Polyimide database construction method, molecular design method and application
By generating polyimide molecular chains through three-level assembly and combining them with machine learning models, the problem of unclear design direction for polyimide materials has been solved, achieving efficient molecular design and rapid screening, shortening the R&D cycle, and promoting the development of flexible electronic devices.
Patent Information
- Application Number
- CN202510520922.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-24
- Publication Date
- 2025-11-25
AI Technical Summary
In the existing technology, the molecular structure design direction of polyimide materials is unclear, and the structure-performance relationship is unclear, resulting in a long development cycle in the field of flexible electronic devices.
A three-level assembly strategy was adopted to generate ABOC polyimide molecular chains, establish a polyimide database, and combine machine learning algorithms and molecular fingerprinting technology to construct a machine learning model. Through cross-validation and hyperparameter tuning, novel polyimide molecules that meet the performance requirements were screened out.
It significantly improves molecular design efficiency, shortens the R&D cycle, reduces experimental costs, enables high-throughput screening and performance prediction, provides a rapid development tool for polyimide materials, and promotes applications in the field of flexible electronic devices.
Smart Images
Figure CN121011271A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data processing, in particular to a polyimide database construction method and a molecular design method and application. BACKGROUND
[0002] Polyimides have shown wide and key application value in high-tech fields such as flexible electronics, display technology, intelligent manufacturing and semiconductor packaging due to their excellent comprehensive performance. In particular, in the field of flexible electronic devices, polyimide films have become an important supporting material for realizing the flexibility and lightweight of devices due to their lightweight, impact resistance and high flexibility. However, with the increasing demand for device performance in the field of flexible electronic devices, not only high flexibility is required, but also the optical properties of colorless transparency in the ultraviolet-visible band and excellent thermal stability are required to meet the stringent requirements of new generation flexible electronic devices on material optical properties. In fact, different dianhydride and diamine monomers can be combined to synthesize polyimides with different performance characteristics. For example, polyimide synthesized by using 4,4'-oxydianiline (ODA) and pyromellitic dianhydride (PMDA) has good thermal stability and mechanical properties, while using other monomer combinations may make the polyimide have special optical properties or solubility. However, due to the lack of a systematic and comprehensive database, researchers and enterprises often need to spend a lot of time and effort to consult scattered literature and experimental data when selecting polyimide materials that meet the property requirements and designing new molecular structures, which is inefficient and difficult to ensure accuracy.
[0003] To address this challenge, researchers have proposed various molecular design strategies, such as introducing fluorine groups with strong electronegativity, constructing bulky non-conjugated structural units, and using aliphatic and flexible connecting bridges to suppress CT effect and optimize energy level distribution. Although these methods have improved the optical properties of polyimides to some extent, most of them still rely on repeated experiments and trial-and-error, resulting in long development cycles, high costs, and low efficiency, making it difficult to achieve rapid material screening and performance-oriented optimization.
[0004] In summary, constructing a systematic, comprehensive and accurate polyimide database and developing an intelligent method system that integrates high-precision prediction and molecular structure analysis can help promote the research and application of polyimide materials and meet the growing demand in the field of flexible electronic devices. SUMMARY
[0005] In order to solve the problem that the molecular structure design direction of existing transparent polyimide materials meeting the application in the field of flexible electronic devices is not clear, the structure-property correlation is not clear, and the development cycle is long, the present application proposes a polyimide database construction method and a molecular design method and application. The technical scheme of the present application is as follows:
[0006] A polyimide database construction method, comprising the following steps:
[0007] S1: collect dianhydride information and diamine monomer information, connect dianhydride and diamine monomer to generate polyimide monomer, and the connection mode is as follows:
[0008] Primary assembly: connect A-B according to the rules of A fragment substituent condensation, if A fragment ends with -HC=O, then directly connect the -NH2 of B fragment for condensation reaction, if A fragment starts with -HC=O, then preposition B fragment;
[0009] Secondary assembly: connect the diamine after hydroxyl treatment as C fragment, and connect C fragment and B fragment through ether bond to generate B-O-C intermediate;
[0010] Tertiary assembly: connect A fragment and B-O-C intermediate according to the rules of primary assembly to form A-B-O-C polyimide molecular chain;
[0011] S2: using os and RDKit library in Python language to identify and standardize A-B-O-C structure, batch generate corresponding SMILES molecular linear expression, and further convert to molecular fingerprint to form polyimide database including polyimide molecular structure information, fragment source and characteristic vector;
[0012] Further, in the tertiary assembly, -NH2 exists at both ends of the B-O-C intermediate, and ring-opening condensation reaction occurs with -HC=O of A fragment to form amide bond;
[0013] Further, the molecular fingerprint in S2 includes Morgan fingerprint, MACCS fingerprint, RDKit fingerprint, Avalon fingerprint, topological fingerprint and atom pair fingerprint.
[0014] A molecular design method, comprising the following steps:
[0015] T1: collect known polyimide molecular structure, calculate the performance parameters of known polyimide molecules such as ultraviolet absorption wavelength, electron transfer number and thermal decomposition temperature by density functional theory, and construct known data set;
[0016] T2: combine the molecular fingerprint in S2 with a machine learning algorithm to build a machine learning model, the machine learning algorithm includes covering gradient boosting, ensemble learning, extreme random tree, AdaBoost, random forest and Bagging, use the data of a known data set as a training set, evaluate the performance of the machine learning model by cross-validation analysis and carry out model parameter adjustment, and screen out the optimal combination of machine learning algorithm and molecular fingerprint; input the SMILES linear expression in the polyimide database into the machine learning model, and output the predicted ultraviolet absorption wavelength, electron transfer number and thermal decomposition temperature of each molecule;
[0017] T3: set performance model screening conditions, screen new molecules in the polyimide database that meet the indicators, perform structure correlation analysis and key substructure contribution evaluation, and guide the structure optimization and directed design of polyimide molecules;
[0018] Further, the number of known polyimide molecules in T1 is not less than 88, wherein the number of known dianhydride is not less than 9;
[0019] Further, the cross-validation in T2 is to use correlation coefficient R 2 , mean absolute error MAE and root mean square error RMSE evaluation index to measure the model, and the formula of the evaluation index is as follows:
[0020]
[0021] In the formula, y i represents the true value of the i-th sample, f i represents the predicted value of the i-th sample, and n is the total number of samples;
[0022] Further, the model parameter adjustment method in T2 includes Bayesian search and grid search;
[0023] Further, the screening condition in T3 is that the ultraviolet absorption wavelength is less than 400 nm, the charge transfer number is less than 0.2e, and the thermal cracking temperature is greater than 500℃;
[0024] Further, the structure correlation analysis and key substructure contribution evaluation in T3 is to perform systematic importance sorting on the molecular fingerprint descriptor by SHAP model, and identify the molecular structure sub-fragment that meets the screening condition.
[0025] The application of a polyimide molecule is applied to the field of flexible electronic devices.
[0026] Compared with the prior art, the present application solves the problem that the existing transparent polyimide material molecular structure design direction is not clear and the structure-performance correlation is not clear, which leads to a long development cycle, and has the specific beneficial effects that:
[0027] 1. Improve the efficiency of molecular design: The present application adopts a three-level assembly strategy to generate A-B-O-C polyimide molecular chains and uses fingerprint technology to establish a real polyimide database, realizing efficient molecular design of polyimide; By combining machine learning and molecular fingerprint technology to build a machine learning model, using known data sets to screen the optimal combination of machine learning algorithms and molecular fingerprints, using cross-validation, hyperparameter tuning and other methods, the machine learning model continuously updates the mapping relationship between molecular structure and properties, significantly improving the ability and accuracy of the machine learning model to predict the properties of newly assembled polyimide molecules.
[0028] 2. High-throughput screening and performance prediction: The present application can quickly screen new polyimide molecular chains that meet the target performance (ultraviolet absorption wavelength λ < 400 nm, charge transfer number CTE < 0.2e, thermal cracking temperature T d >500℃) by setting the screening parameter range of the machine learning model for polyimide ultraviolet absorption wavelength, electron transfer number and thermal stability, providing a theoretical calculation tool for the rapid development of high-performance polyimide materials in the field of flexible electronic devices, and significantly reducing the experimental cost through machine learning model-driven virtual screening, solving the industry problem of unclear structure-property correlation in the development of transparent polyimide.
[0029] 3. Shorten the development cycle: The present application establishes an intelligent path from molecular design to performance prediction, and the screening results of new polyimide molecules synthesized by the polyimide database through the machine learning model can directly guide the directed synthesis process, significantly reducing the trial-and-error link in traditional experimental synthesis, shortening the development cycle, saving a lot of time and cost, and promoting the research and development process of colorless and transparent polyimide materials and their application in the field of flexible electronic devices.
[0030] 4. Enhance the interpretability of the model: The present application combines the SHAP model to rank the importance of molecular fingerprint descriptors, revealing the key substructure sites that affect the performance of polyimide molecules. Through this method, the relationship between the structure and performance of polyimide molecules can be deeply understood, providing theoretical guidance for their directed synthesis. BRIEF DESCRIPTION OF DRAWINGS
[0031] Figure 1 Flowchart of polyimide database construction and polyimide molecular design method;
[0032] Figure 2 Polyimide dianhydride unit type and target performance normal distribution chart in known data set, wherein, figure (a) is 9 kinds of dianhydride unit type in known data set; figure (b-c) is the normal distribution chart of ultraviolet-visible spectrum, charge transfer number and thermal decomposition temperature of known polyimide;
[0033] Figure 3 R of the prediction results of the prediction of the ultraviolet absorption wavelength, the electron transfer number and the thermal decomposition temperature based on the combination of 6 kinds of machine learning algorithms and 6 kinds of molecular fingerprints 2 and the results of MAE are shown in the schematic diagram; wherein, figures (a-c) are R of the prediction results of the prediction of the ultraviolet absorption wavelength, the electron transfer number and the thermal decomposition temperature 2 schematic diagram; figures (d-f) are MAE schematic diagram of the prediction results of the prediction of the ultraviolet absorption wavelength, the electron transfer number and the thermal decomposition temperature; figures (g-i) are R of the prediction results of the prediction of the ultraviolet absorption wavelength, the electron transfer number and the thermal decomposition temperature predicted by the combination of GB and Rdkit 2 and MAE schematic diagram;
[0034] Figure 4 it is a schematic diagram of the analysis results of the SHAP model on the substructure of the ultraviolet absorption wavelength, the electron transfer number and the thermal decomposition temperature. DETAILED DESCRIPTION
[0035] In order to make the technical solutions of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings of the specification of the present application. It should be noted that the following embodiments are only used to better understand the technical solutions of the present application, and should not be understood as a limitation on the present application.
[0036] Example 1.
[0037] In this embodiment, a multi-level modular splicing method is used for polyimide molecular structure construction, a "three-level assembly" polyimide main chain assembly scheme is proposed, and a polyimide database construction method is provided based on the fragment synthesis idea. Specifically, it includes:
[0038] S1: collect dianhydride information and diamine monomer information, connect dianhydride and diamine monomer to generate polyimide monomer, and the connection mode is as follows:
[0039] First assembly: take dianhydride as A fragment and diamine as B fragment, connect A-B according to the condensation rule of A fragment substituent group; if A fragment ends with -HC=O, then directly connect -NH2 of B fragment for condensation reaction; if A fragment starts with -HC=O, then preposition B fragment;
[0040] Second assembly: take the diamine treated with terminal hydroxylation as C fragment, connect C fragment and B fragment through ether bond to generate B-O-C intermediate;
[0041] Third assembly: connect A fragment and B-O-C intermediate according to the first assembly rule, -NH2 exists at both ends of the B-O-C intermediate, and ring-opening condensation reaction occurs between -NH2 and -HC=O of A fragment to form amide bond, and finally form A-B-O-C polyimide molecular chain;
[0042] This embodiment constructs over 200,000 candidate polyimide molecular structures through a full combination of 9 dianhydrides (fragment A), 178 diamines (fragment B), and 126 monomers (fragment C). S2: The ABOC structures are identified and standardized using the os and RDKit libraries in Python, generating corresponding linear expressions for SMILES molecules in batches. Approximately 170,000 valid SMILES sequences are generated and further converted into molecular fingerprints (including Morgan fingerprints, MACCS fingerprints, RDKit fingerprints, Avalon fingerprints, topological fingerprints, and atom-pair fingerprints) to form a polyimide molecular database, including structural information, fragment origins, and feature vectors of polyimide molecules.
[0043] Example 2.
[0044] Based on the polyimide database established in Example 1, this embodiment constructs a machine learning model combining machine learning and molecular fingerprinting techniques, achieving efficient molecular design of polyimides and the ability to predict the properties of newly assembled polyimide molecules. The polyimide molecular design method provided in this embodiment includes the following steps:
[0045] T1: Collect known polyimide molecular structures (in this example, 88 polyimide molecules reported in experimental literature were selected, composed of 9 typical dianhydrides (A fragments) and 50 diamines (B fragments), and calculate the ultraviolet absorption wavelength (λ), electron transfer number (CTE), and thermal decomposition temperature (T) of the known polyimide molecules using density functional theory. d The performance parameters of the model are used to construct a known "structure-performance" dataset, which is then used as the basis for training and validating machine learning models.
[0046] T2: Combine the molecular fingerprint described in S2 with machine learning algorithms (including GB, EL, ADA, ET, RF and Bagging) to construct a machine learning model. Use the data from the known dataset constructed in S3 as the training set. Use cross-validation analysis to evaluate the performance of the machine learning model. Obtain the dominant ultraviolet absorption wavelength, charge transfer number and thermal stability parameters of each molecule in the known dataset. The results constitute the "structure-performance" known dataset, which is used as the basis for training and validating the machine learning model.
[0047] The cross-validation is performed using the correlation coefficient R. 2 The Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used to measure the evaluation model. The formulas for the evaluation indicators are as follows:
[0048]
[0049] In the formula, y i f represents the true value of the i-th sample.i This represents the predicted value of the i-th sample, where n is the total number of samples;
[0050] After cross-validation analysis, model parameter tuning (including Bayesian search and grid search) was performed. The optimal combination of machine learning algorithm and molecular fingerprint was selected as GB and Rdkit, with an accuracy of 99%. This combination was used as the input to the learning model. The linear expression of SMILES molecules in the polyimide molecular database was input into the machine learning model, and the predicted UV absorption wavelength, electron transfer number and thermal decomposition temperature of each molecule were output.
[0051] T3: The performance model screening criteria are set as follows: UV absorption wavelength less than 400 nm, charge transfer quantity less than 0.2e, and thermal decomposition temperature greater than 500℃. A trained GB model is used to predict and score all generated structures in the database, screening out potential colorless, transparent, and highly thermally stable polyimide candidate molecules. New molecules in the selected polyimide database that meet the criteria are then subjected to structural correlation analysis and key substructure contribution assessment. The molecular fingerprint descriptors are systematically ranked by importance using the SHAP model to identify molecular structural sub-fragments that meet the screening criteria. For example... Figure 1 A flowchart for constructing a polyimide database and a method for designing polyimide molecules.
[0052] like Figure 2 Figure 1 shows the normal distribution of polyimide dianhydride unit types and target performance in the known dataset. Figure 2(a) shows the nine dianhydride unit types in the known dataset, including PMDA, BTDA, BPDA, OPDA, 6FDA, DTDA, NTDA, HPMDA, and DSDA. Figure 3(b) shows the UV-Vis spectrum of the known polyimide, which shows a significant absorption peak in the region λ < 400 nm, reflecting the optical absorption characteristics of polyimide molecules in the UV region. Figure 4(c) shows the normal distribution of polyimide dianhydride unit types and target performance in the known dataset. The normal distribution of charge transfer amount (CTE) of imide molecules shows that the CTE value reaches its peak in the range of 0.4-0.6. In the range of CTE < 0.2e, the electron cloud inside the polyimide molecule is highly localized, and electrons are difficult to migrate, resulting in extremely weak charge transfer ability, which meets the application requirements of the field of flexible electronic devices. Figure (d) shows the normal distribution of the thermal decomposition temperature of known polyimide molecules. It can be seen from the figure that the thermal decomposition temperature is concentrated in the range of 418-528℃, which lays the foundation for subsequent screening of thermal decomposition temperatures.
[0053] Figure 3 R is used to predict the ultraviolet absorption wavelength, electron transfer number, and thermal decomposition temperature based on six machine learning algorithms and six combinations of molecular fingerprints. 2 A schematic diagram of the MAE results; where Figure (a) shows the predicted R value of the predicted ultraviolet absorption wavelength.2 Schematic diagram; Figure (b) shows the R value of the predicted number of electrons transferred (i.e., the number of electrons transferred from excited-state electrons). 2 Schematic diagram; Figure (c) shows the predicted R value for the predicted thermal decomposition temperature. 2 Schematic diagrams; Figure (d) shows the MAE diagram of the predicted ultraviolet absorption wavelength; Figure (e) shows the MAE diagram of the predicted number of electrons transferred (i.e., the number of electrons transferred from excited states); Figure (f) shows the MAE diagram of the predicted thermal decomposition temperature; Figures (g)-(i) show the R... 2 The diagram shows that the combination of GB and Rdkit provides the best prediction results. It is used to predict the UV absorption wavelength, electron transfer number, and thermal decomposition temperature of new molecules in the polyimide database constructed in this invention that meet the specified criteria. This can promote the research and development of colorless and transparent polyimide materials and their application in flexible electronic devices. Figure 4 This is a schematic diagram of the analysis results of the SHAP model on the substructure of ultraviolet absorption wavelength, electron transfer number, and thermal decomposition temperature. As can be seen from the figure, the SHAP model clarifies the direction and degree of influence of different features on each target value, and can identify molecular structure sub-fragments that meet the screening conditions.
[0054] In summary, this invention employs a three-stage assembly strategy to generate ABOC polyimide molecular chains and utilizes fingerprinting technology to establish a real polyimide database, achieving efficient molecular design of polyimides. Through the optimal combination of machine learning algorithms and molecular fingerprinting, the ability and accuracy of machine learning models to predict the properties of newly assembled polyimide molecules are significantly improved, and molecules that meet the target properties (UV absorption wavelength λ < 400 nm, charge transfer quantity CTE < 0.2e, thermal decomposition temperature T0) are rapidly screened. d The novel polyimide molecular chain (>500℃) significantly reduces the trial-and-error steps in traditional experimental synthesis, shortens the research and development cycle, saves a lot of time and costs, and promotes the research and development process of colorless and transparent polyimide materials and their application in the field of flexible electronic devices.
[0055] The above description of the embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make several improvements and modifications to the present invention without departing from the principles of the present invention, and these improvements and modifications also fall within the protection scope of the claims of the present invention.
[0056] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for constructing a polyimide database, characterized in that, Includes the following steps: S1: Collect dianhydride and diamine monomer information, and link the dianhydride and diamine monomer to generate polyimide monomers. The linking method is as follows: Primary assembly: Using dianhydride as fragment A and diamine as fragment B, connect A and B according to the rules of substituent condensation of fragment A; if fragment A ends with -HC=O, then directly connect to -NH2 of fragment B for condensation reaction; if fragment A starts with -HC=O, then place fragment B first. Secondary assembly: The terminally hydroxylated diamine is used as the C fragment, and the C fragment is linked to the B fragment via an ether bond to generate the BOC intermediate; Three-level assembly: Following the rules of primary assembly, fragment A is linked to the BOC intermediate to form the ABOC polyimide molecular chain; S2: The os and RDKit libraries in Python are used to identify and standardize the ABOC structure, generate corresponding linear expressions for SMILES molecules in batches, and further convert them into molecular fingerprints to form a polyimide database that includes polyimide molecular structure information, fragment sources, and feature vectors.
2. The method for constructing a polyimide database according to claim 1, characterized in that, In the three-stage assembly, the BOC intermediate has -NH2 at both ends, which undergoes a ring-opening condensation reaction with -HC=O of segment A to form an amide bond.
3. The method for constructing a polyimide database according to claim 1, characterized in that, The molecular fingerprints described in S2 include Morgan fingerprints, MACCS fingerprints, RDKit fingerprints, Avalon fingerprints, topological fingerprints, and atom-pair fingerprints.
4. A molecular design method, characterized in that, Based on the polyimide database of any one of claims 1-3, the molecular design method includes the following steps: T1: Collect known polyimide molecular structures, calculate the performance parameters of known polyimide molecules such as ultraviolet absorption wavelength, electron transfer number and thermal decomposition temperature using density functional theory, and construct a known dataset; T2: Combine the molecular fingerprint described in S2 with machine learning algorithms to construct a machine learning model. The machine learning algorithms include gradient boosting, ensemble learning, extreme random trees, AdaBoost, random forest, and Bagging. Use the data from the known dataset as the training set, and use cross-validation analysis to evaluate the performance of the machine learning model and perform model parameter tuning to select the optimal combination of machine learning algorithms and molecular fingerprints. Input the linear expression of SMILES molecules from the polyimide database into the machine learning model, and output the predicted ultraviolet absorption wavelength, electron transfer number, and thermal decomposition temperature of each molecule. T3: Set performance model screening conditions, screen new molecules that meet the indicators in the polyimide database, perform structural correlation analysis and key substructure contribution assessment, and guide the structural optimization and directional design of polyimide molecules.
5. The molecular design method according to claim 4, characterized in that, The number of known polyimide molecules described in T1 is no less than 88.
6. The molecular design method according to claim 4, characterized in that, The cross-validation described in T2 uses the correlation coefficient R. 2 The Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) are used to measure the evaluation model. The formulas for the evaluation indicators are as follows: In the formula, y i f represents the true value of the i-th sample. i Let represent the predicted value of the i-th sample, where n is the total number of samples.
7. The molecular design method according to claim 4, characterized in that, The model parameter tuning methods described in T2 include Bayesian search and grid search.
8. The molecular design method according to claim 4, characterized in that, The screening conditions described in T3 are: ultraviolet absorption wavelength less than 400 nm, charge transfer quantity less than 0.2e, and thermal decomposition temperature greater than 500℃.
9. The molecular design method according to claim 4, characterized in that, The structural correlation analysis and key substructure contribution assessment described in T3 involve systematically ranking the importance of molecular fingerprint descriptors using the SHAP model to identify molecular structural sub-fragments that meet the screening criteria.
10. An application of a polyimide molecule, characterized in that, The polyimide molecule is provided by the database described in claims 1-3 or designed by the molecular design method described in claims 4-9, and is applied in the field of flexible electronic devices.