Multi-modal deep learning traditional Chinese medicine formula screening method, device, equipment, medium and product

Abstract

The invention discloses a multi-mode deep learning traditional Chinese medicine formula screening method, device, equipment, medium and product, and relates to the field of traditional Chinese medicine artificial intelligence. Firstly, on the basis of single-cell RNA sequencing data of a disease, through differential expression analysis, protein interaction network construction and cell communication analysis, a traditional Chinese medicine formula is screened; constructing a multi-target collaborative regulation network covering a disease core mechanism; secondly, fusing clinical experience, traditional Chinese medicine component structure information and a target protein sequence, predicting affinity of high-throughput traditional Chinese medicine small molecules and a target network by utilizing a trained and optimized multi-modal fusion model, and automatically screening an optimal traditional Chinese medicine formula which is high in synergy and covers a key target based on traditional Chinese medicine-component-target network topology parameters; and finally, performing multi-dimensional efficacy and mechanism verification on the candidate formula through an animal model and a cell experiment to form a dry-wet closed loop of calculation prediction-experimental verification-target point backtracking. According to the method, the accuracy, efficiency and mechanism interpretability of traditional Chinese medicine formula discovery are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence in traditional Chinese medicine, and in particular to a method, apparatus, equipment, medium and product for screening traditional Chinese medicine prescriptions using multimodal deep learning. Background Technology

[0002] Traditional Chinese medicine (TCM) compound formulation development relies heavily on physician experience, leading to problems such as empirical medication, strong subjectivity, unclear mechanisms, lengthy development cycles, and poor reproducibility. Existing computational-aided methods are mostly limited to single chemical components or simple target matching, failing to fully integrate TCM clinical experience, disease phenotypic characteristics (such as cellular heterogeneity revealed by single-cell sequencing), and systematic information on disease target networks. For example: The data dimensions are limited: it only uses chemical components or simple targets, and lacks a systematic analysis of the complex network regulatory mechanisms between diseases and drugs at the levels of clinical human experience with traditional Chinese medicine, disease phenotypic characteristics (such as cellular heterogeneity revealed by single-cell sequencing), and disease target characteristics (differential gene expression).

[0003] Insufficient model accuracy: The AI ​​model used is simple and cannot accurately represent the complex network of interactions between multiple components of traditional Chinese medicine and multiple targets of diseases. The prediction results cannot reflect the characteristics of multiple components and multiple targets in traditional Chinese medicine prescriptions. Furthermore, it cannot achieve high-throughput (up to millions of data points) data processing, which may result in data omissions.

[0004] Lack of experimental closed loop: Most of them are pure computer simulations, which fail to form a complete R&D closed loop of "dry and wet" combination with in vitro and in vivo efficacy experiments, and their predicted results cannot be effectively confirmed.

[0005] Therefore, there is an urgent need for a systematic solution that can integrate multimodal data, be driven by advanced AI, and be rigorously verified by experiments, in order to achieve the precision, standardization, and clarification of mechanisms in traditional Chinese medicine formulation. Summary of the Invention

[0006] The purpose of this application is to provide a method, device, equipment, medium and product for screening traditional Chinese medicine prescriptions using multimodal deep learning, which is a systematic solution that integrates multimodal data, is driven by advanced AI and has been rigorously verified by experiments, and achieves the precision, standardization and mechanism clarification of traditional Chinese medicine prescriptions.

[0007] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a method for screening traditional Chinese medicine prescriptions based on multimodal deep learning, including: Construct a multi-target synergistic regulatory network covering the core mechanisms of disease; Disease target proteins are obtained based on the multi-target synergistic regulatory network; the disease target proteins include multiple key targets; Construct a priori knowledge base; the priori knowledge base includes: a TCM clinical human experience base, a TCM component structure base, and a molecular-protein affinity knowledge base; A multimodal fusion model, EMS-T5, was constructed; the multimodal fusion model includes: the molecular language model MolTo5 and the protein language model EMS-65; The deep features of the target protein were extracted using the protein language model ESM-65. Using the aforementioned Chinese herbal medicine component structure library, the SMILES structural features of small molecules of Chinese herbal medicine were extracted by combining the molecular language model MolT5 with the RDKit language recognition tool. The deep features and SMILES structural features are input into a trained and optimized multimodal fusion model to predict the affinity values ​​of small molecules of traditional Chinese medicine to each key target. The set of high-affinity molecules is determined based on the affinity values; Based on the set of high-affinity molecules, a complex network of "traditional Chinese medicine-component-target" is constructed, and the optimal traditional Chinese medicine formula that covers the key targets and has strong synergistic effects is screened according to the network topology parameters.

[0008] Optionally, the method for screening traditional Chinese medicine prescriptions based on multimodal deep learning further includes the following after the last step: The selected optimal traditional Chinese medicine formula was then subjected to in vitro and in vivo efficacy verification.

[0009] Optionally, the in vitro and in vivo efficacy verification of the selected optimal traditional Chinese medicine formula specifically includes: Serum FSH and E2 levels, ovarian index, and follicle count were measured in a cyclophosphamide-induced rat model of ovarian aging. Cell proliferation activity, γH2AX, p-FOXO3a, MKi67 and SASP factor expression were detected in a D-galactose-induced KGN ovarian granulosa cell senescence model. The effect of the optimal traditional Chinese medicine formula on the expression level of the key target was detected to verify the accuracy of the calculation and prediction, forming a closed-loop feedback from virtual screening to experimental verification.

[0010] Optionally, the construction of a multi-target synergistic regulatory network covering the core mechanisms of the disease specifically includes: Obtain single-cell RNA sequencing data for specific diseases; Based on the RNA sequencing data, data quality control, cell clustering, and differential expression analysis were performed to screen differentially expressed genes that met the criteria of |log2FC|>1 and a corrected P-value <0.05. Functional scoring was performed on the differentially expressed genes using an aging-related database to obtain the scoring results. A protein interaction network was constructed based on the scoring results; Based on the protein interaction network, key ligand-receptor pairs are identified through cell communication analysis, thereby constructing a multi-target synergistic regulatory network covering the core mechanism of the disease.

[0011] Optionally, the construction of the multimodal fusion model EMS-T5 specifically includes the following steps: Obtain results from multiple model combinations; these multiple model combinations include: ProtT5+MolT5, ProtT5+Bert, ProtT5+Mol-former, ESM-65+MolT5, ESM-65+Bert, and ESM-65+Mol-former. Obtain known molecular-protein interaction data from the molecular-protein affinity knowledge base; The known molecular-protein interaction data are input into the results of multiple model combinations for training, prediction, and validation. The combination of high-throughput models with the best prediction performance was selected as the multimodal fusion model EMS-T5.

[0012] Optionally, the construction of the traditional Chinese medicine component structure library specifically includes the following steps: High-frequency TCM components were obtained from the TCMSP and HERB databases, and drug-like molecules were screened according to the Lipinski five rules and the Veber rule. The Lipinski five rules include: molecular weight < 500, number of hydrogen bond donors < 5, number of hydrogen bond acceptors < 10, lipid-water partition coefficient < 5, and number of rotatable bonds ≤ 10. The Veber rule includes: polar surface area ≤ 140 Å. 2 ; A small molecule database, namely a traditional Chinese medicine component structure library, is constructed based on the screened medicinal molecules.

[0013] Secondly, this application provides a traditional Chinese medicine prescription screening device based on multimodal deep learning, comprising: A multi-target synergistic regulation network construction module is used to construct a multi-target synergistic regulation network covering the core mechanisms of diseases; The target protein acquisition module is used to acquire disease target proteins based on the multi-target synergistic regulatory network; the disease target proteins include multiple key targets. A prior knowledge base construction module is used to construct a prior knowledge base; the prior knowledge base includes: a TCM clinical human experience base, a TCM component structure base, and a molecular-protein affinity knowledge base; The EMS-T5 multimodal fusion model building module is used to construct the EMS-T5 multimodal fusion model; the multimodal fusion model includes: the protein language model EMS-65, the molecular language model MolTo5, and the RDKit language recognition tool; The deep feature extraction module is used to extract the deep features of the target protein using the protein language model ESM-65. The SMILES structural feature extraction module is used to extract the SMILES structural features of small molecules of traditional Chinese medicine by using the traditional Chinese medicine component structure library and combining the molecular language model MolT5 with the RDKit language recognition tool. The affinity prediction module is used to input the deep features and SMILES structural features into a trained and optimized multimodal fusion model to predict the affinity values ​​between small molecules of traditional Chinese medicine and each key target. A high-affinity molecule set determination module is used to determine a high-affinity molecule set based on the affinity value; The optimal traditional Chinese medicine (TCM) formulation determination module is used to construct a complex "TCM-component-target" network based on the set of high-affinity molecules, and to screen the optimal TCM formulation that covers the key targets and has strong synergistic effects according to the network topology parameters.

[0014] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multimodal deep learning-based traditional Chinese medicine formulation screening method described in any one of the above.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multimodal deep learning-based traditional Chinese medicine prescription screening method described above.

[0016] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the multimodal deep learning-based traditional Chinese medicine prescription screening method described above.

[0017] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a method, apparatus, equipment, medium, and product for screening traditional Chinese medicine prescriptions based on multimodal deep learning, which has the following beneficial effects: A leap in precision: By defining disease target networks through multi-omics (especially single-cell omics) data and using multimodal AI models for precise matching, the formulation discovery process is more closely integrated with clinical diseases and has more evidence to rely on, greatly improving the accuracy and success rate of candidate Chinese medicine formulations.

[0018] Doubled R&D efficiency: This application shortens the traditionally years-long exploratory research to a few months, achieving high-throughput candidate formulation screening and prioritization, and significantly improving R&D efficiency.

[0019] The mechanism is clearly explained: This application outputs not only the candidate Chinese medicine formulation, but also its potential "component-target-pathway" network of action, providing a molecular-level scientific explanation for the mechanism of action of Chinese medicine formulations.

[0020] Reliable efficacy verification: By integrating in vitro and in vivo efficacy verification modules, it is ensured that the results of AI calculations and predictions are firmly supported by experimental data, forming a reliable closed loop from "virtual screening" to "experimental confirmation." Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of a method for screening traditional Chinese medicine prescriptions based on multimodal deep learning in one embodiment of this application; Figure 2 A flowchart illustrating a method for screening traditional Chinese medicine prescriptions based on multimodal deep learning, provided in one embodiment of this application; Figure 3 This is a schematic diagram showing the comparison results of various models provided in an embodiment of this application; Figure 4 A schematic diagram illustrating the in vivo animal efficacy verification results of screening traditional Chinese medicine formulations according to an embodiment of this application; Figure 5 A schematic diagram illustrating the in vitro cell efficacy verification results of screening traditional Chinese medicine formulations according to an embodiment of this application; Figure 6 This is a schematic diagram illustrating the verification results of screening traditional Chinese medicine formulations to predict multiple core targets (CYP19A1, SERPINE1, MDK-NCL) provided in an embodiment of this application. Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0023] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0025] In one exemplary embodiment, such as Figure 1 and Figure 2 As shown, a method for screening traditional Chinese medicine prescriptions based on multimodal deep learning is provided. This method is executed by a computer device, specifically by a terminal or server alone, or by both a terminal and a server. In this embodiment, it includes the following steps 101 to 110. Wherein: Step 101: Construct a multi-target synergistic regulatory network covering the core mechanisms of the disease.

[0026] In this embodiment, ovarian aging (PCOA) is used as an example. First, public data of single-cell RNA sequencing (scRNA-seq) for specific diseases such as ovarian aging are obtained (this public data can come from the GEO database or other databases). Then, based on the RNA sequencing data, automatic data quality control, cell population (e.g., obtaining 8 ovarian cell subpopulations) and differential expression analysis were performed (screening criteria: |log2FC|>1, adj. P<0.05). We constructed protein-protein interaction (PPI) networks and performed pathway enrichment analysis on differentially expressed genes. Functional scoring was then performed using aging-related databases to identify several key targets, including CYP19A1, SERPINE1, NK / T cells, M cell communication ligand-receptor pairs, and related genes. Further cell communication analysis identified key ligand-receptor pairs, and a multi-target synergistic regulatory network covering the core mechanisms of the disease was constructed using Cytoscape.

[0027] More specifically, a protein-protein interaction network (PPI network) refers to a complex network structure formed within a cell or organism where proteins interact through physical contacts (such as binding, modification, and complex formation) to jointly perform a specific biological function (such as signal transduction, metabolic regulation, and gene expression regulation). Its structure includes nodes, edges, and combinations of nodes and edges. Each node represents a protein. An edge represents a known or predicted interaction between two proteins. The network can be undirected (indicating only the existence of an interaction) or directed (indicating the direction of regulation, such as phosphorylation, activation / inhibition).

[0028] The PPI network serves as a bridge connecting "disease phenotype" and "traditional Chinese medicine targets": First, it integrates a large number of differentially expressed genes into biologically meaningful functional modules, avoiding the isolation of individual genes. Second, it identifies proteins that occupy "hub" positions in the network, which are more likely to be key drivers of disease. Third, it provides a structured and systematic set of multi-targets for subsequent AI models, enabling precise matching of traditional Chinese medicine ingredients. Finally, it supports the scientific expression of the mechanism of action of traditional Chinese medicine compound prescriptions, namely "multi-target synergistic regulation".

[0029] Step 102: Obtain disease target proteins based on the multi-target synergistic regulatory network; in this embodiment, the disease target proteins include a set of key targets including CYP19A1, SERPINE1, and MDK-NCL.

[0030] Step 103: Construct a prior knowledge base; the prior knowledge base includes: a clinical human experience base of traditional Chinese medicine, a structural base of Chinese herbal medicine components, and a molecular-protein affinity knowledge base.

[0031] Among them, the TCM clinical experience database collects ancient books and modern clinical prescriptions from databases such as CNKI, Wanfang, VIP, and PubMed, and statistically analyzes high-frequency Chinese medicines to provide prior knowledge guidance for AI screening.

[0032] Traditional Chinese Medicine (TCM) Ingredient Structure Database: Utilizing high-frequency TCM ingredients from a clinical human experience database, this database integrates TCMSP, HERB, and other databases, and applies drug-likeness rules (such as Lipinski's five rules: molecular weight <500; number of hydrogen bond donors (HBD) <5; number of hydrogen bond acceptors (HBA) <10; lipid-water partition coefficient (LogP) <5; number of rotatable bonds ≤10) and Veber's rules (polar surface area PSA ≤140 Å). 2 (Molecular screening) Screening is performed to select small molecules that meet the above rules, forming a high-quality small molecule database.

[0033] Affinity Knowledge Base: Acquires known molecular-protein interaction data from BindingDB, ChEMBL, etc., for use in training subsequent multimodal fusion models.

[0034] Step 104: Construct the multimodal fusion model EMS-T5; the multimodal fusion model includes: the molecular language model MolTo5 and the protein language model EMS-65.

[0035] Specifically, this involves: acquiring known molecular-protein interaction data from the aforementioned affinity knowledge base, such as BindingDB and ChEMBL, and inputting it into six model combinations (MolT5+ProtT5, MolT5+ESM-65, Bert+ESM-65, Bert+ProtT5, Mol-former+ProtT5, Mol-former+ESM-65) for training, testing, and validation. This training aims to develop an AI model that predicts molecular-target affinity, capable of running on millions of data points. Internal benchmark testing showed that the "MolT5+ESM-65" multimodal fusion model (named the ESM-T5 model) exhibited the best predictive performance (see prediction results). Figure 3 MolT5+ESM-65 is significantly superior to single models or other combinations, therefore MolT5+ESM-65 is chosen as the final multimodal fusion model.

[0036] Figure 3 To compare the performance of different combinations of molecular and protein representation learning models on regression tasks, this bar chart compares six combined models (MolT5+ProtT5, MolT5+ESM-65, Bert+ESM-65, Bert+ProtT5, mol_former+ProtT5, mol_former+ESM-65) in terms of mean squared error (MSE), mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (R²). 2 Performance on the four evaluation metrics. Lower values ​​(for MSE, MAE, RMSE) or higher values ​​(for R) indicate better performance. 2 A higher number indicates better performance. Overall, the MolT5+ESM-65 model combination exhibits the best performance.

[0037] Step 105: Extract the deep features of the target protein using the protein language model ESM-65.

[0038] Extract the protein sequence corresponding to the target from uniprot, and input the protein sequence into a protein language model (such as ESM-65) to extract deep feature representation.

[0039] Step 106: Using the aforementioned Chinese medicine component structure library, extract the SMILES structural features of small molecules of Chinese medicine through the molecular language model MolT5 combined with the RDKit language recognition tool.

[0040] Using a database of Chinese herbal medicine components, SMILES (Simplified Molecular Input Line Entry System, a method for linearly representing molecular structures) are generated through RDKit standardization, and then MolT5 molecular language model is used to extract multidimensional feature vectors SMILES numbers of small molecules.

[0041] Step 107: Input the deep features and SMILES structural features into the trained and optimized multimodal fusion model to predict the affinity values ​​of small molecules of traditional Chinese medicine to each key target.

[0042] When combined with the MolT5+RDKit language recognition tool, ESM-65 automatically predicts the pKD values ​​of high-throughput (500,000-1,000,000) small molecules of traditional Chinese medicine and target proteins.

[0043] Step 108: Determine the set of high-affinity molecules based on the affinity values; Step 109: Based on the set of high-affinity molecules, construct a complex network of "traditional Chinese medicine-component-target", and screen the optimal traditional Chinese medicine formula that covers the key targets and has strong synergistic effects according to the network topology parameters.

[0044] Molecules from a traditional Chinese medicine (TCM) component structure library are input into the ESM-T5 model to predict their affinity with disease target networks. Higher affinity indicates stronger drug-target interaction. Components and targets with high affinity are selected, and a "TCM-component-target" network is constructed using complex network analysis tools (such as Cytoscape's automated interface). Based on network topology parameters (such as degree value and betweenness centrality), the optimal TCM formulation with strong synergistic effects and coverage of key targets is automatically selected (e.g., Rehmannia glutinosa-Lycium barbarum, with dosage derived from clinical experience).

[0045] Step 110: Verify the in vitro and in vivo efficacy of the selected optimal Chinese medicine formula.

[0046] The candidate formulations output from step 109 above are then entered into a standardized verification process to ensure a closed loop of "computation guiding experiment, experiment verifying computation": In vivo animal experiments: A rat model of ovarian aging induced by cyclophosphamide was used. Candidate formulations (such as Rehmannia glutinosa-Lycium barbarum extract) were administered by gavage. Key efficacy indicators such as serum hormone levels (FSH, E2), ovarian index, and follicle count were measured, and immunohistochemical experiments were performed on the ovaries. After cyclophosphamide modeling, SD rats exhibited hair loss and transient hematuria, which significantly improved after HXF treatment. Rat body weight and ovarian index tended to return to normal. Compared to the model, HXF administration decreased serum FSH levels, normalized the FSH / LH ratio, increased E2, restored normal estrous cycles, and reduced atretic follicles, indicating that HXF improved ovarian damage. See details... Figure 4 .

[0047] In vitro cell experiments: After galactose-induced senescence of KGN cells, treatment with HXF revealed that the expression of FOXO3, MKI67, FSHR, and AMH decreased after modeling, and significantly improved after treatment; the expression of SERPINE1, IL6, and IL18 significantly increased after modeling, indicating cell senescence and the production of inflammatory factors, which also significantly improved after treatment. These experimental results demonstrate that HXF improves D-gal-induced senescence in KGN cells. (See details...) Figure 5 .

[0048] The impact of drug administration on the expression of the three core targets (CYP19A1, SERPINE1, and MDK-NCL) virtually screened in step 101 was validated, and it was found that these indicators significantly tended to return to normal after drug administration. See details below. Figure 6 .

[0049] In practical applications, the methods described above in this application can be applied to the formulation screening of innovative traditional Chinese medicines and health products. No specific limitations are made in this embodiment. Those skilled in the art can also apply the above methods to other fields. No specific limitations are made in this embodiment.

[0050] Based on the same inventive concept, this application also provides a multimodal deep learning-based traditional Chinese medicine (TCM) prescription screening device for implementing the aforementioned multimodal deep learning-based TCM prescription screening method. The solution provided by this device is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more multimodal deep learning-based TCM prescription screening device embodiments provided below can be found in the limitations of the multimodal deep learning-based TCM prescription screening method described above, and will not be repeated here.

[0051] In one exemplary embodiment, a multimodal deep learning-based traditional Chinese medicine prescription screening device is provided, comprising: A multi-target synergistic regulation network construction module is used to construct a multi-target synergistic regulation network covering the core mechanisms of diseases; The target protein acquisition module is used to acquire target proteins based on the multi-target synergistic regulatory network; the target proteins include a set of key targets including CYP19A1, SERPINE1, and MDK-NCL. A prior knowledge base construction module is used to construct a prior knowledge base; the prior knowledge base includes: a TCM clinical human experience base, a TCM component structure base, and a molecular-protein affinity knowledge base; The EMS-T5 multimodal fusion model building module is used to construct the EMS-T5 multimodal fusion model; the multimodal fusion model includes: the protein language model EMS-65, the molecular language model MolTo5, and the RDKit language recognition tool; The deep feature extraction module is used to extract the deep features of the target protein using the protein language model ESM-65. The SMILES structural feature extraction module is used to extract the SMILES structural features of small molecules of traditional Chinese medicine by using the traditional Chinese medicine component structure library and combining the molecular language model MolT5 with the RDKit language recognition tool. The affinity prediction module is used to input the deep features and SMILES structural features into a trained and optimized multimodal fusion model to predict the affinity values ​​between small molecules of traditional Chinese medicine and each key target. A high-affinity molecule set determination module is used to determine a high-affinity molecule set based on the affinity value; The optimal traditional Chinese medicine (TCM) formulation determination module is used to construct a complex "TCM-component-target" network based on the set of high-affinity molecules, and to screen the optimal TCM formulation that covers the key targets and has strong synergistic effects according to the network topology parameters.

[0052] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database is used for screening traditional Chinese medicine (TCM) prescriptions using multimodal deep learning. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network. When the computer program is executed by the processor, it implements a multimodal deep learning method for screening TCM prescriptions.

[0053] Those skilled in the art will understand that Figure 7The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0054] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0055] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0056] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0057] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0058] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0059] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0060] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for screening traditional Chinese medicine prescriptions based on multimodal deep learning, characterized in that, The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning includes: Construct a multi-target synergistic regulatory network covering the core mechanisms of disease; Disease target proteins are obtained based on the multi-target synergistic regulatory network; the disease target proteins include: multiple key targets; Construct a priori knowledge base; the priori knowledge base includes: a TCM clinical human experience base, a TCM component structure base, and a molecular-protein affinity knowledge base; A multimodal fusion model, EMS-T5, was constructed; the multimodal fusion model includes: the molecular language model MolTo5 and the protein language model EMS-65; The deep features of the target protein were extracted using the protein language model ESM-65. Using the aforementioned Chinese herbal medicine component structure library, the SMILES structural features of small molecules in Chinese herbal medicines were extracted by combining the molecular language model MolT5 with the RDKit language recognition tool. The deep features and SMILES structural features are input into a trained and optimized multimodal fusion model to predict the affinity values ​​of small molecules of traditional Chinese medicine to each key target. The set of high-affinity molecules is determined based on the affinity values; Based on the set of high-affinity molecules, a complex network of "traditional Chinese medicine-component-target" is constructed, and the optimal traditional Chinese medicine formula that covers the key targets and has strong synergistic effects is screened according to the network topology parameters.

2. The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning according to claim 1, characterized in that, The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning also includes the following after the last step: The selected optimal traditional Chinese medicine formula was then subjected to in vitro and in vivo efficacy verification.

3. The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning according to claim 2, characterized in that, The in vitro and in vivo efficacy verification of the selected optimal traditional Chinese medicine formula specifically includes: Serum FSH and E2 levels, ovarian index, and follicle count were measured in a cyclophosphamide-induced rat model of ovarian aging. Cell proliferation activity, γH2AX, p-FOXO3a, MKi67 and SASP factor expression were detected in a D-galactose-induced KGN ovarian granulosa cell senescence model. The effect of the optimal traditional Chinese medicine formula on the expression level of the key target was detected to verify the accuracy of the calculation and prediction, forming a closed-loop feedback from virtual screening to experimental verification.

4. The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning according to claim 1, characterized in that, The construction of a multi-target synergistic regulatory network covering the core mechanisms of the disease specifically includes: Obtain single-cell RNA sequencing data for specific diseases; Based on the RNA sequencing data, data quality control, cell clustering, and differential expression analysis were performed to screen differentially expressed genes that met the criteria of |log2FC|>1 and a corrected P-value <0.

05. Functional scoring was performed on the differentially expressed genes using an aging-related database to obtain the scoring results. A protein interaction network was constructed based on the scoring results; Based on the protein interaction network, key ligand-receptor pairs are identified through cell communication analysis, thereby constructing a multi-target synergistic regulatory network covering the core mechanism of the disease.

5. The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning according to claim 1, characterized in that, The construction of the multimodal fusion model EMS-T5 specifically includes the following steps: Obtain results from multiple model combinations; these multiple model combinations include: ProtT5+MolT5, ProtT5+Bert, ProtT5+Mol-former, ESM-65+MolT5, ESM-65+Bert, and ESM-65+Mol-former. Obtain known molecular-protein interaction data from the molecular-protein affinity knowledge base; The known molecular-protein interaction data are input into the results of multiple model combinations for training, prediction, and validation. The combination of high-throughput models with the best prediction performance was selected as the multimodal fusion model EMS-T5.

6. The method for screening traditional Chinese medicine prescriptions based on multimodal deep learning according to claim 1, characterized in that, The construction of the traditional Chinese medicine component structure library specifically includes the following steps: High-frequency TCM components were obtained from the TCMSP and HERB databases, and drug-like molecules were screened according to the Lipinski five rules and the Veber rule. The Lipinski five rules include: molecular weight < 500, number of hydrogen bond donors < 5, number of hydrogen bond acceptors < 10, lipid-water partition coefficient < 5, and number of rotatable bonds ≤ 10. The Veber rule includes: polar surface area ≤ 140 Å. 2 ; A small molecule database, namely a traditional Chinese medicine component structure library, is constructed based on the screened medicinal molecules.

7. A traditional Chinese medicine prescription screening device based on multimodal deep learning, characterized in that, The traditional Chinese medicine prescription screening device based on multimodal deep learning includes: A multi-target synergistic regulation network construction module is used to construct a multi-target synergistic regulation network covering the core mechanisms of diseases; The target protein acquisition module is used to acquire disease target proteins based on the multi-target synergistic regulatory network; the disease target proteins include multiple key targets. A prior knowledge base construction module is used to construct a prior knowledge base; the prior knowledge base includes: a TCM clinical human experience base, a TCM component structure base, and a molecular-protein affinity knowledge base; A multimodal fusion model EMS-T5 construction module is used to construct the multimodal fusion model EMS-T5; the multimodal fusion model includes: the molecular language model MolTo5 and the protein language model EMS-65; The deep feature extraction module is used to extract the deep features of the target protein using the protein language model ESM-65. The SMILES structural feature extraction module is used to extract the SMILES structural features of small molecules of traditional Chinese medicine by using the traditional Chinese medicine component structure library and combining the molecular language model MolT5 with the RDKit language recognition tool. The affinity prediction module is used to input the deep features and SMILES structural features into a trained and optimized multimodal fusion model to predict the affinity values ​​between small molecules of traditional Chinese medicine and each key target. A high-affinity molecule set determination module is used to determine a high-affinity molecule set based on the affinity value; The optimal traditional Chinese medicine (TCM) formulation determination module is used to construct a complex "TCM-component-target" network based on the set of high-affinity molecules, and to screen the optimal TCM formulation that covers the key targets and has strong synergistic effects according to the network topology parameters.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the multimodal deep learning method for screening traditional Chinese medicine prescriptions according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the multimodal deep learning method for screening traditional Chinese medicine prescriptions according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the multimodal deep learning method for screening traditional Chinese medicine prescriptions according to any one of claims 1-6.

Images