Intelligent prediction method and system for performance of tissue regeneration material, equipment and medium

Abstract

The invention provides an intelligent prediction method and system for performance of a tissue regeneration material, and the method comprises the steps: efficiently and accurately extracting qualitative and quantitative data of material design parameters and performance from a large number of literatures through an intelligent data extraction tool and a fine-tuning optimized large language model according to a structured data label template designed by an expert; a structured data set in a JSON format is established, and the problem that data fragmentation and qualitative description are dominated in a traditional method is solved; establishing a multi-dimensional feature vector comprising chemical composition, physicochemical characteristics, structural information and a synthesis method and a paired sample set reflecting relative advantages and disadvantages of material performance, wherein the paired sample set is used for training a twin network of a radial basis function neural network based on support feature weight learning; the constructed radial basis function neural network model has the advantage of accurately predicting the performance of the tissue regeneration material, the automatically learned feature weight value can quantitatively reflect the influence of different features on the performance of the material, the research and development efficiency of the tissue regeneration material is improved, and the trial and error cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of biomaterial research and development, and in particular to an intelligent prediction method, system, equipment, and medium for the performance of tissue regeneration materials. Background Art

[0002] Tissue regeneration materials are a type of biomaterial used to promote and guide the repair and regeneration of damaged tissues in vivo and are widely used in the field of regenerative medicine. These materials typically require good biocompatibility, degradability, and mechanical properties.

[0003] The development of traditional tissue regeneration materials (such as silicate bioceramics) relies primarily on trial-and-error experiments, requiring extensive testing to screen material components and ratios. A single synthesis-testing cycle can take weeks, resulting in high R&D costs, long cycles, and low success rates. Artificial intelligence-based performance prediction methods can learn from large-scale data to predict performance based on information such as material composition, ratio, structure, and synthesis method, thereby reducing trial-and-error experiments and improving R&D efficiency.

[0004] At present, relevant methods have made great progress in alloys, catalysts, and battery materials. However, in the field of tissue regeneration materials, the lack of large-scale public data has hindered the research and application of intelligent performance prediction methods. Although a large number of relevant scientific literatures contain rich data, relying solely on manual extraction is not only time-consuming and labor-intensive, but also highly dependent on expert experience and difficult to form a structured data. In addition, the literature on tissue regeneration materials lists the specific chemical formula, crystal structure and other information of the materials, but for biological effect-related properties, such as the key biological information such as the effects of bioceramic materials on cell proliferation, differentiation, and tissue regeneration and repair, the performance descriptions in many cases are qualitative descriptions (such as "Material A has significantly better osteogenic ability than Material B"), and lack directly comparable quantitative values. These problems further increase the difficulty of achieving intelligent and accurate prediction of tissue regeneration material performance.

[0005] Therefore, there is an urgent need for an intelligent prediction method for the performance of tissue regeneration materials to quickly screen out excellent performance schemes from a large number of material design schemes, reduce the number of experimental trial and error, and thus reduce R&D costs. Summary of the Invention

[0006] In order to achieve the above-mentioned purpose and other advantages of the present invention, the first purpose of the present invention is to provide an intelligent prediction method for the performance of tissue regeneration materials, comprising the following steps:

[0007] Obtain and preprocess academic literature in the field of tissue regeneration materials;

[0008] Design a structured data label template based on the characteristics of tissue regeneration material experimental data;

[0009] Manually annotate the literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template. Each sample includes the original literature and its structured data label, forming a literature annotation dataset.

[0010] Select a large language model and perform supervised learning fine-tuning based on the document annotation dataset;

[0011] Input the document to be extracted into the fine-tuned large language model, extract the structured information that conforms to the structured data label template, and output it;

[0012] Optimizing the extracted structured data, combining manually annotated literature data with literature data automatically annotated by the large language model, and establishing a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance;

[0013] Based on the structured document annotation dataset, a multi-dimensional feature vector is generated. Based on the performance parameters and test conditions in different documents, paired samples that can clearly distinguish the relative performance of the two are screened out with experimental evidence, and a paired sample set for the target performance is constructed.

[0014] Constructing a twin network and training the twin network using the paired sample set;

[0015] After the twin network training is completed, the radial basis function neural network model is used to predict the material properties, and the influence of different features on the material properties is quantitatively evaluated based on the feature weight values ​​automatically learned by the radial basis function neural network.

[0016] Furthermore, the pre-processing step:

[0017] Use intelligent data extraction tools to convert text from PDF documents into Markdown documents.

[0018] Furthermore, the structured data tag template is configured as a structured data tag template in JSON format, which at least includes the name of the tissue regeneration material, chemical molecular formula, synthesis method, structural characteristics, performance parameters and test conditions, application scenarios and original sentence source fields.

[0019] Furthermore, the steps of manually annotating the literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template, each sample including the original literature and its structured data label, to form the literature annotation dataset include:

[0020] If the document does not contain certain field information in the structured data tag template, the result of the corresponding field is filled in with null;

[0021] The relative performance of different tissue regeneration materials is annotated, including all pairwise combinations of materials that have experimental evidence in the literature.

[0022] Furthermore, the step of selecting a large language model and performing supervised learning fine-tuning based on the document annotation dataset includes:

[0023] Automatic extraction of structured data of documents is achieved through a phased large language model agent chain, which specifically includes classification agents, recognition agents and extraction agents;

[0024] The classification agent classifies sentences in the Markdown document into the types defined in the structured data tags based on the fine-tuned large language model. If the sentence does not belong to any type defined in the structured data tags, it is classified as "other".

[0025] The recognition agent is used to automatically identify specific parameters and other qualitative descriptions in the classified literature sentences;

[0026] The extraction agent converts the numerical values ​​and units in the text into structured data according to the structured data tag template in JSON format, and converts the qualitative description into the relative advantages and disadvantages of the two sets of material properties.

[0027] Furthermore, the step of inputting the document to be extracted into the fine-tuned large language model, extracting structured information that conforms to the structured data tag template, and outputting the structured information includes:

[0028] The Markdown-formatted document to be extracted is input into the fine-tuned large language model. Based on the prompt engineering, the large language model is guided through the classification agent, recognition agent and extraction agent to extract structured information that conforms to the data label template and output it in JSON format.

[0029] Furthermore, the steps of optimizing the extracted structured data and combining the manually annotated literature data with the literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance include:

[0030] Verify the extracted JSON structured data, correct incorrect extraction results, and select the corrected structured data and its corresponding documents as example samples. Use a small number of sample prompts to guide the large language model to automatically extract more accurate structured information.

[0031] A small amount of annotated literature data is combined with a large amount of literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance.

[0032] Furthermore, the step of generating a multi-dimensional feature vector based on the structured document annotation dataset includes:

[0033] Based on the structured document annotation dataset, the structural features and the category text in the synthesis method are encoded using one-hot encoding, and the resulting feature vector is recorded as x t ; Quantitative parameters are used as numerical features, and the resulting feature vector is recorded as x v ; Determine the chemical elements that may be included according to the type of tissue regeneration material; perform atom bag encoding on the chemical molecular formulas in the literature, and construct a fixed-length feature vector by counting the number of atoms of each element, where the length of the feature vector is equal to the number of chemical elements included, and the resulting feature vector x b ;

[0034] Use physical descriptors to characterize the physical and chemical properties of different atoms in the chemical composition of materials, including descriptors that reflect the basic properties of elements in the periodic table, descriptors that quantify the characteristics of atomic collections, and descriptors that characterize the van der Waals interactions of atoms;

[0035] The physical descriptor value of each atom is obtained from the Xenonpy library. The physical descriptor value of the entire chemical formula is equal to the weighted average of the physical descriptor vector of each atom according to its quantity ratio in the molecular formula. The resulting characteristic vector is recorded as x p ;

[0036] The multidimensional feature vector that quantitatively reflects the chemical composition, physicochemical properties, structural information and synthesis method of tissue regeneration materials is expressed as X = [x b ,x p ,x t ,x v ].

[0037] Furthermore, based on the performance parameters and test conditions in different literatures, the step of screening out paired samples that can clearly distinguish the relative performance advantages and disadvantages supported by experimental evidence and constructing a paired sample set for target performance also includes:

[0038] The form of paired samples between different materials in the same document is {X A ,X B ,X A >X B}, X A represents the eigenvector corresponding to material A, X B represents the eigenvector corresponding to material B, X A >X B This indicates that the osteogenic ability of material A is better than that of material B;

[0039] The form of paired samples between different materials from different literature is {X A ,XB ,X A <X B}, X A <X B This indicates that the osteogenic ability of material A is inferior to that of material B.

[0040] Furthermore, the twin network includes two radial basis function neural networks that share parameters and support feature weight learning. The input-output mapping relationship corresponding to the radial basis function neural network that supports feature weight learning is:

[0041]

[0042] in, is the parameter to be optimized for the network, K is the number of hidden layer neurons, ω k represents the connection weight between the kth hidden layer neuron and the output neuron, ω0 represents the bias weight, ω=[ω0,ω1,,ω K ] represents the connection weight vector of K hidden layer neurons and output neurons, represents the center of the kth hidden layer neuron, σ k is the width of the kth hidden layer neuron, I represents the dimension of the feature vector corresponding to the material, υ=[υ1,,υ I ] is the feature weight vector, υ i Represents the weight of the i-th dimension feature.

[0043] Furthermore, the step of training the twin network using the paired sample set includes:

[0044] Using the paired sample set, {X A ,X B ,X A >X B} form of sample pairs as positive sample pairs, label y = 1; {X A ,X B ,X A <X B} form as negative sample pairs, label y = -1; use the marginal ranking loss function combined with the L2 regularization term:

[0045]

[0046] Among them, f W (X A ) and f W (X B ) represent the performance scores of the model for material A and material B respectively, margin is a hyperparameter that controls the minimum score difference between positive and negative sample pairs; λ is the regularization coefficient;

[0047] The Siamese network is trained using the Adam optimizer and the marginal ranking loss function combined with an L2 regularization term.

[0048] During the training process, balanced sampling is used to ensure a balanced ratio of positive and negative samples, and a curriculum learning strategy is adopted. The network is first trained with simple samples, and difficult samples are gradually introduced and the margin value is reduced during the training process to refine the model's discrimination ability.

[0049] Furthermore, the step of predicting material properties using a radial basis function neural network model includes:

[0050] For a group of materials, the performance score of each material is predicted based on the trained radial basis function neural network, and the performance scores of this group of materials are ranked. The top-ranked materials are selected for synthesis and testing, thereby improving the efficiency of material research and development and reducing trial and error costs.

[0051] The second object of the present invention is to provide an intelligent prediction system for the performance of tissue regeneration materials, which applies the above method and includes a literature collection and preprocessing module, a literature structured data label definition module, a small amount of annotated data set construction module, a large language model fine-tuning module, a literature structured information automatic extraction module, an extraction result optimization and literature annotated data set construction module, a multi-dimensional feature vector and paired sample set construction module, and a tissue regeneration material performance prediction model training and application module based on a twin network; wherein,

[0052] The literature collection and preprocessing module is used to obtain academic literature in the field of tissue regeneration materials and perform preprocessing;

[0053] The structured data label definition module of the literature is used to design a structured data label template based on the characteristics of tissue regeneration material experimental data;

[0054] The small amount of annotated data set construction module is used to manually annotate literature in the field of tissue regeneration materials and generate training samples that conform to the structured data label template. Each sample includes the original document and its structured data label to form a literature annotated data set;

[0055] The large language model fine-tuning module is used to select a large language model and perform supervised learning fine-tuning based on the document annotation dataset;

[0056] The document structured information automatic extraction module is used to input the document to be extracted into the fine-tuned large language model, extract the structured information that conforms to the structured data label template, and output it;

[0057] The extraction result optimization and literature annotation dataset construction module is used to optimize the extracted structured data, combine the manually annotated literature data with the literature data automatically annotated by the large language model, and establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance;

[0058] The multi-dimensional feature vector and paired sample set construction module is used to generate a multi-dimensional feature vector based on the structured document annotation dataset, and based on the performance parameters and test conditions in different documents, screen out paired samples that can clearly distinguish the relative performance of the two with experimental evidence, and construct a paired sample set for the target performance;

[0059] The twin network-based tissue regeneration material performance prediction model training and application module is used to construct a twin network and train the twin network using the paired sample set; after the twin network training is completed, a radial basis function neural network model is used to predict material properties, and based on the feature weight values ​​automatically learned by the radial basis function neural network, the influence of different features on material properties is quantitatively evaluated.

[0060] A third object of the present invention is to provide a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above method when executing the computer program.

[0061] A fourth object of the present invention is to provide a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the above method when executed by a processor.

[0062] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0063] The present invention provides an intelligent prediction method and system for the performance of tissue regeneration materials. By using intelligent data extraction tools and a fine-tuned and optimized large language model, and based on an expert-designed structured data labeling template, the system efficiently and accurately extracts qualitative and quantitative data on material design parameters and performance from a large amount of literature, establishes a structured data set in JSON format, and solves the problems of data fragmentation and qualitative description in traditional methods. The present invention establishes a multi-dimensional feature vector including chemical composition, physicochemical properties, structural information, and synthesis method, and a paired sample set reflecting the relative merits of material performance, which is used to train a twin network based on a radial basis function neural network that supports feature weight learning. The radial basis function neural network model constructed by the present invention has the advantage of accurately predicting the performance of tissue regeneration materials, and the automatically learned feature weight values ​​can quantitatively reflect the impact of different features on material performance, thereby improving the efficiency of tissue regeneration material research and development and reducing trial and error costs.

[0064] 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 and to implement it according to the contents of the description, the following preferred embodiments of the present invention are described in detail with reference to the accompanying drawings. The specific implementation methods of the present invention are given in detail by the following embodiments and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0066] Figure 1 A flow chart of the intelligent prediction method for tissue regeneration material performance;

[0067] Figure 2 Schematic diagram of the twin network structure constructed for the performance prediction model of tissue regeneration materials;

[0068] Figure 3 Schematic diagram of the radial basis function neural network structure that supports feature weight learning;

[0069] Figure 4 Schematic diagram of the intelligent prediction system for tissue regeneration material performance;

[0070] Figure 5 A schematic diagram of a computer device;

[0071] Figure 6 A schematic diagram of a computer-readable storage medium. DETAILED DESCRIPTION

[0072] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. It should be noted that, without conflict, the embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0073] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0074] The figure numbers in this application are only used to distinguish the various steps in the scheme and are not used to limit the execution order of the various steps. The specific execution order is subject to the description in the specification.

[0075] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0076] Example 1

[0077] An intelligent prediction method for tissue regeneration material performance, such as Figure 1 As shown, the following steps are included:

[0078] S100, obtain academic literature in the field of tissue regeneration materials and perform preprocessing;

[0079] In some embodiments, publicly published academic literature in the field of tissue regeneration materials is obtained from databases such as PubMed, Web of Science, and CNKI. Further, the pretreatment step includes:

[0080] Use the intelligent data extraction tool MinerU to efficiently convert the text in PDF format documents into Markdown format documents that are easy to analyze.

[0081] S110. Design a structured data label template based on the characteristics of tissue regeneration material experimental data;

[0082] In some embodiments, the structured data tag template is configured as a JSON-formatted structured data tag template, including at least the following fields: tissue regeneration material name, chemical formula, synthesis method, structural characteristics, performance parameters and test conditions, application scenarios, and original text source. Taking silicate bioceramic materials as an example, the synthesis method must include information such as raw materials, concentration, preparation method, reaction temperature, reaction time, and ceramic components and content in silicate ceramic composite materials; structural characteristics must include information such as macrostructure, micromorphology, size, and porosity; performance parameters must include mechanical properties, degradation rate, biocompatibility, bioactivity, and the relative advantages and disadvantages of different tissue regeneration materials for each performance parameter. Mechanical properties include elastic modulus, stress-strain, yield stress, and shear-thinning properties; biocompatibility includes cell viability and cell migration levels; and bioactivity includes expression levels of osteogenic differentiation-related genes, osteogenic differentiation-related proteins, neural cell-directed differentiation-related genes, neural-specific protein expression levels, as well as the percentage of new bone formation in animals and expression of bone-related protein markers. The test conditions must include information such as instruments and equipment, sample preparation, test methods and parameters.

[0083] S120, manually annotating literature in the field of tissue regeneration materials to generate training samples that conform to a structured data label template, where each sample includes the original literature and its structured data label, forming a literature annotation dataset;

[0084] In this embodiment, domain experts manually annotate a small amount of literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template. Each sample includes the original document and its structured data label in JSON format, forming a document annotation dataset.

[0085] Furthermore, the steps of manually annotating the literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template, each sample including the original literature and its structured data label, to form the literature annotation dataset include:

[0086] If a document does not contain certain fields in the structured data label template, the corresponding fields should be marked as null. The relative performance of different tissue regeneration materials should include all pairwise combinations of materials supported by experimental evidence in the literature, such as "Material A has better osteogenic ability than Material B," "Material A has better osteogenic ability than Material C," and "Material B has better osteogenic ability than Material C."

[0087] S130. Select a large language model (LLM), such as DeepSeek-70B or Llama3.3-70B, and perform supervised learning fine-tuning based on the document annotation dataset; for example, use LoRa technology to implement lightweight fine-tuning of the large language model.

[0088] Furthermore, the step of selecting a large language model and performing supervised learning fine-tuning based on the document annotation dataset includes:

[0089] Automatic extraction of structured data of documents is achieved through a phased large language model agent chain, which specifically includes classification agents, recognition agents and extraction agents;

[0090] The classification agent classifies sentences in Markdown format documents into types defined in structured data tags such as "synthesis method", "structural features" and "performance parameters" based on the fine-tuned large language model. If it does not belong to any type defined in the structured data tags, it is classified as "other".

[0091] The recognition agent was used to automatically identify specific parameters in the classified literature statements, such as elastic modulus 3.6 kPa, cell viability >90%, and other qualitative descriptions, such as BSP protein expression in the GelMA-2CS group was significantly higher than that in the GelMA group.

[0092] The extraction agent converts the numerical values ​​and units in the text into structured data based on the structured data label template in JSON format, and converts the qualitative description into the relative advantages and disadvantages of the two groups of material properties, such as the osteogenic ability of GelMA-2CS is better than that of GelMA.

[0093] S140: Input the document to be extracted into the fine-tuned large language model, extract the structured information that conforms to the structured data label template, and output it;

[0094] Furthermore, the step of inputting the document to be extracted into the fine-tuned large language model, extracting structured information that conforms to the structured data tag template, and outputting the structured information includes:

[0095] The Markdown-formatted document to be extracted is input into the fine-tuned large language model. Based on the prompt engineering, the large language model is guided through the classification agent, recognition agent and extraction agent to extract structured information that conforms to the data label template and output it in JSON format.

[0096] S150, optimizing the extracted structured data, combining manually annotated literature data with literature data automatically annotated by the large language model, and establishing a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance;

[0097] Furthermore, the steps of optimizing the extracted structured data and combining the manually annotated literature data with the literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance include:

[0098] The extracted JSON-formatted structured data was manually verified, and incorrect extraction results were corrected. The corrected structured data and its corresponding literature were selected as sample examples. Using a small number of sample prompts, the LLM was guided to automatically extract more accurate structured information. By combining a small amount of manually annotated literature data with a large amount of literature data automatically annotated by the LLM, a structured literature annotation dataset for intelligent prediction of tissue regeneration material properties was established.

[0099] S160: Generate a multi-dimensional feature vector based on the structured document annotation dataset, and screen out paired samples that can clearly distinguish the relative performance of the two based on experimental evidence based on performance parameters and test conditions in different documents, and construct a paired sample set for target performance.

[0100] Furthermore, the step of generating a multi-dimensional feature vector based on the structured document annotation dataset includes:

[0101] Based on the structured document annotation dataset, the structural features and the category text in the synthesis method are one-hot encoded, and the resulting feature vector is recorded as x t ; Quantitative parameters are used as numerical features, and the resulting feature vector is recorded as x v ; The chemical elements that may be included are determined according to the type of tissue regeneration material. For example, the design elements of silicate bioceramic materials include silicon (Si), oxygen (O), calcium (Ca), hydrogen (H), sodium (Na), magnesium (Mg), titanium (Ti), manganese (Mn), aluminum (Al), iron (Fe), molybdenum (Mo), zinc (Zn), strontium (Sr), copper (Cu), europium (Eu), lithium (Li), cobalt (Co), phosphorus (P), etc. Arranged in ascending order of atomic number, the list of chemical elements considered is [H, Li, O, Na, Mg, Al, Si, P, Ca, Ti, Mn, Fe, Co, Cu, Zn, Sr, Mo, Eu]. The chemical formulas in the literature are encoded by atom bags, and a fixed-length feature vector is constructed by counting the number of atoms of each element. The length of the feature vector is equal to the number of chemical elements included. The resulting feature vector is denoted as x. b With the chemical formula Ca6Si6O 17 Taking (OH)2 as an example, which contains 6 Ca atoms, 6 Si atoms, 19 O atoms and 2 H atoms, the value of the dimension corresponding to Ca in the corresponding eigenvector is 6 / 33, the value of the dimension corresponding to Si is 6 / 33, the value of the dimension corresponding to O is 19 / 33, the value of the dimension corresponding to H is 2 / 33, and the value of the dimension corresponding to other elements not included is 0.

[0102] Furthermore, physical descriptors are used to characterize the physical and chemical properties of different atoms in the chemical composition of the material, including descriptors that reflect the basic properties of elements in the periodic table, such as atomic number and principal quantum number; descriptors that quantify the characteristics of atomic collections, such as the average radius and atomic volume of atoms; descriptors that characterize the van der Waals interactions of atoms, such as the effective radius of atoms under the action of van der Waals forces and the van der Waals radius parameter based on the molecular force field. The values ​​of the physical descriptors of each atom are obtained from the Xenonpy library. The physical descriptor value of the entire chemical formula is equal to the weighted average of the physical descriptor vectors of each atom according to the proportion of its number in the molecular formula. The resulting characteristic vector is recorded as x p Finally, the multidimensional feature vector that quantitatively reflects the chemical composition, physicochemical properties, structural information, and synthesis methods of tissue regeneration materials is expressed as X = [x b ,x p ,x t ,x v ].

[0103] In order to obtain more paired samples reflecting the relative performance of materials, in addition to the paired samples that are limited to the material performance comparison within a single document based on the structured document annotation dataset, we further screened out paired samples that can clearly distinguish the relative performance based on experimental evidence based on the performance parameters and test conditions in different documents, and constructed a paired sample set for the target performance (such as osteogenic ability). Among them, the form of paired samples between different materials in the same document is {X A ,X B ,X A >X B}, X A >X B Indicates that the osteogenic ability of material A is better than that of material B; the paired samples of different materials from different literatures are in the form of {X A ,X B ,X A <X B}, X A <X B Indicates that the bone formation ability of material A is inferior to that of material B. X A represents the eigenvector corresponding to material A, X B Represents the eigenvector corresponding to material B.

[0104] S170: Construct a twin network, and train the twin network using the paired sample set;

[0105] Further, if Figure 2 As shown, the twin network includes two radial basis function neural networks that support feature weight learning and share parameters. W represents the parameters of the radial basis function neural network that supports feature weight learning. The structure of the radial basis function neural network that supports feature weight learning is shown in FIG. Figure 3 As shown, the corresponding input-output mapping relationship is:

[0106]

[0107] in, is the parameter to be optimized for the network, K is the number of hidden layer neurons, ω k represents the connection weight between the kth hidden layer neuron and the output neuron, ω0 represents the bias weight, ω=[ω0,ω1,,ω K ] represents the connection weight vector of K hidden layer neurons and output neurons, represents the center of the kth hidden layer neuron, σ k is the width of the kth hidden layer neuron, I represents the dimension of the feature vector corresponding to the material, υ=[υ1,...,υ I ] is the feature weight vector, υ i Represents the weight of the i-th dimension feature.

[0108] Furthermore, the step of training the twin network using the paired sample set includes:

[0109] Using the paired sample set based on large-scale literature data established above, {X A ,X B ,X A >X B} form of sample pairs as positive sample pairs, label y = 1; {X A ,X B ,X A <X B} form as negative sample pairs, label y = -1; use the marginal ranking loss function combined with the L2 regularization term:

[0110]

[0111] Among them, f W (X A ) and f W (X B ) represent the performance scores of the model for material A and material B respectively. Margin is a hyperparameter that controls the minimum score difference between positive and negative sample pairs, and its initial value is set to 1.0. λ is the regularization coefficient, which is generally set to 0.01.

[0112] The Siamese network is trained using the Adam optimizer and the marginal ranking loss function combined with an L2 regularization term.

[0113] During the training process, balanced sampling is used to ensure a balanced ratio of positive and negative samples, and a curriculum learning strategy is adopted. The network is first trained with simple sample pairs (sample pairs with large performance differences). During the training process, difficult samples (sample pairs with similar performance) are gradually introduced and the margin value is reduced to refine the model's discrimination ability.

[0114] S180. After the twin network training is completed, the radial basis function neural network model is used to predict the material properties, and the influence of different features on the material properties is quantitatively evaluated based on the feature weight values ​​automatically learned by the radial basis function neural network.

[0115] Furthermore, the step of predicting material properties using a radial basis function neural network model includes:

[0116] When applying the model, input the characteristic vectors X1 and X2 of the two materials to be compared. If f W (X1)>f W(X2), then material 1 has better predicted performance; otherwise, material 2 has better predicted performance. For a group of materials, the performance score of each material is predicted based on the trained radial basis function neural network model. The performance scores of this group of materials are then ranked, and the top-ranked materials are selected for synthesis and testing, thereby improving material R&D efficiency and reducing trial-and-error costs.

[0117] Example 2

[0118] An intelligent prediction system for the performance of tissue regeneration materials, using the above method, for a detailed description of the method, please refer to the corresponding description in the embodiment of the intelligent prediction method for the performance of tissue regeneration materials, which will not be repeated here. Figure 4 As shown, the system 200 includes a document collection and preprocessing module 210, a document structured data label definition module 220, a small amount of annotated data set construction module 230, a large language model fine-tuning module 240, a document structured information automatic extraction module 250, an extraction result optimization and document annotated data set construction module 260, a multi-dimensional feature vector and paired sample set construction module 270, and a twin network-based tissue regeneration material performance prediction model training and application module 280; wherein,

[0119] The literature collection and preprocessing module is used to obtain academic literature in the field of tissue regeneration materials and perform preprocessing;

[0120] The structured data label definition module of the literature is used to design a structured data label template based on the characteristics of tissue regeneration material experimental data;

[0121] The small amount of annotated data set construction module is used to manually annotate literature in the field of tissue regeneration materials and generate training samples that conform to the structured data label template. Each sample includes the original document and its structured data label to form a literature annotated data set;

[0122] The large language model fine-tuning module is used to select a large language model and perform supervised learning fine-tuning based on the document annotation dataset;

[0123] The document structured information automatic extraction module is used to input the document to be extracted into the fine-tuned large language model, extract the structured information that conforms to the structured data label template, and output it;

[0124] The extraction result optimization and literature annotation dataset construction module is used to optimize the extracted structured data, combine the manually annotated literature data with the literature data automatically annotated by the large language model, and establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance;

[0125] The multi-dimensional feature vector and paired sample set construction module is used to generate a multi-dimensional feature vector based on the structured document annotation dataset, and based on the performance parameters and test conditions in different documents, screen out paired samples that can clearly distinguish the relative performance of the two with experimental evidence, and construct a paired sample set for the target performance;

[0126] The twin network-based tissue regeneration material performance prediction model training and application module is used to construct a twin network and train the twin network using the paired sample set; after the twin network training is completed, a radial basis function neural network model is used to predict material properties, and based on the feature weight values ​​automatically learned by the radial basis function neural network, the influence of different features on material properties is quantitatively evaluated.

[0127] Based on the technical solution of the above embodiment, optionally, the preprocessing step includes:

[0128] Use intelligent data extraction tools to convert text from PDF documents into Markdown documents.

[0129] Based on the technical solution of the above embodiment, optionally, the structured data label template is configured as a structured data label template in JSON format, including at least the name of the tissue regeneration material, chemical molecular formula, synthesis method, structural characteristics, performance parameters and test conditions, application scenarios and original sentence source fields.

[0130] Based on the technical solution of the above embodiment, optionally, the steps of manually annotating the literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template, each sample including the original literature and its structured data label, to form the literature annotation dataset include:

[0131] If the document does not contain certain field information in the structured data tag template, the result of the corresponding field is filled in with null;

[0132] The relative performance of different tissue regeneration materials is annotated, including all pairwise combinations of materials that have experimental evidence in the literature.

[0133] Based on the technical solution of the above embodiment, optionally, the step of selecting a large language model and performing supervised learning fine-tuning based on the document annotation dataset includes:

[0134] Automatic extraction of structured data of documents is achieved through a phased large language model agent chain, which specifically includes classification agents, recognition agents and extraction agents;

[0135] The classification agent classifies sentences in the Markdown document into the types defined in the structured data tags based on the fine-tuned large language model. If the sentence does not belong to any type defined in the structured data tags, it is classified as "other".

[0136] The recognition agent is used to automatically identify specific parameters and other qualitative descriptions in the classified literature sentences;

[0137] The extraction agent converts the numerical values ​​and units in the text into structured data according to the structured data tag template in JSON format, and converts the qualitative description into the relative advantages and disadvantages of the two sets of material properties.

[0138] Based on the technical solution of the above embodiment, optionally, the step of inputting the document to be extracted into the fine-tuned large language model, extracting structured information that conforms to the structured data label template, and outputting the structured information includes:

[0139] The Markdown-formatted document to be extracted is input into the fine-tuned large language model. Based on the prompt engineering, the large language model is guided through the classification agent, recognition agent and extraction agent to extract structured information that conforms to the data label template and output it in JSON format.

[0140] Based on the technical solution of the above embodiment, optionally, the step of optimizing the extracted structured data and combining the manually annotated literature data with the literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material properties includes:

[0141] Verify the extracted JSON structured data, correct incorrect extraction results, and select the corrected structured data and its corresponding documents as example samples. Use a small number of sample prompts to guide the large language model to automatically extract more accurate structured information.

[0142] A small amount of annotated literature data is combined with a large amount of literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance.

[0143] Based on the technical solution of the above embodiment, optionally, the step of generating a multi-dimensional feature vector based on the structured document annotation dataset includes:

[0144] Based on the structured document annotation dataset, the structural features and the category text in the synthesis method are encoded using one-hot encoding, and the resulting feature vector is recorded as x t ; Quantitative parameters are used as numerical features, and the resulting feature vector is recorded as x v; Determine the chemical elements that may be included according to the type of tissue regeneration material; perform atom bag encoding on the chemical molecular formulas in the literature, and construct a fixed-length feature vector by counting the number of atoms of each element, where the length of the feature vector is equal to the number of chemical elements included, and the resulting feature vector x b ;

[0145] Use physical descriptors to characterize the physical and chemical properties of different atoms in the chemical composition of materials, including descriptors that reflect the basic properties of elements in the periodic table, descriptors that quantify the characteristics of atomic collections, and descriptors that characterize the van der Waals interactions of atoms;

[0146] The physical descriptor value of each atom is obtained from the Xenonpy library. The physical descriptor value of the entire chemical formula is equal to the weighted average of the physical descriptor vector of each atom according to its quantity ratio in the molecular formula. The resulting characteristic vector is recorded as x p ;

[0147] The multidimensional feature vector that quantitatively reflects the chemical composition, physicochemical properties, structural information and synthesis method of tissue regeneration materials is expressed as X = [x b ,x p ,x t ,x v ].

[0148] Based on the technical solution of the above embodiment, optionally, the step of screening out paired samples that can clearly distinguish the relative performance advantages and disadvantages supported by experimental evidence based on performance parameters and test conditions in different literatures, and constructing a paired sample set for target performance further includes:

[0149] The form of paired samples between different materials in the same document is {X A ,X B ,X A >X B}, X A represents the eigenvector corresponding to material A, X B represents the eigenvector corresponding to material B, X A >X B This indicates that the osteogenic ability of material A is better than that of material B;

[0150] The form of paired samples between different materials from different literature is {X A ,X B ,X A <X B}, X A <X B This indicates that the osteogenic ability of material A is inferior to that of material B.

[0151] Based on the technical solution of the above embodiment, optionally, the twin network includes two radial basis function neural networks that support feature weight learning and share parameters, and the input-output mapping relationship corresponding to the radial basis function neural network that supports feature weight learning is:

[0152]

[0153] in, is the parameter to be optimized for the network, K is the number of hidden layer neurons, ω k represents the connection weight between the kth hidden layer neuron and the output neuron, ω0 represents the bias weight, ω=[ω0,ω1,,ω K ] represents the connection weight vector of K hidden layer neurons and output neurons, represents the center of the kth hidden layer neuron, σ k is the width of the kth hidden layer neuron, I represents the dimension of the feature vector corresponding to the material, υ=[υ1,...,υ I ] is the feature weight vector, υ i Represents the weight of the i-th dimension feature.

[0154] Based on the technical solution of the above embodiment, optionally, the step of training the twin network using the paired sample set includes:

[0155] Using the paired sample set, {X A ,X B ,X A >X B} form of sample pairs as positive sample pairs, label y = 1; {X A ,X B ,X A <X B} form as negative sample pairs, label y = -1; use the marginal ranking loss function combined with the L2 regularization term:

[0156]

[0157] Among them, f W (X A ) and f W (X B ) represent the performance scores of the model for material A and material B respectively, margin is a hyperparameter that controls the minimum score difference between positive and negative sample pairs; λ is the regularization coefficient;

[0158] The Siamese network is trained using the Adam optimizer and the marginal ranking loss function combined with an L2 regularization term.

[0159] During the training process, balanced sampling is used to ensure a balanced ratio of positive and negative samples, and a curriculum learning strategy is adopted. The network is first trained with simple samples, and difficult samples are gradually introduced and the margin value is reduced during the training process to refine the model's discrimination ability.

[0160] Based on the technical solution of the above embodiment, optionally, the step of predicting material properties using a radial basis function neural network model includes:

[0161] For a group of materials, the performance score of each material is predicted based on the trained radial basis function neural network, and the performance scores of this group of materials are ranked. The top-ranked materials are selected for synthesis and testing, thereby improving the efficiency of material research and development and reducing trial and error costs.

[0162] Example 3

[0163] A computer device 300, such as Figure 5 As shown, the present invention includes a memory 310, a processor 320, and a computer program 330 stored in the memory and executable on the processor. When the processor executes the computer program, the steps of a method for intelligently predicting the properties of tissue regeneration materials are implemented. For a detailed description of the method, please refer to the corresponding description in the above method embodiment and will not be repeated here.

[0164] Example 4

[0165] A computer-readable storage medium such as Figure 6 As shown, a computer program is stored thereon, and when the computer program is executed by the processor, the steps of a method for intelligently predicting the performance of tissue regeneration materials are implemented. For a detailed description of the method, reference can be made to the corresponding description in the above method embodiment, and no further details will be given here.

[0166] The number of devices and processing scales described herein are intended to simplify the description of the present invention. Applications, modifications, and variations of the present invention will be readily apparent to those skilled in the art.

[0167] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

[0168] The apparatus, computer device, non-volatile computer storage medium, and method provided in the embodiments of this specification correspond to each other. Therefore, the apparatus, computer device, and non-volatile computer storage medium also have similar beneficial technical effects as the corresponding method. Since the beneficial technical effects of the method have been described in detail above, the beneficial technical effects of the corresponding apparatus, computer device, and non-volatile computer storage medium will not be repeated here.

[0169] Those skilled in the art will also appreciate that, in addition to implementing the controller in pure computer-readable program code, it is entirely possible to implement the same functionality by programming the method steps logically, such as through logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered structures within the hardware component. Alternatively, the devices for implementing various functions can be considered both software units implementing the method and structures within the hardware component.

[0170] The systems, devices, or units described in the above embodiments can be implemented by computer chips or physical devices, or by products with certain functions. For ease of description, the above devices are described separately by function, with each unit described separately. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware components.

[0171] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] This specification is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0173] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 The steps for the function specified in one or more boxes.

[0175] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0176] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program units. Generally, program units include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program units may be located in local and remote computer storage media, including storage devices.

[0177] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0178] The foregoing is merely an example of the present invention and is not intended to limit the present invention to one or more embodiments. It will be apparent to those skilled in the art that various modifications and variations may be made to the present invention to one or more embodiments. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention to one or more embodiments shall be included within the scope of the claims of the present invention to one or more embodiments.

Claims

1. An intelligent prediction method for tissue regeneration material performance, characterized in that: The following steps are involved: Obtain and preprocess academic literature in the field of tissue regeneration materials; Design a structured data label template based on the characteristics of tissue regeneration material experimental data; Manually annotate the literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template. Each sample includes the original literature and its structured data label, forming a literature annotation dataset. Select a large language model and perform supervised learning fine-tuning based on the document annotation dataset; Input the document to be extracted into the fine-tuned large language model, extract the structured information that conforms to the structured data label template, and output it; Optimizing the extracted structured data, combining manually annotated literature data with literature data automatically annotated by the large language model, and establishing a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance; Based on the structured document annotation dataset, a multi-dimensional feature vector is generated. Based on the performance parameters and test conditions in different documents, paired samples that can clearly distinguish the relative performance of the two are screened out with experimental evidence, and a paired sample set for the target performance is constructed. Constructing a twin network and training the twin network using the paired sample set; After the twin network training is completed, the radial basis function neural network model is used to predict the material properties, and the influence of different features on the material properties is quantitatively evaluated based on the feature weight values ​​automatically learned by the radial basis function neural network.

2. The intelligent prediction method for tissue regeneration material performance according to claim 1, characterized in that: The pre-processing step comprises: Use intelligent data extraction tools to convert text from PDF documents into Markdown documents.

3. The method for intelligently predicting the performance of tissue regeneration materials according to claim 2, wherein: The structured data tag template is configured as a structured data tag template in JSON format, and at least includes the name of the tissue regeneration material, chemical molecular formula, synthesis method, structural characteristics, performance parameters and test conditions, application scenarios and original statement source fields.

4. The method for intelligently predicting the performance of tissue regeneration materials according to claim 1, wherein: Manually annotating the literature in the field of tissue regeneration materials to generate training samples that conform to the structured data label template, each sample includes the original literature and its structured data label, and the steps of forming the literature annotation dataset include: If the document does not contain certain field information in the structured data tag template, the result of the corresponding field will be filled in with null; The relative performance of different tissue regeneration materials is annotated, including all pairwise combinations of materials that have experimental evidence in the literature.

5. The intelligent prediction method for tissue regeneration material performance according to claim 3, characterized in that: The step of selecting a large language model and performing supervised learning fine-tuning based on the document annotation dataset includes: Automatic extraction of structured data of documents is achieved through a phased large language model agent chain, which specifically includes classification agents, recognition agents and extraction agents; The classification agent classifies sentences in the Markdown document into the types defined in the structured data tags based on the fine-tuned large language model. If the sentence does not belong to any type defined in the structured data tags, it is classified as "other". The recognition agent is used to automatically identify specific parameters and other qualitative descriptions in the classified literature sentences; The extraction agent converts the numerical values ​​and units in the text into structured data according to the structured data tag template in JSON format, and converts the qualitative description into the relative advantages and disadvantages of the two sets of material properties.

6. The intelligent prediction method for tissue regeneration material performance according to claim 5, characterized in that: The step of inputting the document to be extracted into the fine-tuned large language model, extracting the structured information that conforms to the structured data label template, and outputting the structured information includes: The Markdown-formatted document to be extracted is input into the fine-tuned large language model. Based on the prompt engineering, the large language model is guided through the classification agent, recognition agent and extraction agent to extract structured information that conforms to the data label template and output it in JSON format.

7. The intelligent prediction method for tissue regeneration material performance according to claim 6, characterized in that: The steps of optimizing the extracted structured data and combining the manually annotated literature data with the literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance include: Verify the extracted JSON structured data, correct incorrect extraction results, and select the corrected structured data and its corresponding documents as example samples. Use a small number of sample prompts to guide the large language model to automatically extract more accurate structured information. A small amount of annotated literature data is combined with a large amount of literature data automatically annotated by the large language model to establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance.

8. The method for intelligently predicting the performance of tissue regeneration materials according to claim 3, wherein: The step of generating a multi-dimensional feature vector based on the structured document annotation dataset includes: Based on the structured document annotation dataset, the structural features and the category text in the synthesis method are encoded using one-hot encoding, and the resulting feature vector is recorded as x t ; Quantitative parameters are used as numerical features, and the resulting feature vector is recorded as x v ; Determine the chemical elements that may be included according to the type of tissue regeneration material; perform atom bag encoding on the chemical molecular formulas in the literature, and construct a fixed-length feature vector by counting the number of atoms of each element, where the length of the feature vector is equal to the number of chemical elements included, and the resulting feature vector x b ; Use physical descriptors to characterize the physical and chemical properties of different atoms in the chemical composition of materials, including descriptors that reflect the basic properties of elements in the periodic table, descriptors that quantify the characteristics of atomic collections, and descriptors that characterize the van der Waals interactions of atoms; The physical descriptor value of each atom is obtained from the Xenonpy library. The physical descriptor value of the entire chemical formula is equal to the weighted average of the physical descriptor vector of each atom according to its quantity ratio in the molecular formula. The resulting characteristic vector is recorded as x p ; The multidimensional feature vector that quantitatively reflects the chemical composition, physicochemical properties, structural information and synthesis method of tissue regeneration materials is expressed as X = [x b ,x p ,x t ,x v ].

9. The method for intelligently predicting the performance of tissue regeneration materials according to claim 8, wherein: The steps of screening out paired samples that can clearly distinguish the relative performance of the target based on the performance parameters and test conditions in different literatures and supported by experimental evidence, and constructing a paired sample set for the target performance also include: The form of paired samples between different materials in the same document is {X A ,X B ,X A >X B }, X A represents the eigenvector corresponding to material A, X B represents the eigenvector corresponding to material B, X A >X B This indicates that the osteogenic ability of material A is better than that of material B; The form of paired samples between different materials from different literature is {X A ,X B ,X A <X B }, X A <X B This indicates that the osteogenic ability of material A is inferior to that of material B.

10. The method for intelligently predicting the performance of tissue regeneration materials according to claim 9, characterized in that: The twin network includes two radial basis function neural networks that share parameters and support feature weight learning. The input-output mapping relationship corresponding to the radial basis function neural network that supports feature weight learning is: in, is the network parameter to be optimized, K is the number of hidden layer neurons, ω k represents the connection weight between the kth hidden layer neuron and the output neuron, ω0 represents the bias weight, ω=[ω0,ω1,…,ω K ] represents the connection weight vector of K hidden layer neurons and output neurons, represents the center of the kth hidden layer neuron, σ k is the width of the kth hidden layer neuron, I represents the dimension of the feature vector corresponding to the material, υ=[υ1,...,υ I ] is the feature weight vector, υ i Represents the weight of the i-th dimension feature.

11. The method for intelligently predicting the performance of tissue regeneration materials according to claim 10, wherein: The step of training the twin network using the paired sample set includes: Using the paired sample set, {X A ,X B ,X A >X B } form of sample pairs as positive sample pairs, label y = 1; {X A ,X B ,X A <X B } form as negative sample pairs, label y = -1; use the marginal ranking loss function combined with the L2 regularization term: Among them, f W (X A ) and f W (X B ) represent the performance scores of the model for material A and material B respectively, margin is a hyperparameter that controls the minimum score difference between positive and negative sample pairs; λ is the regularization coefficient; The Siamese network is trained using the Adam optimizer and the marginal ranking loss function combined with an L2 regularization term. During the training process, balanced sampling is used to ensure a balanced ratio of positive and negative samples, and a curriculum learning strategy is adopted. The network is first trained with simple samples, and difficult samples are gradually introduced and the margin value is reduced during the training process to refine the model's discrimination ability.

12. The intelligent prediction method for tissue regeneration material performance according to claim 11, characterized in that: The step of using the radial basis function neural network model to predict material properties includes: For a group of materials, the performance score of each material is predicted based on the trained radial basis function neural network, and the performance scores of this group of materials are ranked. The top-ranked materials are selected for synthesis and testing, thereby improving the efficiency of material research and development and reducing trial and error costs.

13. An intelligent prediction system for tissue regeneration material properties, applying the method according to any one of claims 1 to 12, characterized in that: It includes a literature collection and preprocessing module, a literature structured data label definition module, a small amount of annotated data set construction module, a large language model fine-tuning module, a literature structured information automatic extraction module, an extraction result optimization and literature annotation data set construction module, a multi-dimensional feature vector and paired sample set construction module, and a twin network-based tissue regeneration material performance prediction model training and application module; among them, The literature collection and preprocessing module is used to obtain academic literature in the field of tissue regeneration materials and perform preprocessing; The structured data label definition module of the literature is used to design a structured data label template based on the characteristics of tissue regeneration material experimental data; The small amount of annotated data set construction module is used to manually annotate literature in the field of tissue regeneration materials and generate training samples that conform to the structured data label template. Each sample includes the original document and its structured data label to form a literature annotated data set; The large language model fine-tuning module is used to select a large language model and perform supervised learning fine-tuning based on the document annotation dataset; The document structured information automatic extraction module is used to input the document to be extracted into the fine-tuned large language model, extract the structured information that conforms to the structured data label template, and output it; The extraction result optimization and literature annotation dataset construction module is used to optimize the extracted structured data, combine the manually annotated literature data with the literature data automatically annotated by the large language model, and establish a structured literature annotation dataset for intelligent prediction of tissue regeneration material performance; The multi-dimensional feature vector and paired sample set construction module is used to generate a multi-dimensional feature vector based on the structured document annotation dataset, and based on the performance parameters and test conditions in different documents, screen out paired samples that can clearly distinguish the relative performance of the two with experimental evidence, and construct a paired sample set for the target performance; The twin network-based tissue regeneration material performance prediction model training and application module is used to construct a twin network and train the twin network using the paired sample set; after the twin network training is completed, a radial basis function neural network model is used to predict material properties, and based on the feature weight values ​​automatically learned by the radial basis function neural network, the influence of different features on material properties is quantitatively evaluated.

14. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 12 are implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 12 are implemented.

Images