A multimodal DNTF eutectic prediction method based on small sample learning

Abstract

The present invention discloses a multimodal DNTF eutectic prediction method based on small sample learning, and relates to the technical field of eutectic material prediction. The present invention uses multimodal data of related crystal compounds similar to DNTF to train a DNTF eutectic prediction model, and then combines the trained DNTF eutectic prediction model with experimental screening to achieve DNTF eutectic prediction, which solves the problem of poor performance of the trained DNTF eutectic prediction model due to the imbalance of positive and negative samples of DNTF eutectic data in the existing database, and by combining the DNTF eutectic prediction model with experimental screening, the loss of resources and time caused by human experiments can be reduced, and the screening efficiency of DNTF eutectic ligand molecules can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of eutectic material prediction, and specifically relates to a multimodal DNTF (3,4-dinitrofurazanyl furazan oxide) eutectic prediction method based on small sample learning. Background Art

[0002] Currently, CL-20 (hexanitrohexaazaisofuran) is the most widely used material in the field of high energy density, but its high cost restricts its widespread application. Recent studies have found that DNTF (3,4-dinitrofurazanyl furazan oxide) has similar performance to CL-20, but its cost is only half of CL-20. Therefore, DNTF has a high research value. The molecular structure of DNTF is shown in the figure below. Figure 1 As shown. After research, it was found that DNTF still has problems such as poor thermal stability, poor mechanical properties, crystallization, compatibility, etc. These problems can be solved by eutectic.

[0003] Eutectic technology is a new technology developed in recent years. Energetic eutectic technology can change the molecular structure composition (molecular arrangement) and crystal structure (crystal packing coefficient, density, etc.) of energetic materials without destroying the original molecular structure of energetic compounds, reduce the free volume of crystals, form a tight stack, and further reduce the probability of hot spot formation. As a result, physical properties such as density, melting point, sensitivity and explosion performance change, achieving a balance between energy and safety. However, due to its huge experimental cost and certain safety issues, in the past few decades, although eutectic materials have gradually shown their due role in their respective application fields, such as electronic devices, semiconductor engineering, chemical engineering, etc., it is still due to the fact that the eutectic system is too complex and the research on eutectic structure is relatively restricted, so it has not been widely used in the field of energetic materials.

[0004] At present, the field of artificial intelligence machine learning is developing rapidly. In the field of deep learning in artificial intelligence, as the performance of deep learning model prediction and classification is improved, deep learning methods have begun to be applied to the field of materials science. Compared with the traditional method of extracting simple underlying features through data mining, the method based on deep learning can learn a lot of high-level features that cannot be recognized by the naked eye from a large amount of data to represent information, which improves the prediction ability of complex data environments. At the same time, the deep learning method is more robust and adaptive through large-scale data training, and can adapt to different types of data forms, avoiding the problem of large differences in prediction effects of traditional methods due to the complexity of material data.

[0005] With the emergence of the new neural network structure Transformer, it has opened up a new path in the field of deep learning. Transformer can achieve good prediction results in both the natural language field and the computer vision field. However, in the research on deep learning methods for DNTF cocrystal ligand prediction, there are still several problems that need to be solved: First, the ratio of positive samples to negative samples in the existing database is unbalanced, and the different dimensions make the attention of the neural network require higher weights for adjusting the neural network during model training. Second, there is currently no research on the deep learning framework established for DNTF cocrystal ligand molecules, which has failed to play the role of deep learning, and the corresponding deep learning model must be re-established for the data model. Third, many current models do not reflect the iterative characteristics of the cocrystal research process when predicting. These problems have led to the inability of deep learning methods to effectively solve the screening problem of DNTF cocrystal ligand molecules. Summary of the invention

[0006] In order to solve the above problems existing in the prior art, the present invention provides a multimodal DNTF eutectic prediction method based on small sample learning. The technical problem to be solved by the present invention is achieved by the following technical solutions:

[0007] The present invention provides a multimodal DNTF eutectic prediction method based on small sample learning, comprising:

[0008] Step 1: According to the chemical parameters of DNTF, a crystalline compound that meets the preset DNTF related conditions is searched in the Cambridge Crystal Library as a related crystalline compound;

[0009] Step 2: constructing a data set based on the relevant crystalline compounds;

[0010] Step 3: using the data set to train the constructed multimodal DNTF eutectic prediction network based on a small sample, to obtain a trained multimodal DNTF eutectic prediction model based on a small sample;

[0011] Step 4: predicting the ligand molecules of DNTF using the multimodal DNTF co-crystal prediction model based on small samples, and retrieving the ligand molecules to be determined from the Cambridge Crystal Library according to the prediction results;

[0012] Step 5: Screen the ligand molecules to be determined through experiments to obtain the co-crystal ligand molecules of DNTF.

[0013] Compared with the prior art, the present invention has the following beneficial effects:

[0014] The multimodal DNTF co-crystal prediction method based on small sample learning of the present invention searches for crystal compounds that meet preset DNTF related conditions as related crystal compounds in the Cambridge crystal library according to the chemical parameters of DNTF; constructs a data set according to the related crystal compounds; uses the data set to train the constructed multimodal DNTF co-crystal prediction network based on small samples to obtain a trained multimodal DNTF co-crystal prediction model based on small samples; uses the multimodal DNTF co-crystal prediction model based on small samples to predict the ligand molecules of DNTF, and retrieves the ligand molecules to be determined from the Cambridge crystal library according to the prediction results; and screens the ligand molecules to be determined through experiments to obtain the co-crystal ligand molecules of DNTF. The present invention utilizes multimodal data of related crystalline compounds similar to DNTF to train a DNTF co-crystal prediction model, and then combines the trained DNTF co-crystal prediction model with experimental screening to realize DNTF co-crystal prediction, thereby solving the problem of poor performance of the trained DNTF co-crystal prediction model due to the imbalance of positive and negative samples of DNTF eutectic data in the existing database. Furthermore, by combining the DNTF eutectic prediction model with experimental screening, the loss of resources and time caused by human experiments can be reduced, thereby improving the screening efficiency of DNTF eutectic ligand molecules.

[0015] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following preferred embodiments are specifically cited and described in detail as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a planar structural diagram of a DNTF provided by an embodiment of the present invention;

[0017] Figure 2 It is a flowchart of a multimodal DNTF eutectic prediction method based on small sample learning provided by an embodiment of the present invention;

[0018] Figure 3 It is a block diagram of a multimodal DNTF eutectic prediction network based on a small sample provided by an embodiment of the present invention;

[0019] Figure 4 It is a flowchart of a multimodal DNTF eutectic prediction method based on small sample learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, a multimodal DNTF eutectic prediction method based on small sample learning proposed by the present invention is described in detail. Through the description of the specific implementation method, the technical means and effects adopted by the present invention to achieve the predetermined purpose can be more deeply and specifically understood, and it is not used to limit the technical solution of the present invention.

[0021] The embodiment of the present invention provides a multi-modal DNTF eutectic prediction method based on small sample learning, see Figure 2 , Figure 2 FIG. 1 is a planar structural diagram of a DNTF provided by an embodiment of the present invention. Figure 2 As shown, the multimodal DNTF eutectic prediction method based on small sample learning in this embodiment includes:

[0022] Step 1: According to the chemical parameters of DNTF, a crystalline compound satisfying the preset DNTF related conditions is searched in the Cambridge Crystal Library as a related crystalline compound.

[0023] No materials that can form substantial co-crystals with DNTF have been found in the current Cambridge Crystallographic Database (CCDC). Therefore, it is impossible to directly obtain DNTF co-crystal data through the existing database as training data for subsequent co-crystal prediction models. To this end, a scheme is proposed to train a co-crystal prediction model using related crystalline compounds similar to DNTF.

[0024] Optionally, the chemical parameters of DNTF include: molecular formula of DNTF, structure of DNTF, number and type of functional groups of DNTF. That is, in this embodiment, the relevant crystalline compounds similar to DNTF are searched in the Cambridge Crystal Library based on the three characteristics of molecular formula, structure, number and type of functional groups.

[0025] Optionally, the preset DNTF related conditions include: the crystalline compound and DNTF have the same molecular formula, the crystalline compound and DNTF have partially the same chemical structure, the crystalline compound and DNTF have the same number of functional groups, and the crystalline compound and DNTF have at least one of the same functional groups.

[0026] like Figure 1 As shown, the DNTF molecule has a five-membered furazan group, and two nitro groups are connected as substituents to the 3rd and 4th carbon atoms of the furazan ring, respectively. In this embodiment, the chemical structure of the crystalline compound and DNTF is partially the same, including: the crystalline compound has a furazan ring and a nitro group is connected to any position of the furazan ring, or the crystalline compound has a benzene ring and corresponding nitro groups are connected to the 3rd and 4th positions of the benzene ring.

[0027] The functional groups of the DNTF molecule include a furazan ring, a nitro group, a carbonyl group and a CN bond. In this embodiment, the crystalline compound and DNTF have partially the same functional groups, which includes: the crystalline compound has at least two functional groups that are the same as the functional groups of DNTF.

[0028] Step 2: Construct a dataset based on related crystal compounds.

[0029] In an optional embodiment, step 2 includes:

[0030] Step 2.1: Use the Cambridge Crystallographic Database to obtain the crystal information and ligand molecular information of relevant crystal compounds.

[0031] In this embodiment, the relevant information of the relevant crystal compound can be called in the Cambridge Crystal Library through the use of the CSD python API through the code. The Cambridge Crystal Library stores the chemical structure information of the crystal compound, such as the coordinates of the atoms, the element type of the atoms, the information of the chemical bonds, the three-dimensional structure of the molecule, the two-dimensional diagram of the molecule and the symmetry of the molecule; crystallographic information, such as unit cell parameters, space group information, crystallographic lattice type and crystallographic symmetry operation; chemical and physical properties, such as molecular weight, molar refractive index, molar volume, melting point, boiling point, density and refractive index; literature and experimental data, such as literature citation information, experimental methods (such as X-ray diffraction, neutron diffraction, etc.) and data accuracy; chemical and biological relationships, such as chemical reaction information, biological activity data and drug-target interactions; chemical nomenclature and identifiers, such as chemical names, chemical formulas, CSD registration numbers and identifiers of other external databases (such as CAS numbers, InChI, etc.).

[0032] In this embodiment, chemical structure information, crystallographic information and chemical-physical properties are mainly used as crystal information for co-crystal research, that is, the crystal information of the relevant crystalline compound includes chemical structure information, crystallographic information and chemical-physical properties.

[0033] In this embodiment, the ligand molecule information of the relevant crystalline compound includes the molecular formula of the ligand molecule and the functional group of the ligand molecule.

[0034] Step 2.2: Obtain multimodal data of the relevant crystalline compound according to the crystal information of the relevant crystalline compound, use the ligand molecule information as the true label of the multimodal data of the relevant crystalline compound, and obtain a multimodal data sample.

[0035] In eutectic studies, atomic coordinates are used to analyze the spatial arrangement and interactions between different components in the eutectic system. Information on chemical bonds helps to understand the chemical bonding between different components, which is crucial for the formation of eutectics. Crystallographic package parameters are used to determine the lattice matching and lattice spacing of the eutectic system, which has an important influence on the formation and stability of the eutectic. Space group information helps to understand the symmetry of the eutectic structure, which is helpful for predicting and explaining the physical properties of the eutectic. The crystallographic lattice type provides basic structural framework information of the eutectic system. The melting point is one of the characteristics of the eutectic system and can be used to confirm the formation of the eutectic. Density is related to the composition and structure of the eutectic and is an important parameter for studying the properties of the eutectic. The refractive index may be used to analyze the phase purity and optical properties of the eutectic system.

[0036] Based on this, in this embodiment, the multimodal data of the relevant crystalline compound obtained according to the crystal information of the relevant crystalline compound includes: atomic coordinate data, chemical bond data, crystal package parameters, space group data, density data and melting point data.

[0037] Step 2.3: Multimodal data samples of all relevant crystal compounds constitute the dataset.

[0038] Step 3: Use the data set to train the constructed small-sample-based multimodal DNTF eutectic prediction network to obtain a trained small-sample-based multimodal DNTF eutectic prediction model.

[0039] In this embodiment, a data set constructed using the above-mentioned related crystalline compounds is mapped into a representation of the feature space through a designed feature extractor, and then ligand molecule prediction is performed based on the extracted features through a designed prediction module. The network parameters are optimized based on the difference between the predicted ligand molecule information and the true label, thereby obtaining a co-crystal prediction model.

[0040] It should be noted that due to the limitation of the number of data samples in the data set and in order to improve the learning ability of the model, in this embodiment, a meta-learning approach is used to train the model so as to make efficient judgments on various data.

[0041] In an optional embodiment, step 3 includes:

[0042] The dataset is divided into a support set and a query set in a ratio of 4:1. A training task is created in the form of 2 way-5 shot. Multimodal data samples are selected from the support set and input into the multimodal DNTF eutectic prediction network based on small samples for training. Multimodal data samples are selected from the query set and input into the multimodal DNTF eutectic prediction network based on small samples for testing. The network parameters of the multimodal DNTF eutectic prediction network based on small samples are updated based on the loss gradient back propagation. After reaching the preset number of training times, a trained multimodal DNTF eutectic prediction model based on small samples is obtained.

[0043] See also Figure 3 , Figure 3 It is a block diagram of a multimodal DNTF eutectic prediction network based on a small sample provided by an embodiment of the present invention. In this embodiment, the multimodal DNTF eutectic prediction network based on a small sample includes a feature extractor, a feature fusion module and a prediction module. Among them, the feature extractor includes an atomic coordinate data encoder, a chemical bond data encoder, a bond angle data encoder, a space group data encoder, and a density and melting point data encoder.

[0044] In this embodiment, the feature extractor is used to extract features from the input multimodal data samples to obtain atomic coordinate data features, chemical bond data features, bond angle data features, space group data features, and density and melting point data features. That is, the atomic coordinate data is extracted by an atomic coordinate data encoder to obtain the corresponding atomic coordinate data features, the chemical bond data is extracted by a chemical bond data encoder to obtain the corresponding chemical bond data features, the bond angle data in three directions of the crystal package parameters is extracted by a bond angle data encoder to obtain the corresponding bond angle data features, the space group data is extracted by a space group data encoder to obtain the corresponding space group data features, and the density data and melting point data are extracted by a density and melting point data encoder to obtain the corresponding density and melting point data features.

[0045] In this embodiment, one-hot encoding is used for the bond types and linked atoms in the chemical bond data, and the bond length is normalized by multiplication. There are three dimensions of information for the chemical bond data mode. For the atomic coordinate data, as a three-dimensional modal data, it is necessary to normalize its data of each dimension through matrix multiplication. The bond angle data in three directions in its crystal package parameters is a three-dimensional modal data, similar to the atomic coordinate data, and also needs to be normalized through processing. The space group data is uniformly one-hot encoded and normalized. Since it has only two dimensions of data, it is expanded in dimension. Finally, the density data and melting point data are expanded from one-dimensional data to three-dimensional data and normalized.

[0046] In this embodiment, the feature fusion module is used to fuse the atomic coordinate data features, chemical bond data features, bond angle data features, space group data features, and density and melting point data features to obtain multimodal features. The prediction module is used to predict the predicted ligand molecule information of the relevant crystalline compound based on the multimodal features.

[0047] Optionally, the prediction module can be composed of multiple linear layers, the main purpose of which is to take over the feature dimensions extracted by the feature extractor, and then predict the ligand molecule through feature mapping to obtain the predicted ligand molecule information. By comparing the predicted ligand molecule information with the real label of the multimodal data sample, it can be determined whether the predicted ligand molecule can achieve co-crystal with the corresponding related crystal compound, and then obtain the co-crystal prediction result, that is, whether it is co-crystal.

[0048] It should be noted that, in this embodiment, the predicted ligand molecule information output by the prediction module can be represented in a coded form. The predicted ligand molecule information is consistent with the ligand molecule information of the real label, including molecular formula and functional group. The molecular formula can be encoded by direct encoding, and the functional group can be encoded by MACCS keys (MolecularACCess System keys). The encoding principle of MACCS keys is described below.

[0049] MACCS bonds are a set of predefined substructure features that are used to encode specific chemical properties in molecules. Each MACCS bond corresponds to a specific chemical environment or functional group and can be used to quickly identify whether a specific structural feature is present in a molecule. The following are the basic steps of MACCS bond encoding:

[0050] Step (1), define the MACCS key. The MACCS key set contains 166 predefined substructure features, each of which is assigned a unique number (usually from 1 to 166).

[0051] Step (2), molecular parsing, uses cheminformatics tools (such as RDKit, OpenBabel, etc.) to parse the molecular structure (usually a SMILES string or SDF file) into an internal representation.

[0052] Step (3), substructure matching, for each substructure defined by the MACCS bond, check whether there is a match in the molecule, which is usually done by a substructure search algorithm.

[0053] Step (4) generates a fingerprint vector. For each MACCS bond, if there is a matching substructure in the molecule, a binary bit (usually 1) is set at the corresponding position, otherwise it remains 0. Then a binary vector with a length of 166 is obtained, which is called a MACCS fingerprint.

[0054] For example, the acetone (C3H6O) molecule is used as an example to illustrate how to encode a molecule using MACCS bonds. The molecular structure of acetone has a carbonyl group (-C=O) in the center and a methyl group (-CH3) on each side of the carbonyl group.

[0055] First, assume that MACCS bond 1 corresponds to a "carbonyl group (C=O)", bond 2 corresponds to a "halogen atom", bond 3 corresponds to an "aromatic ring", and so on.

[0056] Then, cheminformatics tools are used to resolve the molecular structure into an internal representation.

[0057] Next, perform substructure matching: Check if there is a carbonyl group (C=O) in the molecule: Yes, there is one. Check if there are halogen atoms in the molecule: No. Check if there are aromatic rings in the molecule: No.

[0058] Finally, generate a fingerprint vector: Based on the above matching results, the following MACCS fingerprint vector [1, 0, 0,..., 0] can be generated, a vector of length 166, where the first position is 1 indicating the presence of a carbonyl group.

[0059] Step 4: Use the multimodal DNTF co-crystal prediction model based on small samples to predict the ligand molecules of DNTF, and retrieve the ligand molecules to be determined from the Cambridge Crystal Library based on the prediction results.

[0060] In an optional embodiment, step 4 includes:

[0061] Step 4.1: Use the Cambridge Crystal Library to obtain multimodal data of DNTF.

[0062] In this embodiment, the multimodal data of DNTF includes: atomic coordinate data, chemical bond data, crystal package parameters, space group data, density data and melting point data of DNTF.

[0063] Step 4.2: Input the multimodal data of DNTF into the multimodal DNTF co-crystal prediction model based on small samples to obtain the predicted ligand molecular information of DNTF.

[0064] In this embodiment, the predicted ligand molecule information of DNTF includes: the molecular formula and functional group of the predicted ligand molecule of DNTF.

[0065] Step 4.3: According to the predicted ligand molecule information of DNTF, the corresponding crystal compound is retrieved in the Cambridge Crystal Library as the ligand molecule to be determined.

[0066] It is understandable that due to the complex structure of DNTF, the representation length of the MACCS key needs to be shortened. The representation length of MACCS can be determined based on the structural types of all the ligand molecules of related crystalline compounds that are similar to DNTF, so that the general structure of the predicted ligand molecule of DNTF can be determined based on this code. Then, based on the structure of the predicted ligand molecule, the corresponding crystalline compound is retrieved in the Cambridge Crystal Library as the ligand molecule to be determined. The CCDC database search is mainly based on the functional group results of the predicted ligand molecule.

[0067] Step 5: Screen the ligand molecules to be determined through experiments to obtain the co-crystal ligand molecules of DNTF.

[0068] In an optional embodiment, step 5 includes:

[0069] Step 5.1: Use simulation software to conduct a simulated co-crystal experiment on DNTF and the ligand molecule to be determined, and screen the ligand molecule that can achieve co-crystal with DNTF from the ligand molecule to be determined based on the simulation experiment results.

[0070] Optionally, software simulation can use Material studio software. This software can create and analyze molecular structures, simulate molecular dynamics and phase diagram calculations, etc., which can well utilize the results of deep learning for screening.

[0071] Step 5.2: Conduct a physical co-crystal experiment with the screened ligand molecules and DNTF, and obtain the co-crystal ligand molecules of DNTF based on the results of the physical experiment.

[0072] See also Figure 4 As shown in the flowchart of the multimodal DNTF co-crystal prediction method based on small sample learning, it can be understood that since the ligand molecule of a related crystalline compound has more than one functional group, the process of training the multimodal DNTF co-crystal prediction model based on small samples can be iterated more than once, and due to the problem of the amount of multimodal data samples in the training set, batch training is more beneficial to the subsequent experimental screening part. By combining continuous machine learning prediction with experimental verification, the gap between the co-crystal prediction results of the present invention and the objective experimental results can be corrected.

[0073] The multimodal DNTF co-crystal prediction method based on small sample learning of the present invention utilizes multimodal data of related crystal compounds similar to DNTF to train the DNTF co-crystal prediction model, and then combines the trained DNTF eutectic prediction model with experimental screening to realize the co-crystal prediction of DNTF, thereby solving the problem of poor performance of the trained DNTF eutectic prediction model due to the imbalance of positive and negative samples of DNTF eutectic data in the existing database. In addition, by combining the DNTF eutectic prediction model with experimental screening, the loss of resources and time caused by human experiments can be reduced, thereby improving the screening efficiency of DNTF eutectic ligand molecules.

[0074] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the term "include", "comprise" or any other variant is intended to cover non-exclusive inclusion, so that the article or device including a series of elements includes not only those elements, but also other elements that are not explicitly listed. In the absence of more restrictions, the elements defined by the sentence "including one..." do not exclude the existence of other identical elements in the article or device including the elements. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. The orientation or position relationship indicated by "up", "down", "left", "right", etc. is based on the orientation or position relationship shown in the accompanying drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention.

[0075] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A multimodal DNTF eutectic prediction method based on small sample learning, characterized in that: include: Step 1: According to the chemical parameters of DNTF, a crystalline compound that meets the preset DNTF related conditions is searched in the Cambridge Crystal Library as a related crystalline compound; The chemical parameters of the DNTF include: the molecular formula of DNTF, the structure of DNTF, the number and type of functional groups of DNTF; the preset DNTF-related conditions include: the crystalline compound and DNTF have the same molecular formula, the crystalline compound and DNTF have partially the same chemical structure, the crystalline compound and DNTF have the same number of functional groups, and the crystalline compound and DNTF have at least one of the same functional groups; the crystalline compound and DNTF have partially the same chemical structure including: the crystalline compound has a furazan ring and a nitro group is connected to any position of the furazan ring or the crystalline compound has a benzene ring and corresponding nitro groups are connected to the 3rd and 4th positions of the benzene ring; the crystalline compound and DNTF have partially the same functional groups including: the crystalline compound has at least two functional groups that are the same as the functional groups of DNTF; Step 2: constructing a data set based on the relevant crystalline compounds; Step 3: using the data set to train the constructed multimodal DNTF eutectic prediction network based on a small sample, to obtain a trained multimodal DNTF eutectic prediction model based on a small sample; Step 4: predicting the ligand molecules of DNTF using the multimodal DNTF co-crystal prediction model based on small samples, and retrieving the ligand molecules to be determined from the Cambridge Crystal Library according to the prediction results; Step 5: Screening the ligand molecules to be determined through experiments to obtain the co-crystal ligand molecules of DNTF; The step 2 comprises: Step 2.1: using the Cambridge Crystal Library to obtain crystal information and ligand molecule information of the relevant crystal compound; the crystal information includes chemical structure information, crystallographic information and chemical and physical properties; the ligand molecule information includes the molecular formula of the ligand molecule and the functional group of the ligand molecule; Step 2.2: obtaining multimodal data of the relevant crystalline compound according to the crystal information of the relevant crystalline compound, using the ligand molecule information as a true label of the multimodal data of the relevant crystalline compound, and obtaining a multimodal data sample; the multimodal data includes atomic coordinate data, chemical bond data, crystal package parameters, space group data, density data and melting point data; Step 2.3: The multimodal data samples of all relevant crystal compounds constitute the data set; The step 3 comprises: The data set is divided into a support set and a query set in a ratio of 4:1, and a training task is created in the form of 2 way-5 shot. Multimodal data samples are selected from the support set and input into the multimodal DNTF eutectic prediction network based on small samples for training. Multimodal data samples are selected from the query set and input into the multimodal DNTF eutectic prediction network based on small samples for testing. The network parameters of the multimodal DNTF eutectic prediction network based on small samples are updated based on loss gradient back propagation. After reaching a preset number of training times, a trained multimodal DNTF eutectic prediction model based on small samples is obtained. Among them, the multimodal DNTF eutectic prediction network based on small samples includes a feature extractor, a feature fusion module and a prediction module, the feature extractor includes an atomic coordinate data encoder, a chemical bond data encoder, a bond angle data encoder, a space group data encoder and a density and melting point data encoder; the feature extractor is used to extract features from the input multimodal data sample to obtain atomic coordinate data features, chemical bond data features, bond angle data features, space group data features and density and melting point data features; the feature fusion module is used to fuse the atomic coordinate data features, the chemical bond data features, the bond angle data features, the space group data features and the density and melting point data features to obtain multimodal features; the prediction module is used to predict the predicted ligand molecule information of the related crystalline compound based on the multimodal features.

2. The multimodal DNTF eutectic prediction method based on small sample learning according to claim 1, characterized in that: The step 4 comprises: Step 4.1: Obtain multimodal data of DNTF using the Cambridge Crystal Library; Step 4.2: inputting the multimodal data of the DNTF into the multimodal DNTF co-crystal prediction model based on small samples to obtain the predicted ligand molecule information of the DNTF; Step 4.3: According to the predicted ligand molecule information of the DNTF, the corresponding crystal compound is retrieved in the Cambridge Crystal Library as the ligand molecule to be determined.

3. The multimodal DNTF eutectic prediction method based on small sample learning according to claim 1, characterized in that: The step 5 comprises: Step 5.1: using simulation software to perform a simulated co-crystal experiment on the DNTF and the ligand molecule to be determined, and screening the ligand molecule that can form a co-crystal with the DNTF from the ligand molecule to be determined according to the simulation experiment results; Step 5.2: Conduct a physical co-crystal experiment on the screened ligand molecules and the DNTF, and obtain the co-crystal ligand molecules of the DNTF according to the results of the physical experiment.

Images