Preparation method and system of reproductive targeted food medicine based on lotus active substance and digital twinborn model

Abstract

The invention relates to the technical field of biomedicine and artificial intelligence crossing, and particularly discloses a preparation method of a reproductive targeted drug based on a lotus active substance and a digital twinborn model, which comprises the following steps: constructing a lotus active ingredient database; collecting and preprocessing multi-source heterogeneous patient data; constructing a personalized reproductive system digital twinborn model; iteratively optimizing the proportion of active components of the lotus by adopting a deep reinforcement learning algorithm; the distributed cache accelerates digital twinborn simulation; carrying out peak elimination and scheduling on an asynchronous message queue; token authentication and authority verification are unified; and the report is exported in a high-efficiency block streaming manner. Comprehensive breakthroughs are achieved in the aspects of database management, multi-modal modeling, intelligent optimization, precise manufacturing, system architecture, safety compliance and the like, the research and development efficiency, precision and reliability of reproductive targeted drugs are remarkably improved, and personalized treatment requirements are met.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary field of biomedicine and artificial intelligence, and specifically to a method and system for preparing a reproductive-targeted food drug based on lotus active substances and a digital twin model. Background Art

[0002] As people's demand for reproductive health continues to increase, research on natural plant active ingredients in the field of reproductive regulation has gradually deepened. Lotus is rich in various active substances such as quercetin and nuciferine, which have been shown to have certain regulatory effects on indicators such as ovarian function and sperm quality. However, existing reproductive health care or treatment plans based on lotus ingredients mainly rely on in vitro cell experiments or animal models, which make it difficult to accurately characterize individual differences in the human body, and have the following shortcomings in drug formulation screening and delivery:

[0003] 1. Limitations of the ingredient database

[0004] Currently, most relevant literature records the information of lotus active ingredients and their targets in a manually compiled form, lacks structured and scalable database support, and cannot meet the needs of large-scale screening and quantitative analysis.

[0005] 2. Lack of personalized digital twin models

[0006] While some research has applied digital twin technology to industrial manufacturing and some medical monitoring, few efforts have achieved comprehensive simulation of the human reproductive system. Existing models of human reproductive dynamics often rely on a single data source (such as hormone levels), making it difficult to integrate multi-omics data such as genomics, metabolomics, and imaging, and unable to dynamically predict the effects of individualized drug interventions.

[0007] 3. Insufficient system performance and security

[0008] Large-scale digital twin simulation and reinforcement learning training both have extremely high requirements for computing performance. Existing platforms mostly use centralized computing, which is prone to simulation throughput bottlenecks, queuing delays and task loss problems. At the same time, biomedical data is sensitive information, but most systems lack a unified authentication and audit mechanism, posing a potential risk of data leakage.

[0009] In summary, there is an urgent need for a method that can build a structured and scalable database of lotus active ingredients; deeply integrate multi-omics and imaging data to build a personalized digital twin model of the reproductive system; and thus form a complete, high-performance, safe and reliable reproductive targeted drug preparation technology solution. Summary of the Invention

[0010] The present invention provides a method and system for preparing reproductive-targeted food drugs based on lotus active substances and digital twin models. The present invention has achieved comprehensive breakthroughs in database management, multimodal modeling, intelligent optimization, precision manufacturing, system architecture, and safety compliance, significantly improving the efficiency, accuracy, and reliability of reproductive-targeted drug research and development, and meeting personalized treatment needs.

[0011] The present invention solves the above technical problems through the following technical solutions:

[0012] The method for preparing reproductive-targeted drugs based on lotus active substances and digital twin models includes the following steps:

[0013] (1) Constructing a database of lotus active ingredients, wherein the database adopts a relational database management system,

[0014] (2) Collect and preprocess multi-source heterogeneous patient data,

[0015] (3) Build a personalized digital twin model of the reproductive system,

[0016] (4) Using deep reinforcement learning algorithm to iteratively optimize the ratio of lotus active ingredients,

[0017] (5) Using three-dimensional bioprinting technology to prepare a multilayer biomimetic delivery carrier, the outer layer is a pH-responsive hydrogel, the middle layer is a thermosensitive degradable polymer, and the inner layer is loaded with liensinine-metal complex nanoparticles to achieve controlled release;

[0018] (6) Distributed caching accelerates digital twin simulation. A cache cluster is deployed near the modeling server. Hotspot model parameter fragments are cached based on a consistent hashing algorithm, and active / standby switching and expiration policies are configured.

[0019] (7) Asynchronous message queue peak elimination and scheduling, pushing reinforcement learning training tasks and simulation computing tasks to distributed message queues, supporting priority, failure retry, and dynamic expansion and contraction;

[0020] (8) Unify token authentication and permission verification, deploy OAuth2.0 authentication gateway plug-ins at both the entry gateway and microservice layer, verify access rights based on JWT tokens and record audit logs;

[0021] (9) Efficient block-by-block streaming export of reports, dividing personalized simulation results and recipe solutions into preset data blocks, writing them in parallel in CSV or spreadsheet format, and finally generating a complete report through the merging interface.

[0022] In a specific embodiment, the tables in the database are associated with each other through foreign keys, and a partition table strategy is applied to the genome data table and the metabolome data table to accelerate large-scale data queries.

[0023] In a specific embodiment, the graph neural network uses a multi-head attention mechanism to fuse different data modalities to enhance the accuracy of structure and function coupling simulation.

[0024] In a specific embodiment, the reinforcement learning algorithm uses entropy regularization during training to balance exploration and exploitation, and introduces a confidence bound algorithm to prevent strategy collapse.

[0025] In a specific embodiment, the database table includes:

[0026] Ingredient table, fields include ingredient ID, name, molecular formula, molecular weight, solubility, and target receptor ID. The primary key is ingredient ID, and a joint unique index of name and target receptor ID is established;

[0027] Receptor table, fields include receptor ID, receptor name, biological pathway description, and the primary key is receptor ID;

[0028] Pharmacokinetic table, fields include component ID, half-life, tissue distribution coefficient, component ID is the foreign key and index is established.

[0029] In a specific embodiment, collecting and preprocessing multi-source heterogeneous patient data includes:

[0030] Genomic data, obtaining a list of mutation sites through whole genome sequencing;

[0031] metabolomic data, quantifying the concentrations of multiple metabolites by mass spectrometry;

[0032] For imaging data, ovarian ultrasound and MRI images were used, stored according to the DICOM standard, and structural parameters were extracted;

[0033] The above data are cleaned, normalized and feature extracted, and then input into the subsequent modeling module.

[0034] In a specific embodiment, the distributed cache cluster uses a TTL mechanism and an LRU strategy to jointly manage data expiration and replacement.

[0035] In a specific embodiment, building a personalized reproductive system digital twin model includes:

[0036] Based on the structural modeling of graph neural networks, reproductive organs and cell populations are represented as graph nodes, and the node features are gene expression and metabolite concentrations;

[0037] The dynamics simulation module uses ordinary differential equations to describe the changing patterns of hormone levels and updates the model state based on node outputs;

[0038] The online calibration module uses the patient's real-time hormone monitoring data to update the model parameters using the Kalman filter method.

[0039] In a specific embodiment, a deep reinforcement learning algorithm is used to iteratively optimize the ratio of active ingredients in lotus. The algorithm includes:

[0040] State space: composed of the reproductive function indicator vector output by the current digital twin model;

[0041] Action space: the adjustment range of the mass ratio of each lotus active ingredient;

[0042] Reward function: R = α·ΔAMH + β·Δfollicle number + γ·Δsperm motility - δ·recipe complexity;

[0043] Policy network: trained based on deep policy gradient or proximal policy optimization;

[0044] Experience replay and target network are used to stabilize the training process.

[0045] A system capable of realizing a method for preparing reproductive-targeted drugs based on lotus active substances and digital twin models.

[0046] The present invention's method for preparing reproductive-targeted drugs based on lotus active substances and digital twin models, combined with database management, artificial intelligence algorithms, distributed architecture, and advanced manufacturing technologies, has the following significant beneficial effects:

[0047] 1. Using a relational database, multi-table structured storage and large-scale data rapid query are achieved through foreign key associations and partitioned table strategies; combined with a unique index and ingredient-target retrieval mechanism, it can support real-time dynamic expansion and batch import, significantly shortening the time required to retrieve and update active ingredients.

[0048] 2. Through preprocessing and feature extraction of genomic, metabolomic, and imaging data, multimodal input is constructed. Graph neural networks and multi-head attention mechanisms are introduced, combined with ordinary differential equation dynamics simulation and Kalman filter online calibration to realistically reproduce the dynamic response of individual patients' reproductive systems and improve simulation accuracy and reliability.

[0049] 3. Based on the deep reinforcement learning (DDPG / PPO) algorithm, a balanced exploration method is designed, utilizing entropy regularization and confidence bound mechanisms to achieve adaptive search for complex formulation-efficacy mapping. The reward function integrates AMH, follicle count, sperm motility, and formulation complexity to ensure that the optimization results balance maximizing efficacy and simplifying formulations, with rapid convergence and a tendency towards global optimization.

[0050] 4. Deploy a TTL+LRU strategy distributed cache cluster near the modeling server, accelerate access to hot parameters based on the consistent hashing algorithm, and significantly improve the throughput of digital twin simulation; perform task scheduling through an asynchronous message queue system, support priority, failure retry, and dynamic scaling, effectively eliminate peak traffic and ensure the stable execution of large-scale simulation and training tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0052] Figure 1 Shown is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] Example 1

[0055] This example uses the treatment of female patients with polycystic ovary syndrome (PCOS) as an example. By combining active ingredients such as liensinine, a digital twin model is constructed. The formula is optimized through deep reinforcement learning, and finally a three-dimensional printed bionic delivery carrier is prepared to complete personalized reproductive targeted drugs.

[0056] 1. Constructing a database of lotus active ingredients

[0057] Ingredients

[0058] Sample Record

[0059]

[0060] Primary key: component ID; joint unique index: name, target receptor ID

[0061] Receptor table

[0062] Receptor ID Receptor name Biological pathway description R01 ERα estrogen receptor Mediates the estrogen-ERα signaling pathway and regulates follicle development R02 FSHR follicle-stimulating hormone receptor Affects the action of follicle-stimulating hormone and changes FSH sensitivity

[0063] Pharmacokinetic table

[0064]

[0065]

[0066] Ingredient ID foreign key association, the ingredient ID has been indexed

[0067] 2. Collect and preprocess multi-source heterogeneous patient data

[0068] Patient information: 30-year-old female with newly diagnosed PCOS;

[0069] Genomic data:

[0070] Whole-genome sequencing was performed using the Illumina NovaSeq 6000 platform at an average depth of 30×.

[0071] Mutation detection: A list of approximately 35,000 SNPs / Indels was obtained, and 56 loci related to ovarian function (CYP19A1, FSHR, ESR1, etc.) were extracted;

[0072] Metabolomics data were analyzed by LC–MS / MS (AB Sciex TripleTOF 6600) to quantify the concentrations of 120 endocrine-related metabolites;

[0073] Image data

[0074] Ultrasound: Collect three-dimensional ultrasound of both ovaries and extract the distribution of follicle diameters;

[0075] Data preprocessing

[0076] Cleaning: remove low-quality sequencing reads (Q < 20) and null values ​​in the metabolome;

[0077] Normalization: TPM was used for gene expression, and Z-score normalization was performed for metabolite concentrations;

[0078] Feature extraction: The first 10 principal components and 10 metabolite principal components were selected, merged into a 20-dimensional feature vector, and input into the modeling module.

[0079] 3. Build a personalized digital twin model of the reproductive system

[0080] Graph Neural Network Structure Modeling

[0081] Node definition: organs such as ovary, endometrium, pituitary gland, and key follicle groups. Each node feature includes gene expression (20 dimensions) and metabolite concentration (20 dimensions).

[0082] Edge relationships: anatomical adjacency and connectivity with hormone signaling pathways;

[0083] GNN selection: Use Graph Attention Network with 8 multi-head attention heads and 64 hidden layer dimensions to fuse information from different modalities;

[0084] Dynamics Simulation Module

[0085] Ordinary differential equations are used to describe the changes in FSH, LH, E2, and P4 concentrations:

[0086] Parameters k1-k4 were fitted based on literature and patient data;

[0087] Online calibration module, real-time collection of patient venous blood FSH, E2, and AMH concentrations (once a week);

[0088] Based on the adaptive Kalman filter, the noise covariance is dynamically adjusted to update the k value in the above formula;

[0089] 4. Deep reinforcement learning to optimize the ratio of lotus active ingredients

[0090] State space: The digital twin model outputs reproductive function indicator vectors, follicle count, sperm motility,

[0091] Action space: The mass ratio of C001 and C002 is adjusted within the range of [0.1%–1.0%] by ±0.05% per step;

[0092] Algorithm selection: PPO algorithm, entropy regularization coefficient 0.01, and the introduction of confidence limit algorithm to prevent strategy collapse;

[0093] 5. 3D Bioprinting to Prepare Bionic Delivery Vehicles

[0094] Materials and layers:

[0095] Outer layer: pH 7.4-responsive polyacrylate hydrogel, 300 μm thick;

[0096] Middle layer: thermosensitive degradable polycaprolactone-polyethylene glycol copolymer, thickness 500 μm;

[0097] Inner layer: loaded with C001-gold nanoparticle complex (particle size 50 nm), particle layer thickness 200 μm;

[0098] 6. Distributed Cache Accelerated Simulation

[0099] Consistent hashing sharding, caching the computational hotspots (node ​​feature matrix) in GNN with a TTL of 10 minutes and LRU replacement;

[0100] 7. Asynchronous message queues and scheduling

[0101] Reinforcement learning training tasks and ODE simulations are delivered to high-priority and medium-priority topics, respectively. The system supports token bucket current limiting and three retries for failed messages before placing them in a dead-letter queue. Dynamic scaling automatically adjusts the number of Pods through Kubernetes HPA.

[0102] 8. Unified token authentication and permission verification, all microservice logs are centrally recorded through ELK Stack and tagged by token ID;

[0103] 9. Efficient block-based streaming export of reports, export format: CSV block (default block size 10MB), a total of 10 blocks, written by ParallelStream; dynamic block adjustment: real-time adjustment to 8-12MB based on I / O throughput (claim 9); merge interface: finally generate a complete Excel report through the back-end merge interface (concurrent merge of 5 channels) and send it to the clinician's email.

[0104] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0105] The above description of the disclosed embodiments will enable one skilled in the art to enable or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be applied to other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein, but is to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for preparing reproductive-targeted drugs based on lotus active substances and digital twin models, characterized in that: The following steps are involved: (1) Constructing a database of lotus active ingredients, wherein the database adopts a relational database management system, (2) Collect and preprocess multi-source heterogeneous patient data, (3) Build a personalized digital twin model of the reproductive system, (4) Using deep reinforcement learning algorithm to iteratively optimize the ratio of lotus active ingredients, (5) Using three-dimensional bioprinting technology to prepare a multilayer biomimetic delivery carrier, the outer layer is a pH-responsive hydrogel, the middle layer is a thermosensitive degradable polymer, and the inner layer is loaded with liensinine-metal complex nanoparticles to achieve controlled release; (6) Distributed caching accelerates digital twin simulation. A cache cluster is deployed near the modeling server. Hotspot model parameter fragments are cached based on a consistent hashing algorithm, and active / standby switching and expiration policies are configured. (7) Asynchronous message queue peak elimination and scheduling, pushing reinforcement learning training tasks and simulation computing tasks to distributed message queues, supporting priority, failure retry, and dynamic expansion and contraction; (8) Unify token authentication and permission verification, deploy OAuth2.0 authentication gateway plug-ins at both the entry gateway and microservice layer, verify access rights based on JWT tokens and record audit logs; (9) Efficient block-by-block streaming export of reports, dividing personalized simulation results and recipe solutions into preset data blocks, writing them in parallel in CSV or spreadsheet format, and finally generating a complete report through the merging interface.

2. The method according to claim 1, characterized in that The tables in the database are associated with each other through foreign keys, and a partition table strategy is applied to the genome data table and the metabolome data table to accelerate large-scale data queries.

3. The method according to claim 1, characterized in that The graph neural network adopts a multi-head attention mechanism to fuse different data modalities to enhance the simulation accuracy of structure and function coupling.

4. The method according to claim 1, wherein The reinforcement learning algorithm adopts entropy regularization during training to balance exploration and exploitation, and introduces a confidence bound algorithm to prevent strategy collapse.

5. The method according to claim 1, characterized in that The database table includes: Ingredient table, fields include ingredient ID, name, molecular formula, molecular weight, solubility, and target receptor ID. The primary key is ingredient ID, and a joint unique index of name and target receptor ID is established; Receptor table, fields include receptor ID, receptor name, biological pathway description, and the primary key is receptor ID; Pharmacokinetic table, fields include component ID, half-life, tissue distribution coefficient, component ID is the foreign key and index is established.

6. The method according to claim 1, characterized in that Collect and preprocess multi-source heterogeneous patient data, including: Genomic data, obtaining a list of mutation sites through whole genome sequencing; metabolomic data, quantifying the concentrations of multiple metabolites by mass spectrometry; For imaging data, ovarian ultrasound and MRI images were used, stored according to the DICOM standard, and structural parameters were extracted; The above data are cleaned, normalized and feature extracted, and then input into the subsequent modeling module.

7. The method according to claim 1, characterized in that The distributed cache cluster uses the TTL mechanism and the LRU strategy to jointly manage data expiration and replacement.

8. The method according to claim 1, characterized in that Build a personalized digital twin model of the reproductive system, including: Based on the structural modeling of graph neural networks, reproductive organs and cell populations are represented as graph nodes, and the node features are gene expression and metabolite concentrations; The dynamics simulation module uses ordinary differential equations to describe the changing patterns of hormone levels and updates the model state based on node outputs; The online calibration module uses the patient's real-time hormone monitoring data to update the model parameters using the Kalman filter method.

9. The method according to claim 1, characterized in that A deep reinforcement learning algorithm is used to iteratively optimize the ratio of active ingredients in lotus. The algorithm includes: State space: composed of the reproductive function indicator vector output by the current digital twin model; Action space: the adjustment range of the mass ratio of each lotus active ingredient; Reward function: R = α·ΔAMH + β·Δfollicle number + γ·Δsperm motility - δ·recipe complexity; Policy network: trained based on deep policy gradient or proximal policy optimization; Experience replay and target network are used to stabilize the training process.

10. A system capable of implementing any one of the methods 1-9.

Images