Pharmaceutical design system and pharmaceutical design method

Abstract

Efficient design and validation of immunosuppressants. [Solution] The immunosuppressant design system 800 includes: a recipient information management unit 10 that manages recipient information; a potential donor information management unit 40 that manages potential donor information; a potential recipient prediction unit 20 that predicts potential recipients, which are a group of individuals who are likely to be registered on an organ transplant waiting list within a certain period in the future; a rejection factor identification unit 840 that identifies rejection factors that cause rejection reactions based on the information managed by the recipient information management unit 10 and the potential donor information management unit and the information of potential recipients predicted by the potential recipient prediction unit 20; a molecular design unit 850 that designs or screens molecular structures that specifically bind to the rejection factors identified by the rejection factor identification unit 840 and inhibit their function; and an in silico verification unit 860 that simulates the in vivo dynamics and immune response of drugs having the molecular structure designed or screened by the molecular design unit 850 and verifies their effectiveness.

Description

Technical Field

[0001] The present invention relates to a technology for designing pharmaceuticals for the treatment or prevention of diseases, and particularly to a technology for designing and verifying in advance individualized or optimized pharmaceuticals based on predictions of future disease demands (pandemics and morbidity risks) using artificial intelligence (AI) and simulation technologies.

Background Art

[0002] In the medical field, the development of new pharmaceuticals is an important issue, but the conventional drug discovery process generally requires a long period of 10 to 15 years from basic research to approval. Also, existing AI drug discovery technologies are effective for screening against static target structures, but have not been established as a system for designing by integrating temporal and dynamic factors such as future virus mutation predictions and future morbidity risks in specific patient populations.

Disclosure of the Invention

Problems to be Solved by the Invention

[0003] Since the conventional drug discovery process requires a long period, there is a problem that it will not be in time for the time limit to treatment if the drug is designed and adjusted after a pandemic occurs or when a patient has resistance to standard treatment in a progressive disease such as cancer. Also, in transplantation medicine, there is a time limit that it will not be in time if the drug is adjusted after the combination of donor and recipient is established. An object of the present invention is to enable efficient design and verification of pharmaceuticals.

Means for Solving the Problems

[0004] One aspect of the present invention is a pharmaceutical design system for designing pharmaceuticals for the treatment or prevention of diseases, a disease-related factor information management unit that manages information on disease-related factors that are factors related to the cause or exacerbation of a disease, A potential patient prediction unit predicts potential patients, which are a group of individuals who are likely to contract or require treatment for the aforementioned disease within a certain period in the future. A patient information acquisition unit that acquires information on at least one of the patient or the potential patient predicted by the potential patient prediction unit, A target factor identification unit identifies target factors that cause the onset or progression of the disease, based on the information managed by the disease-related factor information management unit and the information acquired by the patient information acquisition unit. A molecular design unit that designs or screens molecular structures that specifically bind to the target factor identified by the target factor identification unit and inhibit or control its function, The pharmaceutical design system is characterized by comprising: an in silico verification unit that simulates the in vivo pharmacokinetics and biological responses of a drug having a molecular structure designed or screened by the molecular design unit, and verifies its efficacy.

[0005] Another aspect of the present invention is, A method for designing pharmaceuticals for the treatment or prevention of a disease, A disease-related factor information management step for managing information on disease-related factors, which are factors associated with the cause or exacerbation of a disease, A potential patient prediction step that predicts a group of individuals who are likely to develop or require treatment for the aforementioned disease within a certain period in the future, A patient information acquisition step that acquires information on at least one of the patient or the potential patient predicted by the potential patient prediction step, A target factor identification step, which identifies target factors that cause the onset or progression of the disease, based on the information managed by the disease-related factor information management step and the information obtained by the patient information acquisition step, A molecular design step involves designing or screening molecular structures that specifically bind to the target factor identified in the target factor identification step and inhibit or control its function. The drug design system is characterized by comprising: an in silico validation step that simulates the pharmacokinetics and biological responses of a drug having a molecular structure designed or screened by the molecular design step, and verifies its efficacy. [Effects of the Invention]

[0006] According to the present invention, pharmaceuticals can be designed and verified efficiently. [Brief explanation of the drawing]

[0007] [Figure 1] This is a diagram illustrating the configuration of the first embodiment. [Figure 2] This is a functional block diagram of the organ donation management system. [Figure 3] This is a hardware configuration diagram of the emergency call center system. [Figure 4] This is a hardware configuration diagram of an emergency call device. [Figure 5] This is the hardware configuration of the organ donation management system. [Figure 6] This figure shows an example of processing by an emergency call center system and an emergency call device. [Figure 7] This figure shows an example of processing performed by an emergency call center system. [Figure 8] This figure shows an example of processing performed by an emergency call center system. [Figure 9] This figure shows an example of processing performed by an emergency call center system. [Figure 10] This figure shows an example of processing performed by an emergency call center system. [Figure 11] This figure shows an example of processing performed by an emergency call center system. [Figure 12] This figure shows an example of processing performed by an emergency call device. [Figure 13] This is a diagram illustrating the configuration of the second embodiment. [Figure 14] This is a diagram illustrating the configuration of the third embodiment. [Figure 15]It is a diagram showing an example of processing by an emergency notification center device and an emergency responder device. [Figure 16] It is a diagram showing an example of processing by an emergency responder device. [Figure 17] It is a diagram showing an example of an emergency response request content display screen. [Figure 18] It is a diagram showing an example of processing by the emergency notification center device of the fourth embodiment. [Figure 19] It is a diagram showing an example of processing by the emergency notification center device of the fifth embodiment. [Figure 20] It is a functional block diagram of an organ donation management device in the sixth embodiment. [Figure 21] It is a diagram showing an example of processing by the emergency notification center device of the sixth embodiment. [Figure 22] It is a functional block diagram of an immunosuppressant design system of the seventh embodiment. [Figure 23] It is a hardware configuration diagram of an immunosuppressant design system. [Figure 24] It is a diagram showing an example of processing by immunosuppressant design processing. [Figure 25] It is a functional block diagram of a pharmaceutical design system of the eighth embodiment. [Figure 26] It is a diagram showing an example of processing by a pharmaceutical design system.

Embodiments for Carrying Out the Invention

[0008] Embodiments of the present invention will be described below with reference to the drawings. This embodiment relates to an immunosuppressant design system that develops personalized or optimized drugs based on future transplant demand, based on the vast amount of data and future prediction capabilities possessed by a life-saving system that applies multi-AI agent technology. Specifically, based on a large-scale dataset integrating medical and genetic information of recipients, potential donors, and predicted potential recipients, the AI ​​identifies factors that cause rejection reactions, and enables the design and verification of molecular structures that specifically act on those factors from a stage before a donor is actually found or before transplantation is performed.

[0009] 1. About Terminology In this embodiment, the meanings of the following terms are generally as follows. Note that duplicate descriptions of the same (similar) content are permitted in the following description.

[0010] Disease-related factors are factors associated with the cause or exacerbation of a disease, and include pathogens (viruses, bacteria, etc.), allergens, environmental pollutants, or carcinogenic gene mutations.

[0011] Potential patients are individuals who are not currently patients but may develop or require treatment for a specific disease within a certain period in the future. Potential patients are predicted based on epidemiological information such as infectious disease surveillance data, lifestyle data, and genetic statistics data.

[0012] A target factor is a molecule or mechanism within a living organism that directly causes the onset or progression of a disease. Target factors are identified through the analysis of integrated data (such as analysis using graph neural networks). Specific examples of target factors include viral spike proteins, specific receptors on cancer cells, and inflammatory signaling molecules.

[0013] A rejection factor is a biological factor (molecule or mechanism) that directly causes rejection in organ transplantation. Specifically, this includes certain proteins or antigenic epitopes (antigenic determinants).

[0014] Integrated data refers to a collection of multifaceted information (dataset) constructed in an immunosuppressant design system (drug design system) to identify "target factors (or rejection factors)" that directly cause the onset or progression of disease through AI analysis. Specifically, the integrated data is based on medical and genetic information of patients currently in need of treatment, information on future patient groups (potential patients) predicted by the Potential Patient Prediction Department, and information on pathogens, environmental factors, and potential donors managed by the Disease-Related Factor Information Management Department. Furthermore, historical data such as past case data, drug response history, and rejection history in past transplant cases are integrated into this data. Another characteristic of this dataset is that it includes not only static medical information, but also dynamic biological data such as current drug use, metabolic status, real-time homeostasis indicators based on inter-organ networks, and micro-inflammation levels. Furthermore, dynamic environmental information such as physical impact data from the accident site at the time of donor generation, disaster simulation results, or predicted transport time and environmental stress data based on geographical distance may also be integrated.

[0015] In silico validation is a method that uses computer simulations to verify the efficacy, safety, and pharmacokinetics of a drug. This system aims to shorten development time and improve safety by performing validation using virtual patient models and nanomachine models on a computer before actually synthesizing and administering the drug.

[0016] A large-scale dataset is a collection of data that integrates medical and genetic information of recipients and potential donors, as well as future predictive information (potential recipient information) and the history of rejection in past transplant cases. This dataset is constructed in a format optimized for AI analysis (for example, a format using data warehouse technology) and is used to extract potential features that determine the presence or absence of rejection. In this specification, a large-scale dataset is treated as a specific embodiment (or synonym) of the above-mentioned "integrated data" constructed and implemented in a format that can be analyzed by AI on the system.

[0017] A graph neural network (GNN) is a type of deep learning model that treats the relationships between data as a graph structure (nodes and edges) and learns their complex interactions. In this system, it is used to learn immunological interactions between donors and recipients, as well as complex networks between genes and proteins, as graph structures, in order to identify factors that cause rejection reactions.

[0018] Cryptic pockets are hidden binding sites (pockets) that temporarily form on the surface of proteins during dynamic transformation processes. While difficult to detect through static structural analysis, they can be predicted using molecular dynamics simulations (MD simulations). By designing compounds that bind to these sites, it becomes possible to develop immunosuppressants targeting previously undruggable targets.

[0019] Protein structure prediction AI is an artificial intelligence program or algorithm that uses machine learning techniques such as deep learning to accurately infer the three-dimensional structure of a protein from amino acid sequence information. A specific example is AlphaFold2, developed by DeepMind Technologies. In this embodiment, it is implemented in the molecular design unit 850 and used to identify the three-dimensional structure of rejection factors, search for cryptic pockets (hidden binding sites) formed during the dynamic transformation of proteins, and design molecular structures that bind to them. Since its execution requires enormous computing resources such as CPUs and GPUs, it may include a configuration that executes in large-scale parallel on a scalable computing pipeline on the cloud (e.g., Vertex AI Pipelines).

[0020] Internal biological signals such as electroencephalograms (EEGs) are physiological signals that directly reflect the functional state of the central and autonomic nervous systems, acquired through head-mounted devices and wearable sensors. Specifically, these include the power spectral density of alpha and theta waves in electroencephalograms (EEGs), the QRS complex and heart rate variability (HRV) in electrocardiograms (ECGs), and mean arterial pressure (MAP). In this system, by integrating these signals with external features (such as facial color and body movement) obtained from surveillance cameras and performing multimodal analysis, it is used to identify mental and physical stress states that are difficult to detect by appearance alone, as well as subtle immunodynamic fluctuations (such as fluctuations in autonomic nervous system activity) that may indicate rejection reactions.

[0021] HLA (Human Leukocyte Antigen) can be described as the blood type of white blood cells, and it is the most important immunological marker for determining the compatibility between a donor and a recipient in organ transplantation. A rare HLA type refers to an HLA type that occurs very infrequently within a general population. "Rare" (extremely low frequency) means, for example, that the number of cases is so small that statistically significant verification cannot be performed using only real data, and that the generation of synthetic data (virtual patients) using AI is essential. Cases where only one in several hundred thousand people, or where only a few individuals are found even after searching through bone marrow banks worldwide (with tens of millions of registered individuals), are also considered "rare."

[0022] Synthetic data is artificially generated data created by learning statistical features from data of real patients (such as electronic health records). While possessing statistically equivalent properties to real data, it does not contain information that could identify individuals, thus protecting privacy and making it suitable for AI training and validation. In this system, it is used to expand data on rare cases and build a virtual patient population.

[0023] A virtual patient is a hypothetical patient in computer simulations that models specific genetic backgrounds, physiological characteristics, and disease states. Composed of synthetic data, it is used to examine rare HLA combinations that are difficult to recruit in real-world clinical practice, or to verify drug responses under specific conditions.

[0024] Regulatory T cells (Tregs) are a type of T cell that suppressively regulates the immune response and maintains immune tolerance to the body. This system aims to suppress only the rejection reaction to transplanted organs and reduce side effects such as the risk of infection by designing vaccines and peptide drugs that induce Tregs in the body that specifically act on identified antigen epitopes.

[0025] Autonomous nanomachines are devices composed at the molecular or microscopic scale that autonomously perform specific tasks within a living organism (such as drug delivery or tissue repair). In this system, they are used to accumulate in specific tissues or vascular endothelial cells of transplanted organs to locally release immunosuppressants (drug delivery) or to repair microscopic damage to organs.

[0026] The Homeostasis Index (HIS) is a dynamic indicator that shows the current state of balance in a living organism. It is calculated based on real-time biological data that integrates not only static test values ​​but also current drug use, metabolic status, and the state of inter-organ networks. It reflects the individual physiological stress state of each patient at the time of transplantation and is used for more precise compatibility assessment and drug design.

[0027] Federated learning is a distributed machine learning method that does not aggregate data in one place, but instead keeps the data on each device or facility (node), while only sharing and updating the learning parameters (gradient information) of the AI ​​model. This makes it possible to perform AI learning on a global scale while protecting sensitive genetic information and highly private medical data whose cross-border transfer is restricted by national laws and regulations.

[0028] A donor database is a system for registering and managing information about people who may be able to donate organs. The donor database registers and manages information on an individual basis for people who have expressed their intention to donate organs. More specifically, the donor database registers and manages information on the types of organs that can be donated after brain death (heart, lungs, liver, kidneys, pancreas, small intestine, eyeballs, etc.) and the types of organs that can be donated after death when the heart has stopped (kidneys, pancreas, eyeballs, etc.), as information for each individual who has expressed their intention to donate organs.

[0029] A potential donor refers to an individual who has the potential to become a donor (organ provider, organ source). Potential donors include both those registered in the donor database and those not registered in the donor database.

[0030] A recipient refers to an individual who is registered on a waiting list as someone who wishes to receive an organ transplant.

[0031] A waiting list is a list of recipients registered to await organ donation. An example of a waiting list is the one used by the Japan Organ Transplant Network (JOTNW). JOTNW centrally manages recipient information and fairly and impartially selects candidates for organ transplants based on medical criteria, urgency, and waiting period.

[0032] Potential recipients refer to individuals or groups of individuals who are not currently on the waiting list but are expected to be admitted to the waiting list within a certain period of time in the future after meeting the criteria for diagnosis, disease progression, and eligibility. The purpose of predicting potential recipients is to facilitate long-term clinical resource allocation, policy effectiveness evaluation, hospital-level capacity planning, and the design of early intervention strategies.

[0033] Referral refers to the act or process of sending a patient who may require organ transplantation from a general medical institution or specialist to a facility or department specializing in organ transplantation for specialized treatment and evaluation. Information regarding referrals is used as input data for potential recipient prediction models (hereinafter also simply referred to as "prediction models"), events in modeling approaches, and items for bias correction of prediction models, etc.

[0034] A predictive model is a model used to calculate (predict / estimate) potential recipients. In predictive models, information on how often patients are referred to specialists is an important factor in estimating the probability of subsequent disease progression and being placed on a waiting list. For example, in the microsimulation (individual-based) probability rule, one of the predictive models, "referral" means that the patient receives evaluation from a specialist as a step before being registered on the transplant waiting list. The "bias" in bias correction of predictive models refers to the practical bias that patients are not registered on the waiting list unless they are "referred" to a specialist facility.

[0035] The organ donation process is the process of providing organs from a donor to a recipient. The organ donation process can include both immediate donation and on-demand donation. Immediate donation involves preserving organs from a donor using standard preservation methods (simple refrigeration or mechanical perfusion) and providing them to the recipient as quickly as possible. On-demand donation involves preserving organs from a donor until a suitable recipient becomes available, then reviving (restoring, recovering) them to a transplantable state and providing them to the recipient as quickly as possible. Preservation methods used in on-demand donation include, for example, ice freezing, supercooling, and vitrification.

[0036] Ice-temperature freezing is a preservation method that cools organs to the lowest temperature possible without causing damage from freezing. Supercooling is a preservation method that supercools organs (cooling them below their freezing point without solidifying them). Ice-temperature freezing and supercooling are effective in suppressing organ damage. In other words, ice-temperature freezing and supercooling can prevent the physical destruction of cells and tissues caused by ice crystals that occur during normal freezing.

[0037] Vitrification is a preservation method that involves perfusing tissue with a high concentration of cryoprotective agents (CPAs) instead of water, followed by cooling, thereby transitioning the tissue to a solid state without the formation of ice crystals. In this preservation method, after vitrification is achieved, the organ is cooled to extremely low temperatures, such as liquid nitrogen temperature, to ensure further physical stability. Therefore, vitrification is suitable for the long-term preservation of organs. On the other hand, applying vitrification as a preservation method requires advanced technologies such as measures to suppress organ damage due to ice crystals and cracks, measures to repair organ damage, and measures to suppress chemical damage due to CPA toxicity. In this regard, molecular nanotechnology (MNT), in which molecular-sized autonomous robots (nanomachines) explore the body's vascular network and cells to detect and repair ischemic damage, chemical damage due to CPA toxicity, physical damage due to ice crystals and cracks, and the patient's underlying disease at the molecular or atomic level, may be able to revive (restore, recover) vitrified organs to a transplantable state.

[0038] EHR stands for "Electronic Health Record," a system that digitizes patients' medical information and shares it among multiple healthcare institutions. EHR information is managed by national systems (e.g., the Information-technology Promotion Agency, Ministry of Health, Labour and Welfare in Japan) and the systems of collaborating organizations. These organizations may include national and international organizations. EHRs centrally manage each patient's medical information throughout their life using a patient ID. The patient ID may include a nationwide common identifier (medical ID) used to link insurance medical information on an individual basis in the medical, nursing care, and health fields.

[0039] The recipient ID is an identifier used by the life-saving system to identify the recipient. A potential donor ID is an identifier used by life-saving systems to identify a potential donor. The recipient ID and potential donor ID are linked to the patient ID in the EHR, respectively.

[0040] The linking process between recipient IDs and potential donor IDs and patient IDs in EHRs managed by external EHR management organizations is carried out under strict security protocols. During the linking process, the patient ID itself is not stored within the life-saving system 1 network; instead, a one-way, irreversible hash value (or cryptographically pseudo-anonymized identifier) ​​generated by the EHR management organization is used as an intermediary key. This intermediary key is, for example, linked to the life-saving system ID using strong AES-256 encryption and stored in an isolated compartment within the memory units of the emergency call center device 100 and the organ donation management device 600. Highly sensitive information, such as biometric data and risk indicators, is processed while linked to this pseudo-anonymized life-saving system ID, and personally identifiable information is disclosed only to authorized operators and medical teams to the minimum extent necessary for life-saving measures. This ensures both highly efficient system operation and robust protection of potential donor personal information.

[0041] The organ preservation specialist team is a team of medical professionals within a life-saving system that provides organ transplantation to quickly and efficiently retrieve, preserve, and transport donor organs.

[0042] Hypothetical information is not definitive, real information at the present time, but rather information that estimates future possibilities.

[0043] A call means conversing using a means of communication (e.g., a telephone). The process for making a call may include the exchange of information other than the call itself between the emergency call device that is the source of the emergency call and the emergency call center device that receives the emergency call. The source of the call may be read as the sender or originator. The exchange of information other than the call itself may include, for example, the exchange of GPS information detected by GPS (Global Positioning System). GPS information is information indicating the coordinate position on Earth, specifically longitude and latitude. GPS information may also include information such as altitude and speed of movement.

[0044] Automatic emergency calls are emergency calls made without human intervention when an emergency or anomaly is detected through automated mechanisms such as sensors or systems. Manual emergency calls are emergency calls made proactively by a human who recognizes an emergency or anomaly and uses a telephone or dedicated application.

[0045] Automated conversation is a type of conversation in which a system or AI autonomously generates a conversation using natural language processing technology and provides contextually appropriate responses, without requiring user input or operation. Automated conversation includes conversations between systems or AI (Artificial Intelligence) and other systems or AIs, as well as conversations between systems or AIs and humans. An automated conversation function is a function that enables automated conversation.

[0046] Direct conversation is a conversation that takes place between people. A direct conversation function is a function that enables direct conversation.

[0047] Emergency call information is essential for responding quickly and accurately to emergencies. This information may include the location of the emergency call device used, the device's unique information, the caller's location, the content of the call, and the caller's contact information. The content of the call may include the location of the person being reported to and information about their condition. Types of emergencies include incidents, accidents, fires, and medical emergencies. Those being reported to emergency services may include injured or ill individuals. Those being reported to emergency services may also be elderly or disabled individuals who require protection during disasters.

[0048] Emergency response request information is provided when necessary responses are requested in an emergency, and it is information that enables a swift and appropriate response. It also serves as information that allows the recipient to decide whether or not to comply with the request. Emergency response request information may include the current location of the person requiring emergency response, information about their condition, and information about the necessary emergency response. Necessary emergency responses may include life-saving and rescue. Depending on the necessary emergency response, the person requiring emergency response may be referred to as a person needing life-saving, a person needing rescue, or a patient.

[0049] 2. First Embodiment 2-1. Composition (Life-saving system 1) As shown in Figure 1, the life-saving system 1 includes an emergency call center device 100, an emergency call device 200, an emergency responder device 300, an organ donation management device 600, and an organ preservation device 700. The life-saving system 1 serves as the foundation for the immunosuppressant design system 800, which is shown as the seventh embodiment. Furthermore, the life-saving system 1 can also serve as the foundation for the pharmaceutical design system 900, which is shown as the eighth embodiment.

[0050] (Emergency call center device 100) The emergency call center device 100 is installed in the emergency call center (also called the emergency communication command center or disaster emergency information center) 1a. The emergency call center device 100 has a communication unit 110.

[0051] (Telephone section 110) The call unit 110 includes an automatic conversation unit 111 having an automatic conversation function and a direct conversation unit 112 having a direct conversation function.

[0052] (Automated conversation unit 111) The automated conversation unit 111 has a function (automatic emergency call determination function) that determines whether an emergency call is an automated emergency call (hereinafter also referred to as "automatic call") or a manual emergency call (hereinafter also referred to as "manual call").

[0053] Emergency calls include 110 (to the police), 119 (to the fire and ambulance services), and 118 (to the Japan Coast Guard). Emergency calls may also include #7119 (emergency telephone consultation), #8000 (pediatric emergency telephone consultation), and #9910 (road emergency hotline). Furthermore, emergency calls may also include emergency contact with power companies (electrical equipment failure, broken power lines, etc.) and gas companies (gas leaks, etc.).

[0054] The method by which the automated conversation unit 111 determines whether an emergency call is an automated or manual call is arbitrary.

[0055] The automated conversation unit 111 can determine whether an emergency call is automated or manual based on the content of the conversation with the caller of the emergency call. More specifically, the automated conversation unit 111 can send a question voice to the caller of the emergency call asking whether it is automated or manual (for example, "Is this an automated call?", "Are you an AI agent?", etc.), and determine whether an emergency call is automated or manual based on the response to the question (for example, "Yes," "No," "It is an automated call," "It is not an automated call," "I am an AI agent," "I am not an AI agent," "I am a human," etc.).

[0056] The automated conversation unit 111 may determine that an emergency call is an automated call if, for example, the emergency call contains information in a predefined format. Information in a predefined format is, for example, information in the format of { "event": "fire", "location": { "lat": 35.6895, "lon": 139.6917}, "time": "2025-03-24T21:45:00Z"}.

[0057] The automated conversation unit 111 can determine, for example, that an emergency call is an automated call if it includes trigger information from a sensor or program. Trigger information may include, for example, "SensorID:67890 has detected smoke."

[0058] The automated conversation unit 111 can determine, for example, that an emergency call is a manual call if it includes a call log indicating that the emergency call is a manual call. A call log indicating that it is a manual call is, for example, log information such as "UserID:12345 pressed the call button".

[0059] The automated conversation unit 111 can determine whether an emergency call is automated or manual, for example, based on the caller's emotions inferred from the words included in the emergency call. More specifically, the automated conversation unit 111 can determine that an emergency call is manual if it includes words that express the caller's feelings, such as "scared" or "anxious."

[0060] The automated conversation unit 111 can determine whether an emergency call is automated or manual, for example, based on the tone and tempo of the words included in the emergency call. More specifically, the automated conversation unit 111 may determine that an emergency call is manual if it is presumed that the caller is in a state of tension or fear. Examples of signs that indicate tension or fear include a trembling voice, rapid speech, and an increase in emphasized expressions.

[0061] The automated conversation unit 111 may determine that an emergency call is a manual call if, for example, the words included in the emergency call are subjectively interpreted descriptions of the situation. More specifically, the automated conversation unit 111 may determine that an emergency call is a manual call if the words included in the emergency call are a mixture of objective facts and the speaker's personal feelings or emotions. Examples of a mixture of objective facts and the speaker's personal feelings or emotions include, "There was a really loud noise, like I've never heard anything like it before," or "There's a fire burning, and it's really scary."

[0062] The automatic conversation unit 111 has a function to communicate with the emergency call device 200 that made the automatic emergency call in high-speed conversation mode when the emergency call is an automatic emergency call. "High-speed conversation mode" may be read as "conversation time reduction mode".

[0063] The automated conversation unit 111 has a function (emergency call information notification function) that notifies a designated emergency responder device 300 of emergency call information based on a conversation in high-speed conversation mode with the emergency call device 200.

[0064] The automated conversation unit 111 has a function (emergency call information notification function) that notifies a designated emergency responder device 300 of emergency call information based on a normal conversation with the emergency call device 200.

[0065] The method by which the automated conversation unit 111 notifies the emergency responder device 300 of emergency call information is arbitrary. The automated conversation unit 111 can notify the emergency responder device 300, which is the recipient of the emergency call information, in a manner appropriate to the device. Such appropriate methods may include notification by automated conversation, notification by text, notification by image, etc.

[0066] The automated conversation unit 111 has a function (person-to-person conversation function) that allows it to communicate with the caller (person) who made the manual emergency call in normal conversation mode when the emergency call is a manual call.

[0067] The automated conversation unit 111 has a function (response instruction function) that, based on a normal conversation with the caller (person), instructs the caller (person) on how to deal with the emergency using machine voice (natural language voice). The instructions for dealing with the emergency may include instructions for initial firefighting, first aid, evacuation, etc.

[0068] The automated conversation unit 111 has a function (telephone exchange function, conversation method change function) that, when an emergency call is a manual call, transfers the call with the caller (person) to the direct conversation unit 112.

[0069] The automated conversation unit 111 has a service-side AI agent (hereinafter also referred to as the "operator agent function" or simply the "operator agent") 101.

[0070] (Direct conversation section 112) The direct conversation unit 112 is a functional unit that allows a human operator to have a direct telephone conversation with the caller (person) who made the emergency call when the emergency call is a manual call. Direct conversation may include the intervention of an interpreter between the two people talking on the telephone. Interpretation may include human interpretation and computer interpretation (machine interpretation).

[0071] The direct conversation unit 112 relays voice communication between the emergency call device 200 and the operator terminal 300D in order to enable direct telephone conversation between the caller (person) and the human operator. The operator terminal 300D is a terminal used by the human operator (person) of the emergency call center 1a.

[0072] (High-speed conversation mode) High-speed conversation modes may include rapid-fire mode, abbreviation mode, machine language mode, etc.

[0073] Fast-talk mode is a mode that allows natural language voice calls to be completed in a shorter time than usual. Fast-talk mode can be achieved, for example, by performing processes such as removing silent parts of the conversation, blocking parts of the conversation, or thinning out parts of the conversation before converting the content of the conversation to be transmitted into audio data. Fast-talk mode can also be called high-speed playback mode.

[0074] Abbreviation mode is a mode in which communication is conducted using abbreviations, shortened words, compound words, etc. Abbreviations, shortened words, compound words, etc., may be expressions that are incomprehensible to humans, as long as a conversation can be established between the emergency call device 200 and the automatic conversation unit 111.

[0075] The machine language mode is a mode in which a conversation is established exclusively between the emergency call device 200 and the automated conversation unit 111. Examples of machine language modes include the giver link mode and the mathematical data sharing mode.

[0076] GiverLink mode is a high-speed conversation mode that uses a voice-based protocol for efficient communication between AI agents. In GiverLink mode, AI agents use a proprietary language that they can understand. This language can be binary code, a data-compressed format, or a special algorithm based on voice signals. GiverLink mode improves communication speed and accuracy, and also ensures noise immunity.

[0077] The mathematical data sharing mode is a high-speed conversation mode in which data exchange and sharing are based on mathematical methods and formats. Through conversations using the mathematical data sharing mode, AI agents can efficiently perform tasks by exchanging, for example, the weights of language models in matrix or vector format. The mathematical data sharing mode enables the transmission of structured information that is useful for coordination between machine learning models in the life-saving system 1 as a multi-AI agent system.

[0078] The mathematical data sharing mode is a high-speed conversation mode that significantly reduces call time by structuring and exchanging the weights and intermediate representations of the language model used for inference in matrix or vector form between the caller agent 201 and the operator agent 101, without using redundant natural language. Specifically, the caller agent 201 performs local inference regarding emergency response (e.g., determining the need for cardiopulmonary resuscitation) based on information integrating the situation at the scene, the patient's vital signs (biometric data stream), and the environmental context. It then encodes the inference result as the difference ΔW of the weight parameters or a feature vector V (e.g., a 1024-dimensional vector showing features extracted from the noise level at the scene and the speaker's emotional tone), and transmits this to the emergency call center device 100 in binary format or a data-compressed high-speed protocol. The operator agent 101 applies the weights WC of the central collaborative model to the received matrix / vector data to make an integrated decision. This data sharing enables critical information exchange with millisecond-order low latency, even under high load conditions such as congested communication bandwidth or large-scale disasters.

[0079] This mathematical data sharing mode, compared to conventional natural language voice communication, reduces communication delay to the millisecond order by exchanging the amount of information necessary for transmitting biometric data and environmental context in emergencies in a matrix-vector format that eliminates redundancy. This low-latency data exchange is an essential technology for critically accelerating the detection of signs of serious conditions in the health condition monitoring unit 60 and minimizing ischemic time from organ retrieval to restoration of blood flow.

[0080] (Emergency call device 200) Examples of emergency call devices 200 include automobiles 200A, smartphones 200B, smartwatches 200C, smart glasses 200D, etc. Emergency call devices 200 also include monitored person terminals (referred to as child GPS, GPS Talk, etc.) carried by the person being monitored (e.g., a child). An automobile (in other words, an in-vehicle computer) 200A is an example of a car device. Smartphones 200B and monitored person terminals are examples of mobile devices. Smartwatches 200C and smart glasses 200D are examples of wearable devices. Emergency call devices 200 may also include autonomous mobile robots (not shown). Autonomous mobile robots may include humanoid robots, animal-type robots, etc.

[0081] (Informant Agent 201) Each emergency call device 200 has a caller agent 201. The caller agent 201 is an AI agent (software agent) that operates within each emergency call device 200. The caller agent 201 is composed of multiple sub-agents and can handle complex tasks and dynamic environments. The caller agent 201 can automatically perform predetermined processes. These predetermined processes include automatic emergency call processing, which involves making an automatic call, and high-speed conversation processing, which involves making a call in high-speed conversation mode.

[0082] (Operator Agent 101) Operator agent 101 is an AI agent (software agent) that operates in the call unit 110. Operator agent 101 is composed of multiple sub-agents and can handle complex tasks and dynamic environments. Operator agent 101 can automatically perform predetermined processes related to emergency response requests. These predetermined processes include emergency call reception processing to receive emergency calls from the emergency call device 200, automatic call determination processing to determine whether the emergency call is an automatic call, call processing in high-speed conversation mode, call processing in normal conversation mode, emergency call information notification processing to notify a predetermined emergency responder device 300 of the emergency call information, response instruction processing to instruct the caller (person) on how to deal with the emergency, and conversation method change processing to change the conversation method.

[0083] The call processing in high-speed conversation mode includes pre-processing to initiate a call in high-speed conversation mode. This pre-processing includes sending a high-speed conversation request and receiving a high-speed conversation acceptance notification. Sending a high-speed conversation request is the process of sending a request for a call in high-speed conversation mode (hereinafter also referred to as "high-speed conversation request") to the emergency call device 200. Receiving a high-speed conversation acceptance notification is the process of receiving an acceptance notification from the emergency call device 200 indicating that the high-speed conversation request has been accepted.

[0084] The response instruction process is the process of instructing the caller (person) who made a manual emergency call to take action in the emergency situation, based on a phone call with the caller. These instructions may include, for example, instructions regarding cardiopulmonary resuscitation, first aid, firefighting, evacuation, accident response, etc.

[0085] (Emergency responder device 300) Examples of emergency response devices 300 include ambulances 300A, fire engines 300B, patrol cars 300C, operator terminals 300D, and terminals installed in police stations and police boxes. Ambulances (in other words, the on-board terminals in ambulances) 300A are an example of emergency vehicle devices. Fire engines (in other words, the on-board terminals in fire engines) 300B are an example of fire vehicle devices. Patrol cars (in other words, the on-board terminals in patrol cars) 300C are an example of police vehicle devices. Operator terminals 300D and terminals installed in police stations and police boxes are examples of emergency command system devices. Emergency response devices 300 may include autonomous mobile robots (not shown). Autonomous mobile robots may include humanoid robots, animal-type robots, etc. Emergency response devices 300 are also devices for organ preservation teams.

[0086] (Organ donation management device 600) As shown in Figure 2, the organ donation management device 600 includes a recipient information management unit 10, a potential recipient prediction unit 20, a potential recipient information management unit 30, a potential donor information management unit 40, a transplant suitability detection unit 50, a health status monitoring unit 60, and a donor selection unit 70. The recipient information management unit 10 has a recipient information management agent 11. The potential recipient prediction unit 20 has a potential recipient prediction agent 21. The potential recipient information management unit 30 has a potential recipient information management agent 31. The potential donor information management unit 40 has a potential donor information management agent 41. The transplant condition match detection unit 50 has a match detection agent 51. The health status monitoring unit 60 has a health status monitoring agent 61. The donor selection unit 70 has a donor selection agent 71.

[0087] The recipient information management unit 10, the potential recipient prediction unit 20, the potential recipient information management unit 30, the potential donor information management unit 40, the transplant suitability detection unit 50, the health status monitoring unit 60, and the donor selection unit 70 are realized, for example, through the cooperation of the hardware and software of the organ donation management device 600 (see Figure 5). The recipient information management unit 10, the potential recipient prediction unit 20, the potential recipient information management unit 30, the potential donor information management unit 40, the transplant suitability detection unit 50, the health status monitoring unit 60, and the donor selection unit 70 can access the EHR database of the EHR management organization and the databases of other information management organizations.

[0088] (Recipient Information Management Department 10) The Recipient Information Management Unit 10 is a functional unit that manages recipient information. The information managed includes the recipient's personal information, as well as the recipient's medical and genetic information. The recipient's personal information includes age, gender, contact information, recipient ID, etc. The recipient's medical and genetic information includes clinical data such as the underlying disease, complications, co-existing diseases, medical condition, medical urgency, presence or absence of infectious diseases, height, and weight, as well as information on the genetic background, such as blood type, HLA type (human leukocyte antigen), or gene polymorphisms related to tissue compatibility, which are central to organ compatibility assessment. In this embodiment, "genetic information" does not necessarily mean the complete DNA base sequence analysis data, but is a broad concept that includes genetically determined biological indicators such as HLA type and blood type. Based on this, the Recipient Information Management Unit 10 performs the "management of genetic information" of the present invention based on data obtained from normal transplant medical examinations.

[0089] (Recipient Information Management Agent 11) The recipient information management agent 11 is a component of the AI ​​agent that functions as part of the recipient information management unit 10. The recipient information management agent 11 provides the recipient detection agent 51 with information about the recipients it manages.

[0090] (Potential recipient prediction unit 20) The potential recipient prediction unit 20 is a functional unit that predicts potential recipients. The potential recipient prediction unit 20 has the function of predicting a group of individuals who are not currently registered on the organ transplant waiting list but who may be diagnosed, experience disease progression, or meet the eligibility criteria and be added to the waiting list within a certain period in the future. The function of predicting potential recipients can also be described as an organ demand prediction function. Therefore, the potential recipient prediction unit may be read as an organ demand prediction unit.

[0091] The potential recipient prediction unit 20 predicts potential recipients by combining multiple specific data with established statistical and computational methods such as hazard models, Markov models, and personalized machine learning.

[0092] Here, we will briefly explain examples of the data set input to the potential recipient prediction unit 20, the processing method (modeling approach) of the potential recipient prediction unit, and methods for ensuring prediction accuracy.

[0093] (Data set input to the potential recipient prediction unit 20) The potential recipient prediction unit 20 can use the following data as input data. Epidemiological data: Incidence and prevalence of diseases (e.g., CKD progression rate, end-stage heart failure prevalence, cirrhosis incidence), as well as disease trends by age, sex, and region. Medical utilization data: Outpatient and inpatient visits, referral rates to specialized departments such as nephrology and cardiac surgery, number of scheduled surgeries, and frequency of examinations (e.g., eGFR trends, liver function indicators). Diagnostic and severity indicators: Clinical scores (e.g., NYHA, MELD, eGFR stage), presence or absence of complications, and medication history. Medical access and system data: Transplant eligibility criteria, application of waiting list registration criteria (selection protocols), number of specialized facilities per region, OPO (organ donor organization) system, insurance coverage status, and referral pathways to the waiting list. Time-series registry data: historical trends in the waiting list, distribution of registered ages, average waiting time, etc. (e.g., national transplant registries). Sociodemographic data: demographic trends (progress of aging), socioeconomic conditions, and lifestyle factors (e.g., obesity rates, diabetes rates). Public health interventions and exogenous shocks: Data to reflect the impact of new treatments, implementation of screening policies, natural disasters, infectious disease outbreaks, and economic factors.

[0094] (Processing method in the potential recipient prediction unit 20) The potential recipient prediction unit 20 makes predictions based on the above data set using one or more of the following specific and feasible modeling approaches.

[0095] Cohort transformation + multistate survival analysis: The study uses local residents and patient populations as the starting point for the patient cohort and estimates the probability of transitions from "healthy → chronic disease → severe illness → waiting list registration" (multiple states) using a hazard model. In this process, age, sex, underlying diseases, and treatment history are incorporated as covariates, and the cumulative number of registrations for each period is estimated.

[0096] Machine learning-based prediction of disease onset and progression risk (personalized): This method utilizes time-series features from EHRs to predict rapid disease progression and transplant suitability in the short to medium term. For example, it can capture a sharp decline in eGFR in CKD patients and estimate the risk of dialysis initiation or transplant waiting list registration within one year. To ensure reliability for policy and clinical use, the interpretability of the model is ensured by using feature importance analysis methods (such as SHARP).

[0097] The potential recipient prediction unit 20 aims to individually predict the probability of an individual being placed on the organ transplant waiting list within a certain future period (e.g., within one year) based on the patient's EHR time-series data. This prediction model employs a hybrid ensemble AI model that combines a recurrent neural network (RNN) type, such as LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit), with a gradient boosting decision tree (XGBoost or LightGBM) in order to integrally process dynamic data (time-series data) and static data (baseline information) obtained from the EHR.

[0098] The features input to the prediction model shall include the following dynamic (time-series) and static data. Dynamic features are extracted from EHR records over the past year, with particular emphasis on the rate of change over the most recent three months. Specifically, in the renal domain (CKD), the absolute value of eGFR (estimated glomerular filtration rate) and the average rate of change over the most recent three months, the fluctuation pattern of serum creatinine levels, and the frequency and positive rate of urine protein and occult blood tests are used. In the cardiac domain (heart failure), the history of deterioration of the NYHA (New York Heart Association) score, the trends in BNP (brain natriuretic peptide) and NT-proBNP levels, and the rate of recent rapid increases are used. In the hepatic domain (cirrhosis), the trends in MELD / MELD-Na scores (end-stage liver disease model), bilirubin levels, and albumin levels are used. In addition, information on medical interventions, such as the history of initiation and discontinuation of specific critical disease medications, and changes in the frequency of hospitalization / outpatient visits, are also incorporated as features. Static features used include age, sex, presence or absence of underlying diseases (such as diabetes, hypertension, and dyslipidemia), HLA type information (if available), and geographical access information (distance to the transplant facility).

[0099] The latent recipient prediction unit 20 uses a hybrid ensemble AI model. This model consists of a first layer (LSTM / GRU layer) and a second layer (XGBoost layer). The first layer takes the dynamic (time-series) features mentioned above as input and generates a "latent vector representation of disease progression" from the time-series pattern. This layer plays a role in capturing the accelerating changes in disease progression. The second layer combines the latent vector representation obtained from the first layer with static features as input. This integrates the time-series progression pattern with the individual's fixed medical and demographic risk factors to make the final prediction.

[0100] In this model, the target variable is defined as "positive" (at risk) for individuals who have been newly registered on an organ-specific waiting list within one year of the prediction reference date. Regarding the loss function, considering the data imbalance where the number of registered individuals on the waiting list is significantly smaller than the number of unregistered individuals, the model uses Focal Loss instead of the usual binary cross-entropy to improve the prediction accuracy of the minority positive sample (patients who are actually registered). For validation, considering the time-series nature of the data, time-series cross-validation (a method that trains on historical data and validates on more recent data) is employed to ensure the model's practical future prediction performance. Furthermore, to ensure the clinical reliability of the prediction results, the importance of features is analyzed using methods such as SHAP (SHapley Additive exPlanations), ensuring interpretability so that physicians and administrators can understand which clinical indicators the model based its predictions on.

[0101] The output of this model includes the following predictive information about individual potential recipient candidates (patients who are not currently registered): The model outputs the risk probability of being placed on the waiting list within one year (from 0 to 100%) as its primary predictive metric. Furthermore, it identifies the top three dynamic / static features that had the greatest impact on the prediction probability (e.g., a sharp drop in eGFR change rate, deterioration of NYHA score) and outputs them as a list ranked by their primary contributing features.

[0102] The model is trained using large-scale linked registry data and anonymized historical EHR data. The training dataset is constructed as time-series data including static features (age, sex, blood type, HLA type, presence or absence of underlying diseases), dynamic features consisting of absolute values ​​of test results over the past three years and the mean rate of change and standard deviation of fluctuation over the most recent year, as well as event history (referral to a specialist, initiation / discontinuation of drug administration). The training data uses the fact that a patient was newly registered on the organ transplant waiting list within one year from the training target as the ground truth label. Training is performed using a two-stage transfer learning approach in which the LSTM / GRU unit learns time-series dependencies and extracts patterns of progression of disease severity, and then the XGBoost unit combines the output of the LSTM and static features to predict the final registration risk probability. Model performance is evaluated based on maintaining an AUC (Area Under Curve) of 0.85 or higher, and accuracy is ensured through periodic retraining.

[0103] The potential recipient prediction unit 20 employs microsimulation (individual-based) as one of its prediction methods. Microsimulation (individual-based) simulates each event, such as disease progression, diagnosis, referral, and registration, on an individual basis based on probability rules. This makes it possible to compare scenarios to see how policy changes, such as enhanced screening or improved referral flows, affect the increase or decrease in the potential list.

[0104] Furthermore, the potential recipient prediction unit 20 employs a combined synthetic cohort method and a Markov model (organ-specific) as another method for modeling the organ donation system. This method estimates the future pool of eligible recipients by using a Markov chain that reflects the eligibility criteria and transplant age for each organ. The system addresses the significant differences in transplant eligibility criteria across organs (e.g., dialysis-dependent pathways for kidneys, and being placed on a waiting list due to acute exacerbations in liver and heart) by modeling each organ separately.

[0105] (Method for ensuring prediction accuracy in the potential recipient prediction unit 20) As a method for ensuring prediction accuracy in the potential recipient prediction unit 20, firstly, registration bias correction is performed. Since registration to the waiting list is influenced by the referral rate and selection criteria, the observed number of registrations is only the lower limit of eligible individuals. Therefore, the distribution of the referral rate and selection criteria is incorporated into the model to correct for unobserved bias. Next, the prediction results are not simply point estimates, but are calculated and presented as prediction intervals (confidence intervals) using bootstrap and Bayesian posterior distributions to evaluate uncertainty. Furthermore, a mechanism for continuous monitoring and retraining is established to periodically evaluate prediction performance even after model implementation and to perform retraining to maintain and improve performance. In addition, to consider regional characteristics where there are large differences in the number of hospitals and OPO systems, a multi-layer model is used to estimate regional differences and a regional hierarchical structure is considered to ensure prediction accuracy for each region.

[0106] The integration of a large-scale linked registry with a real-time EHR pipeline enables the detection of early disease progression signals and more accurately predicts the risk of short-term waiting registration. Furthermore, molecular and biomarker integration, incorporating individual genetic risks and biomarkers (such as inflammation indicators), improves the accuracy of individualized disease progression risk predictions. In addition, automated screening functions, integrated with an operational flow that automatically detects patients exceeding risk thresholds within the EHR and encourages early referral, further enhance prediction accuracy.

[0107] (Potential recipient prediction agent 21) The potential recipient prediction agent 21 is a component of the AI ​​agent that functions as part of the potential recipient prediction unit 20. The potential recipient prediction agent 21 provides information on the predicted potential recipients to the potential recipient information management agent 31.

[0108] (Potential Recipient Information Management Unit 30) The Potential Recipient Information Management Unit 30 is a functional unit that manages information about potential recipients. The information managed includes the personal information of potential recipients, as well as their medical and genetic information. The personal information of potential recipients includes age, gender, contact information, and potential recipient ID. The medical and genetic information of potential recipients includes clinical data such as the underlying disease, complications, comorbidities, medical condition, medical urgency, physique (height, weight), and presence or absence of infectious diseases, as well as information on genetic background such as blood type, HLA type, or genetic predisposition.

[0109] Here, the "medical and genetic information" managed by the potential recipient information management unit 30 does not necessarily have to be actual measured values, but may be "virtual information (probabilistic genetic profile)" generated by the prediction model of the potential recipient prediction unit 20. For example, if actual HLA type data does not exist, the "most likely HLA type probability distribution" statistically estimated from the individual's race, region of origin, or family history information is managed as genetic information. This makes it possible for the system to start simulations of future suitability assessments and drug design (predictive drug discovery) even at a stage where definitive genetic information is not available. The function of managing information on potential recipients can also be said to be a function of managing information on organs for which demand is predicted (organs that will be needed in the future). Therefore, the potential recipient information management unit may be read as the demand-predicted organ information management unit.

[0110] (Potential recipient information management agent 31) The potential recipient information management agent 31 is a component of the AI ​​agent that functions as part of the potential recipient information management unit 30. The potential recipient information management agent 31 receives and manages the prediction information (demand) generated by the potential recipient prediction agent 21. The potential recipient information management agent 31 provides the information of the potential recipients it manages to the suitability detection agent 51.

[0111] (Potential Donor Information Management Department 40) The Potential Donor Information Management Unit 40 is a functional unit that manages information on potential donors who may become donors in the future. Specifically, the Potential Donor Information Management Unit 40 has the function of acquiring and managing registration information (such as the types of organs that can be donated) of individuals who have expressed their intention to donate organs, in conjunction with external databases, including the donor database. The information managed includes the personal information and medical information of potential donors. The personal information of potential donors includes age, gender, contact information, and potential donor ID. The medical information of potential donors includes information necessary for organ compatibility assessment, such as blood type, physique (height, weight), HLA type, medical history, pre-existing conditions, presence or absence of infectious diseases, and lymphocyte crossmatch results. The information of potential donors is updated periodically based on the individual's consent, through data linkage with a personal AI agent (caller agent 201) via the emergency call device 200.

[0112] The collection of biometric data (heart rate, brain waves, etc.) by the potential donor information management unit 40 is contingent on electronic consent (e-Consent) based on the individual's clear expression of will. This consent is obtained through the interface of the emergency call device 200 (caller agent 201) held by the potential donor, verified by, for example, an electronic signature or equivalent authentication means, and recorded in a distributed ledger technology (DLT)-based consent management module in the storage unit 650 along with an immutable timestamp. The potential donor can confirm the scope of their consent and immediately withdraw it at any time by operating the emergency call device 200. If the consent is withdrawn, the potential donor information management unit 40 immediately stops collecting the relevant biometric data and anonymizes the existing data.

[0113] (Potential donor information management agent 41) The potential donor information management agent 41 is a component of the AI ​​agent that functions as part of the potential donor information management unit 40. Based on the individual's consent, the potential donor information management agent 41 periodically updates the information it manages through data linkage via the informant agent 201. The potential donor information management agent 41 provides the information of the potential donors it manages to the transplant suitability detection unit 50 (suitability detection agent 51).

[0114] (Transplantation condition suitability detection unit 50) The transplant compatibility detection unit 50 is a functional unit that detects potential donors who meet the transplant conditions for each recipient or potential recipient in advance, based on the recipient information managed by the recipient information management unit 10, the potential recipient information managed by the potential recipient information management unit 30, and the potential donor information managed by the potential donor information management unit 40. This detection is performed continuously by algorithmically evaluating factors such as blood type match, HLA type compatibility, and similarity in body size.

[0115] The transplant suitability detection unit 50 continuously calculates a Comprehensive Suitability Score (CSS) for all detected potential donor-recipient pairs to maximize the success rate of transplantation and long-term prognosis.

[0116] The CSS is determined by multiplying the scores of three key endpoints—the medical criteria score (S_Med), the organ / timing optimization score (S_Org), and the recipient prognosis score (S_Prog)—by weighting them according to their respective importance, and then summing them up.

[0117] The organ / timing optimization score (S_Org) evaluates the optimal timing of organ donation based on real-time dynamic data from the health monitoring unit 60, including the risk of sudden changes in the potential donor's health (low probability of transition) and the possibility of preserving the extracted organ (short ischemic time).

[0118] The recipient prognosis score (S_Prog) uses the AI ​​model of the potential recipient prediction unit 20 to predict and evaluate the recipient's long-term graft survival rate after transplantation, as well as the post-transplant prognosis considering the urgency and duration of the waiting period, if this organ is transplanted.

[0119] The Medical Reference Score (S_Med) assesses static, basic medical fit, such as blood type, HLA type, and body size similarity.

[0120] The donor selection unit 70 identifies the recipient with the highest CSS score as the best match when an opportunity for organ donation arises, such as when a potential donor reaches a critical health condition. This identification triggers the initiation of a rapid organ donation process.

[0121] (Suitable person detection agent 51) The match detection agent 51 is a component of the AI ​​agent that functions as part of the transplant condition match detection unit 50. The match detection agent 51 receives recipient information from the recipient information management agent 11. The match detection agent 51 receives potential recipient information from the potential recipient information management agent 31. The match detection agent 51 receives potential donor information from the potential donor information management agent 41. Based on the information received, the match detection agent 51 pre-detects potential donors that meet the transplant conditions for each recipient or potential recipient, and provides the detection results to the donor selection agent 71.

[0122] (Health Monitoring Department 60) The health monitoring unit 60 is a functional unit that continuously monitors the health status of each potential donor. The health monitoring unit 60 receives biometric data (heart rate, respiratory irregularities, blood pressure, etc.) transmitted from the caller agent 201 of the emergency call device 200 (car device, mobile device, wearable device, etc.) associated with the potential donor via the communication unit 160, and evaluates whether the potential donor is in a serious health condition.

[0123] If the emergency notification device 200 is a device (head-worn device) capable of measuring the brainwaves of a potential donor, the health monitoring unit 60 receives biometric data, including changes in the potential donor's brainwaves, via the communication unit 160. The health monitoring unit 60 then uses this biometric data, including changes in the potential donor's brainwaves, as vital signs to detect early signs of a serious condition that could lead to brain death in the potential donor.

[0124] (Health monitoring agent 61) The health status monitoring agent 61 is a component of the AI ​​agent that functions as part of the health status monitoring unit 60. The health status monitoring agent 61 works closely with the operator agent 101 and the caller agent 201. An advanced AI agent-to-agent communication protocol, including a high-speed conversation mode, is used between these agents to respond to particularly urgent situations. This protocol goes beyond mere data transfer, enabling the acquisition and analysis of information regarding the health status of potential donors in an extremely efficient and near real-time manner.

[0125] Specifically, the health monitoring agent 61 utilizes highly efficient information generated or collected while the operator agent 101 is communicating with the caller agent 201 in high-speed conversation mode.

[0126] In this embodiment, "highly efficient information" refers to structured real-time health risk data acquired and analyzed with low latency between the caller agent 201 mounted on the emergency call device 200 and the operator agent 101 of the emergency call center device 100 via a high-speed conversation mode (giver link mode, mathematical data sharing mode, etc.). The real-time health risk data is a collection of data obtained by AI that instantly structures voice analysis of the call content, transcribed conversation logs, and related biometric data, and converts them into potential health risk indicators.

[0127] The health monitoring agent 61 obtains structured information, which is acquired as features and situation context vectors extracted from the real-time biometric data stream, and evaluates it as an anomaly score (CSS_Org).

[0128] The features extracted from the real-time biometric data stream are derived from biosignals such as electroencephalograms (EEGs) and electrocardiograms (ECGs) of potential donors. Specifically, these include rapid fluctuations in low-frequency components indicating impaired brain function and aperiodicity in heart rate variability indicating a cardiovascular crisis.

[0129] The structured information obtained as a situation context vector includes the type of emergency self-diagnosed by the caller agent 201, and environmental sensor data (impact, acceleration, etc.).

[0130] The extracted features are input into an individualized risk model and converted into risk indicators such as the probability of difficulty in recovering brain function (P_BrainLoss) and the probability of irreversible organ damage (P_OrganDamage). The health status monitoring agent 61 integrates these risk indicators with the situational context to calculate an anomaly score (CSS_Org) indicating the life-threatening condition of potential donors in real time and transmits it to the donor selection unit 70.

[0131] The donor selection unit 70 determines that a "predetermined health condition" (donor selection trigger) has been reached when, in addition to the potential donor meeting the transplantation criteria, all of the following composite threshold conditions are met.

[0132] The composite threshold condition includes (1) to (3) below. (1) AI risk assessment: The probability of difficulty in recovering brain function (P_BrainLoss) continuously exceeds a predetermined threshold (e.g., 90%). (2) Confirmation of life-saving measures: The situation context is linked to whether life-saving measures are underway at the scene or whether the predetermined time limit has already been exceeded. (3) Organ preservation conditions: The abnormal score (CSS_Org) is below the critical level that causes irreversible damage to organ function, and the organ is in a state where it is medically possible to donate the organ.

[0133] Objective judgments based on these combined conditions ensure the maximum continuation of life-saving efforts and a subsequent rapid and efficient transition to an organ donation process.

[0134] The donor selection unit 70 selects a potential donor as a donor only if all of the following composite threshold conditions are met: AI risk assessment that P_BrainLoss continuously exceeds a predetermined threshold, exceeding the limit time for on-site life-saving measures, and organ preservation conditions that CSS_Org is below the critical level for irreversible damage, based on the progress of life-saving measures and the results of biometric data analysis. This objective and comprehensive judgment criterion is at the core of this system, which balances bioethical considerations (maximum continuation of life-saving efforts) and medical efficiency (maximization of organ donation opportunities).

[0135] The "Probability of Impossible Brain Function Recovery (P_BrainLoss)" and "Probability of Irreversible Organ Damage (P_OrganDamage)" calculated by the health monitoring agent 61 are calculated in real time using the following algorithm. In calculating P_BrainLoss, a feature combining the power spectral density of alpha and theta waves extracted from electroencephalogram (EEG) data, and mean arterial pressure (MAP) and pupillary reflex data is input to a pre-trained recurrent neural network (RNN). The RNN is trained using data from past severe trauma patients and outputs a temporal probability distribution leading to irreversible brain function failure. P_BrainLoss is calculated as the complement of the "probability that brain function will recover within one hour if life-saving measures are continued at the scene." On the other hand, in the calculation of P_OrganDamage, the asymmetry of the periodicity of the QRS complex in the electrocardiogram (ECG), the rate of rapid decrease in blood pressure (BP), and ischemic time are used as the main input features. These features are input into a Markov decision process model that predicts the rate of progression of irreversible ischemic damage to a specific organ (e.g., kidney, liver), and P_OrganDamage is defined as the "probability that the organ's function at the time of retrieval reaches a critical level that makes it unsuitable for transplantation," as predicted by the model.

[0136] The health monitoring agent 61, by continuously monitoring and analyzing this information flow, has the ability to detect signs of a rapid deterioration in the health of a potential donor or a life-threatening situation much earlier than a human would notice. This early detection capability enables decisively accelerating the direction of rapid medical intervention and the next steps in the organ donation process (e.g., preparing to dispatch a specialized medical team, preparing to match a suitable donor and recipient), which is essential in maximizing the chances of saving a donor's life or preserving available organs. This entire collaborative and analytical system forms the foundation for achieving the best possible medical outcomes while ensuring that organ donation opportunities are not missed and while adhering to ethical constraints.

[0137] (Donor Selection Section 70) The donor selection unit 70 is a functional unit that selects potential donors who have been detected as suitable by the transplant suitability detection unit 50 and who have reached a predetermined health condition (e.g., a severe condition in which recovery of brain function is predicted to be extremely difficult) by the health condition monitoring unit 60. This selection serves as a trigger to initiate a rapid organ donation process and is used to request the emergency response device 300, etc., to prepare for appropriate organ preservation and transport.

[0138] (Donor Selection Agent 71)

[0139] The donor selection agent 71 is a component of the AI ​​agent that functions as part of the donor selection unit 70. The donor selection agent 71 receives information on suitability to transplant conditions from the match detection agent 51. The donor selection agent 71 receives evaluation results from the health status monitoring agent 61 to determine whether the potential donor has reached a predetermined health state. Based on the information received from the match detection agent 51 and the health status monitoring agent 61, the donor selection agent 71 makes a decision on donor selection. The donor selection agent 71 cooperates with the operator agent 101 (emergency call center device 100) to notify emergency response devices of emergency response request information containing information on the selected donor and matched recipient.

[0140] (Organ preservation device 700) The organ preservation device 700 (Figure 1) functions as an organ preservation unit 80 for preserving organs extracted from a donor in a way that allows them to be restored to a transplantable state. The organ preservation device 700 is located in an organ preservation facility separate from the emergency call center device 100.

[0141] The organ preservation device 700 is primarily used in the on-demand donation process (the process of preserving organs until a suitable recipient becomes available). In other words, the organ preservation device 700 is a device for preserving organs extracted from a donor in a state that can be revived (restored, recovered) to a transplantable state until a suitable recipient becomes available. Specifically, the organ preservation device 700 preserves organs in a state that can be revived (restored, recovered) to a transplantable state by methods such as ice freezing, supercooling, and vitrification.

[0142] (Configuration of emergency call center device 100) As illustrated in Figure 3, the emergency call center device 100 is composed of a control unit 120, an information display unit 130, an input unit 140, a storage unit 150, a communication unit 160, and the like. Each functional part of the emergency call center device 100 is realized through the cooperation of hardware and software.

[0143] The control unit 120 includes a processor 121 and a memory 122. The processor 121 is responsible for the overall control of the emergency call center device 100. The processor 121 can be implemented by, for example, a CPU (Central Processing Unit), MCU (Micro Controller Unit), MPU (Micro Processor Unit), GPU (Graphics Processing Unit), NPU (Neural Processing Unit), or TPU (Tensor Processing Unit). The memory 122 includes, for example, ROM (Read Only Memory), RAM (Random Access Memory), and flash ROM. For example, flash ROM and ROM store various programs (including data used by the programs), and RAM is used as the work area of ​​the processor 121. Programs stored in memory 122 are loaded into the processor 121, causing the processor 121 to execute the coded processes. The programs stored in memory 122 include agent programs for implementing one or more operator agents 101. The agent programs include agent learning programs.

[0144] The information display unit 130 is a functional unit for presenting various information to operators of the emergency call center device 100 (hereinafter also referred to as "center administrators"). The information display unit 130 includes a display 131 and a speaker 132. The display 131 visually presents various information to the center administrators. The speaker 132 audibly presents various information to the center administrators. The display 131 displays the results of processing performed by the control unit 120, etc. The speaker 132 outputs the results of processing performed by the control unit 120, etc., as audio. The center administrators may include system administrators (people) responsible for monitoring and maintaining the emergency call center 1a system, data administrators (people) responsible for managing and analyzing call history and system logs, human operators responsible for responding to emergency calls, etc.

[0145] The input unit 140 is a functional unit for the center administrator to input information to the emergency call center device 100. The input unit 140 includes an operation input unit 141 that accepts operations from the center administrator, an image input unit (camera) 142 for inputting images, and an audio input unit (microphone) 143 for inputting voice. If a touch panel is used for the operation input unit 141, the operation input unit 141 can be integrated with the display 131. Furthermore, if the operation input unit 141 can accept voice operations, the operation input unit 141 and the audio input unit 143 can be integrated.

[0146] The storage unit 150 includes auxiliary storage devices such as an HDD (Hard Disk Drive) or SSD (Solid State Drive) and stores various types of data. The storage unit 150 may also store programs executed by the processor 121 (including data used by the programs).

[0147] The communication unit 160 is a functional unit that transmits and receives data with the organ donation management device 600 and external devices via the communication network 400. The communication network 400 may include multiple types of networks of different scales, such as PAN (Personal Area Network), LAN (Local Area Network), MAN (Metropolitan Area Network), WAN (Wide Area Network), and the Internet. The communication network 400 may also include multiple types of networks using different communication methods, such as wired networks and wireless networks.

[0148] (Configuration of Emergency Call Device 200) As illustrated in Figure 4, the emergency call device 200 is composed of a control unit 220, an information display unit 230, an input unit 240, a storage unit 250, a communication unit 260, and the like. Each functional part of the emergency call device 200 is realized through the cooperation of hardware and software.

[0149] The control unit 220 includes a processor 221 and a memory 222. The processor 221 is responsible for the overall control of the emergency call device 200. The processor 221 is implemented by, for example, a CPU, MCU, MPU, GPU, NPU, TPU, etc. The memory 222 includes, for example, ROM, RAM, and flash ROM. For example, flash ROM and ROM store various programs (including data used by the programs), and RAM is used as the work area of ​​the processor 221. Programs stored in memory 222 are loaded into the processor 221, causing the processor 221 to execute the coded processes. The programs stored in memory 222 include an agent program for implementing the caller agent 201. The agent program includes an agent learning program.

[0150] The information display unit 230 is a functional unit for presenting various information to the user of the emergency call device 200. The information display unit 230 includes a display 231 and a speaker 232. The display 231 visually presents various information to the user of the emergency call device 200. The speaker 232 audibly presents various information to the user of the emergency call device 200. The display 231 displays the results of processing performed by the control unit 220, etc. The speaker 232 outputs the results of processing performed by the control unit 220, etc., as audio.

[0151] The input unit 240 is a functional unit for the user to input information to the emergency call device 200. The input unit 240 includes an operation input unit 241 that accepts user input, an image input unit (camera) 242 for inputting images, and an audio input unit (microphone) 243 for inputting voice. When a touch panel is used for the operation input unit 241, the operation input unit 241 is integrated with the display 231. Furthermore, if the operation input unit 241 can accept voice input, the operation input unit 241 and the audio input unit 243 can be integrated. Hereinafter, the operation input unit 241 integrated with the display 231 will be referred to as the touch panel 244.

[0152] The storage unit 250 includes, for example, an auxiliary storage device such as an HDD or SSD, and stores various types of data. The storage unit 250 may also store programs executed by the processor 221 (including data used by the programs).

[0153] The communication unit 260 is a functional unit that transmits and receives data with external devices via the communication network 400.

[0154] In addition to the functional units illustrated in Figure 3, the emergency call device 200 also has functional units to realize its intended function as a car device, mobile device, wearable device, etc. For example, if the emergency call device 200 is a car 200A, the emergency call device 200 has functional units to realize the function of a car device. If the emergency call device 200 is a smartphone 200B or a monitored person's terminal, the emergency call device 200 has functional units to realize the function of a mobile device. If the emergency call device 200 is a smartwatch 200C or smart glasses 200D, the emergency call device 200 has functional units to realize the function of a wearable device.

[0155] (Configuration of organ donation management device 600) As illustrated in Figure 5, the organ donation management device 600 is composed of a control unit 620, an information display unit 630, an input unit 640, a storage unit 650, a communication unit 660, and the like. Each functional unit of the organ donation management device 600 is realized through the cooperation of hardware and software.

[0156] The control unit 620 includes a processor 621 and a memory 622. The processor 621 is responsible for the overall control of the organ donation management device 600. The processor 621 can be implemented by, for example, a CPU, MCU, MPU, GPU, NPU, TPU, etc. The memory 622 includes, for example, ROM, RAM, and flash ROM. For example, flash ROM and ROM store various programs (including data used by the programs), and RAM is used as the work area of ​​the processor 621. Programs stored in the memory 622 are loaded into the processor 621, causing the processor 621 to execute the coded processes.

[0157] The programs stored in memory 622 include a recipient information management program, a potential recipient prediction program, a potential recipient information management program, a potential donor information management program, a transplant suitability assessment program, a health status monitoring and biometric data analysis program, and a donor selection and process initiation trigger program.

[0158] The recipient information management program is a program for managing the recipient's personal information and medical information (blood type, HLA type, medical condition, urgency, etc.). In other words, the recipient information management program is a program that realizes the functions of the recipient information management unit 10.

[0159] The potential recipient prediction program is a program that uses statistical and computational methods such as hazard models, Markov models, and personalized machine learning to predict a group of individuals (potential recipients) who are likely to be added to the waiting list within a certain period in the future. In other words, the potential recipient prediction program is a program that implements the functions of the potential recipient prediction unit 20.

[0160] The potential recipient information management program is a program for managing predictive information (including hypothetical information) of potential recipients. In other words, the potential recipient information management program is a program for realizing the functions of the potential recipient information management unit 30.

[0161] The potential donor information management program manages the personal and medical information (blood type, HLA type, medical history, etc.) of potential donors and periodically updates this information through data linkage via the emergency call device 200 (caller agent 201), etc. In other words, the potential donor information management program is a program that realizes the functions of the potential donor information management unit 40.

[0162] The transplant compatibility evaluation program is a program that algorithmically and continuously evaluates and detects the compatibility (blood type match, HLA type match, similarity in physique, etc.) between recipients, potential recipients, and potential donors. In other words, the transplant compatibility evaluation program is a program that realizes the functions of the transplant compatibility detection unit 50.

[0163] The health status monitoring and biometric data analysis program receives and analyzes biometric data (heart rate, blood pressure, electroencephalogram, etc.) transmitted from the potential donor's emergency call device 200, and continuously monitors and evaluates whether or not the donor is in a serious health condition. In other words, the health status monitoring and biometric data analysis program is a program that enables the functions of the health status monitoring unit 60.

[0164] The donor selection and process initiation trigger program is a program that activates a trigger to select a potential donor who has been confirmed to be suitable and has reached a predetermined critical health condition, and to initiate a rapid organ donation process. In other words, the donor selection and process initiation trigger program is a program that realizes the functions of the donor selection unit 70.

[0165] Furthermore, the programs stored in memory 622 include a recipient information management agent program, a potential recipient prediction agent program, a potential recipient information management agent program, a potential donor information management agent program, a suitable candidate detection agent program, a health status monitoring agent program, a donor selection agent program, and the like.

[0166] The recipient information management agent program is a program that implements the functions of the recipient information management agent 11. The potential recipient prediction agent program is a program that implements the functions of the potential recipient prediction agent 21. The potential recipient information management agent program is a program that implements the functions of the potential recipient information management agent 31. The potential donor information management agent program is a program that implements the functions of the potential donor information management agent 41. The conformer detection agent program is a program that implements the functions of the conformer detection agent 51. The health status monitoring agent program is a program that enables the functions of the health status monitoring agent 61 (including high-speed data exchange with operator agent 101, etc.). The donor selection agent program is a program designed to implement the functions of the donor selection agent 71.

[0167] Furthermore, in order for the organ donation management device 600 to function as part of the life-saving system 1, the memory 622 also stores AI agent communication protocol programs, system linkage and data access programs, and other similar programs.

[0168] The AI ​​agent-to-agent communication protocol program is designed to enable efficient data exchange between the operator agent 101, the caller agent 201, and other agents. If an emergency response agent (e.g., emergency response agent 301) is present, the AI ​​agent-to-agent communication protocol program includes a program to enable efficient data exchange between the operator agent 101, the caller agent 201, and the emergency response agent.

[0169] System integration and data access programs are programs that enable interfaces and processes for accessing databases of external EHR management organizations and other information management organizations.

[0170] The information display unit 630 is a functional unit for presenting various information to operators of the organ donation management device 600 (hereinafter also referred to as "organ donation managers"). The information display unit 630 includes a display 631 and a speaker 632. The display 631 visually presents various information to organ donation managers. The speaker 632 audibly presents various information to center administrators. The display 631 displays the results of processing performed by the control unit 620, etc. The speaker 632 outputs the results of processing performed by the control unit 620, etc., as sound. Organ donation managers may include system administrators (people) responsible for monitoring and maintaining the organ donation management device 600 system, data administrators (people) responsible for managing and analyzing various histories and system logs related to organ donation, and human operators responsible for various actions to realize organ donation.

[0171] The input unit 640 is a functional unit for organ donation managers to input information to the organ donation management device 600. The input unit 640 includes an operation input unit 641 that accepts operations from the organ donation manager, an image input unit (camera) 642 for inputting images, and an audio input unit (microphone) 643 for inputting voice. If a touch panel is used for the operation input unit 641, the operation input unit 641 can be integrated with the display 631. Furthermore, if the operation input unit 641 can accept voice operations, the operation input unit 641 and the audio input unit 643 can be integrated.

[0172] The storage unit 650 includes, for example, an auxiliary storage device such as an HDD or SSD, and stores various types of data. The storage unit 650 may also store programs executed by the processor 621 (including data used by the programs).

[0173] The communication unit 660 is a functional unit that transmits and receives data with the emergency call center device 100 and external devices via the communication network 400.

[0174] (Processes performed by the emergency call center device 100 and the emergency call device 200) Referring to Figures 6 to 12, an example of the processing performed by the emergency call center device 100 and the emergency call device 200 will be described.

[0175] (Processed by the emergency call center system) As shown in Figure 6, the emergency call center device 100 receives an emergency call (step S11) and performs emergency call response processing (step S12).

[0176] As shown in Figure 7, in the emergency call response process (step S12), the emergency call center device 100 determines whether the emergency call is an automatic call or not (step S13: automatic emergency call determination process). If the emergency call center device 100 determines that the emergency call is an automatic call (YES in step S13), it executes the automatic call response process (step S100, Figure 8). On the other hand, if it determines that the emergency call is a manual call (NO in step S13), the emergency call center device 100 executes the manual call response process (step S110, Figure 11).

[0177] As shown in Figure 8, in the automatic notification response process (step S100), the emergency notification center device 100 requests the emergency notification device 200 to communicate in high-speed conversation mode (step S101: high-speed conversation request process).

[0178] The emergency call center device 100 determines whether the emergency call device 200 has accepted the request based on the response from the emergency call device 200 (step S102).

[0179] If the emergency call center device 100 determines that the emergency call device 200 has accepted the request (YES in step S102), it performs high-speed conversation processing (step S120, Figure 9).

[0180] On the other hand, if the emergency call device 200 determines that it will not accept the request (NO in step S102), it performs normal conversation processing (step S130, Figure 10).

[0181] As shown in Figure 9, in high-speed conversation processing (step S120), the emergency call center device 100 communicates with the emergency call device 200 in high-speed conversation mode (step S121: conversation processing in high-speed conversation mode).

[0182] Then, based on the communication with the emergency call center device 200, the emergency call center device 100 notifies a designated emergency responder device 300 of the emergency call information (Step S122: Emergency call information notification process). As mentioned above, the method of notifying the emergency call information is arbitrary.

[0183] For example, if a fire truck or ambulance is needed, the emergency call center device 100 can, based on communication with the emergency call device 200, automatically notify a human operator at the emergency call center 1a via operator terminal 300D of emergency call information such as the location and situation at the scene and the condition of the injured or ill person.

[0184] Furthermore, the emergency call center device 100 can, for example, notify emergency response devices 300 (such as ambulances 300A, fire trucks 300B, patrol cars 300C, etc.) located closer to the current location of the emergency call device 200 using any method of choice.

[0185] As shown in Figure 10, in normal conversation processing (step S130), the emergency call center device 100 communicates with the emergency call device 200 in normal conversation mode (step S131: conversation processing in normal conversation mode).

[0186] Then, based on the communication with the emergency call center device 200, the emergency call center device 100 notifies a designated emergency responder device 300 of the emergency call information (Step S132: Emergency call information notification process).

[0187] As shown in Figure 11, in the manual call response process (step S110), the emergency call center device 100 communicates with the caller (person) who made the manual call using the emergency call device 200 in normal conversation mode (step S111: person-to-person conversation process).

[0188] Then, based on the content of the conversation with the caller (person), the emergency call center device 100 attempts to give the caller instructions on how to deal with the emergency using an automated voice (Step S112: Action instruction processing).

[0189] Next, the emergency call center device 100 determines, based on the response from the caller (person), whether the caller understood the automated voice instructions regarding how to deal with the emergency (Step S113: Understanding Determination Process).

[0190] The method by which the emergency call center device 100 determines whether the caller (person) understood the machine-voiced instructions regarding how to respond to the emergency is arbitrary. For example, the determination can be based on whether the caller (person) performed the expected actions in accordance with the machine-voiced instructions. Alternatively, the determination can be based on the caller's (person's) response to confirmation questions after the instructions. The caller's (person's) response can also be analyzed using natural language processing technology, and the determination can be based on the analysis results.

[0191] If the emergency call center device 100 determines that the caller (person) does not understand the automated voice instructions on how to deal with the emergency (NO in step S113), it changes the communication method with the caller to a direct conversation method (step 114) (conversation method change process).

[0192] More specifically, the emergency call center device 100 relays voice communication between the emergency call device 200 and the operator terminal 300D. This allows the caller who made the emergency call to communicate directly with the human operator at the emergency call center 1a.

[0193] (Processing by emergency call device 200) As shown in Figure 6, the emergency call device 200 makes an emergency call (step S21: emergency call processing) and performs emergency call device-side call processing (step S22, Figure 12).

[0194] As shown in Figure 12, in the emergency call device side call processing (step S22), the emergency call device 200 determines whether the emergency call it made is an automatic call or not (step S23).

[0195] If it is determined that the emergency call is an automatic call (YES in step S23), the emergency call device 200 determines whether or not it has received a high-speed conversation request from the emergency call center device 100 (step S24: high-speed conversation request reception determination process).

[0196] If it is determined that no request for high-speed conversation has been received (NO in step S24), the emergency call device 200 communicates with the emergency call center device 100 in normal conversation mode (step S27: normal conversation processing).

[0197] On the other hand, if it is determined that a request for high-speed conversation has been received (YES in step S24), the emergency call device 200 determines whether or not it is possible to make a call in high-speed conversation mode (step S25: high-speed conversation feasibility determination process).

[0198] If the emergency call device 200 determines that it is possible to make a call in high-speed conversation mode (YES in step S25), it will make a call with the emergency call center device 100 in high-speed conversation mode (step S26: high-speed conversation processing). When initiating a call in high-speed conversation mode, the emergency call device 200 sends an acceptance notification to the emergency call center device 100 and then starts the call in high-speed conversation mode.

[0199] On the other hand, if it is determined that it is not possible to make a call in high-speed conversation mode (NO in step S25), the emergency call device 200 will make a call with the emergency call center device 100 in normal conversation mode (step S27: normal conversation processing).

[0200] Furthermore, if it is determined that the emergency call is a manual call (NO in step S23), the emergency call device 200 changes the communication method with the emergency call center device 100 to a direct communication method (step S28: direct conversation processing). In this case, the caller (person) who made the emergency call using the emergency call device 200 and the human operator at the emergency call center 1a can communicate directly.

[0201] 2-3. Operation of the First Embodiment The first embodiment of the life-saving system 1 is based on multi-AI agent technology and exerts the following effects in order to improve the efficiency of life-saving medical care, including organ transplantation.

[0202] (Expediting responses to emergency calls) In the first embodiment of the life-saving system 1, the emergency call center device 100 can communicate with the emergency call device 200 that made the automatic emergency call (step S121 in Figure 9) in high-speed conversation mode when the emergency call is an automatic call (YES in step 13 in Figure 7). By communicating in high-speed conversation mode, it is possible to receive the emergency call in a shorter time and notify the designated emergency responder device 300 of the emergency call information in a shorter time compared to communicating in normal mode (step S122 in Figure 9).

[0203] Therefore, this life-saving system 1 can suppress the decrease in the response speed to emergency calls when the number of automatic calls increases.

[0204] More specifically, in the event of a large-scale disaster such as an earthquake, tsunami, or typhoon, a very large number of automatic notification devices issue automatic notifications almost simultaneously. Conventional life-saving systems cannot efficiently process these automatic notifications due to the concentration of emergency calls and a shortage of operators (communication commanders) to respond to them. In contrast, this life-saving system 1 can efficiently process such a large number of automatic notifications even when a very large number of automatic notification devices issue automatic notifications almost simultaneously, through high-speed serial processing by a high-capacity operator agent 101 or parallel processing by multiple operator agents 101.

[0205] Furthermore, in the first embodiment of the life-saving system 1, the emergency call center device 100 can communicate in normal conversation mode with the caller (person) who has made a manual call using the emergency call device 200 (step S111 in Figure 11).

[0206] The emergency call center device 100 can then attempt to give instructions to the caller (person) regarding how to deal with the emergency situation using an automated voice, based on the content of the conversation with the caller (person) (step S112 in Figure 11).

[0207] Therefore, according to the first embodiment of the life-saving system 1, in normal conversation mode, it is possible to give instructions to the caller (person) on how to deal with the emergency using machine voice.

[0208] Furthermore, in the first embodiment of the life-saving system 1, if the emergency call center device 100 determines that the caller (person) does not understand the machine-voiced instructions (explanation, persuasion) on how to deal with the emergency (NO in step S113 of Figure 11), it changes the communication method with the caller to a direct conversation method (step S114 of Figure 11).

[0209] Therefore, according to the first embodiment of the life-saving system 1, for callers (people) who cannot understand machine-generated instructions in normal conversation mode, a human operator at the emergency call center 1a can provide instructions (explanations, persuasion) using a human voice (the voice of the human operator).

[0210] Thus, in the first embodiment of the life-saving system 1, the emergency call center device 100 can communicate with the emergency call device 200 that made the automatic emergency call (YES in step S13 of Figure 7) in high-speed conversation mode (step S121 of Figure 9), and if the emergency call is a manual emergency call, the person who made the manual emergency call can communicate directly with the human operator (step S114 of Figure 11).

[0211] Therefore, according to the first embodiment of the life-saving system 1, while suppressing a decrease in the response speed to emergency calls when automatic calls increase, it is possible for the caller (person) and the human operator to communicate directly when the emergency call is a manual call.

[0212] Incidentally, the life-saving system 1 of the first embodiment can be considered as a multi-AI agent learning system that includes one or more (a small or large number) operator agents 101 and multiple (a large number) caller agents 201.

[0213] Therefore, by using, for example, a mathematical data sharing mode for communication between AI agents in the life-saving system 1 as a multi-AI agent learning system, data exchange between AI agents can be made more efficient, distributed learning can be accelerated, and the overall performance of the life-saving system 1 can be rapidly improved. This makes it possible to more effectively suppress the decrease in response speed to notifications when the number of automated notifications increases.

[0214] (Maintaining up-to-date information) The recipient information management unit 10, the potential recipient information management unit 30, and the potential donor information management unit 40 always maintain up-to-date information.

[0215] (Forecasting potential demand and continuous hypothetical suitability assessment) The potential recipient prediction unit 20 continuously predicts future organ demand (potential recipients) based on machine learning models (such as cohort transformation models and personalized machine learning models) using real-time EHRs and large-scale registry data.

[0216] The transplant suitability detection unit 50 continuously evaluates the compatibility between a large number of recipients and potential donors based on recipient information managed by the recipient information management unit 10 and potential donor information managed by the potential donor information management unit 40, and selects pairs with high compatibility.

[0217] Furthermore, the transplant suitability detection unit 50 continuously evaluates the suitability of potential donors to future recipients, even if there are no recipients at present, based on information on potential recipients managed by the potential recipient information management unit 30 and information on potential donors managed by the potential donor information management unit 40, and selects pairs with high suitability.

[0218] (Advanced monitoring of health status) When an abnormal situation occurs in the emergency call device 200 of a potential donor, and the caller agent 201 issues an automatic emergency call, the operator agent 101 of the emergency call center device 100 communicates with the caller agent 201 in high-speed conversation mode. The health status monitoring agent 61 works closely with the operator agent 101 and the caller agent 201 in high-speed conversation mode to collect emergency call information, including biometric data (EEG, heart rate, etc.), location information, and circumstances of the incident of the seriously injured or ill person, in an extremely short time without human intervention.

[0219] The health monitoring unit 60 immediately analyzes the collected emergency call information and determines whether the potential donor has reached a specified health condition (e.g., severe trauma, cardiac arrest, severe decline in brain function).

[0220] (Rapid donor selection and initiation of the organ donation process) If a potential donor falls into a critical health condition (a specified health condition) and their suitability is confirmed by the transplant suitability detection unit 50, the donor selection unit 70 selects that potential donor as a donor. This selection decision is made instantaneously through high-speed data exchange and judgment between AI agents, without waiting for manual judgment by a human operator.

[0221] After selection, the operator agent 101 of the emergency call center device 100 quickly notifies designated emergency responder devices 300 (e.g., ambulance devices, emergency command system devices, organ preservation team devices, etc.) of emergency response request information containing information on the selected donor and suitable recipients (including potential recipients). This allows preparations for organ retrieval, preservation, and transport of the donor to begin quickly and efficiently, simultaneously with the occurrence of a donor (or a condition likely to lead to death).

[0222] (Organ preservation) If there is no suitable recipient for the extracted organ, or if the most suitable recipient is a potential recipient (future patient), the donation process is initiated as needed, and the organ is transported to the organ storage unit 80. In the organ storage unit 80, the organ is stored in a state where it can be restored (resuscitated) to a transplantable state until a predicted recipient is registered on the waiting list.

[0223] 2-4. Effects of the First Embodiment According to the first embodiment of the life-saving system 1, the following effects can be obtained in life-saving medical care, including organ transplantation.

[0224] (Minimizing ischemic time and improving survival rates) AI instantly detects sudden changes in the health status of potential donors and automatically completes the matching and donor selection process with highly compatible recipients, significantly reducing ischemia time (the time from donor identification to restoration of blood flow) compared to traditional processes that rely on human judgment. This improves recipient survival rates and organ graft survival rates.

[0225] (Maximizing the use of organ resources and meeting future demand) The potential recipient prediction unit 20 predicts future organ demand, the transplant suitability detection unit 50 continuously evaluates potential suitability in advance, and the organ preservation unit 80 preserves organs in a transplantable state until recipients predicted as potential recipients are registered on the waiting list. This maximizes organ donation opportunities and enables an optimal matching strategy that takes into account not only current but also future recipient demand.

[0226] (Ensuring high efficiency and objectivity in emergency response) Even in the event of an increase in automated calls or a large-scale disaster, the AI ​​agent's high-speed conversation mode and parallel processing can suppress a decrease in the response speed to emergency calls. Furthermore, by continuously evaluating compatibility such as blood type and HLA type in advance and selecting donors based on objective data, a rapid and objective organ donation process is ensured, leading to increased efficiency in life-saving medical care.

[0227] 2-5. Example Operation Scenario An example of an operational scenario for the life-saving system 1 of the first embodiment will be described.

[0228] (Premise) The assumptions for this operational scenario are as follows: Recipient Information: Information on recipient A, who has a high level of urgency, is registered in the recipient information management unit 10. Pre-selection of potential donors: The transplant suitability detection unit 50 pre-selects potential donors (donor candidates) B, C, and D that are suitable for recipient A. Initial status of the potential donor: Potential donor B is healthy. Post-donor candidate's condition: Potential donor D later progresses through bradycardia (a condition where the heart rate is significantly below the normal range) to a critical condition in which recovery of brain function is predicted to be extremely difficult. Post-donor status: Potential donor C later becomes medically unsuitable (e.g., elevated body temperature, high likelihood of active infection), but does not reach the severe condition seen in candidate D.

[0229] (process) The process performed by this life-saving system 1 is as follows: Continuous monitoring of biometric data: The health status monitoring unit 60 continuously monitors biometric data from the emergency notification devices 200 associated with potential donors B, C, and D. Detection of changes in health status: The health status monitoring unit 60 detects that potential donor B's vital signs are stable. The health status monitoring unit 60 detects that potential donor C has become medically unsuitable. The health status monitoring unit 60 detects a sign of a serious health condition, bradycardia, in potential donor D. Suspension and exclusion of the donor selection process: The donor selection unit 70 immediately excludes potential donor C, who is medically unsuitable, from the selection process. The donor selection unit 70 immediately suspends the donor selection process in order to prioritize life-saving measures for potential donor D. Notification to emergency responders and initiation of life-saving measures: Operator agent 101 prioritizes saving the life of potential donor D, and while gathering detailed information in high-speed conversation mode, immediately notifies the nearest emergency responder device 300 of D's location and condition. Based on this notification, the emergency response team immediately rushes to the scene and begins life-saving measures such as cardiopulmonary resuscitation (CPR) on candidate D. Continuous monitoring of condition and confirmation of transition to critical condition: While the emergency response team is carrying out life-saving measures, the health monitoring unit 60 continuously monitors the condition of potential donor D. Based on the monitoring results, it is confirmed from changes in biometric data such as electroencephalograms that, despite life-saving measures, the condition of potential donor D has further deteriorated and has transitioned to a critical condition in which recovery of brain function is predicted to be extremely difficult. Donor selection: At this point, the donor selection unit 70 confirms that potential donor D meets the transplant criteria and is in a predetermined severe health condition that exceeds the limits of life-saving measures, and selects potential donor D as the donor for recipient A. (effect) This process ensures that the opportunity for organ donation is only realized after every possible effort has been made to save the life of a potential donor.

[0230] More specifically, a "predetermined severe health condition beyond the limits of life-saving measures" refers to a technical state in which, after maximum medical intervention to save the life of a potential donor has been exhausted, recovery is deemed extremely difficult based on objective AI analysis. This determination is automatically made by a combination of threshold conditions constructed in accordance with current legal and medical guidelines (e.g., brain death determination protocols and standard time limits for cardiopulmonary resuscitation). Specifically, the predetermined health condition is determined to have been reached when the probability of difficulty in brain function recovery (P_BrainLoss), calculated in real time by the health status monitoring agent 61, continuously exceeds a predetermined threshold (e.g., 90%) despite attempts at resuscitation by the emergency response team, and the time elapsed for life-saving measures linked from the emergency responder device exceeds the time limit of the resuscitation protocol. As a result, the donor selection unit 70 automatically triggers the donor selection and organ donation process for highly suitable recipients (including potential recipients) by the transplant suitability detection unit 50, based on objective AI judgment, without waiting for a final human decision, thereby minimizing ischemic time and increasing the efficiency of life-saving medical care.

[0231] The composite threshold conditions for "specified health conditions" that trigger donor selection are set to ensure a balance between maximum life-saving efforts and organ preservation, and the specific thresholds are based on the following medical and engineering rationale. The threshold of 90% for the probability of difficulty in recovering brain function (P_BrainLoss) is based on the engineering of the statistical limit of resuscitation success rates in current life-saving medical protocols (data where the resuscitation success rate drops to less than 10%), and the technical basis for determining "extremely difficult recovery" is when the recovery probability calculated by the AI ​​exceeds 90%, which correlates with this limit. The specified limit time for "exceeding the limit time for life-saving measures" is based on the standard resuscitation attempt time recommended in cardiopulmonary resuscitation (CPR) guidelines (e.g., 20 to 30 minutes), and is dynamically adjusted by an individualized algorithm that takes into account the age of the potential donor and the risk of the underlying disease, depending on the situational context (e.g., hypothermia). Furthermore, the organ preservation criteria (CSS_Org) define a probability of irreversible organ damage (P_OrganDamage) of less than 5% as "below critical level." This 5% threshold is a medical and engineering criterion derived from past clinical data on the functional prognosis of organs in the field of transplant surgery, and is calculated backward from the maximum acceptable range of ischemic injury to ensure long-term graft survival.

[0232] 3. Second Embodiment In the following explanation, components that are functionally common to those already described will be denoted by the same symbols, and their explanations will be omitted as appropriate.

[0233] The emergency call center device 100 of the life-saving system 1 shown in Figure 13 has a communication unit 110 and an emergency call determination unit 170. The emergency call determination unit 170 has a function to determine whether the emergency call is an automatic or manual call. The automatic conversation unit 111 does not need to have a function to determine whether the emergency call is an automatic or manual call.

[0234] The call unit 110 makes a call using either the automated conversation unit 111 or the direct conversation unit 112, based on the result of the determination made by the emergency call determination unit 170.

[0235] According to this second embodiment, similar to the first embodiment, while suppressing a decrease in the response speed to emergency calls when automated calls increase, it is possible for the caller (person) and the human operator to communicate directly when the emergency call is a manual call.

[0236] 4. Third Embodiment 4-1. Composition The overall configuration of the third embodiment of the life-saving system 1 shown in Figure 14 is the same as that of the first embodiment. However, in the third embodiment of the life-saving system 1, the emergency call device 200 also functions as an emergency responder device 500. That is, each emergency call device 200 in the third embodiment of the life-saving system 1 has a caller agent 201 and an emergency responder agent 301. The program for realizing the emergency responder agent 301 is stored in the memory 222, storage unit 250, etc., of the emergency call device 200 as the emergency responder device 500.

[0237] In the third embodiment, the emergency call center device (in other words, the operator agent 101) 100 has a function to transmit emergency response request information to a predetermined emergency responder device 500.

[0238] The designated emergency responder device 500 is the emergency responder device 500 (in other words, the emergency call device 200) of a user who is registered as having the skills required for emergency response (hereinafter also referred to as a "skilled worker registered user"). Information on the designated emergency responder device 500 and its skilled worker registered user is stored in a database (not shown) accessible by the emergency call center device 100.

[0239] Users who can be registered as having the skills required for emergency response vary depending on the content (category) of the emergency response. For example, when the content of the emergency response is life-saving, medical workers such as doctors and nurses, and those who have completed a first aid training course and obtained a certification can be registered as users with life-saving skills. Also, for example, when the content of the emergency response is rescue, those with relevant rescue qualifications such as water rescue operator qualifications, mountain rescue operator qualifications, international rescue qualifications, and firefighter qualifications, as well as those with experience in these operations, can be registered as users with rescue skills.

[0240] For example, when the content of the emergency response is life-saving, when a predetermined emergency responder device 500 is detected within a predetermined distance from the location of the person in need of life-saving, the emergency reporting center device 100 (in other words, the operator agent 101) transmits emergency response request information to the detected emergency responder device 500.

[0241] The emergency responder device 500 (in other words, the emergency responder agent 301) has a function of receiving emergency response request information from the emergency reporting center device 100 and notifying the registered users with skills of the emergency responder device 500 of the content of the emergency response request information.

[0242] (Operator agent 101) The operator agent 101 can automatically execute predetermined processes related to the emergency response request. The predetermined processes include an emergency response request information transmission process of transmitting the emergency response request information to one or more predetermined emergency responder devices 500, a process of determining whether there is a response indicating that the emergency response is possible (hereinafter referred to as "response possible response") from one or more of the emergency responder devices 500 that are the transmission destinations of the emergency response request information, a retransmission process of repeating the emergency response request information transmission process until there is at least one response possible response, and the like.

[0243] (Emergency responder agent 301) The emergency response agent 301 is an AI agent (software agent) that operates in each emergency response device 500. The emergency response agent 301 is composed of multiple sub-agents and can handle complex tasks and dynamic environments. The emergency response agent 301 can automatically execute predetermined processes related to emergency response. These predetermined processes include emergency response request information reception processing, emergency response related processing, location information acquisition processing, emergency response request content presentation processing, response feasibility determination processing, response feasibility transmission processing, etc. Hereinafter, the emergency response device 500 on which the agent 301 operates will also be referred to as the device 500.

[0244] The emergency response request information reception process is the process of receiving emergency response request information. The emergency response-related processing is the process of enabling emergency response based on the emergency response request information. The location information acquisition process is the process of acquiring the current location information of the device 500. The emergency response request content presentation process is the process of presenting the content of the emergency response request to the registered skilled user of the device 500 using the information presentation unit 230. The response feasibility determination process is the process of determining whether the registered skilled user of the device 500 will respond to the emergency response request. The response feasibility response transmission process is the process of transmitting a response feasibility response to the emergency notification center device 100.

[0245] 4-2. Processes performed by the emergency call center system and emergency call device. (Emergency call center device 100) In addition to the processes shown in Figures 6 to 12, the emergency call center device 100 can also perform the processes illustrated in Figure 15 (processing by the emergency call center device).

[0246] As shown in Figure 15, the emergency call center device 100 transmits emergency response request information to a predetermined emergency responder device 500 (Step S141: Emergency response request information transmission process). The emergency call center device 100 preferentially transmits the emergency response request information to one or more predetermined emergency responder devices 500 located closer to the location of the person requiring emergency response (hereinafter referred to as "person in need of life" in the description of the third embodiment).

[0247] Next, the emergency call center device 100 determines whether or not there is a response from the emergency responder device 500 to which the emergency response request information has been sent (Step S142: Determination of whether or not a response is possible). For example, if there is a response within a predetermined time (for example, within 15 seconds) after sending the emergency medical assistance request information, the emergency call center device 100 determines that there is a response, and if there is no response within the predetermined time, it determines that there is no response.

[0248] If it is determined that there was no response (NO in step S142), the emergency call center device 100 performs a process to resend the emergency response request information (step S141: retransmission process). The retransmission process includes the process of resending the emergency response request information to the emergency responder device 500 to which the emergency response request information has already been sent, and the process of sending the emergency response request information to an emergency responder device 300 other than the emergency responder device 500 to which the emergency response request information has already been sent.

[0249] If it is determined that a response is possible (YES in step S142), the emergency call center device 100 terminates the current process (END).

[0250] In the third embodiment, since the predetermined emergency responder device 500 is also a predetermined emergency notification device 200, the emergency response request information transmission process (step S141) can be included, for example, in at least one of the emergency notification information notification process (step S122) in Figure 9 and the emergency notification information notification process (step S132) in Figure 10.

[0251] Alternatively, the human operator who has spoken directly with the caller through the direct conversation processing (step S28) in Figure 12 may operate a predetermined device in the emergency call center 1a to transmit emergency response request information to a predetermined emergency responder device 500.

[0252] (Emergency responder device 500) In addition to the emergency call device-side call processing (step S22) shown in Figure 12, the emergency responder device 500 can also perform the processing illustrated in Figure 16 (processing by the emergency responder device).

[0253] As shown in Figure 15, the emergency responder device 500 receives emergency response request information (step S31: emergency response request information reception process) and performs emergency response-related processing (step S32).

[0254] As shown in Figure 16, in the emergency response-related processing (step S32), the emergency responder device 500 acquires its current location information using its GPS function (step S321: location information acquisition processing).

[0255] Next, the emergency responder device 500 performs the emergency response request content presentation process (step S322). The emergency response request content presentation process (step S322) includes the process of displaying the emergency response request content display screen G, as illustrated in Figure 17, on the display 231 (touch panel 244) based on the current location information and emergency response request information of the device 500.

[0256] Next, the emergency responder device 500 determines whether the user of the device 500 will respond to the emergency response request (step S323: response availability determination process). If the user of the device 500 determines that they will respond to the emergency response request (YES in step S323), it sends a response availability message to the emergency call center device 100 (step S324).

[0257] As illustrated in Figure 17, the emergency response request display screen G shows a map M indicating the current location P1 of the device 500 and the current location P2 of the person in need of rescue, text information T about the person in need of rescue, and a button B that allows the user of the device 500 to select whether or not to respond to this emergency response request. The map M also shows the location P3 of the AED.

[0258] Button B includes a YES button B1 and a NO button B2. The YES button B1 is an operator that the user operates (for example, a tap operation) when responding to an emergency response request. The NO button B2 is an operator that the user operates when not responding to an emergency response request.

[0259] When the YES button B1 is operated, the emergency responder device 500 determines that the user of the device 500 responds to the emergency response request (YES in step S323, FIG. 16), and transmits a response available response to the emergency notification center device 100 (step S324, FIG. 16).

[0260] 4-3. Function According to the third embodiment, the emergency response request information can be transmitted to a predetermined emergency responder device 500. Therefore, according to the third embodiment, in addition to the function of the first embodiment, the following functions are realized.

[0261] (Quick identification and request of skilled registered users) The emergency notification center device 100 (operator agent 101) detects a sudden change of a potential donor (person in need of rescue), and generates emergency response request information. At this time, the emergency response request information is transmitted to the emergency responder device 500 (emergency notification device 200) of a user (skilled registered user, e.g., doctor, nurse, person who has completed a first aid training course) registered as a person having skills required for rescue, etc., existing within a predetermined distance from the position of the person in need of rescue.

[0262] (Acceleration of initial first aid response) When the emergency responder device 500 receives the emergency response request information, it presents to the user the content including the current position of the person in need of rescue and the necessary emergency response (e.g., first aid treatment, location of AED). Thereby, it becomes possible to quickly respond to an emergency situation by obtaining the cooperation of skilled registered users in the vicinity of the potential donor without waiting for donor selection or the arrival of a specialized team.

[0263] 4-4. Effects related to organ transplantation According to the third embodiment, in addition to accelerating donor selection, the following effects can be obtained by strengthening the response in the initial stages.

[0264] (Accelerating initial life-saving measures before ischemic time occurs) To minimize ischemic time (the time from donor identification to the restoration of blood flow), which significantly impacts the success rate of organ transplantation, early response before donor identification is crucial. In the third embodiment, in parallel with a rapid donor selection process using AI, initial life-saving measures before ischemic time occurs can be accelerated with the cooperation of nearby registered skilled users. This increases the likelihood that essential initial treatments for maintaining organ function can be performed before the organ preservation specialist team arrives at the scene.

[0265] (Improving the efficiency and ensuring safety of the entire life-saving process) Emergency response requests are sent to the devices of individuals with the necessary skills for emergency response (registered skilled users), enabling rapid response to emergencies with the cooperation of a wide range of skilled individuals, not just firefighters and police officers. This early intervention contributes to the survival of potential donors, ultimately maximizing organ transplant opportunities and achieving greater efficiency in life-saving medical care.

[0266] 5. Fourth Embodiment 5-1. Composition The overall configuration of the life-saving system 1 in the fourth embodiment is the same as in the third embodiment. However, in the life-saving system 1 of the fourth embodiment, the operator agent 101 functions as a meta-AI (meta-agent). The operator agent 101 monitors, coordinates, manages, optimizes, etc., other AI agents (e.g., health status monitoring agent 61, donor selection agent 71, caller agent 201) in the life-saving system 1 as a multi-AI agent system.

[0267] More specifically, the operator agent 101 has functions such as behavior monitoring, evaluation, higher-order judgment, adjustment, risk management, learning process optimization, and correction. These functions work in conjunction with each other.

[0268] The behavior monitoring function is a feature that monitors the behavior of other AI agents in real time.

[0269] The evaluation function analyzes the behavior of other AI agents and assesses risk factors. For example, the evaluation function assesses the possibility of goal inconsistencies or conflicts between AI agents, and the likelihood of AI agents deviating from their respective set goals. These possibilities are evaluated as values ​​for predetermined risk factors.

[0270] The higher-order decision function is a function that makes sophisticated decisions to optimize the life-saving system 1. More specifically, the main role of the higher-order decision function is to make various decisions from the perspective of optimizing the life-saving system 1, based on the analysis and evaluation results from the evaluation function. For example, if problems arise such as inconsistencies or conflicts in goals among AI agents, or if there are AI agents whose likelihood of deviating from the goal is greater than a threshold (risk threshold), the higher-order decision function makes decisions regarding the redistribution of goals and tasks to each AI agent, intervention, adaptive enhancement, and modification in order to resolve these problems. The adjustment function, risk management function, learning process optimization function, and modification function operate based on the decisions made by the higher-order decision function.

[0271] The coordination function adjusts the goals and tasks of each AI agent so that multiple AI agents can work together and cooperate with one another. More specifically, the coordination function sets the task priorities of each AI agent and strengthens the coordination between AI agents to ensure harmonious operation, in order to maximize the probability of achieving the goals in the life-saving system 1. This coordination function distributes processing power and computing resources among multiple AI agents, improving the efficiency of the life-saving system 1.

[0272] The risk management function is designed to prevent malfunctions by limiting the operating range of each AI agent to prevent other AI agents from taking inappropriate actions. Specifically, the risk management function immediately intervenes in other AI agents that have detected predetermined behavior (AI agents whose predetermined risk factors are judged to be greater than the risk threshold). Intervention in other AI agents may include intervention through communication with other AI agents (persuasion, training), modification of algorithms and parameters through hacking of other AI agents, and termination of agent functions.

[0273] The learning process optimization function is a feature that evolves the learning algorithms of other AI agents to lead to better results. The main roles of the learning process optimization function are learning support, environmental adaptation promotion, and knowledge sharing. Through the learning process optimization function, the adaptive capabilities of other AI agents are promoted, and other AI agents evolve to respond more quickly to new environments and challenges. It also promotes the improvement of the adaptive capabilities of operator agent 101, which is an element of life-saving system 1.

[0274] The correction function identifies malfunctions and errors that occur in the life-saving system 1 and attempts to correct them automatically. The main roles of the correction function are self-diagnosis, fault recovery, and stability maintenance.

[0275] 5-2. Processing by the Operator Agent In addition to the processing already described, the operator agent 101 can perform the meta-processing shown in Figure 18 (step S300: processing by the meta-agent, processing by the emergency call center device 100). As shown in Figure 18, the operator agent 101 performs a behavior monitoring process (step S301) to monitor the behavior of other AI agents. Based on the results of the behavior monitoring process (step S301), operator agent 101 determines whether or not there are other AI agents that have detected a predetermined behavior (step S302: Intervention Target Determination Process). If operator agent 101 determines that there are other AI agents that have detected a predetermined behavior (YES in step S302), it performs an intervention process to intervene in the other AI agent and suppress the predetermined behavior (step S303).

[0276] For example, operator agent 101 audits data sharing between informant agent 201, which notifies of the availability of a potential donor, and health status monitoring agent 61, which collects physiological data. The purpose of this is to detect fraudulent information manipulation or sharing (fraudulent sharing) that unfairly accelerates the organ donation process rather than providing life-saving treatment to potential donors. Fraudulent sharing can result from human intent or algorithmic bias in the system and can lead to ethical risks such as delays in life-saving treatment or inappropriate decisions. If operator agent 101 detects that data indicating the possibility of recovery is being deliberately ignored and that the transition to organ donation is being forced, it immediately determines this to be an ethical risk and performs a mandatory intervention (S303).

[0277] This intervention is an emergency measure to prevent data distorted or accelerated processes due to fraudulent collaboration from influencing the decisions of downstream donor selection agents 71. Specifically, operator agent 101 automatically filters or invalidates fraudulent data and forces donor selection agents 71 to re-evaluate based on sound data. Operator agent 101 also temporarily and completely halts the organ donation process by donor selection agents 71 to allow time for a detailed investigation and re-evaluation.

[0278] Furthermore, for example, if operator agent 101 determines that caller agent 201 is behaving inappropriately by prioritizing communication speed over information integrity while executing high-speed conversation mode, it will perform an intervention process (S303) with caller agent 201. In this case, operator agent 101 sends a control signal to caller agent 201 to switch modes, even if it slightly increases the call duration, in order to ensure that the necessary data points are reliably transmitted, thereby correcting and suppressing the caller agent's behavior.

[0279] Furthermore, for example, operator agent 101 detects an ethical risk between emergency response agent 301 and the agent on the organ donation management device 600 side that disregards the principle of prioritizing saving lives and unduly accelerates the organ donation process. Upon detecting this ethical risk, operator agent 101 immediately performs an intervention process (S303) as a meta-AI. This process is carried out by modifying or suppressing the actions of donor selection agent 71, which is the final decision-making body for fraudulent collaboration.

[0280] Furthermore, for example, if operator agent 101 determines that emergency responder agent 301 is prioritizing nearby resources such as AEDs over responders with more appropriate expertise in terms of overall suitability for emergency response, operator agent 101 will perform an intervention process (S303) to emergency responder agent 301. In this case, operator agent 101 sends a control signal to emergency responder agent 301 to correct and suppress its behavior, thereby changing the priority of the emergency response request from nearby but less skilled resources to more specialized and appropriate responders, even if they are further away.

[0281] Furthermore, if, for example, the health monitoring agent 61 or the donor selection agent 71 detects an action that excessively prioritizes organ donation (e.g., prioritizing the initiation of the donor selection process before sufficient life-saving measures have been taken for potential donors), the operator agent 101 evaluates this as a predetermined behavior (ethical risk factor).

[0282] When operator agent 101 detects this ethical risk, it immediately performs an intervention process (S303) to modify or suppress the actions of health status monitoring agent 61 and donor selection agent 71 (S303). For example, operator agent 101 temporarily suspends or suppresses the notification of monitoring results by health status monitoring agent 61 to donor selection unit 70. Operator agent 101 also prompts escalation to a human operator if necessary.

[0283] Furthermore, if, for example, the evaluation algorithm in the transplant suitability detection agent 51 is found to have been influenced by human or data-driven bias, and a risk is detected that ethical standards such as medical urgency and fair waiting periods may be compromised, the operator agent 101 will suppress the suitability detection process (S303).

[0284] Furthermore, if, for example, the prediction model of the potential recipient prediction agent 21 is influenced by medically inappropriate factors such as socioeconomic status or regional characteristics, resulting in ethical bias where a specific group of individuals is unfairly excluded from the benefits of the prediction, the operator agent 101 will suppress the generation of the prediction results and their provision to the potential recipient information management agent 31 (S303).

[0285] 5-3. Effects According to the life-saving system 1 of the fourth embodiment, the following effects and benefits can be obtained. (1) Minimize ischemic time and ensure absolute priority is given to saving lives. In the process from detecting sudden changes in potential donors to selecting suitable donors, the meta-AI (operator agent 101) monitors and adjusts the behavior of other agents in real time, maximizing the accuracy and speed of decision-making, thereby significantly reducing ischemia time (the time from donor identification to restoration of blood flow). Furthermore, the ethical risk management function immediately corrects and suppresses the behavior of agents that deviate from the priority principles of life-saving measures, ensuring not only an improvement in recipient survival rates but also absolute priority in life-saving efforts for donors.

[0286] (2) Optimal utilization of organ resources and strengthening response to future demand The potential recipient prediction agent 20 and the transplantation criterion suitability detection agent 51 continuously perform predictions and evaluations based on real-time EHRs and large-scale registry data. In addition, meta-AI optimizes the learning process of each agent and corrects data bias, improving prediction accuracy and the objective fairness of suitability evaluation. This strengthens the feasibility of realizing an optimal matching strategy that takes into account not only current but also future recipient demand.

[0287] (3) Extreme efficiency and ethical assurance of objectivity in emergency response. Even during automated notifications or large-scale disasters, Meta AI coordinates and optimizes the coordination and communication between multiple AI agents, maximizing the efficiency of the high-speed conversation mode and maximizing the speed of response to emergency calls. In addition to selection based on objective data such as blood type and HLA type, Meta AI's function of actively detecting and correcting ethical risks (such as fraudulent collaboration and inappropriate prioritization) ensures the fairness and objectivity of the entire organ donation process from an ethical standpoint.

[0288] Thus, the fourth embodiment makes it possible to achieve the goal of "saving more lives" while ensuring the safety and ethical integrity (minimizing ethical issues) of the entire life-saving system 1.

[0289] Furthermore, if a multi-AI agent system (humanitarian system) aiming to "save more lives," as in the fourth embodiment, can be realized as ASI (Artificial Super Intelligence) faster than, for example, a multi-AI agent system (inhumane system) aiming to "take more lives," then it may be possible to preemptively suppress or restrain the latter (inhumane system) with the former (humanitarian system). This has the potential to help address problems such as multi-AI agent technology being used in war or terrorism, resulting in the loss of human lives. According to the fourth embodiment, in addition to the effects and advantages of the third embodiment, the safety of the life-saving system 1 as a multi-AI agent system is improved.

[0290] 6. Fifth Embodiment 6-1. Composition (Lifesaving method) The fifth embodiment of the life-saving method is a life-saving method using a multi-AI agent system for providing donor organs to a recipient. In this life-saving method, agents included in the multi-AI agent system (life-saving system 1) perform the processes shown in Figure 19.

[0291] Recipient information management step (S401) in which the recipient information management agent 11 manages the recipient information. The potential recipient prediction step (S402) involves the potential recipient prediction agent 21 predicting the potential recipient. Potential recipient information management step (S403): The potential recipient information management agent 31 manages information about potential recipients. Potential donor information management step (S404) involves the potential donor information management agent 41 managing the information of potential donors. Match detection step (S405): The match detection agent 51 detects potential donors (matched individuals) that meet the transplantation conditions for each recipient or potential recipient, based on the recipient information, potential recipient information, and potential donor information. Health status monitoring step (S406): Health status monitoring agent 61 monitors the health status of each potential donor. The donor selection step (S407) involves a donor selection agent 71 selecting potential donors who meet the transplant criteria and are in a specified state of health. Emergency response request information transmission step (S408): Operator agent 101 transmits emergency response request information, including information on selected donors and suitable recipients and / or potential recipients, to emergency response agent 301 working on a device associated with the emergency responder, based on a call with caller agent 201.

[0292] 6-2. Action and Effects (Pre-assessment of suitability based on future demand and maximization of resources) Firstly, based on the future transplant demand forecast by the potential recipient prediction agent 21 (S402) and the management of the latest information by the information management agents 11, 31, and 41 (S401, S403, S404), the match detection agent 51 continuously and proactively evaluates the suitability of current and future recipient candidates (S405). This allows for a decision to transition to the donation process (organ preservation) at any time, even if there are no current recipients with high urgency when a donor becomes available, as long as the existence of a potential recipient who will be suitable in the future is confirmed. This prevents the waste of valuable organ resources and maximizes their utilization.

[0293] (Minimizing ischemic time through multi-AI agent collaboration) Secondly, based on the real-time monitoring of the potential donor's health status by the health status monitoring agent 61 (S406) and the results of the pre-evaluation by the match detection agent 51, the donor selection agent 71 immediately performs donor selection the moment a potential donor reaches a predetermined critical health condition (S407). Since the selection is completed through high-speed data exchange between AI agents without human judgment, the waiting time from donor discovery to organ retrieval is shortened to the absolute minimum, and the recipient survival rate and transplant organ graft survival rate are improved by minimizing ischemic time.

[0294] (Immediate processing through autonomous emergency response commands) Thirdly, after the selection by the donor selection agent 71 is completed, the operator agent 101 immediately and automatically transmits emergency response request information, including information on the selected donor and the most suitable recipient (including potential recipients), to the emergency response agent 301 (S408). This ensures a seamless process from donor selection to dispatching the on-site response team, allowing preparations for organ retrieval, preservation, and transport to begin without delay, thereby enabling a rapid and efficient response in time-constrained emergency medical care.

[0295] 7. Sixth Embodiment 7-1. Composition (Life-saving system 1) The sixth embodiment encompasses the life-saving method of the fifth embodiment. In the sixth embodiment, blockchain technology is applied to the life-saving system 1 of the first embodiment. More specifically, the life-saving system 1 of the first embodiment generates a digital identifier on the blockchain corresponding to each organ (heart, liver, kidney, etc.) when a decision is made to remove an organ or when potential donor information is registered in the system, in order to manage the organs of potential donors "organ by organ".

[0296] Organs are managed on an organ-by-organ basis using organ NFTs. Organ NFTs are non-fungible tokens (NFTs) that serve as digital identifiers linking unique attributes and history to each organ. Organ NFTs can also be interpreted as "organ IDs."

[0297] As shown in Figure 20, the organ donation management device 600 of the sixth embodiment of the life-saving system 1 includes, in addition to the configuration of the first embodiment, a blockchain recording unit 81, an identifier generation unit 82, a history recording unit 83, and a storage treatment data recording unit 84.

[0298] The hardware configuration of the organ donation management device 600 in the sixth embodiment is the same as in Figure 5. Each functional part of the organ donation management device 600 is realized through the cooperation of hardware and software.

[0299] The blockchain recording unit 81 is a functional unit for recording the fact of donor selection by the donor selection unit 70 and the information that forms the basis of that objective judgment in an immutable format.

[0300] The identifier generation unit 82 is a functional unit that generates a unique digital identifier, including a nonfungible token (NFT), for each type of organ provided by a potential donor and registers it on the blockchain.

[0301] The history recording unit 83 is a functional unit for recording the results of suitability detection, the results of donor selection, and the entire physical and medical history from organ extraction to donation to the recipient, linked to a digital identifier and recorded on the blockchain.

[0302] The preservation treatment data recording unit 84 is a functional unit for recording preservation treatment data, including the history of temperature, pressure, or drug application during long-term preservation treatment of organs by the organ preservation device (organ preservation unit) 700, on a blockchain, linked to a digital identifier assigned to the organ.

[0303] 7-2. Effects (Part 1) With the above configuration, the life-saving system 1 of the sixth embodiment provides the following functions and effects. By recording the facts and criteria for donor selection (suitability, serious health conditions) on an immutable blockchain, it is possible to contribute to ensuring ethical objectivity and improving social trust in the rapid AI-driven donor selection process.

[0304] By generating a unique digital identifier, including NFTs, for each organ type and linking and recording the entire physical and medical history from extraction to donation, ultimate traceability in the organ donation process can be achieved, eliminating the risk of data tampering and dramatically improving organ quality control and transparency.

[0305] By recording and verifying detailed data such as temperature and chemical solutions used in long-term organ preservation procedures on a blockchain, it becomes possible to objectively prove the quality assurance of long-term preserved organs. This will increase the reliability of transplant decisions for future recipients and allow for the management of the risks of long-term organ preservation in the ongoing donation process.

[0306] To elaborate on the hardware implementation of each functional unit in the sixth embodiment, the blockchain recording unit 81, identifier generation unit 82, history recording unit 83, and storage treatment data recording unit 84 shown in Figure 20 are realized through the cooperation of the control unit 620, storage unit 650, and communication unit 660 shown in Figure 5. Specifically, the memory 622 additionally stores dedicated programs for executing communication protocols with the blockchain network and smart contracts, and the processor 621 executes these programs to perform hash value calculation and NFT generation processing as the identifier generation unit 82. Furthermore, the transaction issuance process to the distributed ledger by the blockchain recording unit 81 is realized by the communication unit 660 transmitting the signed data generated by the processor 621 to an external blockchain node via the communication network 400. In addition, a private key and wallet information that manages the right to write to the blockchain are stored in an isolated secure area within the storage unit 650 and are used for authentication when the history recording unit 83 and storage treatment data recording unit 84 record data. This allows for the physical support of the sixth embodiment, which requires tamper-proof recording and high security, while using a general-purpose hardware configuration.

[0307] 7-3. Operation of Life-Saving System 1 The life-saving system 1 of the sixth embodiment performs, for example, the organ NFT lifecycle recording process (S500) shown in Figure 21. The organ NFT lifecycle recording process (S500) is a process to record the history from organ extraction to transplantation using NFTs and to prevent tampering with the history. The organ NFT lifecycle recording process (S500) includes identifier generation processing (S501), suitability assessment recording processing (S502), ethical transparency assurance processing (S503), and provision / storage recording processing (S504).

[0308] Various agents, such as the potential donor information management agent 41, the donor selection agent 71, the match detection agent 51, the health status monitoring agent 61, the operator agent 101, and the emergency response agent 301, work together to perform the necessary processing by coordinating the functions of the blockchain recording unit 81, the identifier generation unit 82, the history recording unit 83, and the storage treatment data recording unit 84.

[0309] Identifier generation process (S501): The potential donor information management agent 41 and / or donor selection agent 71 generate organ NFTs for each organ (e.g., kidney R / L, liver) based on the potential donor information (whether consent was given, HLA type, blood type, etc.) and record the first timestamp.

[0310] In other words, the identifier generation unit 82 generates unique identifiers (organ NFTs) for each organ (heart, kidney, etc.) based on information about potential donors. The history recording unit 83 creates these NFTs and an initial timestamp indicating the date and time of generation as recorded data. The blockchain recording unit 81 writes this initial recorded data to the blockchain, initiating the tracking history of the organs.

[0311] Suitability assessment record processing (S502): The suitability detection agent 51 records the results of the suitability assessment (e.g., suitability scores for recipient A and potential recipient X) in the organ NFT, and persists the basis for its objective judgment.

[0312] In other words, the history recording unit 83 adds the evaluation results, such as the compatibility score with the recipient calculated by the transplantation condition suitability detection unit 50, the evaluation criteria, and the ID of the suitable recipient, to the recorded data as an event log. The blockchain recording unit 81 adds this history of evaluation results to the blockchain, serving as evidence of the fairness and objectivity of the suitability evaluation.

[0313] Ethical Transparency Assurance Process (S503): Health status monitoring agent 61 and / or donor selection agent 71 record the facts of the detection of a serious health condition that triggered organ harvesting (e.g., S405) and the final decision on donor selection (e.g., S406), thereby demonstrating that the ethical process was objectively carried out.

[0314] In other words, the history recording unit 83 adds the final donor selection decision by the donor selection unit 70 and proof that ethical processes such as prioritizing life-saving measures were properly implemented (such as monitoring records by operator agent 101) as important events in the recorded data. The blockchain recording unit 81 writes this record proving ethical transparency to the blockchain, irreversibly ensuring the legitimacy of the process.

[0315] Provision / Storage Record Processing (S504): Operator agent 101 and / or emergency responder agent 301 record the entire history of physical events and procedures in real time, including the time of organ retrieval, storage method, transport time, start and completion of resuscitation, and the receiving medical institution.

[0316] In other words, after organ retrieval, the organ preservation data recording unit 84 collects and records detailed data of the preservation procedures performed by the organ preservation specialist team (e.g., application of cryogenic freezing / supercooling / vitrification, temperature history, CPA concentration, etc.) as technical data. The history recording unit 83 records the event that the organ was provided and the preservation process was initiated. The blockchain recording unit 81 writes both this technical preservation data and the procedure event to the blockchain, ensuring the quality control history of the organ. This record is later used to determine whether the organ is suitable for resuscitation and transplantation.

[0317] 7-4. Effects (Part 2) The life-saving system 1 of the sixth embodiment, which operates as described above, achieves the following effects through the application of blockchain. Achieving ultimate traceability: Organ NFTs record the entire history of an organ's movement, preservation procedures, and suitability assessments, from transplantation from the donor to the recipient, or long-term storage in the Organ Preservation Device 700, in an unalterable manner. This enables highly transparent management that leaves no room for doubt.

[0318] Ensuring objectivity and ethical integrity in the process: Because the time and facts of when objective criteria such as "transplant suitability" and "specified health condition" are met by the donor selection agent 71 are recorded, ethical concerns and the legitimacy of decisions in the organ donation process are objectively proven, and social trust is enhanced.

[0319] Risk management for long-term storage (on-demand provision process): By recording all data on storage and resuscitation processes in the organ storage unit 80 (temperature, application status of resuscitation agents, etc.), the quality of the organs can be guaranteed, and the reliability of future transplant decisions for predicted potential recipients can be maximized.

[0320] Enhanced information security: By recording only the hash values ​​(encrypted data fingerprints) of highly sensitive patient medical information (such as EHR linked data) on the blockchain, and limiting actual data access to authorized medical institutions, it is possible to achieve both data security and privacy.

[0321] 8. Seventh Embodiment (Immunosuppressant Design System) The seventh embodiment relates to a system for designing drugs to suppress rejection reactions in organ transplantation, and more particularly to a technology for designing and validating personalized or optimized immunosuppressants based on future transplant demand using artificial intelligence (AI) and cloud-based machine learning pipelines.

[0322] (Background technology) Organ transplantation is an effective treatment for end-stage organ failure, but post-transplant rejection remains a major challenge. Traditionally, nonspecific immunosuppressants have been used to suppress rejection, but these can have side effects such as increased risk of infection and nephrotoxicity.

[0323] Meanwhile, the use of AI and machine learning has been rapidly advancing in the drug discovery process in recent years. For example, in the design of antibody drugs, there are attempts to use machine learning to search for the optimal combination of amino acid sequences, and to accelerate structural analysis using protein three-dimensional structure prediction software such as "AlphaFold2" provided by DeepMind Technologies.

[0324] However, while AlphaFold2 can estimate protein structures with extremely high accuracy from amino acid sequences, it presents technical challenges such as the enormous computational demands on both CPUs and GPUs, making resource procurement and implementation difficult. In traditional on-premise environments, resource procurement took several months and construction incurred significant costs, hindering rapid research and development (R&D).

[0325] (Problems that the invention aims to solve) Furthermore, in conventional drug discovery processes, it is not uncommon for it to take 10 to 15 years to create a single drug, and there is a time constraint that prevents drug adjustments from being made after a donor-recipient combination has been established.

[0326] The objective of the seventh embodiment is to integrate future transplant demand forecasting data with the latest cloud-based AI drug discovery technology, enabling the efficient design and verification of immunosuppressants that specifically act against potential rejection reactions in a scalable computing environment. 8-1. Composition The seventh embodiment constructs an immunosuppressant design system 800 for designing and verifying immunosuppressants, based on the vast amount of data and future prediction functions possessed by the life-saving system 1 described in the first to sixth embodiments. Here, the immunosuppressant design system 800 is described as a subsystem that constitutes a part of the life-saving system 1. More specifically, the immunosuppressant design system 800 is described as encompassing the organ donation management device 600. In addition, other AI agents that are subject to monitoring, adjustment, management, optimization, etc. by the operator agent 101 as a meta-AI as shown in the fourth embodiment include the AI ​​agents in the seventh embodiment (e.g., recipient information management agent 11, potential recipient prediction agent 21, potential donor information management agent 41, dataset construction agent 821, rejection factor identification agent 841, molecular design agent 851, in silico verification agent 861, etc.). Furthermore, the potential recipient information management agent 31 may provide information on the potential recipients it manages to the recipient information acquisition agent 811.

[0327] (Functional configuration of the immunosuppressant design system 800) As shown in Figure 22, the immunosuppressant design system 800 comprises a recipient information management unit 10, a potential recipient prediction unit 20, a potential recipient information management unit 30, a potential donor information management unit 40, a recipient information acquisition unit 810, a dataset construction unit 820, a large-scale dataset storage unit 830, a rejection factor identification unit 840, a molecular design unit 850, and an in silico verification unit 860. These functional units are built on an integrated platform, for example, on the cloud, and work in conjunction with each other to ensure scalability and computing power. Each functional unit is realized through the collaboration of hardware and software that constitute the cloud.

[0328] (Recipient information acquisition unit 810) The recipient information acquisition unit 810 acquires information on at least one of the recipient or the potential recipient predicted by the potential recipient prediction unit 20. More specifically, in this embodiment, the recipient information acquisition unit 810 acquires information on the recipient managed by the recipient information management unit 10 and / or information on the potential recipient managed by the potential recipient information management unit 30.

[0329] The recipient information acquisition agent 811 is a component of the AI ​​agent that functions as part of the recipient information acquisition unit 810. The recipient information acquisition agent 811 acquires at least one of the following: information on recipients managed by the recipient information management agent 11, or information on potential recipients managed by the potential recipient information management agent 31. Specifically, the recipient information acquisition agent 811 collects medical and genetic information of patients currently registered on the waiting list, as well as information on future patient groups predicted by the potential recipient prediction agent 21. By providing this information to the dataset construction agent 821, the agent performs processing that supports the construction of a large dataset encompassing current and future transplant demand.

[0330] (Dataset construction unit 820) The dataset construction unit 820 integrates information (medical information, genetic information) managed by the potential donor information management unit 40 with information (medical information, genetic information) of at least one of the recipient or potential recipients acquired by the recipient information acquisition unit 810. Furthermore, the dataset construction unit 820 integrates the history of rejection in past transplant cases and constructs a large-scale dataset optimized for AI analysis.

[0331] (Dataset construction agent 821) The dataset construction agent 821 is a component of the AI ​​agent that functions as part of the dataset construction unit 820. The dataset construction agent 821 integrates information provided by the recipient information acquisition agent 811 and the potential donor information management agent 41, as well as the history of rejection in past transplant cases, to construct a large dataset (integrated data) optimized for AI analysis. The dataset construction agent 821 also performs the process of storing the constructed large dataset in the large dataset storage unit 830.

[0332] The large-scale dataset storage unit 830 is a functional unit that holds the large-scale dataset constructed by the dataset construction unit 820. The data infrastructure for the dataset construction unit 820 and the large-scale dataset storage unit 830 employs, for example, data warehouse technology. BigQuery is an example of data warehouse technology. By utilizing BigQuery, it is possible not only to store data but also to maintain a state where data can be immediately analyzed once it is loaded, and to accelerate integrated analysis with a wide range of data other than R&D data (such as external omics data).

[0333] (Rejection factor identification section 840) The rejection factor identification unit 840 analyzes the data stored in the large-scale dataset storage unit 830 using a graph neural network (GNN) or a deep learning model. Specifically, the rejection factor identification unit 840 learns the complex interactions between potential donors and recipients (and potential recipients) as a graph structure and extracts potential features that determine the presence or absence of rejection. Based on the extracted features, the rejection factor identification unit 840 identifies rejection factors (specific proteins or antigenic epitopes) that are the direct cause of the rejection.

[0334] Furthermore, the rejection factor identification unit 840 has the function of identifying in advance antigen epitopes common to donor groups that are highly likely to be transplanted into a specific potential recipient, based on the prediction data from the potential recipient prediction unit 20. This makes it possible to narrow down the target even before the donor is confirmed.

[0335] (Rejection Factor Identification Agent 841) The rejection factor identification agent 841 is a component of the AI ​​agent that functions as part of the rejection factor identification unit 840. The rejection factor identification agent 841 analyzes the large dataset held in the large dataset storage unit 830 using a graph neural network (GNN) or a deep learning model. Specifically, the rejection factor identification agent 841 learns the complex interactions between potential donors and recipients (and potential recipients) as a graph structure and extracts potential features that determine the presence or absence of rejection. Based on the extracted features, the rejection factor identification agent 841 identifies rejection factors (specific proteins or antigen epitopes) that are the direct cause of rejection and provides the identified information to the molecular design agent 851. In addition, based on the prediction data from the potential recipient prediction agent 21, the rejection factor identification agent 841 performs a process to pre-identify antigen epitopes common to donor groups that have a high probability of being transplanted in the future to a specific potential recipient.

[0336] (Molecular Design Department 850) The molecular design unit 850 designs molecular structures that specifically bind to identified rejection factors and inhibit or control their function.

[0337] The Molecular Design Unit 850 implements an AI for predicting protein structures. An example of such an AI is AlphaFold, an artificial intelligence program for predicting protein structures. While AlphaFold can accurately estimate three-dimensional structures from amino acid sequences, it suffers from extremely high computational load. To address this challenge, the Molecular Design Unit 850 is built as a scalable structure estimation system utilizing, for example, Vertex AI Pipelines. This allows for flexible allocation of resources such as CPUs and GPUs according to research needs, enabling highly efficient, computationally driven R&D that can infer protein structures (for example, on a scale of 1,000 or more structures per day).

[0338] Furthermore, the molecular design unit 850 predicts cryptic pockets (hidden binding sites) that are temporarily formed during the dynamic transformation of proteins, by combining static structural prediction using protein structure prediction AI with molecular dynamics calculations (MD simulations) and machine learning. By designing small or medium-sized molecule compounds that bind to these pockets and allosterically inhibit the function of rejection factors, it enables drug development against targets that were previously considered undruggable.

[0339] Furthermore, the molecular design unit 850 designs the molecular structure of a vaccine or peptide drug that induces specific regulatory T cells (Tregs) in the body in response to antigen epitopes common to future donor groups identified by the rejection factor identification unit 840.

[0340] Furthermore, the molecular design unit 850 performs optimization processing on the designed drug molecule, adding linker structures and release control sites suitable for mounting onto autonomous nanomachines that specifically accumulate in vascular endothelial cells and other tissues of transplanted organs. The molecular structure data designed by the molecular design unit 850 (molecular structure data) is converted into an API and made immediately available to other components within the immunosuppressant design system 800 (e.g., the in silico validation unit 860) or to external researchers, for example, through serverless technology. Examples of serverless technologies include Cloud Run and Cloud Functions.

[0341] (Molecular design agent 851) The molecular design agent 851 is a component of the AI ​​agent that functions as part of the molecular design unit 850. Based on the rejection factor identification information provided by the rejection factor identification agent 841, the molecular design agent 851 predicts the three-dimensional structure of the rejection factor using a protein structure prediction AI (e.g., AlphaFold2). The molecular design agent 851 then designs or screens molecular structures that specifically bind to the predicted three-dimensional structure or cryptic pockets (hidden binding sites) predicted by molecular dynamics calculations, thereby inhibiting or controlling their function. Furthermore, the molecular design agent 851 designs Treg-inducing vaccines against antigen epitopes common to future donor groups, performs structural optimization processing suitable for deployment on autonomous nanomachines, and provides data of the designed molecular structures to the in silico validation agent 861.

[0342] (In silico verification unit 860) The in silico verification unit 860 uses calculations to verify the efficacy and safety of the designed drug. Specifically, the in silico validation unit 860 uses generative AI, including GANs (Generative Adversarial Networks) or diffusion models, to learn statistical features from EHRs of real patients and generate statistically equivalent synthetic data while protecting privacy. In generating this synthetic data, the in silico validation unit 860 maps the complex correlations (copulas) between multidimensional data of real patients (age, sex, genotype, medical history, test results, etc.) onto a latent space using a variational autoencoder (VAE). The in silico validation unit 860 samples from this continuous latent space and reconstructs the data through a decoder, thereby generating an infinite number of "unknown patient profiles" that do not exist in the real data but are statistically and medically possible. This allows for the virtual construction of populations with rare genetic backgrounds where real data is lacking, or patient groups with complex disease conditions that may appear in the future, and comprehensively validates the efficacy and side effects of drugs for these virtual patient groups. Using this synthesized data, the in silico validation unit 860 constructs a virtual patient population, such as a group with rare HLA types that have a small number of cases in real data, and verifies side effects and efficacy.

[0343] The in silico validation unit 860 performs immune tolerance induction simulations and / or nanomachine DDS simulations.

[0344] The immune tolerance induction simulation is a simulation used to predict the efficiency of inducing immune tolerance when a Treg-inducing vaccine is administered during the waiting period before transplantation.

[0345] Nanomachine DDS simulation is a simulation used to predict local drug concentrations and systemic leakage rates when drugs are released locally in organs by autonomous nanomachines, and to assess the risk of systemic side effects.

[0346] The results of these simulations can be, for example, converted into an API using serverless technology and made immediately available to other components within the immunosuppressant design system 800 (e.g., the donor selection unit 70).

[0347] Furthermore, the in silico validation unit 860 has the function of selecting drug molecules that meet predetermined criteria for efficacy and safety scores as final candidates, based on the results of the above validation and simulation. The selected candidate data is output to improve the accuracy of prognosis prediction in the donor selection process described later.

[0348] Information validated by the in silico validation unit 860 (such as immunosuppressant information and efficacy prediction data) is provided to the donor selection agent 71 via API. The donor selection agent 71 uses this information to calculate the recipient prognosis score (S_Prog). Specifically, the donor selection agent 71 revises the score upward if rejection is predicted to be highly controllable. This expands the possibility of determining compatibility even for donor-recipient combinations that were previously considered high-risk medically, provided that optimized drug combinations are used, thereby contributing to maximizing transplant opportunities.

[0349] (In silico verification agent 861) The in silico validation agent 861 is a component of the AI ​​agent that functions as part of the in silico validation unit 860. Based on molecular structure data provided by the molecular design agent 851, the in silico validation agent 861 utilizes synthetic data (virtual patients) and nanomachine models generated using generative AI (such as GANs and diffusion models) to simulate the in vivo pharmacokinetics and immune response of drugs with the given molecular structure. Specifically, the in silico validation agent 861 performs efficacy validation on a virtual patient group with rare HLA types for which real-world data is scarce, simulations of immune tolerance induction for Treg-inducing vaccines, and simulations of local drug release by autonomous nanomachines. In addition, the in silico validation agent 861 selects drug candidates that meet efficacy and safety criteria based on the validation results and provides their prognosis prediction information to the donor selection agent 71, thereby contributing to the refinement of the recipient prognosis score (S_Prog) in donor selection.

[0350] (Hardware configuration of the Immunosuppressant Design System 800) As illustrated in Figure 23, the immunosuppressant design system 800 is composed of a control unit 1010, an information display unit 1020, an input unit 1030, a storage unit 1040, a communication unit 1050, and the like. Each functional unit of the immunosuppressant design system 800 is realized through the cooperation of hardware and software.

[0351] The control unit 1010 includes a processor 1011 and a memory 1012. The processor 1011 is responsible for the overall control of the immunosuppressant design system 800. The processor 1011 is implemented by, for example, a CPU, as well as a GPU or TPU to handle the computational load of deep learning and structure prediction. The memory 1012 includes, for example, ROM, RAM, and flash ROM. For example, flash ROM and ROM store various programs (including data used by the programs), and RAM is used as the work area of ​​the processor 1011. Programs stored in the memory 1012 are loaded into the processor 1011, causing the processor 1011 to execute the coded processes.

[0352] The programs stored in memory 1012 include a recipient information acquisition program, a dataset construction program, a rejection factor identification program, a molecular design program, and an in silico validation program. Furthermore, since the immunosuppressant design system 800 incorporates or integrates with the functions of the organ donation management device 600, it may also include a recipient information management program, a potential recipient prediction program, and a potential donor information management program.

[0353] The recipient information acquisition program is a program for acquiring information from the recipient information management unit 10 and / or the potential recipient information management unit 30 and passing it to the dataset construction unit 820. In other words, the recipient information acquisition program is a program for realizing the functions of the recipient information acquisition unit 810.

[0354] The dataset construction program is designed to integrate recipient and potential donor information, potential recipient prediction information, and past rejection history to build a large-scale dataset optimized for AI analysis. In other words, the dataset construction program is the program that implements the functions of the dataset construction unit 820.

[0355] The rejection factor identification program is a program that uses a graph neural network (GNN) or deep learning model to extract latent features that determine the presence or absence of rejection reactions from a large dataset, and to identify rejection factors. In other words, the rejection factor identification program is a program that realizes the function of the rejection factor identification unit 840.

[0356] The molecular design program uses protein structure prediction AI (e.g., AlphaFold2) to predict the three-dimensional structure of rejection factors and designs molecular structures that specifically bind to them, as well as compounds that bind to cryptic pockets. This program can be executed in large-scale parallel execution by leveraging scalable computing resources (e.g., Vertex AI Pipelines). In other words, the molecular design program is a program that realizes the functions of the molecular design unit 850.

[0357] The in silico validation program is a program that simulates and validates the efficacy and safety of a designed drug using synthetic data (virtual patients) and nanomachine models generated using generative AI (such as GANs and diffusion models). In other words, the in silico validation program is a program that enables the functions of the in silico validation unit 860.

[0358] Furthermore, the programs stored in memory 1012 may also include recipient information management agent programs, potential recipient prediction agent programs, potential donor information management agent programs, recipient information acquisition agent programs, dataset construction agent programs, rejection factor identification agent programs, molecular design agent programs, and in silico validation agent programs.

[0359] The recipient information management agent program is a program that implements the functions of recipient information management agent 11. The potential recipient prediction agent program is a program that implements the functions of potential recipient prediction agent 21. The potential donor information management agent program is a program that implements the functions of potential donor information management agent 41. The recipient information acquisition agent program is a program that implements the functions of recipient information acquisition agent 811. The dataset construction agent program is a program that implements the functions of dataset construction agent 821. The rejection factor identification agent program is a program that implements the functions of rejection factor identification agent 841. The molecular design agent program is a program that implements the functions of molecular design agent 851. The in silico validation program is a program that implements the functions of in silico validation agent 861.

[0360] Furthermore, to enable the immunosuppressant design system 800 to function on a cloud platform, memory 1012 also stores API integration programs, data warehouse management programs, and other similar programs. The API integration program is designed to make the functions of the Molecular Design Unit 850 and the In Silico Validation Unit 860 available as APIs, making them accessible to other components and external researchers through serverless technologies (such as Cloud Run and Cloud Functions). The data warehouse management program uses technologies such as BigQuery to manage the large-scale dataset storage unit 830 and is designed to accelerate integrated data analysis. The information display unit 1020 is a functional unit for presenting various information to operators and researchers of the immunosuppressant design system 800. The information display unit 1020 includes a display 1021 and a speaker 1022. The display 1021 visually presents operators and others with 3D models of molecular structures, simulation results, and the like. The input unit 1030 is a functional unit for operators and others to input information into the immunosuppressant design system 800. The input unit 1030 includes an operation input unit 1031, an image input unit 1032, an audio input unit 1033, and the like. The memory unit 1040 includes, for example, cloud storage or a data warehouse (such as BigQuery) and stores large datasets (integrated data), trained models, designed molecular structure data, etc. The communication unit 1050 is a functional unit that transmits and receives data with the organ donation management device 600 (or its internal components), external EHR management organizations, or affiliated pharmaceutical company systems via the communication network 400.

[0361] 8-2. Operation of the Immunosuppressant Design System 800 The immunosuppressant design system 800 of the seventh embodiment performs, for example, the immunosuppressant design process (S800) shown in Figure 24. The immunosuppressant design process (S800) is a process that includes steps corresponding to each functional part of the immunosuppressant design system 800.

[0362] The immunosuppressant design process (S800) includes a recipient information management step (S401), a potential donor information management step (S402), a potential recipient prediction step (S403), a potential recipient information management step (S404), a recipient information acquisition step (S805), a dataset construction step (S806), a large dataset storage step (S807), a rejection factor identification step (S808), a molecular design step (S809), and an in silico validation step (S810).

[0363] Recipient information management step (S401): The recipient information management agent 11 manages the recipient's information (medical information and genetic information). Potential donor information management step (S402): Potential donor information management agent 41 manages information on potential donors. Potential recipient prediction step (S403): The potential recipient prediction agent 21 predicts the potential recipient. Potential recipient information management step (S404): The potential recipient information management agent 31 manages information about potential recipients (medical information and genetic information).

[0364] Recipient information acquisition step (S805): The recipient information acquisition agent 811 acquires at least one of the recipient information managed by the recipient information management agent 11, or the potential recipient information managed by the potential recipient information management agent 31.

[0365] Dataset construction step (S806): The dataset construction agent 821 integrates the information managed by the potential donor information management agent 41 and the information obtained in the recipient information acquisition step (S805), as well as the history of rejection in past transplant cases, to construct a large-scale dataset. Large-scale dataset storage step (S807): The dataset construction agent 821 stores (retains) the constructed large-scale dataset in the large-scale dataset storage unit 830.

[0366] Rejection factor identification step (S808): The rejection factor identification agent 841 extracts potential features that determine the presence or absence of a rejection reaction from a large dataset using a graph neural network or a deep learning model, and identifies rejection factors that cause the rejection reaction based on the extracted features. Molecular design step (S809): Molecular design agent 851 designs or screens molecular structures that specifically bind to the identified rejection factor and inhibit its function. This step may include predicting the three-dimensional structure using protein structure prediction AI and searching for cryptic pockets. In silico validation step (S810): The in silico validation agent 861 simulates the pharmacokinetics and immune response of drugs with the designed or screened molecular structure to validate their efficacy. Based on the validation results, drug candidates whose efficacy has been confirmed are selected and output.

[0367] 8-3. Effects According to the immunosuppressant design system 800 of the seventh embodiment, the following effects are achieved. (Realization of predictive drug discovery) The functionality of the first embodiment enables the construction of a large-scale dataset integrating information on predicted future transplant demand (potential recipients) with information on potential donors. This eliminates the time constraint that arises when it is too late once a donor-recipient pairing is actually established, making it possible to design and prepare optimized immunosuppressants that will be needed in the future, even before a patient is placed on a waiting list or before a donor appears.

[0368] (Identification of individualized rejection factors) Based on the medical and genetic information of the recipient (including potential recipients) and potential donors managed in the first embodiment, it is possible to identify in advance the specific risk of rejection (rejection factors) that may occur in individual combinations and design drugs that act with pinpoint accuracy.

[0369] (Utilization of a rapid data collection infrastructure) The seventh embodiment is based on the vast amount of data held by the life-saving system 1 of the first and second embodiments. Real-time biometric data of potential donors and emergency situation data collected through the communication infrastructure streamlined by the second embodiment are used as analytical data for identifying rejection factors in the immunosuppressant design system, contributing to improved accuracy in drug design that reflects dynamic biological responses.

[0370] (Robust drug design that reflects on-site treatment history) Real-time on-site data collected by emergency response agents (such as treatments performed by the general public and specialists, and the transport environment) is integrated into a large-scale dataset via the dataset construction unit 820. This enables the construction of highly accurate virtual patient models that reflect changes in organ inflammation and immunogenicity, which fluctuate depending on the life-saving measures taken. This allows for the design of immunosuppressants that are effective even against complex rejection risks resulting from differences in emergency treatment.

[0371] (Ensuring the quality and improving the security of training data) Meta-AI ensures the quality (ground truth) of large datasets by monitoring and correcting biases and inconsistencies in the biometric and environmental data collected by each agent. As a result, the rejection factor identification unit and the molecular design unit can learn based on highly accurate data with noise removed, enabling the design of highly safe and effective immunosuppressants with minimal discrepancies between simulations and real-world data.

[0372] (Expansion of compatibility criteria in donor selection) The drug efficacy prediction data validated by the in silico validation unit 860 is provided to the donor selection agent 71 and used to calculate the recipient prognosis score (S_Prog). If rejection is predicted to be controllable by the drug, the score is revised upward, thus expanding the possibility of matching even donor-recipient combinations that were previously considered high-risk medically, and contributing to maximizing transplant opportunities.

[0373] (Precise molecular design based on environmental stress history) The entire history of organs (temperature deviations, storage time, physical stress during transport, etc.), which is immutably recorded on the blockchain, is integrated as training data through the dataset construction unit 820. This makes it possible to dynamically predict the probability of cryptic pockets (hidden binding sites) forming spontaneously only when protein structures change due to physical and chemical stress, and to design highly accurate specialized molecular targeted drugs that are only needed for organs that have gone through that specific transport environment.

[0374] 9. Eighth Embodiment (Pharmaceutical Design System) 9-1. Composition The eighth embodiment relates to a drug design system 900 that extends the basic architecture of the immunosuppressant design system 800 of the seventh embodiment to drug design for the treatment or prevention of common diseases. A life-saving system 1 having the drug design system 900 is configured by replacing the immunosuppressant design system 800 with the drug design system 900.

[0375] (Pharmaceutical Design System 900) The functional configuration of the pharmaceutical design system 900 according to the eighth embodiment is shown in Figure 25. The pharmaceutical design system 900 is built on a cloud infrastructure and hardware configuration similar to that of the seventh embodiment (see Figure 23), but its functional blocks include a patient information acquisition unit 910, a potential patient prediction unit 920, an integrated dataset construction unit 925, a large-scale dataset storage unit 930, a disease-related factor information management unit 940, a target factor identification unit 945, a molecular design unit 950, and an in silico verification unit 960. Each of these functional units is configured to include an AI agent that autonomously executes its respective function (patient information acquisition agent 911, potential patient prediction agent 921, integrated dataset construction agent 926, disease-related factor information management agent 941, target factor identification agent 946, molecular design agent 951, and in silico verification agent 961), and they operate in cooperation with each other.

[0376] (Patient information acquisition department 910) The patient information acquisition unit 910 has the function (patient information acquisition function) to acquire information on future patient groups (potential patients) predicted by the potential patient prediction unit 920, in addition to information on patients who currently require treatment for a disease. The information acquired here includes the patient's medical information and genetic information, and if actual measurement data is not available, it also includes hypothetical information based on a probabilistic estimation model.

[0377] (Potential patient prediction unit 920) The potential patient prediction unit 920 has the function of predicting potential patients, which are a group of individuals who are likely to contract a specific disease or require treatment within a certain period in the future, based on epidemiological information such as infectious disease surveillance data, lifestyle data, and genetic statistics data. This prediction function allows the system to proactively grasp not only the demand that is currently apparent, but also the medical demand that may arise in the future.

[0378] (Disease-Related Factor Information Management Department 940) The Disease-Related Factor Information Management Unit 940 extends the concept of "potential donors" in the seventh embodiment to manage information on factors related to the cause or exacerbation of disease (disease-related factors). Disease-related factors include pathogens (viruses, bacteria, etc.), allergens, environmental pollutants, or carcinogenic gene mutations. This management unit collects the latest pathogen information and environmental factor data from external research institutions and environmental monitoring systems and keeps it constantly up-to-date.

[0379] (Integrated Dataset Construction Unit 925) The integrated dataset construction unit 925 integrates current and future patient information acquired by the patient information acquisition unit 910 with disease-related factor information managed by the disease-related factor information management unit 940, and further integrates past case data and drug response history to construct a large-scale dataset optimized for AI analysis. The constructed data is stored in the large-scale dataset storage unit 930 and used for subsequent AI analysis.

[0380] (Target factor identification unit 945) The target factor identification unit 945 analyzes the integrated data stored in the large-scale dataset storage unit 930 using deep learning models such as graph neural networks (GNNs) to identify molecules and mechanisms (target factors) within the body that directly cause the onset or progression of disease. For example, spike proteins of viruses that are predicted to mutate, or surface receptors of specific cancer cells may be identified as target factors.

[0381] More specifically, the target factor identification unit 945 uses a gene language model (genome language model) that applies a natural language processing model with a transformer architecture to gene sequence analysis in predicting viral mutations. The target factor identification unit 945 tokenizes the RNA or DNA sequences of a vast number of viral strains collected in the past and learns them as if they were a language, thereby probabilistically generating next-generation base sequence patterns that are biologically and grammatically valid.

[0382] Furthermore, the target factor identification unit 945 performs a filtering process to evaluate the binding affinity of the spike protein structure of the generated predicted mutant strain to human cell receptors (such as ACE2) through structural simulation (molecular docking). This identifies only the three-dimensional structures of "future mutant strains" that are likely to have high infectivity, rather than simply mathematically generated sequences, as target factors, and outputs them to the molecular design unit 950.

[0383] (Molecular Design Department 950) The molecular design unit 950 designs or screens molecular structures that specifically bind to identified target factors and inhibit or control their function. Here, precise molecular design is performed, targeting cryptic pockets (hidden binding sites) and other areas formed in response to dynamic structural changes of the target factor, utilizing protein structure prediction AI and molecular dynamics calculations.

[0384] (In silico verification unit 960) The in silico verification unit 960 uses calculations to verify the efficacy and safety of the designed drug. Specifically, the in silico validation unit 960 uses generative AI, including GANs (Generative Adversarial Networks) or diffusion models, to learn statistical features from EHRs of real patients and generate statistically equivalent synthetic data while protecting privacy. In generating this synthetic data, the in silico validation unit 960 maps the complex correlations (copulas) between multidimensional data of real patients (age, sex, genotype, medical history, test results, etc.) onto a latent space using a variational autoencoder (VAE). The in silico validation unit 960 samples from this continuous latent space and reconstructs the data through a decoder, thereby generating an infinite number of "unknown patient profiles" that do not exist in the real data but are statistically and medically possible. This allows for the virtual construction of populations with rare genetic backgrounds where real data is lacking, or patient groups with complex disease conditions that may appear in the future, and comprehensively validates the efficacy and side effects of drugs for these virtual patient groups. Using this synthesized data, the in silico validation unit 960 constructs a virtual patient population, such as a group with rare HLA types that have a small number of cases in real data, and verifies side effects and efficacy.

[0385] Furthermore, the memory unit of the drug design system 900 (corresponding to the memory 1012 in the eighth embodiment) stores various programs for realizing the functions of the drug design system 900. Specifically, it stores a patient information acquisition program, a potential patient prediction program, a disease-related factor information management program, an integrated dataset construction program, a target factor identification program, a molecular design program, and an in silico validation program. The disease-related factor information management program is a program for realizing the functions of the disease-related factor information management unit 940. The target factor identification program is a program for identifying target factors that cause the onset or progression of disease using a graph neural network or a deep learning model on integrated data, and realizes the functions of the target factor identification unit 945. Furthermore, the memory unit of the pharmaceutical design system 900 (corresponding to the memory 1012 in the eighth embodiment) also stores agent programs for realizing AI agents corresponding to each of these functional units (patient information acquisition agent 911, potential patient prediction agent 921, integrated dataset construction agent 926, disease-related factor information management agent 941, target factor identification agent 946, molecular design agent 951, and in silico validation agent 961).

[0386] 9-2.Operation Next, with reference to Figure 26, the flow of the drug design process performed by the drug design system 900 will be explained. First, in the disease-related factor information management step (S901), the disease-related factor information management unit 940 (disease-related factor information management agent 941) collects and manages disease-related factor information such as the latest pathogen information and environmental factors. In parallel, in the potential patient prediction step (S902), the potential patient prediction unit 920 predicts future disease demand (occurrence of potential patients) based on epidemiological data, etc.

[0387] Next, in the patient information acquisition step (S903), the patient information acquisition unit 910 acquires current patient information and information on predicted potential patients. Then, in the integrated dataset construction step (S904), the integrated dataset construction unit 925 (integrated dataset construction agent 926) integrates this patient information, disease-related factor information, and historical data to construct a large dataset, which is then stored in the large dataset storage unit 830.

[0388] Next, in the target factor identification step (S905), the target factor identification unit 945 (target factor identification agent 946) analyzes the constructed large-scale dataset and identifies target factors for drug discovery based on the interaction between predicted patient groups and disease-related factors. For example, it identifies the most effective inhibition points from combinations of viral variants predicted to cause future epidemics and the genetic characteristics of populations at high risk of infection.

[0389] Subsequently, in the molecular design step (S906), the molecular design unit 950 (molecular design agent 951) designs the optimal molecular structure for the identified target factor. Finally, in the in silico validation step (S907), the in silico validation unit 960 (in silico validation agent 961) simulates the designed drug candidate on a virtual patient model, and outputs the one whose efficacy and safety are confirmed as the final drug candidate.

[0390] 9-3. Action and Effects According to the drug design system 900 of this embodiment, the potential patient prediction unit 920 predicts future disease demand, and the target factor identification unit 945 and molecular design unit 950 perform analysis and design based on integrated data including this information. This realizes "predictive drug discovery," which involves designing and preparing optimized drugs that will be needed in the future, even before a pandemic actually occurs or cancer progresses and becomes resistant to standard treatments. As a result, the conventional time constraint that arises when development begins only after a need becomes apparent is eliminated.

[0391] Furthermore, by using large-scale datasets that integrate environmental stress history and dynamic biological data, it becomes possible to design drugs that target cryptic pockets, which are formed only when protein structures change due to physical and chemical stress. This makes it possible to provide highly effective drugs even for highly mutable viruses and intractable cancers, which were previously difficult to treat with conventional technologies. In addition, in silico validation using virtual patients enables rapid and safe development even for novel and rare diseases for which real-world data is scarce.

[0392] Furthermore, the reliability of future predictions by the potential patient prediction unit 920 and the target factor identification unit 945 is ensured by a retrospective validation process incorporated into the system. Specifically, the system periodically runs simulations that predict the pandemic situation and disease distribution in the "future" (i.e., the present) as seen from a specific point in the past (e.g., three years ago) using only data from that point in time as training data. The system evaluates the discrepancy (prediction error) between this prediction result and the currently observed real-world data, and after confirming that the error is within a predetermined acceptable range, it performs a prediction of the "true future" based on current data. This self-validation loop suppresses hallucination and overfitting of AI models in predictive drug discovery, and continuously guarantees technical validity.

[0393] 9-4. Application Examples of the Pharmaceutical Design System 900 9-4-1. Pandemic countermeasures for infectious diseases (antiviral drugs, etc.) In its application to infectious disease pandemic response, the drug design system 900 enables proactive medical intervention using a predictive drug discovery approach against the threat of viruses that have not yet spread or emerged in human populations. First, the disease-related factor information management unit 940 extends the concept of "potential donors" to wild animals, livestock, and viral reservoirs present in the environment that are natural hosts of zoonotic diseases, and continuously monitors the vast amount of genomic data obtained from them. Specifically, the disease-related factor information management unit 940 collaborates with international virus surveillance networks and environmental DNA monitoring systems to continuously collect and accumulate viral gene sequences contained in samples taken from wild animals (bats, birds, etc.) and sewage epidemiological data as "disease-related factors."

[0394] The target factor identification unit 945 predicts the direction of viral evolution and mutation by applying a gene language model with a Transformer architecture to the accumulated current viral genome data. The target factor identification unit 945 learns RNA or DNA sequences as language and interprets natural selection pressure and host immune evasion behavior as context to probabilistically generate next-generation mutation patterns that are biologically viable and have a high probability of acquiring the ability to infect and transmit to humans. Furthermore, based on the predicted mutation sequence information, the target factor identification unit 945 uses protein structure prediction AI to construct a three-dimensional model of the spike protein and other structures of future mutant strains, and identifies structural sites with high binding affinity to human cell receptors (such as ACE2) as specific "target factors" on which the drug should act.

[0395] In parallel, the potential patient prediction unit 920 performs epidemiological simulations using human movement data such as global air travel data, climate change data, and demographic data for each region to predict areas that will become the epicenter of infection spread and groups at high risk of severe illness when the variant strain spills over (transmits between species) to humans, and identifies these as "potential patients." The patient information acquisition unit 910 acquires the genetic characteristics of this predicted potential patient group (such as the distribution of HLA types and polymorphisms in immune-related genes), and integrates this with the predicted viral mutation data in the integrated dataset construction unit 925.

[0396] Based on this integrated data, the molecular design unit 950 designs molecular structures for small molecules, antibodies, or mRNA vaccines that physically or chemically inhibit the function of predicted mutant target factors (such as spike proteins) by targeting dynamic structural changes and cryptic pockets (hidden binding sites) revealed by molecular dynamics calculations. The molecules designed here undergo multi-objective optimization using AI to ensure cross-reactivity not only with currently circulating strains but also with predicted future mutant strains.

[0397] Finally, the in silico validation unit 960 uses a virtual patient population created by generative AI, which mimics the genetic background and immune profiles of predicted potential patient groups, to simulate and validate the efficacy and safety of the designed drug candidate. This process enables the system to obtain statistically significant efficacy data in the pre-pandemic stage, when clinical trials using real patients are impossible, and to present optimized drug candidates that can be immediately manufactured and supplied the moment a viral outbreak becomes apparent.

[0398] 9-4-2. Cancer Treatment (Molecular Targeted Therapies) In its application to cancer treatment (molecular targeted drugs), the drug design system 900 predicts resistant mutant strains that may emerge in the future, even before cancer develops or before cancer cells acquire resistance to existing chemotherapy or first-line treatments, and realizes predictive drug discovery by designing drugs that target these specific structural changes. First, the disease-related factor information management unit 940 integrates and manages a database of known driver gene mutations, along with environmental stressors that increase cancer risk (such as ultraviolet radiation, chemical exposure, and smoking history), the specific mutational signatures they leave on the genome, and other known driver gene mutations as "disease-related factors." Meanwhile, the potential patient prediction unit 920 predicts "potential patients"—risk groups with a high probability of developing a specific cancer in the future, or patients currently undergoing treatment who are at high risk of recurrence or metastasis in the future—based on genetic tendencies such as polygenic scores and time-series analysis of lifestyle logs.

[0399] Next, the target factor identification unit 945 uses an evolutionary algorithm that simulates cancer evolution to predict the evolutionary trajectory of cancer cell clones when controlled environmental stress factors and selective pressures from standard therapeutic drugs are applied to the predicted genetic background (germline mutations) of potential patients. Through this simulation, the target factor identification unit 945 identifies somatic mutations (e.g., specific secondary mutations such as EGFR and KRAS) that have the highest probability of appearing in the future, and defines proteins with these mutations as "target factors." Furthermore, the target factor identification unit 945 applies a combination of long-term molecular dynamics (MD) simulations and deep learning analysis to the three-dimensional structure of the predicted mutant protein to identify the probability and location of "cryptic pockets" (hidden binding sites) that are exposed on the surface only at the moment the protein undergoes a structural change under thermal fluctuations or specific environmental stress, and which cannot be discovered by static crystal structure analysis.

[0400] The molecular design unit 950 designs molecular structures for small or medium-sized compounds that are compatible with the dynamic shape and physicochemical properties (hydrophobicity, charge distribution, etc.) of identified cryptic pockets and that allosterically inhibit protein function by strongly binding the moment the pocket opens. In this molecular design process, AI uses generative models to create compounds with novel scaffolds not found in existing compound libraries, and performs multi-objective optimization to block cancer cell survival signals while minimizing toxicity to normal cells.

[0401] The in silico validation unit 960 generates a virtual patient population that reflects the genetic diversity of potential patient groups and validates the tumor reduction effect and the incidence of side effects when a designed molecularly targeted drug is administered through simulation. This process eliminates the conventional time constraint of treatment stalling due to the emergence of resistance mutations, enabling the development and preparation of optimal molecularly targeted drugs that proactively contain cancer evolution before clinical needs become apparent.

[0402] 10. Alternative Examples The present invention is not limited to the embodiments described above. It will be obvious to those skilled in the art that various modifications or alterations can be conceived within the scope of the claims, and these will naturally also fall within the technical scope of the present invention. Furthermore, the components of the above embodiments may be combined in any way without departing from the spirit of the invention.

[0403] (Substitution using a probabilistic gene model) In the above embodiment, the potential donor information management unit 40 manages genetic information. However, for potential donors who do not have actual genetic information, a probabilistic genetic model (Synthetic Genetic Profile) generated by the immunosuppressant design system 800 based on their residential area, race, or phenotype data may be managed as a substitute for actual genetic information. In this case, the rejection factor identification unit 840 performs risk assessment and drug design based on genetic background information as a probability distribution, rather than definitive genetic information. This makes it possible to start the operation of the entire system and predictive drug discovery without collecting highly private genetic information of individuals in advance.

[0404] (Fit scoring based on dynamic information) In the first embodiment, the transplant suitability detection unit 50 may be configured to score the suitability of a recipient or potential recipient and a potential donor by taking into account not only static information such as blood type and HLA type, but also dynamic information such as the recipient's or potential recipient's current use of immunosuppressants and the donor's current organ function status (e.g., renal function, hepatic function). This enables the immunosuppressant design system 800 to learn the correlation between these dynamic physiological parameters and the risk of rejection through a large dataset, resulting in the design of more accurate immunosuppressants that are dynamically adapted not only to genetic compatibility but also to the individual patient's metabolic state and drug sensitivity at the time of transplantation.

[0405] (Donor selection trigger combining on-site information) The health status monitoring unit 60 can combine not only the biometric information of potential donors but also on-site information from traffic accidents and disasters (e.g., car collision sensor data, disaster simulation results) to trigger donor selection. This allows the immunosuppressant design system 800 to learn the effects of physical impact from trauma and environmental stress on the immunogenicity of transplanted organs as part of a large dataset, enabling it to design more powerful and precise immunosuppressants specifically designed to suppress acute inflammatory responses unique to traffic accidents and disasters.

[0406] (Extraction of vital signs through video analysis) The health status monitoring unit 60 can monitor the healthy state of potential donors based on data acquired from a non-contact vital sign sensing system that uses images from surveillance cameras and other sources. This system uses video analysis to detect subtle color changes, body movements, and posture changes in the potential donor's face and skin, and continuously monitors non-contact vital signs such as heart rate, respiratory rate, and state of wakefulness. As a result, the immunosuppressant design system 800 can integrate autonomic nervous system activity in a natural state and changes in the skin surface associated with subtle inflammatory responses, which were difficult to obtain with conventional invasive examinations, into a large dataset and train the AI. This has the effect of designing highly accurate personalized immunosuppressants that also take into account the patient's daily physiological state and the patterns of immune dynamics fluctuations associated with stress load.

[0407] More specifically, vital sign sensing using video analysis identifies skin areas (mainly the face and hands) of the subject (potential donor) from video devices such as surveillance cameras and smartphones. By analyzing the average brightness and color information of these areas over time and removing noise, periodic signals associated with the heartbeat are extracted. This makes it possible to continuously and non-contact measure heart rate (pulse rate) and respiratory rate from breathing movements accompanied by body movements. Furthermore, the immunosuppressant design system 800 integrates the non-contact vital data acquired in this way into a large-scale dataset. This allows for the analysis of the correlation between subtle autonomic nervous system activity and fluctuations in hemodynamics in daily life, which were difficult to measure with conventional contact-type examinations, and signs of immune rejection, using deep learning models. This enables the design of more accurate and personalized immunosuppressants that can adapt to changes in immune dynamics according to the patient's mental or physical stress levels.

[0408] Vital sign sensing using video analysis allows for simultaneous, unconscious monitoring of multiple potential donors for any signs of abnormality, for example, by using multiple surveillance cameras installed in a large space. In addition, by acquiring brainwave data from potential donors using devices capable of measuring brainwaves (such as head-mounted devices), and having AI recognize changes in the potential donor's posture, apnea, and abnormal facial color, it becomes possible to detect signs of abnormality early. Furthermore, the immunosuppressant design system 800 integrates and analyzes this multimodal information (external features from video and internal biological signals such as brainwaves), thereby identifying complex rejection triggers that cannot be detected by a single indicator. This dramatically improves the quality and quantity of training data for designing and validating immunosuppressants with novel mechanisms of action that block the process of abnormal biological reactions chaining together and leading to rejection at the molecular level.

[0409] (Acquisition of detailed medical information through integration with electronic medical records) In addition to acquiring data from wearable devices, the health monitoring unit 60 may be configured to link with the electronic medical record system of the medical institution where the potential donor regularly receives treatment, and to acquire detailed medical information such as blood test data and imaging diagnostic results.

[0410] This allows for the acquisition of detailed and objective clinical information, including not only biometric data from wearable devices (heart rate, blood pressure, etc.) but also blood test data and imaging diagnostic results. This significantly improves the accuracy of evaluating the health status of potential donors, particularly in determining organ condition and compatibility. Furthermore, it enables comprehensive monitoring using medical information, not only in emergencies but also during normal times, based on the long-term medical history and health trends of potential donors, contributing to more reliable donor selection and preservation of organ donation opportunities. In addition, the immunosuppressant design system 800 integrates the morphological and functional characteristics of organs obtained from long-term clinical data and imaging diagnostics into a large-scale dataset and trains AI on it. This allows for the highly accurate identification of specific rejection risk factors that depend on the individual patient's medical history and the physiological state of organs, which cannot be predicted by genetic information alone. This enables the design of immunosuppressants with molecular structures optimized for the complex in vivo environment.

[0411] (Donor selection considering geographical distance) The donor selection unit 70 may have a function that considers the geographical distance between the potential donor and the recipient, in addition to the compatibility score and health status. This makes it possible to shorten the organ transport time and further shorten the ischemic time. In addition, the immunosuppressant design system 800 integrates and analyzes predicted transport time data based on geographical distance into a large dataset, quantitatively learning the impact of ischemic time during transport on organ tissue damage and changes in immunogenicity. This has the effect of designing an immunosuppressant that is optimized to strongly block specific rejection signals amplified due to ischemia-reperfusion injury, even when long transport times are unavoidable.

[0412] (Centralized information management through an integrated database) The recipient information management unit 10 or the potential recipient information management unit 30 and the potential donor information management unit 40 may be configured as a management unit using a single integrated database, and the transplant condition match detection unit 50 may access this integrated database to select highly matched pairs. In addition, by configuring the dataset construction unit 820 of the immunosuppressant design system 800 to directly access this integrated database, it becomes possible to comprehensively collect the multimodal data necessary for AI learning.

[0413] By centrally managing recipient, potential recipient, and potential donor information in a single integrated database, the information retrieval and matching process by the transplant suitability detection unit 50 is accelerated compared to when multiple systems are involved. This improves the performance of the life-saving system 1, enabling rapid organ extraction from donors and provision of those organs to recipients in a transplantable state. Furthermore, in the immunosuppressant design system 800, a large amount of heterogeneous data, including genetic information, past rejection history, and current physiological status, is aggregated on a single platform. This dramatically speeds up the construction and updating of large datasets necessary for training graph neural networks (GNNs) and deep learning models, significantly reducing the lead time of the computational drug discovery process from rejection factor identification to molecular design.

[0414] Furthermore, since all relevant information (such as the current recipient's medical urgency, predictive information on potential recipients, and the latest biometric data of potential donors) is managed on a single platform, data duplication, inconsistencies, and discrepancies in freshness can be prevented. This ensures that information on sudden changes in potential donors detected by the health status monitoring unit 60 is reflected in the transplant suitability assessment without delay and reliably, improving the accuracy and reliability of the assessment. At the same time, the immunosuppressant design system 800 can train its AI model using a high-quality dataset free of data inconsistencies, reducing the risk of mislearning due to noise and enabling the design and in silico validation of highly accurate molecular structures that are truly suited to the complex in vivo environment of each individual patient.

[0415] Furthermore, by handling future organ demand forecasts (hypothetical information) from the potential recipient prediction unit 20, current waiting list information (actual demand), and a list of potential donors on the same platform, the optimal organ resource allocation strategy can be dynamically determined across immediate and on-demand donation processes. This maximizes organ donation opportunities and makes life-saving medical care through organ transplantation extremely efficient. In addition, the immunosuppressant design system 800 integrates and analyzes this future demand forecast and current supply information to proactively identify potential supply-demand gaps and increasing trends in specific rejection risks, enabling highly accurate "predictive drug discovery" that designs and develops necessary immunosuppressants even before patients are actually registered on the waiting list.

[0416] Furthermore, the AI ​​agent group that forms the core of the life-saving system 1 (operator agent, caller agent, health status monitoring agent, etc.) can directly exchange and reference highly efficient information (such as real-time health risk indicators) via a single database. This simplifies the data flow in complex AI agent communication protocols (such as giverlink mode), improving the overall system's coordination and response speed. In addition, the in silico verification unit 860 of the immunosuppressant design system 800 can continuously improve the prediction accuracy of the simulation by taking real-time feedback data (such as biological responses and compatibility changes after drug administration) obtained from AI agents in the clinical setting without delay through this simple data flow, and autonomously correcting and evolving the parameters of the virtual patient model.

[0417] (Global centralized management of EHRs) The life-saving system 1 will be linked to a global, centralized management system for EHRs (Electronic Health Records). Specifically, it will be possible to realize a highly reliable medical data linkage platform using My Number (Japan), National Health Insurance Number (UK), Social Security Number (US), etc. Establishing such a global medical data linkage platform will enable the realization of a more efficient life-saving system that takes into account the global supply and demand balance of organs. Through this linkage, the organ donation management device 600 will be able to continuously grasp and process virtual information regarding the global demand and supply of organs. In addition, the dataset construction unit 820 of the immunosuppressant design system 800 will be able to comprehensively collect vast amounts of clinical data on diverse races, genetic backgrounds, and rare HLA types that are difficult to collect in a single country or region, and will be able to construct an unbiased, high-quality, large-scale international dataset that is essential for training deep learning models such as graph neural networks (GNNs) through this global linkage platform.

[0418] Specifically, the potential recipient prediction unit 20 utilizes global EHR and registry data to predict with extremely high accuracy the risk and demand for future waitlist registration across national borders for specific organ types (global demand forecasting). Furthermore, based on this global demand forecasting, the immunosuppressant design system 800 can identify early trends in specific immunological risks and rejection factors that are expected to increase in particular regions or ethnic groups in the future, and realize "global predictive drug discovery" by designing and developing immunosuppressants with molecular structures tailored to those populations in advance, even before patients actually fill the waitlists.

[0419] The transplant suitability detection unit 50 continuously matches demand information of recipients and potential recipients worldwide with information on potential donor supply candidates. This makes it possible to identify the most suitable supply-demand pairs on a global level, transcending regional constraints, and contributes to maximizing transplant opportunities and improving transplant outcomes. Furthermore, the in silico validation unit 860 of the immunosuppressant design system 800 utilizes global matching data to construct a virtual patient population that mimics rare donor-recipient combinations that are difficult to validate with real data alone. This allows for precise simulation and verification using generative AI to determine what side effects and efficacy the designed drug will exhibit in a variety of genetic combinations.

[0420] The organ preservation unit 80 enables the long-term storage of excised organs in a transplantable state, particularly when suitable recipients are unavailable or anticipated to be potential recipients in the future. This plays a crucial role in preventing the waste of globally scarce organ resources and increasing the certainty of supply to meet future demand. In addition, the immunosuppressant design system 800 utilizes the long-term storage period of the organ to design a dedicated immunosuppressant optimized for the physiological state of the organ and the immune profile of the potential recipient expected to receive the transplant in the future. By completing in silico efficacy validation, it is possible to immediately provide optimal drug therapy to minimize rejection when the preserved organ is actually transplanted.

[0421] (Real-time multilingual communication using predictive translation functionality) In the first embodiment, the interpretation function in the call unit 110 has an advanced real-time interpretation function (predictive interpretation function) that operates when the language of the caller and the language of the human operator differ. This interpretation function is executed by the AI ​​agent predicting (predicting) the content of the utterance before a unit of conversation (utterance) is completed. This minimizes call delays and enables rapid and smooth communication that overcomes language barriers, contributing to accurate information transmission and faster response in emergencies. This is one specific example of how computer-based interpretation (machine interpretation) in the direct conversation unit 112 can be efficiently realized by applying multi-AI agent technology.

[0422] Furthermore, in the immunosuppressant design system 800, this predictive translation function enables the accurate and real-time collection and integration of unstructured data, such as subjective symptom reports and detailed medical histories, from potential donors and recipients around the world speaking different languages, without loss of nuance. As a result, the dataset construction unit 820 can construct an extremely diverse and high-quality global dataset free from linguistic and regional biases. Consequently, the rejection factor identification unit 840 and the molecular design unit 850 can learn complex immunological features associated with rare HLA types and genetic backgrounds specific to particular racial and linguistic regions, enabling the design of highly versatile and personalized immunosuppressants with high accuracy.

[0423] (Grief care function provided by AI agent) The operator agent 101 may have a grief care function that, when interacting with the donor's family in connection with the donor selection process, estimates the family's emotional state and level of grief from their tone of voice and speech content, and dynamically adjusts the speed of the conversation, tone of voice, and vocabulary used according to the estimation results. This function allows the operator agent 101 to smoothly convey necessary information and confirm intentions while being mindful of the family's psychological distress, thereby increasing the family's acceptance of the system.

[0424] Furthermore, in the immunosuppressant design system 800, the grief care function enables the early establishment of psychological trust with the family, making it possible to smoothly and reliably obtain consent for detailed genetic analysis of the donor and provision of specific medical history data. The dataset construction unit 820 can then integrate high-quality data on rare HLA types and immunological characteristics, which were previously difficult to obtain due to family refusal or confusion, into a large dataset without any data loss. As a result, the rejection factor identification unit 840 can perform highly accurate AI learning based on a more diverse and comprehensive genetic background, which ultimately results in the design of optimized immunosuppressants that are suitable for a wider range of patients and rare cases.

[0425] (Utilization of organs and cells as foundational resources for regenerative medicine) In the first embodiment, the life-saving system 1 that optimizes the organ donation process has the function of utilizing preserved organs or cells as foundational resources for regenerative medicine, and the function of integrating these technologies. The purpose of this alternative example is that preserved organs and cells are being utilized in regenerative medicine as "transplant resources" and "seeds of regeneration," and in particular, when combined with iPS cell and stem cell preservation technologies, it has the potential to directly lead to resolving the shortage of organ transplants and realizing personalized medicine.

[0426] This alternative approach is applicable to the replacement and complementation of organ transplantation. While organ transplantation suffers from a severe shortage of donors, reliance on transplantation can be reduced by repairing and regenerating organs or their constituent cells stored in the organ preservation unit 80 of the life-saving system 1 using regenerative medicine technology. Examples include cardiomyocyte sheet transplantation after myocardial infarction and hepatocyte transplantation for liver disease.

[0427] This alternative is applicable to the preservation and reuse of iPS cells and stem cells. iPS cells are pluripotent cells and, if preserved, can differentiate into necessary organs and tissues in the future. Creating them from the patient's own cells reduces rejection and enables personalized treatment. By linking with potential donor and recipient information managed by the organ donation management device 600, it becomes possible to centrally manage these cells as a stable long-term supply resource using preservation technologies (vitrification and other preservation techniques). In addition, the immunosuppressant design system 800 analyzes the HLA type and genetic characteristics of these centrally managed cell resources as part of a large-scale dataset, predicting in advance the risk of specific rejection reactions that may occur when transplanting allogeneic iPS cells or stem cells. This makes it possible to design and prepare immunosuppressants and Treg-inducing vaccines suitable for that particular cell line before cell donation, resulting in the provision of rapid and safe regenerative medicine treatment with minimized rejection risk, even when the patient's own cells are not used.

[0428] This alternative is applicable to tissue engineering and 3D bioprinting. Specifically, it becomes possible to incorporate the function of artificially printing organs and tissues (3D bioprinting technology) using cells from preserved organs as bioink into a life-saving system. In addition to the regeneration of cartilage and skin, complex organs such as the heart and kidneys will be targeted in the future. Furthermore, the molecular design unit 850 of the immunosuppressant design system 800 simulates the subtle inflammatory and immune responses that living organisms exhibit to the unique structure and scaffold material of bioprinted artificial tissues. By designing locally acting drugs to suppress these responses and increase the engraftment rate, as well as drug molecules for nanomachine DDS (drug delivery systems) to be pre-integrated into the tissue, it supports the long-term maintenance of the function of artificial organs with complex three-dimensional structures.

[0429] This alternative approach can also be applied to disease modeling and drug discovery research. Specifically, disease models can be created from cells of preserved organs and used to elucidate disease mechanisms and develop new drugs. More specifically, the in silico validation unit 860 of the immunosuppressant design system 800 utilizes organoids and disease models created from preserved cells as a biological assay system (wet validation environment) to verify the efficacy of novel drug candidates designed by AI. This allows for highly accurate verification and feedback of computer-predicted binding affinity to cryptic pockets and pharmacological effects using actual human cell-derived tissue models, dramatically shortening the development cycle of innovative immunosuppressants with fewer side effects while reducing reliance on animal experiments.

[0430] This alternative method can also be applied to improving beauty and quality of life. Specifically, it can be used for tooth regeneration, hair follicle regeneration, and skin regeneration using cells from preserved organs. This has the potential to expand its application beyond life-saving medicine to areas aimed at improving quality of life.

[0431] Preserved organs are crucial as foundational resources for regenerative medicine, and when combined with iPS cell and stem cell preservation technologies, they directly contribute to resolving organ transplant shortages, personalized medicine, and drug discovery research. The integration of organ preservation technology, regenerative medicine, AI, and 3D printing will greatly contribute to the advancement of life-saving medical technology. In particular, integration with the Immunosuppressant Design System 800 will establish molecular-level control technologies that induce immunological tolerance, which is essential for tissues and organs created by regenerative medicine to function in vivo over the long term. In other words, by combining the preservation of physical "seeds of regeneration" with the design of chemical "rejection control," it will be possible to maximize the effects of regenerative medicine and simultaneously solve the dual challenges of donor shortages and rejection reactions in transplant medicine.

[0432] (Functionality of the AI ​​agent in the Emergency Response Device 300) In the third embodiment, the emergency responder agent 301 may be an AI agent that operates in each emergency responder device 500 and emergency responder device 300. With this configuration, it becomes possible to respond quickly to emergencies not only with the cooperation of registered skilled users of automobiles 200A, smartphones 200B, monitored terminals, smartwatches 200C, and smart glasses 200D, but also with the cooperation of registered skilled users riding in ambulances 300A, fire trucks 300B, patrol cars 300C, etc.

[0433] Furthermore, in the immunosuppressant design system 800, real-time on-site data collected by emergency response agents 301 through these diverse devices (detailed intervention history from initial treatment by ordinary citizens to advanced life-saving treatment by specialists, as well as transport environment data) is comprehensively integrated into a large-scale dataset via the dataset construction unit 820. As a result, the in silico validation unit 860 can construct a highly accurate virtual patient model that reflects even the subtle effects of differences in the quality and content of life-saving measures on the inflammatory state and immunogenicity of organs. This enables the design of optimized immunosuppressants that demonstrate robust efficacy even against complex rejection risks that fluctuate depending on the emergency treatment provided.

[0434] (Monitoring and coordination of multiple agents by an independent meta-AI) In the fourth embodiment, the operator agent 101 functions as a meta-AI, but the life-saving system 1 may have a separate meta-AI that monitors, coordinates, manages, optimizes, etc., the operator agent 101, the caller agent 201, and the emergency responder agent 301. In this case, the operator agent 101 also works under the command of the meta-AI together with the emergency caller agent 201 and the emergency responder agent 301.

[0435] In addition, in the immunosuppressant design system 800, this independent meta-AI plays a role in ensuring the quality (Ground Truth) of the large-scale dataset generated by the dataset construction unit 820 by comprehensively monitoring and correcting biases and inconsistencies contained in the biological and environmental data collected by each agent. As a result, the rejection factor identification unit 840 and the molecular design unit 850 can correctly learn complex immunological interactions based on highly accurate data from which noise caused by learning biases and malfunctions of individual agents has been eliminated. Consequently, it is possible to design highly safe and effective immunosuppressants with very little discrepancy between simulation and real-world data.

[0436] (Advanced suitability assessment through ultra-early precision medical collaboration) The transplant suitability detection unit evaluates compatibility by considering recipient information, potential donor information, and real-time homeostasis indicators based on a comprehensive understanding of inter-organ networks for early disease prediction and prevention. This enables ultra-precise scoring to predict the long-term success rate of transplantation by comprehensively analyzing dynamic biological data such as the recipient's current immune response state and the micro-inflammation level of the donor organ, in addition to static information such as blood type and HLA type. Furthermore, the immunosuppressant design system 800 integrates this dynamic biological data (homeostasis indicators and micro-inflammation levels) into a large-scale dataset and trains AI on it. This allows it to identify specific rejection factors that are spontaneously activated due to the physiological stress state of the organ at the time of transplantation and the micro-inflammatory environment, which cannot be predicted by static genetic information alone. It then enables the design of highly effective immunosuppressants with fewer side effects by targeting cryptic pockets (hidden binding sites) that are formed on the protein surface only under those specific conditions.

[0437] (Maximizing information gathering through high-speed AI agent communication protocols) The operator agent 101 of the emergency call center device and the caller agent 201 of the emergency call device 200 communicate in ultra-fast conversation modes such as giverlink mode and mathematical data sharing mode, which are designed to reduce conversation time during emergencies. As a result, from the moment of the call, the AI ​​instantly structures and analyzes the potential donor's biometric data (heart rate, blood pressure, brain waves, etc.), and transmits highly efficient information that detects signs of serious health conditions to the organ donation management device in real time, faster than human intervention, enabling rapid donor selection and preparation of the medical team.

[0438] In addition, in the immunosuppressant design system 800, dynamic multimodal data showing the rapid physiological changes and progression of inflammatory responses in donors under emergency conditions are directly integrated into the dataset construction unit 820 as structured data in matrix or vector format without delays or loss of information associated with natural language processing, through this mathematical data sharing mode, etc. As a result, the in silico validation unit 860 can accurately simulate the harsh microenvironment unique to life-threatening organs, which is difficult to capture with normal clinical tests, and can design robust immunosuppressants that can control the risk of rejection that only appears under such extreme conditions.

[0439] (Organ resuscitation using molecular nanotechnology (NMT)) The Organ Preservation Device 700 preserves organs through vitrification, a method suitable for long-term storage, and then revives (restores) them to a transplantable state by applying molecular nanotechnology (MNT), an autonomous robotic technology at the molecular level. During this revitalization process, MNTs explore the body's vascular network and cells, detecting and repairing not only ischemic damage, CPA (cryoprotective agent) toxicity, and physical damage caused by ice crystals and cracks, but also signs of chronic inflammation (senoinflammation), a common underlying factor in aging and major diseases, at the molecular level. This maximizes the function and health of the organ immediately before transplantation.

[0440] In addition, the immunosuppressant design system 800 utilizes this precise repair process by MNT to locally deliver and release immunosuppressants optimized for deployment on autonomous nanomachines designed by the molecular design unit 850 to target sites within organs. At this time, based on nanomachine DDS simulations by the in silico verification unit 860, the risk of drug leakage throughout the body is minimized while maintaining drug concentrations within the therapeutic range at the organ site. This has the effect of completing immunological environmental adjustment (immunological resuscitation) that specifically blocks rejection factors, simultaneously with physical functional repair, at the pre-transplant stage.

[0441] (A donor selection process that integrates ethical considerations) The donor selection unit 70 acts as a trigger to initiate a rapid organ donation process, but requires confirmation of the donor's intention to donate organs. If a serious health condition is detected in a potential donor whose intention to donate is unknown, the system automatically initiates a process of contacting the family and a pre-mortem intention confirmation process estimated based on the EHR, prioritizing this over donor selection. This ensures both highly efficient rapid life-saving medical care and compliance with ethical and legal constraints. In addition, the immunosuppressant design system 800 utilizes the waiting time while family contact and intention confirmation are being made. The in silico validation unit 860 generates synthetic data based on statistical features extracted from the EHR, allowing the system to begin preliminary design and simulation of immunosuppressants suited to the donor candidate's genetic background without using personally identifiable biometric data, while fully protecting privacy. This means that the optimal drug candidate, whose efficacy has already been verified on a computer, can be presented the moment formal consent is obtained. This offsets the time lost due to adherence to ethical processes by shortening the lead time for drug discovery and verification, thus maintaining the speed and quality of transplant medicine as a whole without compromising it.

[0442] (Application of dynamic stress data derived from life-saving processes to drug discovery) The health monitoring unit 60 collects real-time data on the donor's death process and rapid physiological changes during life-saving measures (such as cytokine storms, ischemia-reperfusion injury, and rapid blood pressure fluctuations), which is then integrated into the immunosuppressant design system 800 as learning data for identifying rejection factors. This dynamic biological data under extreme conditions provides crucial information for elucidating specific inflammatory markers that trigger acute rejection reactions immediately after transplantation, which cannot be captured by static EHRs or normal laboratory values, as well as hidden antigen presentation processes that are expressed only under stressful conditions.

[0443] (Ensuring the authenticity of training data through blockchain) In the sixth embodiment, the entire history of the organ (biological response at the time of extraction, preservation treatment data, transportation history, and post-transplant prognosis) recorded immutably by the blockchain recording unit 81 is utilized as ground truth data to guarantee the quality of the AI ​​drug discovery training dataset. This sets it apart from general medical big data, where the origin and processing history of the data are unclear, and dramatically improves the accuracy of molecular design of immunosuppressants and prediction of side effects based on highly reliable data with minimized noise and bias, while also strengthening the explainability regarding the safety of the developed drugs.

[0444] (A federative learning platform that applies privacy protection technology) The Recipient Information Management Department 10 and the Potential Donor Information Management Department 40 will use zero-knowledge proofs and federated learning technologies to build a decentralized drug discovery platform that shares and updates only the learning parameters (gradient information) of AI models, while keeping sensitive genetic and biometric data, which are difficult to transfer across borders due to national regulations, within the original medical institutions and devices. This will balance the protection of personal information with the utilization of data on a global scale, and promote the development of universal and personalized immunosuppressants that can address rare HLA types and genetic backgrounds specific to particular races and regions.

[0445] (Privacy protection compliance assessment using zero-knowledge proof) The transplant suitability detection unit 50 may be configured to determine suitability without disclosing specific personal information or raw data of the recipient and potential donor to each other, using zero-knowledge proof (ZKP) technology. Specifically, the suitability detection agent 51 receives only cryptographic proof (Proof) generated from attribute information such as the potential donor's HLA type and blood type, and verifies the validity of the proof, thereby confirming only the fact that the donor "suits the transplant conditions" without decrypting the actual medical data. This makes it possible to achieve both a high level of privacy protection and rapid suitability detection.

[0446] Furthermore, in the immunosuppressant design system 800, by applying this zero-knowledge proof technology to the data collection process of the dataset construction unit 820, it becomes possible to incorporate into a large-scale dataset only by cryptographically verifying the authenticity of attributes such as possessing specific genetic characteristics (e.g., rare HLA types or specific rejection-related genes), without directly sharing sensitive genetic information itself that is restricted in transfer by the laws and regulations of each country. As a result, the rejection factor identification unit 840 can safely utilize highly sensitive and diverse global data, which was previously difficult to access due to privacy protection barriers, as a learning base. Consequently, it can accurately identify highly versatile and reliable molecular targets for immunosuppressants that also cover rejection reaction mechanisms specific to patient groups with rare genetic backgrounds.

[0447] (Autonomous decision mode using edge AI) The caller agent 201 of the emergency call device 200 may have an autonomous determination mode that functions when the communication network 400 is shut down due to a large-scale disaster or the like. In this autonomous determination mode, the caller agent 201 uses a lightweight inference model (edge ​​AI) that is completed within the device to autonomously detect the transition of a potential donor to a critical condition from the acquired biometric data. The detected critical condition and the time of detection are logged and temporarily stored in an immutable area (secure element) within the device, and are preferentially transmitted to the emergency call center device 100 when communication is restored. This prevents the loss of donor occurrence information during periods when the communication infrastructure is down, and enhances the robustness of the life-saving system.

[0448] Furthermore, in the immunosuppressant design system 800, this autonomous judgment mode allows the donor's biological data (physiological changes in the process leading to death and rapid stress responses) recorded under extreme conditions such as communication disruption to be retrospectively integrated into the dataset construction unit 820 without any data loss. This makes it possible to secure organ inflammation and immune response data under the harsh environment unique to disasters, which is difficult to obtain under normal circumstances, as a learning resource. As a result, the rejection factor identification unit 840 can comprehensively analyze rejection factors that are specifically expressed only when the donor is subjected to physical and psychological stress so severe that communication infrastructure is non-functional, as well as the dynamics of cryptic pockets formed on the protein surface due to rapid environmental changes. Consequently, it is possible to design highly robust immunosuppressants that exhibit high efficacy not only in stable environments under normal circumstances but also for organs extracted under unstable supply environments such as those during large-scale disasters.

[0449] (Application of Dynamic NFTs and Non-Transferable Tokens) In the sixth embodiment, the organ NFT generated by the identifier generation unit 82 may be implemented as a dynamic NFT (dNFT) in which metadata is automatically updated according to changes in the organ's state, such as temperature and elapsed time, obtained from an external oracle via a smart contract. Furthermore, it is desirable that this organ NFT be designed as a SoulBound Token (SBT) that cannot be transferred between wallets. This makes it possible to reflect changes in the organ's state in the digital identifier in real time and to prevent the resale or transfer of organs, which would be ethically unacceptable, at the system architecture level.

[0450] Furthermore, in the immunosuppressant design system 800, the environmental stress history of organs (temperature deviations and subtle changes due to the passage of storage time), which is recorded moment by moment on the blockchain by this dynamic NFT (dNFT), is integrated as time-series learning data accurately linked to post-transplant prognosis data through the dataset construction unit 820. As a result, the rejection factor identification unit 840 and the molecular design unit 850 can dynamically predict not only the probability of genetic incompatibility but also the probability of the occurrence of cryptic pockets (hidden antigen-presenting sites) that are spontaneously formed due to changes in the three-dimensional structure of proteins caused by physical and chemical stress during transport. This has the effect of enabling the design of special immunosuppressive protocols and molecular targeted drugs that are only necessary for organs that have gone through that specific transport environment, with high accuracy based on noise-free "ground truth data" whose authenticity is mathematically guaranteed by Soulbound Tokens (SBTs).

[0451] (Safety device using a physical shutoff mechanism) The output path of the donor selection signal from the donor selection unit 70 may be equipped with a physical blocking mechanism (hardware kill switch) that can only be deactivated by a human operator or authorized medical professional through physical operation (such as operating a hardware switch with a physical key or biometric authentication). With this configuration, even if an AI agent such as the donor selection agent 71 malfunctions or is hacked and makes an incorrect donor selection decision, as long as the physical blocking mechanism is deactivated, the final actions on the real world, such as dispatching an organ retrieval team or issuing medical intervention orders, will be prevented. This ensures the ultimate safety of a system that involves human lives at the hardware level.

[0452] Furthermore, in the immunosuppressant design system 800, this physical blocking mechanism also functions as a final safety device that blocks the transmission of design data to external sources (e.g., automated compound synthesizers or databases of partner pharmaceutical companies) at the physical layer, rather than the network layer, when the molecular design unit 850 and the in silico verification unit 860 attempt to output dangerous molecular structures with unknown serious side effects or toxicity due to hallucination of the AI ​​model or malicious external intervention. This ensures that immunosuppressants designed on a computer always undergo a physical approval process by authorized human experts before being synthesized as actual substances. This effectively maintains the speed of AI drug discovery while reliably preventing the leakage and manufacture of biohazardous or ethically problematic drugs.

[0453] The process described in the above embodiment can be realized by executing a pre-prepared program on a computer. This program is, for example, stored on a computer-readable storage medium and executed by being read from the storage medium. Alternatively, this program may be provided in the form of a non-volatile (non-transient) storage medium such as flash memory, or it may be provided via a network such as the Internet.

[0454] 11. Summary This specification discloses at least the following configuration: (1) An immunosuppressant design system for designing immunosuppressants to suppress rejection reactions after organ transplantation, The Recipient Information Management Department manages recipient information (medical information and genetic information), The Potential Donor Information Management Department manages information on potential donors (medical information and genetic information), A potential recipient prediction unit predicts potential recipients, which are a group of individuals who are likely to be registered on the organ transplant waiting list within a certain period in the future. A potential recipient information management unit manages information (medical information and genetic information) of the potential recipient predicted by the potential recipient prediction unit, A recipient information acquisition unit that acquires recipient information managed by the recipient information management unit and / or potential recipient information managed by the potential recipient information management unit, A rejection factor identification unit identifies rejection factors that cause rejection reactions based on the information managed by the potential donor information management unit and the information acquired by the recipient information acquisition unit, A molecular design unit that designs or screens molecular structures that specifically bind to the rejection factor identified by the rejection factor identification unit and inhibit its function, An immunosuppressant design system comprising: an in silico verification unit that simulates the in vivo pharmacokinetics and immune response of a drug having a molecular structure designed or screened by the molecular design unit, and verifies its efficacy. Here, if "information" includes genetic information, that genetic information is not limited to definitive genetic information; it may also be statistically estimated genetic information. In other words, with respect to genetic information, the requirement of "managing information" is satisfied even if there is no actual measured data.

[0455] (1) Effect According to the immunosuppressant design system of (1), by comprising a potential recipient prediction unit that predicts potential recipients who may be registered on a waiting list within a certain period in the future, and a recipient information acquisition unit that acquires information on at least one of the current recipient or the predicted potential recipient, it becomes possible to realize predictive drug discovery that identifies factors causing rejection reactions in advance according to future demand, and initiates drug development, not only for patients currently on the waiting list but also for patient groups who will require transplantation in the future, even before a donor-recipient combination is actually established. Furthermore, based on information from potential donors and acquired recipient information, the rejection factor identification unit identifies rejection factors, and the molecular design unit designs or screens molecular structures that specifically bind to those factors. This makes it possible to design optimized drugs that act precisely against the risk of specific rejection reactions that may occur in individual combinations of donors and recipients (including future patients), unlike conventional nonspecific immunosuppressants. Furthermore, by incorporating an in silico validation unit that simulates the pharmacokinetics and immune response of designed or screened drugs, efficacy and safety can be rapidly and efficiently verified on a computer without going through the physical experimental process of actually synthesizing and administering compounds. This dramatically shortens the development period and has the effect of designing and providing drugs in time for the time constraints of transplant medicine.

[0456] (2) A dataset construction unit constructs a large-scale dataset that integrates information on at least one of the recipient or potential recipient and information on the potential donor, as well as the history of rejection in past transplant cases. It has a large-scale dataset storage unit that holds the aforementioned large-scale dataset, The immunosuppressant design system according to (1), characterized in that the rejection factor identification unit extracts potential features that determine the presence or absence of a rejection reaction from the large dataset using a graph neural network or a deep learning model, and identifies the rejection factors based on the extracted features.

[0457] (2) Effect According to (2), by constructing a large dataset that includes information on either the recipient or potential recipient, or both, as well as information on potential donors, and the history of rejection in past transplant cases, and then analyzing this dataset using a graph neural network or deep learning model, in addition to the effects of (1), it becomes possible to extract latent features such as complex correlations and nonlinear characteristics that determine the presence or absence of rejection, which were difficult to discover by humans or conventional statistical methods. As a result, a data-driven approach based on a vast amount of historical data makes it possible to identify rejection factors that directly cause rejection with extremely high accuracy, contributing to the design of more reliable immunosuppressants.

[0458] (3) The immunosuppressant design system according to (1), characterized in that the in silico verification unit verifies the effectiveness of the drug for a population with a specific genetic background using a model of a virtual patient composed of synthetic data generated by learning statistical features from the electronic health records of real patients using a generative AI.

[0459] (3) Effects (3) has the effect of (1) plus the following effect: The in silico validation department generates synthetic data based on statistical features learned from the electronic health records of real patients using generative AI, and constructs this as a model of a virtual patient. This makes it possible to validate the efficacy of drugs while ensuring privacy is protected without directly using specific sensitive personal information. Furthermore, by using this virtual patient model constructed from synthetic data, it is possible to statistically reproduce on a computer population with specific genetic backgrounds for which it is difficult to secure a sufficient sample size using only real data. This enables statistically significant simulation-based efficacy evaluation for subjects that were difficult to validate in conventional clinical trials, such as patient groups with rare HLA types, and can improve the accuracy of developing immunosuppressants that address diverse genetic backgrounds.

[0460] (4) The immunosuppressant design system according to (1), characterized in that the molecular design unit predicts the three-dimensional structure of the rejection factor identified by the rejection factor identification unit using a protein structure prediction AI, and designs the molecular structure based on said three-dimensional structure.

[0461] (4) Effect (4) has the effect of (1) plus the following effect: Firstly, by using protein structure prediction AI to predict the three-dimensional structure of rejection factors, it becomes possible to quickly obtain the three-dimensional structure of identified rejection factors without waiting for experimental structural determination (such as X-ray crystallography). Secondly, by designing the molecular structure based on the predicted three-dimensional structure, it becomes possible to rationally and precisely design molecules (drugs with high binding affinity) that are physically and chemically compatible with the shape and properties of rejection factors.

[0462] (5) The immunosuppressant design system according to (1), characterized in that the molecular design unit has the function of predicting cryptic pockets, which are potential binding sites that are temporarily formed during the dynamic transformation of the protein structure of the rejection factor, by combining molecular dynamics calculations (MD simulations) and machine learning, and designs small or medium-sized molecules that allosterically inhibit the function of the rejection factor by binding to the predicted cryptic pockets.

[0463] (5) Effects (5) has the effect of (1) plus the following effect: Firstly, by combining molecular dynamics (MD) simulations and machine learning with the protein structure of rejection factors, it becomes possible to predict cryptic pockets, which are temporarily formed only during dynamic processes rather than when the protein is stationary, as potential binding sites. Secondly, by targeting the predicted cryptic pocket, it becomes possible to design small or medium-sized molecules that have a mechanism of action that allosterically (non-competitively or through structural changes) inhibits the function of rejection factors.

[0464] (6) The immunosuppressant design system according to (1), characterized in that the in silico validation unit has the function of learning statistical features from electronic health records (EHRs) of real patients using generative AI including a generative adversarial network (GAN) or diffusion model, and generating statistically equivalent synthetic data while protecting individual privacy, and using a virtual patient group composed of the generated synthetic data to validate the side effects and efficacy of the drug for a population with a specific genetic background or rare HLA type.

[0465] (6) Effects (6) has the effect of (1) plus the following effect: Firstly, by using generative AI, including generative adversarial networks or diffusion models, to learn statistical features from the electronic health records of real patients and generate statistically equivalent synthetic data, it becomes possible to secure the data necessary for drug validation while protecting individual privacy. Secondly, by using a virtual patient population composed of generated synthetic data, it becomes possible to virtually construct populations with specific genetic backgrounds or rare HLA types for which it would be difficult to secure a sufficient sample size using only real data. Thirdly, by using the constructed virtual patient population, it becomes possible to verify the side effects and efficacy of drugs in silico (on a computer) for rare populations that were previously difficult to verify due to insufficient data.

[0466] (7) The rejection factor identification unit identifies antigen epitopes common to donor groups that have a high probability of being transplanted in a specific potential recipient, based on data from the potential donor information management unit and the potential recipient prediction unit. The molecular design unit designs the molecular structure of a vaccine or peptide drug that induces specific regulatory T cells (Tregs) in the body in response to the identified antigen epitope. The immunosuppressant design system according to (1), characterized in that the in silico verification unit simulates the efficiency of inducing immune tolerance by administering the vaccine or peptide drug during the period prior to transplant surgery.

[0467] (7) Effects (7) has the effect of (1) plus the following effect: Firstly, by identifying antigenic epitopes common to donor groups that are highly likely to be transplanted into a specific potential recipient, based on data from the Potential Donor Information Management Department and the Potential Recipient Prediction Department, it becomes possible to predictively identify antigenic sites that are highly likely to cause rejection in future transplants, even before a specific donor is confirmed. Secondly, by designing the molecular structure of a vaccine or peptide drug that induces specific regulatory T cells in the body in response to the identified antigen epitope, it becomes possible to design a drug that specifically induces immune tolerance (a state in which the immune system does not respond) to a particular antigen, rather than suppressing the immune system in general. Thirdly, by simulating the efficiency of inducing immune tolerance through the administration of the vaccine or peptide drug during the period prior to transplant surgery, it becomes possible to verify in silico (on a computer) the extent to which immune tolerance can be induced by pre-transplant medication during the period before the transplant surgery.

[0468] (8) The molecular design unit performs an optimization process on the designed molecular structure of the immunosuppressant, adding a linker structure or release control site suitable for mounting or binding to autonomous nanomachines that specifically accumulate in vascular endothelial cells or tissues of transplanted organs. The immunosuppressant design system according to (1), characterized in that the in silico verification unit simulates the local drug concentration in an organ and the rate of systemic leakage when the drug is locally released by the autonomous nanomachine.

[0469] (8) Effect (8) has the effect of (1) plus the following effect: Firstly, by performing an optimization process on the molecular structure of the designed immunosuppressant to add a linker structure or release control site suitable for loading or binding to autonomous nanomachines that specifically accumulate in vascular endothelial cells or tissues of the transplanted organ, it becomes possible not only to design the drug itself, but also to design a structure (a structure optimized for loading and binding to nanomachines) that allows the drug to be specifically delivered and act locally in the transplanted organ. Secondly, by simulating the local drug concentration in an organ and the systemic leakage rate when a drug is released locally by the autonomous nanomachine, it becomes possible to verify the balance between the local efficacy (drug concentration) in the target organ and the risk of systemic side effects (leakage rate) in silico (on a computer) without actually administering the drug to a living organism.

[0470] (9) The dataset construction unit integrates static medical information of the recipient and potential donor, as well as dynamic biological data including current drug use, metabolic status, and real-time homeostasis index or microinflammation level based on inter-organ networks, into the large dataset. The immunosuppressant design system according to (1), characterized in that the rejection factor identification unit identifies specific rejection factors that are suddenly activated due to a physiological stress state or a subtle inflammatory environment at the time of transplantation.

[0471] (9) Effects (9) has the effect of (1) plus the following effect: The dataset construction unit integrates dynamic biological data, such as current drug use, metabolic status, and real-time homeostatic indicators and micro-inflammation levels showing the state of inter-organ networks, in addition to static medical information of recipients and potential donors, into a large-scale dataset. This enables the AI ​​to learn about the physiological stress state and subtle inflammatory environment changes of individual patients at the time of transplantation, which cannot be predicted by genetic information alone. As a result, the rejection factor identification unit can identify rejection factors that are activated spontaneously only under specific physiological conditions, enabling the design of highly personalized immunosuppressants that are precisely adapted to the patient's unique internal environment and dynamic immune state at the time of transplantation, thereby preventing rejection.

[0472] (10) The dataset construction unit integrates physical impact data from the accident site at the time of donor generation, disaster simulation results, or predicted transport time and environmental stress data based on the geographical distance from the donor to the recipient into the large dataset. The immunosuppressant design system according to claim 1, characterized in that the molecular design unit designs molecular structures that target cryptic pockets or amplified inflammatory signals formed on the protein surface of organs due to physical impact caused by trauma, ischemia-reperfusion injury, or prolonged transport stress.

[0473] (10) Effect (10) has the effect of (1) plus the following effect: The dataset construction unit integrates data such as physical impact data from accident sites at the time of donor generation, disaster simulation results, and environmental stress data such as predicted transport time based on geographical distance. This enables a quantitative understanding of the impact of trauma and prolonged ischemia-reperfusion injury on organ immunogenicity. As a result, the molecular design unit can design molecular structures that target cryptic pockets formed only when protein structures change due to physical damage or transport stress, or inflammatory signals that are specifically amplified by environmental factors. This allows for the provision of immunosuppressants that can suppress the risk of rejection not only from the donor's genetic background but also from the physical and environmental history of the donation process.

[0474] (11) The aforementioned potential donor information management unit acquires data on subtle color changes, body movements, and posture changes of the donor's face and skin obtained from a non-contact vital sensing system using video analysis. The immunosuppressant design system according to claim 1, characterized in that the rejection factor identification unit performs multimodal analysis integrating external characteristics obtained from the non-contact vital sensing system and internal biological signals such as electroencephalograms, and identifies factors that are precursors to rejection reactions based on autonomic nervous system activity and patterns of fluctuations in immune dynamics associated with subtle inflammatory responses.

[0475] (11) Effect (11) has the effect of (1) plus the following effect: The Potential Donor Information Management Department uses a non-contact vital sensing system based on video analysis to acquire external characteristics of donors, such as their complexion and subtle body movements. By integrating this with internal biological signals such as electroencephalograms (EEGs) and performing multimodal analysis, it becomes possible to understand the fluctuation patterns of autonomic nervous system activity and immune dynamics associated with subtle inflammatory responses in a natural state, which were difficult to capture with conventional invasive examinations. As a result, the Rejection Factor Identification Department can identify predictive factors of rejection reactions that occur in conjunction with the patient's mental or physical stress state and fluctuations in vital signs with high accuracy. This has the effect of designing highly compatible immunosuppressants that take into account the donor's daily physiological state and unconscious stress responses.

[0476] (12) The immunosuppressant design system works in conjunction with a biological assay system using organoids or disease models created from preserved organs or cells. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit receives measured data obtained by applying the drug candidate designed by the molecular design unit to the biological assay system, and corrects or verifies the virtual patient model or simulation parameters based on the measured data.

[0477] (12) Effect (12) has the effect of (1) plus the following effect: The immunosuppressant design system, in conjunction with biological assay systems using organoids or disease models created from preserved organs or cells, and by feeding back the measured data obtained there to the in silico validation unit, makes it possible to correct the discrepancy between computer simulation results and actual biological responses. This allows for highly accurate verification of whether drug candidates designed by the molecular design unit exhibit the predicted pharmacological effects in actual living tissues, thereby reducing reliance on animal experiments and enabling the efficient development of immunosuppressants with a high probability of success in clinical trials.

[0478] (13) The molecular design unit designs the molecular structure of an immunosuppressant optimized for local release, which is mounted on an autonomous nanomachine that repairs minute damaged or chronically inflammatory sites within a vitrified organ during the process of reviving the organ to a transplantable state using molecular nanotechnology. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit simulates the behavior of the autonomous nanomachine and the control of drug release during the resuscitation process to verify the immunological resuscitation effect on the organ immediately before transplantation.

[0479] (13) Effects (13) has the effect of (1) plus the following effect: The Molecular Design Department designs drug molecules optimized for autonomous nanomachines that function in the process of resuscitating vitrified organs using molecular nanotechnology, and the In Silico Validation Department simulates their local release behavior. This makes it possible to deliver high concentrations of the drug only to microscopic damaged or inflamed areas that require repair, without circulating the drug throughout the body. This minimizes the risk of systemic side effects, while simultaneously completing physical repair and immunological rejection control (immunological resuscitation) of organs immediately before transplantation, dramatically increasing the graft survival rate after transplantation.

[0480] (14) The immunosuppressant design system according to claim 1, characterized in that, using Federated Learning technology, sensitive genetic information and biodata whose cross-border transfer is restricted by national laws and regulations or privacy policies are kept within the system, while only the gradient information of the learning parameters of the AI ​​model in the rejection factor identification unit or the molecular design unit is shared and updated, thereby designing an immunosuppressant that reflects the genetic background specific to a particular region, race, or population.

[0481] (14) Effect (14) has the effect of (1) plus the following effect: By constructing a decentralized drug discovery platform that uses federative learning technology to physically retain sensitive genetic and biometric data within its own system (within each management unit or device), while only sharing and updating gradient information of AI model learning parameters, it becomes possible to safely use data for learning even if cross-border transfer and aggregation are restricted due to strict legal regulations and privacy policies in each country. This enables the construction of unbiased global models that encompass rare genetic backgrounds and immunological characteristics specific to particular races, regions, or populations, resulting in the design of effective and safe immunosuppressants for diverse patient populations worldwide.

[0482] (15) A computer program for implementing the immunosuppressant design system described in any one of items (1) to (14) using one or more computers.

[0483] (15) Effects By running this computer program on one or more computers, any of the immunosuppressant design systems described in (1) through (14) can be realized. More specifically, each of the functional units described in (1) to (14), such as the recipient information management unit, potential donor information management unit, potential recipient prediction unit, dataset construction unit, large-scale dataset storage unit, rejection factor identification unit, molecular design unit, and in silico verification unit, can be implemented and realized as software running on one or more computers (hardware).

[0484] (16) A method for designing immunosuppressants to suppress rejection reactions after organ transplantation, Recipient information management step for managing the recipient's medical and genetic information, A potential donor information management step involves managing the medical and genetic information of potential donors, A potential recipient prediction step that predicts potential recipients, which are a group of individuals who are likely to be placed on an organ transplant waiting list within a certain period in the future, A recipient information acquisition step that acquires information about the recipient managed by the recipient information management step and / or information about a potential recipient managed by the potential recipient information management step, Based on the information managed by the aforementioned potential donor information management step and the information obtained by the aforementioned recipient information acquisition step, a rejection factor identification step is performed to identify rejection factors that cause rejection reactions. A molecular design step involves designing or screening molecular structures that specifically bind to the rejection factor identified in the rejection factor identification step and inhibit its function, An immunosuppressant design method characterized by comprising: an in silico validation step for simulating the pharmacokinetics and immune response of a drug having a molecular structure designed or screened by the molecular design step, and verifying its efficacy.

[0485] (16) Effects Firstly, by predicting a group of individuals who may be placed on a waiting list within a certain period in the future during the potential recipient prediction step, and then performing the rejection factor identification step and molecular design step based on that information, it becomes possible to realize a "predictive drug discovery" process in which immunosuppressants that meet future demands are designed and prepared in advance, even before patients are actually placed on the waiting list or before donors appear. Secondly, by using information on potential recipients, which represent a future patient population, in addition to information on recipients and potential donors, it is possible to comprehensively identify not only currently apparent risks but also risk factors for specific rejection reactions that may occur in the future, thereby designing optimized drugs with broader compatibility. Thirdly, by simulating and verifying the pharmacokinetics and immune response in the in silico validation step for the molecular structure designed in the molecular design step, efficacy and safety can be rapidly evaluated on a computer without going through the physical experimental process of actually synthesizing and administering the compound. This dramatically shortens the development period for immunosuppressants and makes it possible to respond immediately to the time-constrained needs of transplant medicine.

[0486] (17) A dataset construction step involves constructing a large dataset that integrates information on at least one of the recipient or potential recipient and information on the potential donor, as well as the history of rejection in past transplant cases. The large-scale dataset storage step includes a step for storing the large-scale dataset, The immunosuppressant design method according to (16), characterized in that the rejection factor identification step involves extracting potential features that determine the presence or absence of a rejection reaction from the large dataset using a graph neural network or a deep learning model, and identifying the rejection factors based on the extracted features.

[0487] (17) Effects According to (17), by constructing a large dataset that includes information on either the recipient or potential recipient, or both, as well as information on potential donors, and the history of rejection in past transplant cases, and then analyzing this dataset using a graph neural network or deep learning model, it becomes possible to extract latent features such as complex correlations and nonlinear characteristics that determine the presence or absence of rejection, which are difficult to discover by humans or conventional statistical methods, in addition to the effects of (16). As a result, a data-driven approach based on a vast amount of historical data can identify rejection factors that directly cause rejection with extremely high accuracy, contributing to the design of more reliable immunosuppressants.

[0488] (18) A drug design system for designing drugs for the treatment or prevention of disease, The Disease-Related Factor Information Management Department manages information on disease-related factors, which are factors associated with the cause or exacerbation of a disease. A potential patient prediction unit predicts potential patients, which are a group of individuals who are likely to contract or require treatment for the aforementioned disease within a certain period in the future. A patient information acquisition unit that acquires information on at least one of the patient or the potential patient predicted by the potential patient prediction unit, A target factor identification unit identifies target factors that cause the onset or progression of the disease, based on the information managed by the disease-related factor information management unit and the information acquired by the patient information acquisition unit. A pharmaceutical design system comprising: a molecular ...

Claims

1. An immunosuppressant design system for designing immunosuppressants to suppress rejection reactions after organ transplantation, The Potential Donor Information Management Department manages information on potential donors, A potential recipient prediction unit predicts potential recipients, which are a group of individuals who are likely to be registered on the organ transplant waiting list within a certain period in the future. A recipient information acquisition unit that acquires information on at least one of the recipient or the potential recipient predicted by the potential recipient prediction unit, A rejection factor identification unit identifies rejection factors that cause rejection reactions based on the information managed by the potential donor information management unit and the information acquired by the recipient information acquisition unit, A molecular design unit that designs or screens molecular structures that specifically bind to the rejection factor identified by the rejection factor identification unit and inhibit its function, An immunosuppressant design system comprising: an in silico verification unit that simulates the in vivo pharmacokinetics and immune response of a drug having a molecular structure designed or screened by the molecular design unit, and verifies its efficacy.

2. A dataset construction unit constructs a large-scale dataset that integrates information on at least one of the recipient or potential recipient and information on the potential donor, as well as the history of rejection in past transplant cases. It has a large-scale dataset storage unit that holds the aforementioned large-scale dataset, The immunosuppressant design system according to claim 1, characterized in that the rejection factor identification unit extracts potential features that determine the presence or absence of a rejection reaction from the large dataset using a graph neural network or a deep learning model, and identifies the rejection factors based on the extracted features.

3. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit verifies the effectiveness of the drug for a population with a specific genetic background using a model of a virtual patient composed of synthetic data generated by learning statistical features from the electronic health records of real patients using a generation AI.

4. The immunosuppressant design system according to claim 1, characterized in that the molecular design unit predicts the three-dimensional structure of the rejection factor identified by the rejection factor identification unit using protein structure prediction AI, and designs the molecular structure based on said three-dimensional structure.

5. The immunosuppressant design system according to claim 1, characterized in that the molecular design unit has the function of predicting cryptic pockets, which are potential binding sites that are temporarily formed during the dynamic transformation of the protein structure of the rejection factor, by combining molecular dynamics calculations and machine learning, and designs a small or medium-sized molecule compound that allosterically inhibits the function of the rejection factor by binding to the predicted cryptic pocket.

6. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit has the function of learning statistical features from the electronic health records of real patients using generative AI including an adversarial generative network or diffusion model, and generating statistically equivalent synthetic data while protecting individual privacy, and using a virtual patient group composed of the generated synthetic data to verify the side effects and efficacy of the drug for a population with a specific genetic background or rare HLA type.

7. Based on the integrated data, the rejection factor identification unit identifies antigenic epitopes common to donor groups that have a high probability of being transplanted in a specific potential recipient in the future. The molecular design unit designs the molecular structure of a vaccine or peptide drug that induces specific regulatory T cells in the body in response to the identified antigen epitope. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit simulates the efficiency of inducing immune tolerance by administering the vaccine or peptide drug during the period prior to transplant surgery.

8. The molecular design unit performs an optimization process on the designed molecular structure of the immunosuppressant, adding a linker structure or release control site suitable for mounting or binding to autonomous nanomachines that specifically accumulate in vascular endothelial cells or tissues of transplanted organs. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit simulates the local drug concentration in an organ and the rate of systemic leakage when the drug is locally released by the autonomous nanomachine.

9. The dataset construction unit integrates static medical information of the recipient and potential donor, as well as dynamic biological data including current drug use, metabolic status, and real-time homeostatic indicators or micro-inflammation levels based on inter-organ networks, into the large dataset. The immunosuppressant design system according to claim 1, characterized in that the rejection factor identification unit identifies specific rejection factors that are suddenly activated due to a physiological stress state or a subtle inflammatory environment at the time of transplantation.

10. The dataset construction unit integrates physical impact data from the accident site at the time of donor generation, disaster simulation results, or predicted transport time and environmental stress data based on the geographical distance from the donor to the recipient into the large dataset. The immunosuppressant design system according to claim 1, characterized in that the molecular design unit designs molecular structures that target cryptic pockets or amplified inflammatory signals formed on the protein surface of organs due to physical impact caused by trauma, ischemia-reperfusion injury, or prolonged transport stress.

11. The aforementioned potential donor information management unit acquires data on subtle color changes, body movements, and posture changes of the donor's face and skin obtained from a non-contact vital sensing system using video analysis. The immunosuppressant design system according to claim 1, characterized in that the rejection factor identification unit performs multimodal analysis integrating external characteristics obtained from the non-contact vital sensing system and internal biological signals such as electroencephalograms, and identifies factors that are precursors to rejection reactions based on autonomic nervous system activity and fluctuation patterns of immune dynamics associated with subtle inflammatory responses.

12. The immunosuppressant design system works in conjunction with a biological assay system using organoids or disease models created from preserved organs or cells. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit receives measured data obtained by applying the drug candidate designed by the molecular design unit to the biological assay system, and corrects or verifies the virtual patient model or simulation parameters based on the measured data.

13. The molecular design unit designs the molecular structure of an immunosuppressant optimized for local release, which is mounted on an autonomous nanomachine that repairs minute damaged or chronically inflammatory sites within a vitrified organ during the process of reviving the organ to a transplantable state using molecular nanotechnology. The immunosuppressant design system according to claim 1, characterized in that the in silico verification unit simulates the behavior of the autonomous nanomachine and the control of drug release during the resuscitation process to verify the immunological resuscitation effect on the organ immediately before transplantation.

14. An immunosuppressant design system according to claim 1, characterized in that, using federated learning technology, sensitive genetic information and biodata whose cross-border transfer is restricted by the laws and regulations or privacy policies of each country are kept within the system, while only the gradient information of the learning parameters of the AI ​​model in the rejection factor identification unit or the molecular design unit is shared and updated, thereby designing an immunosuppressant that reflects the genetic background specific to a particular region, race, or population.

15. A computer program for implementing the immunosuppressant design system described in any one of claims 1 to 14 using one or more computers.

16. A method for designing immunosuppressants to suppress rejection reactions after organ transplantation, The potential donor information management step involves managing information about potential donors, A potential recipient prediction step that predicts potential recipients, which are a group of individuals who are likely to be placed on an organ transplant waiting list within a certain period in the future, A recipient information acquisition step that acquires information on at least one of the recipient or the potential recipient predicted by the potential recipient prediction step, Based on the information managed by the aforementioned potential donor information management step and the information obtained by the aforementioned recipient information acquisition step, a rejection factor identification step is performed to identify rejection factors that cause rejection reactions. A molecular design step involves designing or screening molecular structures that specifically bind to the rejection factor identified in the rejection factor identification step and inhibit its function, An immunosuppressant design method characterized by comprising: an in silico validation step for simulating the pharmacokinetics and immune response of a drug having a molecular structure designed or screened by the molecular design step, and verifying its efficacy.

17. A dataset construction step involves constructing a large dataset that integrates information on at least one of the recipient or potential recipient and information on the potential donor, as well as the history of rejection in past transplant cases. The large-scale dataset storage step includes a step for storing the large-scale dataset, The immunosuppressant design method according to claim 16, characterized in that the rejection factor identification step involves extracting potential features that determine the presence or absence of a rejection reaction from the large dataset using a graph neural network or a deep learning model, and identifying the rejection factors based on the extracted features.

18. A drug design system for designing drugs for the treatment or prevention of disease, The Disease-Related Factor Information Management Department manages information on disease-related factors, which are factors associated with the cause or exacerbation of a disease. A potential patient prediction unit predicts potential patients, which are a group of individuals who are likely to contract or require treatment for the aforementioned disease within a certain period in the future. A patient information acquisition unit that acquires information on at least one of the patient or the potential patient predicted by the potential patient prediction unit, A target factor identification unit identifies target factors that cause the onset or progression of the disease, based on the information managed by the disease-related factor information management unit and the information acquired by the patient information acquisition unit. A pharmaceutical design system comprising: a molecular design unit that designs or screens molecular structures that specifically bind to the target factor identified by the target factor identification unit and inhibit or control its function; and an in silico verification unit that simulates the in vivo dynamics and biological response of a drug having the molecular structure designed or screened by the molecular design unit and verifies its efficacy.

19. The aforementioned disease is an infectious disease. The aforementioned disease-related factor information management unit manages information regarding viruses or bacteria of wild animal origin, or variant strains of known pathogens prevalent in a specific region, as the aforementioned disease-related factors. The aforementioned potential patient prediction unit predicts residents of areas where future infection spread is expected to occur as potential patients, based on epidemiological simulations or human flow data. The target factor identification unit identifies molecules related to the mutation patterns or mechanisms of entry into human cells of pathogens predicted to cause future epidemics, based on the integrated data of disease-related factors and potential patients, as target factors. The pharmaceutical design system according to claim 18, characterized in that the molecular design unit pre-designs the molecular structure of a vaccine or antiviral drug that acts on the predicted target factor before the pathogen mutates and causes an epidemic.

20. The aforementioned disease is cancer. The aforementioned disease-related factor information management unit manages information regarding gene mutations or environmental stress factors that increase the risk of cancer as disease-related factors. The aforementioned potential patient prediction unit predicts a group of people at risk of developing the cancer in the future, based on genetic tendencies or lifestyle data, as the aforementioned potential patients. The target factor identification unit identifies cryptic pockets formed on the protein surface by future gene mutations or environmental stress as target factors, based on the interaction between the predicted genetic characteristics of the potential patient and the disease-related factors. The pharmaceutical design system according to claim 18, characterized in that the molecular design unit pre-designs the molecular structure of a molecularly targeted drug that acts on the target factor before the onset of cancer or before resistance to existing drugs is acquired.

21. A computer program for implementing the pharmaceutical design system described in any one of claims 18 to 20 using one or more computers.

22. A method for designing pharmaceuticals for the treatment or prevention of a disease, A disease-related factor information management step for managing information on disease-related factors, which are factors associated with the cause or exacerbation of a disease, A potential patient prediction step that predicts a group of individuals who are likely to develop or require treatment for the aforementioned disease within a certain period in the future, A patient information acquisition step that acquires information on at least one of the patient or the potential patient predicted by the potential patient prediction step, A target factor identification step, which identifies target factors that cause the onset or progression of the disease, based on the information managed by the disease-related factor information management step and the information obtained by the patient information acquisition step, A molecular design step involves designing or screening molecular structures that specifically bind to the target factor identified in the target factor identification step and inhibit or control its function. A method for designing a pharmaceutical product, comprising: an in silico validation step for simulating the pharmacokinetics and biological responses of a drug having a molecular structure designed or screened by the molecular design step, and verifying its efficacy.

Images