the molecular guidance experiments with stability >2.78.
[0118] Embodiment ten:
[0119] This embodiment is basically the same as the above embodiment, with the following features:
[0120] In this embodiment, a system for implementing a quantitative structural relationship method to predict the hole mobility and stability of an organic small molecule hole transport material in a perovskite solar cell implements the QSPR method for predicting hole mobility and stability based on the SHAP value constructed by an interpretable GBRT regression model in the above embodiment, including:
[0121] Input module: Search the molecular structure of organic SM-HTMs in perovskite solar cells, the hole mobility and stability calculated theoretically from the published literature, and use them as input data;
[0122] Data analysis module: Using the data obtained from the input module, the QSPR method based on the SHAP value is constructed to construct an interpretable GBRT regression model to predict the hole mobility and stability of organic SM-HTMs.
[0123] Output module: Output the hole mobility and stability of organic SM-HTMs.
[0124] The system of this embodiment can quickly predict the hole mobility and stability of organic SM-HTMs in perovskite solar cells (PSCs), especially constructing an interpretable GBRT regression model based on SHAP value to predict the quantitative structural relationship (QSPR) of hole mobility and stability, which is helpful for low-cost and high-efficiency discovery of efficient organic small molecule hole transport materials.
[0125] In summary, the above embodiment constructs an interpretable GBRT regression model based on SHAP value to accelerate the discovery of QSPR method of organic SM-HTMs with high hole mobility and stability, including the following steps: (1) searching the structure of organic SM-HTMs in perovskite solar cells, hole mobility and stability obtained by theoretical calculation from literature; (2) drawing SM-HTMs molecules with ChemDraw and generating descriptors with Dragon software; (3) taking hole mobility and stability as target variables and descriptors generated by Dragon as independent variables, data preprocessing is performed, and the preprocessed data set samples are used as data set samples for subsequent modeling; (4) randomly dividing the preliminarily screened data set into training set and test set, and the proportion of the test set accounts for 15% of the entire data set; (5) using SHAP embedding The GBRT method is used to screen variables according to the "contribution value" of the target variable, and the optimal variable subset selection for GBRT modeling is obtained by comparison and leave-one-out cross-validation; (6) A rapid prediction model for the hole mobility and stability of organic SM-HTMs molecules is established by GBRT regression to predict the hole mobility and stability of the test set; (7) The literature is consulted and combined with SHAP to explain the features, and the relationship between the features and the target features is further obtained, a QSPR model is constructed, and new core groups and side groups are designed; (8) The newly designed core groups and side groups are merged into the original data set to form a core group library and a side group library, and the groups in the two libraries are randomly combined to generate a large number of virtual samples, and these virtual samples are put into the constructed QSPR model for high-throughput screening to quickly predict the hole mobility and stability of organic SM-HTMs. The above embodiment is based on reliable literature real values and modeling methods, and the prediction model of organic SM-HTMs constructed has the advantages of convenience, speed, and no chemical pollution.
[0126] The above embodiment method avoids repeated experiments and the process of trial and error. The structure descriptor is generated by Dragon software. After SHAP nested GBRT method feature screening and modeling, the hole mobility and stability of organic SM-HTMs can be predicted in advance. The operation process is simple, low-cost and pollution-free, and only one person is required to complete it. In the above embodiment method, SHAP and references are used to explain the features and construct a QSPR model, which better explains the relationship between the structure and function of the dye molecule and helps solve the shortcomings that most methods cannot explain. The above embodiment method uses GBRT to establish a QSPR model for quickly predicting the hole mobility and stability of organic SM-HTMs, generates a large number of virtual samples by manual drawing, and puts these samples into the constructed QSPR model for high-throughput screening, and quickly predicts the hole mobility and stability of organic SM-HTMs with new combinations of fragments, providing a reference for PSCs researchers, shortening research and development time, and reducing research and development costs.
[0127] The above describes the embodiments of the present invention in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments. Various changes can be made according to the purpose of the invention. Any changes, modifications, substitutions, combinations or simplifications made according to the spirit and principle of the technical solution of the present invention should be equivalent replacement methods. As long as they meet the purpose of the invention and do not deviate from the technical principles and inventive concepts of the present invention, they belong to the protection scope of the present invention.
Claims
1. A QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values, characterized in that: The following steps are involved: 1) Use computer systems to search for literature, and search the literature for the structure, theoretically calculated hole mobility and stability of organic small molecule hole transport materials in perovskite solar cells; 2) Draw the molecular structure of organic SM-HTMs using ChemDraw and generate descriptors using Dragon software; 3) For the initially screened data set, randomly divide it into training set and test set, and the proportion of the test set accounts for 15% of the entire data set; 4) The SHAP-nested GBRT method was used for feature screening, and the optimal variable subset selection for GBRT modeling was obtained by comparative leave-one-out cross-validation; 5) Establish a rapid prediction model for hole mobility and stability of organic SM-HTMs using GBRT regression; 6) Based on the established rapid prediction model of hole mobility and stability of organic SM-HTMs, predict the hole mobility and stability of the molecules to be detected; 7) The relationship between the variable and the target variable is further obtained based on the positive and negative impact of the SHAP value; 8) Based on the influence of the positive and negative SHAP values of the variables on the target variables in step 7), consult the literature, use the explanations of the features in the references, construct a QSPR model, and design new core groups and side groups; 9) The newly designed core groups and side groups are merged into the original data set to form a core group library and a side group library. The groups in the two libraries are randomly combined to generate a large number of virtual samples. These virtual samples are put into the constructed QSPR model for high-throughput screening to quickly predict the hole mobility and stability of organic SM-HTMs.
2. The QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values according to claim 1, characterized in that: In the step 1), after searching the literature, the obtained data samples are preprocessed, including unifying the calculation method of sample molecules, sorting out the molecular structure of the samples, sorting out the target variables hole mobility and stability, and determining the number of sample data.
3. The QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values according to claim 1, characterized in that: In step 4), the SHAP nested GBRT method is used to screen variables according to the "contribution value" of the target variable, and the optimal variable subset selection for GBRT modeling is obtained by comparative leave-one-out cross-validation.
4. The QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values according to claim 1, characterized in that: In the step 9), the high-throughput screening method comprises the following steps: The design conditions are as follows: select organic SM-HTMs molecules studied theoretically, divide the core groups and side groups into fragments, randomly combine the core groups and side groups, and draw a large number of virtual samples through ChemDraw; Then the large number of virtual samples are put into the constructed prediction model for forecasting.
5. A QSPR system for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values, executing the QSPR method for predicting hole mobility and stability by constructing an interpretable GBRT regression model based on SHAP values as described in any one of claims 1 to 4, specifically comprising: Input module: Search the molecular structure, theoretically calculated hole mobility and stability of organic SM-HTMs in perovskite solar cells from publicly available literature and use them as input data; Data analysis module: Using the data obtained from the input module, the QSPR method based on the SHAP value is constructed to construct an interpretable GBRT regression model to predict the hole mobility and stability of organic SM-HTMs. Output module: Output the hole mobility and stability information of organic SM-HTMs.
