A method for expressing a unique fingerprint of a porous metal organic framework material

Abstract

This invention discloses a proprietary fingerprint representation method for porous metal-organic framework (MOF) materials, belonging to the technical field of digital R&D of MOF materials. First, the crystallographic information file of MOFs is used as input, and the three strongest XRD peaks are selected to represent the structural information fingerprint of the MOFs. Second, the chemical information fingerprints of MOF substructures, including SBUs and ligands, are obtained and combined with the structural information fingerprint to construct the proprietary fingerprint of the MOFs. Based on the proprietary fingerprint of MOF materials, the geometric structure and physicochemical characteristics of MOFs are analyzed. Finally, data access, export, and user feedback interfaces are provided on a database website. The advantages of this invention are: 1) All program packages are independently written in 100% Python, allowing the technology to be applied to existing R&D processes; 2) The proprietary fingerprint of MOF materials is unique, concise, and reusable, reducing the search difficulty for material scientists; 3) The proposed MOF feature calculation workflow is helpful for the construction of various performance prediction models and the mining of related descriptors.

Description

Technical Field

[0001] This invention belongs to the technical field of digital research and development of MOFs materials, and mainly relates to a technology for digitally representing MOFs materials based on Python methods. Background Technology

[0002] Metal-organic frameworks (MOFs) are a class of crystalline materials composed of metal ions or clusters and organic ligands. Their unique structural features and excellent porosity have attracted widespread attention in the field of catalysis. The tunability and diversity of MOFs make them an ideal platform for catalyst design and development. To date, nearly 70,000 MOFs have been experimentally synthesized, and hundreds of thousands of hypothetical MOFs have been predicted through computation. Databases such as the Materials Project and COD databases have also emerged, providing free computational materials data and research tools to materials scientists worldwide. However, in the development of MOF materials, the digital representation and processing of MOF structures and properties still face challenges. Traditionally, researchers use MOFs' CCDC numbers, COD IDs, QMOF IDs, CSD Refcodes, chemical formulas, and minimum and maximum values ​​of different elemental quantities to search for MOF crystal structure information, and use specialized software and tools to process MOF structures. These methods have some limitations. First, using cumbersome processes and complex programming languages ​​can lead to longer development cycles and reduced efficiency. Secondly, the lack of standardized identifiers increases the difficulty of research for researchers, creating data format barriers and significantly reducing data flow and reusability. Finally, the lack of sufficient characteristic parameters may limit the accurate description and comparison of MOF materials and restrict the mining of MOF material properties. Therefore, a simple and efficient method is needed to achieve the digital representation of MOF materials, while providing unique identifiers and sufficient characteristic parameters. Summary of the Invention

[0003] The purpose of this invention is to address the aforementioned problems by establishing a proprietary fingerprint representation method for porous metal-organic framework (MOF) materials. This method involves developing unique, concise, and reusable proprietary fingerprints for MOF materials using 100% Python, and integrating high-throughput materials computing software packages such as RDkit, mofdscribe, and matminer to enrich the characteristic information of MOF materials. This standardizes the identifier format for MOF materials and promotes data sharing and utilization in the field of digital R&D of MOF materials.

[0004] To achieve the above objectives, the present invention employs the following technical methods:

[0005] First, the standard SMILES format of SBU and Linker in MOF materials and the three strongest peaks of their X-ray diffraction patterns are used as identifiers for SBU, ligand, and topology, respectively, and sorted in the order of SBU.Linker.XRD to construct a proprietary fingerprint representation for MOF materials. Then, based on the above proprietary fingerprint of MOFs, the geometric structure and physicochemical characteristics of MOFs are fully calculated. Finally, interfaces for material data access, export, and user feedback are provided.

[0006] The specific steps are as follows:

[0007] S1: Based on the MOFs database, X-ray diffraction pattern calculations were performed on its crystallographic information files to initially obtain the three strongest XRD peaks as structural information fingerprints of MOFs.

[0008] S2: Using the standard SMILES format, further obtain the chemical information fingerprints of MOF substructures, including SBU and ligands, and combine them with the structural information fingerprints obtained in S1 to form the proprietary fingerprint expression of MOF materials.

[0009] S3: Based on the proprietary fingerprint of MOFs materials in S2, calculate the geometric structure and physicochemical characteristics of MOFs using Python algorithms, in preparation for the construction of various performance prediction machine learning models and related descriptor mining work.

[0010] S4: Provide data access, export, and user feedback interfaces on the database website to promote the continuous improvement and optimization of the digital representation of MOF materials.

[0011] Furthermore, the MOFs database mentioned in S1 uses the 2019-ASR CoRE and h-MOFs database to verify the uniqueness of the MOFs proprietary fingerprint representation method, and uses the QMOF database to verify whether the feature subset calculated in S3 can effectively cover the global MOFs information.

[0012] Furthermore, the X-ray diffraction pattern calculation process described in S1 is based on the pymatgen.core.lattice module traversing all reciprocal lattice points within the confined sphere, without requiring multiplicity correction.

[0013] Furthermore, the X-ray diffraction pattern calculation process described in S1 involves setting the peak position range to 0°-90°, the step size to 0.02°, the peak position matching error of the 2019-ASR CoRE database to ±0.3°, and the peak position matching error of the h-MOFs database to ±0.2°.

[0014] Furthermore, the proprietary fingerprint representation of MOFs materials described in S2 uses the three strongest peaks of the SBU and Linker in the MOFs material, as well as the canonical SMILES format and X-ray diffraction patterns, as identifiers for SBU, ligand, and topology, respectively, and sorts them in the order of SBU.Linker.XRD, with different components separated by a "." separator.

[0015] Furthermore, the SBU and ligand specification SMILES mentioned in S2 depend on the Open Babel core and Pybel libraries.

[0016] Furthermore, in S3, when calculating the geometric structure and physicochemical features of MOFs based on MOFs proprietary fingerprints, Python algorithm libraries such as mofdscribe, matminer, and RDkit are used. The calculated feature categories are divided into BU_Featurizer, Chemistry, Pore, Topology, and Host_Guest, totaling 5 categories and 158-dimensional feature sets.

[0017] Furthermore, the data access and export process described in S4 involves setting a Custom (upload ciffile) tab, which allows users to customize and upload ciff files, and provides channels for downloading data in either JSON or CSV format.

[0018] Furthermore, the user feedback interface described in S4 uses 3Dmol.js to visualize MOFs and their substructures, and supports users to submit their own experimental data in the Data Upload tab, thereby improving the quality of data availability.

[0019] The advantages of this invention are:

[0020] 1. This invention creates unique, concise, and reusable proprietary fingerprints for MOF materials using 100% Python language, and integrates high-throughput materials computing packages such as pymatgen, mofdscribe, and matminer to enrich the characteristic information of MOF materials. It also standardizes the format of MOF material identifiers, reduces the search difficulty for materials scientists, and promotes data sharing and utilization in the field of digital R&D of MOF materials.

[0021] 2. This invention innovatively proposes a computational workflow that can effectively cover the full-domain information features of MOF materials, which is helpful for the construction of various performance prediction machine learning models and related descriptor mining work.

[0022] 3. This invention provides a platform for accessing, exporting, and providing user feedback on MOF material data, which meets the search needs of materials scientists, improves the credibility of data-driven R&D models for materials synthesis workers, significantly reduces the R&D cycle and cost of MOF materials, and promotes the continuous improvement and optimization of the digital representation of MOF materials.

[0023] 4. All program packages in this invention are independently written in 100% Python, which simplifies the process of digital representation and has better portability and integration. They can be easily integrated with other tools and software, which makes it easier for researchers to apply this technology to existing R&D processes. Attached Figure Description

[0024] Figure 1 This is a workflow framework diagram of a Python-based method for representing proprietary fingerprints of MOFs materials proposed in this invention.

[0025] Figure 2 This is an example of a proprietary fingerprint for MOFs materials constructed according to the present invention.

[0026] Figure 3 This is a schematic diagram illustrating the MOFs proprietary fingerprint feature display and calculation method provided by the present invention in a database website. Detailed Implementation

[0027] The technical method of the present invention will be further described below with reference to specific embodiments and accompanying drawings.

[0028] Example 1: Verifying the uniqueness of the MOFs proprietary fingerprint representation method based on the 2019-ASR CoRE database.

[0029] The 2019-ASR CoRE database has removed atomic misalignments and repetitive structures, and currently contains 12,000 MOF data points. Corresponding MOF materials can be fabricated based on relevant reports. Example 1 describes the fingerprinting of all MOFs in the 2019-ASR CoRE database. The overall process is as follows: Figure 1 As shown, it includes the following steps:

[0030] S1. Utilizing Python algorithm modules such as pandas, os, and pymatgen.io.cif, the custom-defined `read_cif_files_to_structure()` function reads CIF files in batches to obtain MOF structure information, saving it as a DataFrame. Then, using the `xrd.get_pattern(zero_background=True).as_dict()` function, it automatically calculates and obtains the peak values ​​from 0° to 90°, with each step representing 0.02° of MOF material X-ray diffraction data. Furthermore, it extracts the XRD peak values ​​and their corresponding angles, storing the top three peak positions in a dictionary format as MOF structure information fingerprints.

[0031] S2. The `decomposition()` function is called to identify the atomic connections in the CIF (Chemical Information Framework). By determining the connectivity, the MOFs are segmented into SBUs and Linkers, stored in .mol ​​file format. The `WriteFile` method of the `OBConversion` object is then used to convert the molecular information in the `OBMol` object into the standard SMILES format, which is used as the chemical fingerprint of the MOFs. Finally, the MOF-specific fingerprints are constructed by sorting them according to the `SBU.Linker.XRD` order, with different components separated by a "." delimiter. A specific example is shown below. Figure 2 As shown, it is stored in the specified file in CSV format.

[0032] S3. Utilizing Python algorithms such as mofdscribe, matminer, and RDkit, it calls the fingerprints from S2 and batch-calculates attributes including AtomicOrbitals, OxidationStates, BandCenter, BranchPointEnergy, DopingFermi, CoordinationNumber, and OPSiteFingerprint. These attributes can be categorized into five main classes: BU_Featurizer, Chemistry, Pore, Topology, and Host_Guest, totaling 158 features. The output is in CSV format. The `featurize_dataframe()` function incorporates the `ignore_errors = True / False` parameter to retain only successfully calculated results, ensuring smooth computation.

[0033] S4. Using the fingerprint from S2 and the feature subset from S3 as source data, a user-friendly GUI interface is created, providing tabs such as Custom (upload cif file), Download, and Data Upload for users to customize operations. Figure 3The search interface provided for the database website.

[0034] To verify the scientific validity of the aforementioned Python-based method for expressing proprietary fingerprints of MOFs, 38 MOFs synthesized experimentally with coordination numbers of 6, 7, 8, and 9 were selected as the test set, with the peak position matching error set to ±0.2°. Then, a custom CIF file and experimental XRD data were uploaded to the Custom (upload CIF file) tab. The generated MOF identifiers were then compared with those in the database. It was found that 35 MOFs had completely identical SBU and Linker SMILES, and the XRD peak position matching error was less than ±0.2°, achieving an accuracy rate of 92.1%. Compared to traditional query methods such as CCDC ID and COD ID, this invention is more accurate and efficient.

[0035] Example 2: Verifying the uniqueness of the MOFs proprietary fingerprint representation method based on the h-MOFs database

[0036] The h-MOFs database is created by combining existing MOF building blocks (SBU, organic ligands, topology) and then selecting structurally sound and energy-stable MOFs, including those from computational simulations and experimentally synthesized models. It contains a total of 137,953 MOF data points. Example 2 describes fingerprinting of all MOFs in the h-MOFs database, the overall process of which is as follows: Figure 1 As shown, it includes the following steps:

[0037] S1. Utilizing Python algorithm modules such as pandas, os, and pymatgen.io.cif, the custom-defined `read_cif_files_to_structure()` function reads CIF files in batches to obtain MOF structure information, saving it as a DataFrame. Then, using the `xrd.get_pattern(zero_background=True).as_dict()` function, it automatically calculates and obtains the peak values ​​from 0° to 90°, with each step representing 0.02° of MOF material X-ray diffraction data. Furthermore, it extracts the XRD peak values ​​and their corresponding angles, storing the top three peak positions in a dictionary format as MOF structure information fingerprints.

[0038] S2. The `decomposition()` function is called to identify the atomic connections in the CIF (Chemical Information Framework). By determining the connectivity, the MOFs are segmented into SBUs and Linkers, stored in .mol ​​file format. The `WriteFile` method of the `OBConversion` object is then used to convert the molecular information in the `OBMol` object into the standard SMILES format, which is used as the chemical fingerprint of the MOFs. Finally, the MOF-specific fingerprints are constructed by sorting them according to the `SBU.Linker.XRD` order, with different components separated by a "." delimiter. A specific example is shown below. Figure 2 As shown, it is stored in the specified file in CSV format.

[0039] S3. Utilizing Python algorithms such as mofdscribe, matminer, and RDkit, it calls the fingerprints from S2 and batch-calculates attributes including AtomicOrbitals, OxidationStates, BandCenter, BranchPointEnergy, DopingFermi, CoordinationNumber, and OPSiteFingerprint. These attributes can be categorized into five main classes: BU_Featurizer, Chemistry, Pore, Topology, and Host_Guest, totaling 158 features. The output is in CSV format. The `featurize_dataframe()` function incorporates the `ignore_errors = True / False` parameter to retain only successfully calculated results, ensuring smooth computation.

[0040] S4. Using the fingerprint from S2 and the feature subset from S3 as source data, a user-friendly GUI interface is created, providing tabs such as Custom (upload cif file), Download, and Data Upload for users to customize operations. Figure 3 The search interface provided for the database website.

[0041] Example 1 has verified the scientific validity of the above-mentioned MOF material-specific fingerprint expression method based on Python. Example 2 aims to verify whether the MOF-specific fingerprint can effectively explore the unknown wide-area chemical space of MOF materials. The specific verification process is as follows: 82 MOFs synthesized in the experiment with coordination numbers of 4, 5, 6, 7, 8, and 9 were selected as the test set, and the peak position matching error was set to ±0.3°. Then, a custom CIF file and experimental XRD data were uploaded in the Custom (upload CIF file) tab. Subsequently, the MOF identifiers in the database were compared with those generated. It was found that 78 MOFs had completely consistent SBU and Linker SMILES, and the XRD peak position matching error was less than 0.3°, with an accuracy rate of 95.1%. This indicates that the MOF material-specific fingerprint expression method proposed in this invention can accurately express MOFs.

[0042] Example 3: Verifying whether a subset of MOF features can effectively cover global MOF information based on the QMOF database.

[0043] QMOF is currently the most comprehensive MOF database, containing 20,375 MOF data points sourced from nine databases including CDS and CORE. The data includes physicochemical properties such as band gap, bond order, density of states, and porosity. Example 3 describes the fingerprinting of all MOFs in the QMOF database, with the overall process as follows: Figure 1 As shown, it includes the following steps:

[0044] S1. Utilizing Python algorithm modules such as pandas, os, and pymatgen.io.cif, the custom-defined `read_cif_files_to_structure()` function reads CIF files in batches to obtain MOF structure information, saving it as a DataFrame. Then, using the `xrd.get_pattern(zero_background=True).as_dict()` function, it automatically calculates and obtains the peak values ​​from 0° to 90°, with each step representing 0.02° of MOF material X-ray diffraction data. Furthermore, it extracts the XRD peak values ​​and their corresponding angles, storing the top three peak positions in a dictionary format as MOF structure information fingerprints.

[0045] S2. The `decomposition()` function is called to identify the atomic connections in the CIF (Chemical Information Framework). By determining the connectivity, the MOFs are segmented into SBUs and Linkers, stored in .mol ​​file format. The `WriteFile` method of the `OBConversion` object is then used to convert the molecular information in the `OBMol` object into the standard SMILES format, which is used as the chemical fingerprint of the MOFs. Finally, the MOF-specific fingerprints are constructed by sorting them according to the `SBU.Linker.XRD` order, with different components separated by a "." delimiter. A specific example is shown below. Figure 2 As shown, it is stored in the specified file in CSV format.

[0046] S3. Utilizing Python algorithms such as mofdscribe, matminer, and RDkit, it calls the fingerprints from S2 and batch-calculates attributes including AtomicOrbitals, OxidationStates, BandCenter, BranchPointEnergy, DopingFermi, CoordinationNumber, and OPSiteFingerprint. These attributes can be categorized into five main classes: BU_Featurizer, Chemistry, Pore, Topology, and Host_Guest, totaling 158 features. The output is in CSV format. The `featurize_dataframe()` function incorporates the `ignore_errors = True / False` parameter to retain only successfully calculated results, ensuring smooth computation.

[0047] S4. Using the fingerprint from S2 and the feature subset from S3 as source data, a user-friendly GUI interface is created, providing tabs such as Custom (upload cif file), Download, and Data Upload for users to customize operations. Figure 3 The search interface provided for the database website.

[0048] To verify the rationality of the above workflow for calculating the full-domain information features of MOFs, the QMOF dataset was first randomly divided into two parts, QMOF-1 and QMOF-2, in a 1:1 ratio, and labeled with hydrogen storage performance using the mature m-CGCNN model. Then, feature subsets were obtained from the MOFs in the QMOF-1 part based on the above workflow, and feature subsets were obtained from the existing data in the database for QMOF-2. Subsequently, QMOF-1 and QMOF-2 were divided into training and testing sets in an 8:2 ratio, respectively, and R-values ​​were calculated using an ANN model with the same parameters. 2 The R of the QMOF-1 dataset 2 =0.92, which is better than the R score on the QMOF-2 dataset.2 =0.80, indicating that the computational workflow covering the full-domain information features of MOF materials provides richer MOF material feature information.

Claims

1. A method for expressing a unique fingerprint of a porous metal-organic framework material, characterized in that, The specific steps are as follows: S1: Based on the MOFs database, X-ray diffraction pattern calculations were performed on its crystallographic information files to initially obtain the three strongest XRD peaks as structural information fingerprints of MOFs. The MOFs database uses the 2019-ASR CoRE and h-MOFs database to verify the uniqueness of the MOFs proprietary fingerprint representation method, and the QMOF database to verify whether the feature subset calculated by S3 can effectively cover the global MOFs information. The X-ray diffraction pattern calculation process described above is based on the pymatgen.core.lattice module traversing all reciprocal lattice points within the confined sphere without requiring multiplicity correction. The calculation peak position range is set to 0°-90°, the step size is set to 0.02°, the peak position matching error of the 2019-ASR CoRE database is set to ±0.3°, and the peak position matching error of the h-MOFs database is set to ±0.2°. S2: Using the standard SMILES format, further obtain the chemical information fingerprints of MOF substructures, including SBU and ligands, and combine them with the structural information fingerprints obtained in S1 to form the proprietary fingerprint expression of MOF materials. The proprietary fingerprint representation of MOF materials uses the canonical SMILES format of SBU and Linker in MOF materials and the three strongest peaks of their X-ray diffraction patterns as identifiers for SBU, ligand, and topology, respectively, and sorts them in the order of SBU.Linker.XRD. Different components are separated by ".". The dependency libraries for SBU and ligand canonical SMILES are Open Babelcore and Pybel. S3: Based on the proprietary fingerprint of MOFs materials in S2, calculate the geometric structure and physicochemical characteristics of MOFs using Python algorithms, in preparation for the construction of various performance prediction machine learning models and related descriptor mining work. S4: Provide data access, export, and user feedback interfaces on the database website.

2. The method of claim 1, wherein the method is a method of porous metal-organic framework material exclusive fingerprint expression. When calculating the geometric structure and physicochemical features of MOFs based on MOFs proprietary fingerprints as described in S3, the Python algorithm library of mofdscribe, matminer and RDkit is used. The calculated feature categories are divided into BU_Featurizer, Chemistry, Pore, Topology and Host_Guest, a total of 5 categories and 158-dimensional feature sets.

3. The method of claim 1, wherein the method is a method of porous metal-organic framework material exclusive fingerprint expression. The data access and export process described in S4 involves setting up a Custom tab, allowing users to customize and upload CIF files, and providing channels for downloading data in JSON or CSV formats.

4. The proprietary fingerprint expression method for porous metal-organic framework materials according to claim 1, characterized in that, The user feedback interface described in S4 uses 3Dmol.js to visualize MOFs and their substructures, and allows users to submit their own experimental data in the Data Upload tab.

Images