A method and system for optimizing the quantitative addition parameters of a medicinal peptide based on machine learning

Abstract

The application discloses a kind of medical peptide quantitative addition parameter optimization method and system based on machine learning, by constructing chemical rule knowledge base and combining the parameter combination of data-driven model output is verified, identify the chemical rule in chemical rule knowledge base does not comply or causes the situation of side reaction and is modified or generates warning, to obtain decision result, and according to decision result obtains feedback and updates knowledge base or adjusts model parameter, and then constructs in medical peptide quantitative addition parameter optimization decision scene, can integrate chemical rule and be modified under parameter verification, provide the decision explanation of chemical logic, in decision result feedback process can be based on decision result and update chemical rule knowledge base or adjust data-driven model parameter with expert participation decision determination, improve the reliability of parameter optimization result, support system knowledge and the iterative improvement of model.

Description

TECHNICAL FIELD

[0001] The present application relates to the parameter optimization decision technology in the field of medical peptide synthesis, and in particular to a medical peptide quantitative addition parameter optimization method and system based on machine learning. BACKGROUND

[0002] In the research and production process of medical peptides, solid-phase synthesis or liquid-phase synthesis is the core process for constructing target peptide chains. These synthesis methods are highly dependent on the accurate and sequential addition of various materials, including different types of amino acid derivatives, activators, coupling agents, protecting group removal reagents, and various reaction solvents. The quantitative addition parameters of each material, such as the molar ratio between each component, the concentration of the solution, the volume of addition, the rate of material addition, and the precise time node, together with the reaction system temperature, reaction duration, and other process conditions, collectively determine the efficiency of peptide chain extension reactions, the accuracy of target sequence construction, the purity level of the final product, and the types and relative contents of by-products.

[0003] Traditional parameter optimization approaches mainly rely on the accumulation of experience by researchers or through the execution of a large number of orthogonal experiments and single-factor variable screening experiments. This approach can be effective when dealing with simple structures or mature synthesis routes of peptides, but when the target is a new type of medical peptide with high structural complexity and a wide variety of types, it exhibits a long research and development cycle, a large consumption of raw materials, and difficulty in systematically revealing the deep connections between parameters and synthesis results, resulting in low overall efficiency of optimization work and an inability to quickly respond to the development needs of new or complex structure peptides.

[0004] To improve optimization efficiency, medical research and development institutions have introduced data analysis systems based on historical synthesis data training to assist in predicting and optimizing quantitative addition parameters in medical peptide synthesis. This system analyzes the polypeptide structure information, detailed process parameters, and corresponding synthesis yield, purity, and other result data recorded in the database to construct a prediction model. However, data analysis models trained solely on historical data statistical rules may deviate significantly from the experienced judgments of senior synthetic chemists based on their understanding of the chemical behavior of these special structures or from the specific synthesis strategies reported in the literature for such special structural units.

[0005] Researchers try to apply these known chemical rules or structure-reactivity relationships to specific peptide structures, however, existing data analysis systems often lack an effective mechanism to smoothly integrate such symbolic chemical knowledge in the design. When the quantitative parameter combination recommended by the model contradicts the professional intuition or long-term accumulated experimental experience of synthetic chemists, due to the lack of transparency of the internal decision-making process of most current data analysis models, researchers have difficulty in clearly and intuitively understanding what input features the model is based on and how these features interact, ultimately leading to the system recommending this specific set of parameters, which may even contradict conventional cognition. SUMMARY

[0006] The purpose of the present application is to provide a machine learning-based medical peptide quantitative addition parameter optimization method and system, which combines chemical rule knowledge and data-driven model for parameter verification, identifies cases where parameter combinations do not conform to chemical rules or trigger side reactions, modifies parameter combinations or generates warnings, updates the chemical rule knowledge base or adjusts data-driven model parameters based on decision results, improves the reliability of parameter optimization results, and supports iterative improvement of system knowledge and models.

[0007] To achieve the above purpose, the technical scheme of the present application has:

[0008] In one aspect of the present application, a machine learning-based medical peptide quantitative addition parameter optimization method is provided, comprising the following steps:

[0009] S1. Obtain raw data associated with the field of medical peptide synthesis from multiple specific channels, and use a data fusion algorithm to fuse the obtained raw data to obtain a chemical rule knowledge base;

[0010] S2. Obtain the structural characteristics of the target peptide, obtain a quantitative addition parameter combination based on the structural characteristics of the target peptide in combination with the parameters input by the data-driven model, verify the obtained quantitative addition parameter combination according to the chemical rule knowledge base to identify verification results that do not conform to the chemical rules in the chemical rule knowledge base or may trigger side reactions, and modify the quantitative addition parameter combination or generate a warning based on the verification results to obtain a modified or warning-attached quantitative addition parameter combination;

[0011] S3. Generate a decision result based on the modified or warning-attached quantitative addition parameter combination;

[0012] S4. Obtain feedback information based on the decision result and update the chemical rule knowledge base or adjust the preset parameters in the data-driven model based on the obtained feedback information, wherein the feedback information includes user-suggested parameter adjustments or supporting chemical principles for their suggestions.

[0013] Further, the step S1 comprises the following steps:

[0014] S11, acquiring chemical application knowledge information related to amino acid sequences, protecting groups and coupling reagents in the field of medical peptide synthesis;

[0015] S12, formulating rules based on chemical application knowledge information according to peptide structure characteristics and process conditions, and potential risks or recommended operations corresponding to peptide structure characteristics and process conditions, and structurally storing the data sets, and collecting the obtained data sets to obtain a chemical rule knowledge base.

[0016] Further, the step S2 specifically comprises the following steps:

[0017] S21, acquiring a quantitative addition parameter combination combined by the structure characteristics of the target peptide and the parameters output by the data-driven model;

[0018] S22, according to the chemical rule knowledge base, performing a preliminary check on the obtained quantitative addition parameter combination to identify a preliminary check result that does not conform to the chemical rules in the chemical rule knowledge base or that may cause side reactions;

[0019] S23, based on the preliminary check result, performing a preliminary correction on the current quantitative addition parameter combination or generating a warning to obtain a modified or warning-attached quantitative addition parameter combination obtained after the preliminary check;

[0020] S24, performing iterative check processing on the target peptide information set with the modified or warning-attached quantitative addition parameter combination obtained after the preliminary check, checking the current quantitative addition parameter combination to obtain a final check result associated with the current quantitative addition parameter combination, and obtaining a modified or warning-attached quantitative addition parameter combination based on the final check result.

[0021] Further, the step S22 specifically comprises:

[0022] S221, acquiring one or more parameter subsets in the current quantitative addition parameter combination that have synergistic effects or antagonistic effects between each member parameter, and for each identified parameter subset, retrieving parameter interaction rules corresponding to the parameter subset from the chemical rule knowledge base or the parameter interaction knowledge database, the parameter interaction rules indicating the influence on the progress, efficiency of the target chemical reaction, or the occurrence probability, degree of side reactions when the member parameters in the parameter subset act together;

[0023] S222, based on the retrieved each parameter interaction rule, performing adjustment on the compliance checking conclusion of the single rule checking result on the single parameter or the whole parameter combination in the parameter subset or evaluating the risk level or the occurrence probability of the side reaction caused by the joint action of the member parameters in the parameter subset, to obtain parameter interaction adjustment information;

[0024] S223, according to the chemical rule knowledge base, performing single rule checking on the obtained quantitative addition parameter combination to identify the parameters that are inconsistent with the chemical rules in the chemical rule knowledge base or that can cause side reactions, to obtain a single rule checking result;

[0025] S224, combining the single rule checking result and the parameter interaction adjustment information to obtain a preliminary checking result.

[0026] Further, the step S223 specifically comprises:

[0027] S2231, retrieving the chemical rules related to the structure characteristics of the current quantitative addition parameter combination and the target peptide from the chemical rule knowledge base, and identifying the rules with their own confidence scores or probability descriptions of applicable conditions in the related chemical rules, to obtain the chemical rules with uncertainty information;

[0028] S2232, for each obtained chemical rule with uncertainty information, based on the confidence score or the probability description of the applicable conditions of the chemical rule itself, and the matching situation of the quantitative addition parameter combination and the prerequisite of the chemical rule, calculating the compliance probability of the current quantitative addition parameter combination to the chemical rule with uncertainty information;

[0029] S2233, for one or more preset side reactions, based on the compliance probability of the current quantitative addition parameter combination to the chemical rule with uncertainty information, and when there are one or more chemical rules related to the same side reaction, using a probability synthesis method to calculate the comprehensive risk probability of the current quantitative addition parameter combination causing the side reaction;

[0030] S2234, using the calculated compliance probability of the current quantitative addition parameter combination to each chemical rule and the calculated comprehensive risk probability of the current quantitative addition parameter combination causing each preset side reaction, to identify the parameters that are inconsistent with the chemical rules in the chemical rule knowledge base or that can cause side reactions, and to obtain a single rule checking result.

[0031] Further, the step S2233 specifically comprises:

[0032] S22331, retrieving and acquiring, from a rule dependency data source, dependency descriptions among a plurality of chemical rules related to the side reaction, the dependency descriptions indicating interaction patterns of the chemical rules in jointly affecting the probability of the side reaction;

[0033] S22332, according to the acquired dependency descriptions, before performing the probability synthesis operation, modifying individual risk contribution of at least one rule of the plurality of chemical rules to the side reaction, to obtain a modified individual risk contribution, and / or constructing a probability synthesis structure integrating the dependency descriptions, and acquiring, as inputs of the probability synthesis structure, individual risk contributions of the plurality of chemical rules related to the side reaction to the side reaction, wherein the individual risk contribution indicates a compliance probability of the current quantitative addition parameter combination for the corresponding rule calculated by the corresponding rule, or is derived from probability information directly contained in the corresponding rule for describing triggering of the side reaction;

[0034] S22333, performing the probability synthesis operation by using the obtained modified individual risk contribution, and / or performing the probability synthesis operation by using the constructed probability synthesis structure and the acquired individual risk contribution as inputs thereof, so as to calculate, through the probability synthesis operation, a comprehensive risk probability of the side reaction combined with the dependency relationships among the chemical rules and the current quantitative addition parameter combination triggering the side reaction.

[0035] Further, the step S24 specifically comprises:

[0036] S241, after obtaining the modified or warning-attached quantitative addition parameter combination after the initial verification, starting an iterative verification process, and in each iteration of the iterative verification process, performing re-verification on the quantitative addition parameter combination of the current iteration round according to the chemical rule knowledge base;

[0037] S242, judging whether the iterative verification process meets a preset iteration termination condition, the iteration termination condition indicating that the updated quantitative addition parameter combination reaches a stable state without further modification after one complete re-verification or the number of iterations of the iterative process reaches a preset upper limit; if the iteration termination condition is met, the updated quantitative addition parameter combination is taken as the modified or warning-attached quantitative addition parameter combination; if the preset iteration termination condition is not met, the updated quantitative addition parameter combination is taken as an input quantitative addition parameter combination of a next iteration round, and the re-verification in each iteration of the iterative process is returned to be performed, and finally a re-verification result is obtained;

[0038] S242、based on the rechecking result, performing re-correction on the quantitative addition parameter combination of the current iteration round to obtain a final corrected quantitative addition parameter combination or a quantitative addition parameter combination with a warning.

[0039] Further, the step S3 specifically comprises:

[0040] S31, obtaining a relevance result between the corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning and the chemical rule knowledge base; S32, based on the relevance result, evaluating the influence of the corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning on one or more preset optimization targets, and obtaining a decision result.

[0041] Further, the step S32 specifically comprises:

[0042] S321, obtaining the selected corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning and user preference information for a plurality of mutually conflicting medical peptide optimization targets;

[0043] S322, for each of the plurality of mutually conflicting medical peptide optimization targets, based on the obtained corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning, calculating a quantitative achievement degree of the medical peptide optimization target; based on the obtained corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning and the plurality of mutually conflicting medical peptide optimization targets, identifying the conflict relationship between the plurality of mutually conflicting medical peptide optimization targets, and quantifying the conflict relationship to obtain a conflict relationship quantification result;

[0044] S323, according to the calculated quantitative achievement degree of each medical peptide optimization target, the obtained conflict relationship quantification result, and the obtained user preference information, using a comprehensive evaluation algorithm to calculate a comprehensive influence evaluation of the selected parameter combination on the plurality of mutually conflicting medical peptide optimization targets;

[0045] S324, obtaining a decision result based on the comprehensive influence evaluation.

[0046] In an aspect of the present application, the present application provides a machine learning-based medical peptide quantitative addition parameter optimization method. By constructing a chemical rule knowledge base and verifying the parameter combination output by the data-driven model, the chemical rule inconsistency or side reaction in the chemical rule knowledge base is identified and corrected or an alert is generated to obtain a decision result. The decision result is fed back to update the knowledge base or adjust the model parameters, thereby constructing a decision-making scenario for medical peptide quantitative addition parameter optimization. The chemical rules are integrated to correct the parameters under verification, provide chemical logic for decision explanation, and update the chemical rule knowledge base or adjust the data-driven model parameters based on the decision result and expert decision-making in the decision result feedback process, thereby improving the reliability of the parameter optimization result and supporting the iterative improvement of system knowledge and models.

[0047] As a second aspect of the present application, a machine learning-based medical peptide quantitative addition parameter optimization system is provided, comprising:

[0048] A chemical rule knowledge base construction module is configured to obtain raw data associated with the medical peptide synthesis field from multiple specific channels, fuse the obtained raw data using a data fusion algorithm to obtain a chemical rule knowledge base;

[0049] A data acquisition and parameter correction warning module is configured to obtain the structural characteristics of the target peptide, obtain a quantitative addition parameter combination based on the structural characteristics of the target peptide and the parameters input by the data-driven model, verify the obtained quantitative addition parameter combination according to the chemical rule knowledge base to identify the verification result of the chemical rule inconsistency or possible side reaction in the chemical rule knowledge base, and correct the quantitative addition parameter combination or generate a warning based on the verification result to obtain a corrected or warning-attached quantitative addition parameter combination;

[0050] A decision basis generation module is configured to generate a decision result based on the corrected or warning-attached quantitative addition parameter combination;

[0051] A feedback and calibration module is configured to obtain feedback information based on the decision result and update the chemical rule knowledge base or adjust the preset parameters in the data-driven model based on the obtained feedback information, wherein the feedback information includes user-suggested parameter adjustments or supporting chemical principles.

[0052] In the two aspects of the present application, the present application provides a medical peptide quantitative addition parameter optimization system based on machine learning, which is constructed based on an optimization method for medical peptide quantitative parameters, and the system framework is completed by configuring a chemical rule knowledge base construction module, a data acquisition and parameter correction warning module, a decision basis generation module, and a feedback and calibration module, to complete the verification of the parameter combination output by the construction of the chemical rule knowledge base and the combination of the data-driven model, identify the chemical rule inconsistency or side reaction in the chemical rule knowledge base and correct or generate a warning, obtain a decision result, and obtain feedback according to the decision result and update the knowledge base or adjust the model parameters, thereby constructing a medical peptide quantitative addition parameter optimization decision-making scenario that can integrate chemical rules for parameter correction under parameter verification, provide chemical logic decision explanation, and update the chemical rule knowledge base or adjust the data-driven model parameters based on the decision result and with the participation of experts in decision-making in the decision result feedback process, improve the reliability of the parameter optimization result, and support the iterative improvement of system knowledge and models.

[0053] For better understanding and implementation, the present application is described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is a flowchart of the optimization method in the present embodiment;

[0055] Figure 2 is a flowchart of step S1 in the optimization method in the present embodiment;

[0056] Figure 3 is a flowchart of step S2 in the optimization method in the present embodiment;

[0057] Figure 4 is a flowchart of step S22 in the optimization method in the present embodiment;

[0058] Figure 5 is a flowchart of step S223 in the optimization method in the present embodiment;

[0059] Figure 6 is a flowchart of step S2233 in the optimization method in the present embodiment;

[0060] Figure 7 is a flowchart of step S24 in the optimization method in the present embodiment;

[0061] Figure 8 is a flowchart of step S3 in the optimization method in the present embodiment;

[0062] Figure 9is a flowchart of the step S32 indicated in the optimization method in the embodiment;

[0063] Figure 10 is a system block diagram of the optimization system in the embodiment. DETAILED DESCRIPTION

[0064] In order to better illustrate the present application, the present application will be further described in detail below with reference to the accompanying drawings.

[0065] It should be clear that the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0066] The terms used in the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "an" and "the" used in the present application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein means and includes any or all possible combinations of one or more associated listed items.

[0067] The following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present application, as detailed in the appended claims. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not necessarily describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0068] In addition, in the description of the present application, "multiple" means two or more, unless otherwise specified. The "and / or" describes the association between the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. The character " / " generally represents that the associated objects before and after are in an "or" relationship.

[0069] In the prior art, it can be known from the content of the background art that the data analysis model trained by simply relying on the statistical law of historical data may deviate from the experience judgment of experienced synthetic chemists based on the understanding of the chemical behavior of these special structures, or from the specific synthesis strategy reported in the literature for such special structural units.

[0070] Therefore, the technical problem actually solved by the present application is how to improve the reliability of parameter optimization results in the field of medical peptide quantitative addition parameters and support iterative improvement of system knowledge and models.

[0071] Based on the above, the present application provides a preferred example for illustration as follows:

[0072] As an aspect of an embodiment of the present application, as shown in Figure 1 A machine learning-based medical peptide quantitative addition parameter optimization method is provided, which includes the following steps:

[0073] S1, obtaining raw data associated with the field of medical peptide synthesis from multiple specific channels, and using a data fusion algorithm to fuse the obtained raw data to obtain a chemical rule knowledge base;

[0074] S2, obtaining the structural characteristics of the target peptide, obtaining the quantitative addition parameter combination based on the structural characteristics of the target peptide and the parameters input by the data-driven model, verifying the obtained quantitative addition parameter combination according to the chemical rule knowledge base to identify the verification results that do not conform to the chemical rules in the chemical rule knowledge base or may cause side reactions, and combining the verification results to modify or generate warnings for the quantitative addition parameter combination, to obtain the quantitative addition parameter combination after modification or with warnings;

[0075] S3, generating a decision result based on the modified or warning-attached quantitative addition parameter combination;

[0076] S4, obtaining feedback information based on the decision result and updating the chemical rule knowledge base or adjusting the preset parameters in the data-driven model based on the obtained feedback information, wherein the feedback information includes user-suggested parameter adjustments or supporting chemical principles for their suggestions.

[0077] Among them, the construction of the chemical rule knowledge base can collect data from channels such as literature databases, experimental records, and expert experience interviews; the data fusion algorithm can use technologies such as ontology matching and rule extraction to integrate data from different sources and in different formats to form a structured set of chemical rules, thereby providing a unified and reliable basis for subsequent parameter verification; the structural characteristics of the target peptide can be realized by inputting amino acid sequences, modification information, etc.; the data-driven model can be a machine learning model, such as a neural network or a decision tree, which predicts parameter combinations based on historical data; the verification process can be a rule engine that compares the parameter combination output by the model with the rules in the knowledge base; verification The result can be a Boolean value (compliant / non-compliant) or a risk level. Based on the verification result, a risk level can be generated or the parameter value can be automatically adjusted to generate an alert; the decision result can be a recommended parameter scheme, a list of alternative schemes, or a parameter combination with a risk assessment. The generation process can be to directly output the revised parameter combination, or to sort and present it in combination with other factors; feedback information can be obtained through user interface input or experimental result analysis, updating the chemical rule knowledge base can be to convert new experimental findings or expert experience into rules and add them to the knowledge base, and adjusting data-driven model parameters can be through retraining or parameter fine-tuning to make the model better adapt to new data and rules.

[0078] In a specific example application, after knowledge extraction in the chemical rule knowledge base, each rule can be structured / vectorized and stored according to "IF [peptide structure feature A AND / OR process condition B] THEN [potential risk C OR recommended operation D]", and experts are allowed to add, delete, modify and query. After inputting the target peptide structure and a set of preliminary parameters corresponding to the target peptide structure, the target peptide structure is parsed, and the rules related to these features and input parameters are retrieved in the rule base. If the parameter triggers the "potential risk C" condition of a rule, the system marks this risk. If the chemical rule knowledge base contains the corresponding "recommended operation D" for the identified risk, the system generates a specific parameter adjustment suggestion warning, or if there is no correction strategy, a risk warning to alert R&D personnel to pay attention, and presents the output in the form of a report to form a decision result. The decision result can be a recommended parameter scheme presented in a list state, a list of alternative schemes, or a parameter combination with risk assessment.

[0079] Based on the results of decision-making, extraction and analysis are carried out. The chemical principles or successful experiences verified by experiments submitted by experts can be formalized as new chemical rule entries and added to the knowledge base after being reviewed and confirmed by a group of experts in the field for their universality and accuracy, or used to modify the applicable conditions or parameter ranges of existing rules.

[0080] In combination with the above, the method stores the chemical principles and expert experience in the field of medicinal peptide synthesis in a structured manner by establishing a chemical rule knowledge base. After the data-driven model generates a preliminary quantitative parameter combination, the parameter combination is verified using the chemical rule knowledge base. The verification process identifies parts of the parameter combination that conflict with known chemical rules or may trigger side reactions. Based on the verification results, the parameter combination is modified, such as adjusting reagent ratios or reaction conditions, or generating warning information to alert users of potential risk points. In this way, the recommended parameter scheme is ensured to be chemically reasonable, avoiding recommendations that may violate basic chemical principles generated by the model. Further, the method introduces a feedback mechanism. Feedback information generated by users or actual synthesis results, including suggestions for parameter adjustment or chemical principles supporting these suggestions, is used to update the chemical rule knowledge base or adjust the internal parameters of the data-driven model. This feedback cycle enables the system to continuously learn and adapt, continuously improving the accuracy and reliability of parameter optimization, solving the problem of existing systems that are difficult to integrate expert knowledge and lack continuous improvement capabilities.

[0081] As a specific example in this embodiment, assume that the target peptide structure contains a sequence known to be prone to beta-elimination under basic conditions.

[0082] Extract and fuse rules from literature and experimental data to form a chemical rule knowledge base, which contains a rule: a specific sequence has a high risk of beta-elimination when the pH is higher than X; the data-driven model recommends a parameter combination based on historical data, which includes the use of a buffer with a pH of Y (Y>X). The system obtains the characteristics of the target peptide structure and the parameter combination. According to the chemical rule knowledge base, the verification module identifies that the pH of the buffer in the parameter combination conflicts with the beta-elimination rule in the knowledge base. The verification result indicates that there is a high risk. The system combines the verification result and automatically modifies the pH of the buffer to Z (Z<X) to obtain the modified parameter combination, or generates a warning message: "Attention: the current buffer pH may cause beta-elimination side reactions"; generates a parameter scheme including the modified pH value as the decision result: if the user adopts the modified scheme and obtains good results in the experiment, the results can be used as feedback to further fine-tune the model.

[0083] As shown in Figure 2 Step S1 of the embodiment includes the following steps:

[0084] S11, obtain chemical application knowledge information related to amino acid sequences, protecting groups, and coupling reagents in the field of medicinal peptide synthesis;

[0085] S12, based on the chemical application knowledge information, rules are formulated according to the peptide structure characteristics and process conditions and the potential risks or recommended operations corresponding to the peptide structure characteristics and process conditions, and are structured and stored according to the data set, and the obtained various data sets are collected to obtain a chemical rule knowledge base.

[0086] In step S11, specific chemical application knowledge directly related to the synthesis of medical peptides is obtained. These knowledge can be collected from public chemical literature, patent database, chemical reaction database, expert experience database or internal experimental data. The acquisition process can adopt information extraction, text mining or manual sorting, etc. The obtained information focuses on the behavior and influence of specific amino acid sequences (for example, amino acids prone to racemization, with steric hindrance or specific reaction activity), commonly used protecting groups (for example, their removal conditions, stability, possible side reactions) and coupling reagents (for example, coupling efficiency, side reaction tendency, applicable range) under different synthesis conditions.

[0087] Further, in step S12, the obtained chemical application knowledge is converted into structured rules. Rule formulation can be based on pre-defined rule templates, and knowledge information is mapped to fields such as "peptide structure characteristics", "process conditions", "potential risks" and "recommended operations". For example: a rule can be described as "when the peptide chain contains a specific sequence fragment and uses a specific coupling reagent, there is a certain side reaction risk". These rules are organized into data sets and stored in a structured manner, for example, stored in a table of a relational database, each rule as a record, containing the values of each field, and finally forming a chemical rule knowledge base, such as the reference structure sequence "[peptide structure characteristics A AND / OR process conditions B] THEN [potential risks COR recommended operations D]".

[0088] In a specific example, consider building a chemical rule knowledge base about the behavior of racemizable amino acids under different coupling conditions. Obtain racemization rate data and related literature reports on common racemizable amino acids (for example, His, Cys, Asp) in solid-phase peptide synthesis using different coupling reagents (for example, HATU, HBTU, PyBOP) and additives (for example, HOBt, HOAt). For example, it is obtained that when HATU is used to couple C-terminal His, the racemization risk is high. Step S12 formulates a rule based on this knowledge: "if the peptide chain C-terminal contains His and the coupling reagent is HATU, there is a high risk of racemization, and it is recommended to use alternative coupling reagents or add HOAt". This rule is structured and stored in a table of a relational database, containing the field "peptide structure characteristics" value "C-terminal His", "process conditions" value "coupling reagent HATU", "potential risks" value "high racemization risk", "recommended operations" value "use alternative coupling reagents / add HOAt".

[0089] As Figure 3 shown, step S2 of the present embodiment includes the following steps:

[0090] S21, obtaining a quantitative addition parameter combination combined by the structural characteristics of the target peptide and the parameters output by the data-driven model;

[0091] S22, according to the chemical rule knowledge base, the obtained quantitative addition parameter combination is subjected to primary verification to identify the primary verification results that are inconsistent with the chemical rules in the chemical rule knowledge base or may cause side reactions;

[0092] S23, based on the primary verification result, the current quantitative addition parameter combination is subjected to primary correction or generates an alert, and the quantitative addition parameter combination obtained after the primary verification is obtained is corrected or accompanied by an alert;

[0093] S24, the obtained quantitative addition parameter combination after the primary verification is subjected to iterative verification processing on the target peptide information set, the current quantitative addition parameter combination is subjected to verification to obtain the final verification result associated with the current quantitative addition parameter combination, and the quantitative addition parameter combination after the correction or accompanied by the alert is obtained based on the final verification result.

[0094] Among them, step 21 is responsible for receiving or generating the quantitative addition parameter combination to be processed, which is based on the structural information of the target peptide and the parameter set calculated by the data-driven model.

[0095] Step S22 is to perform the first rule check, and compare with the chemical rule knowledge base to identify the part of the parameter combination that conflicts with the known chemical rules or has potential side reaction risk. The primary verification result is obtained thereby.

[0096] Step S23 is to adjust or mark the warning according to the primary verification result, which is a direct response to the preliminary discovery of the problem.

[0097] Step S24 Step S23 is an iterative verification process. The parameter combination obtained after the correction or accompanied by the alert in step S23 is taken as input, and the loop verification is started, and in each loop, the current parameter combination is rechecked according to the chemical rule knowledge base.

[0098] Specifically, by introducing an iterative verification process in this process step, the verification and correction capabilities of the quantitative addition parameter combination are enhanced, solving the problem that a single verification and correction process may introduce new situations that are inconsistent with chemical rules or trigger new side reaction risks. During the iterative process, the parameter combination after the initial correction is used as input and verified again according to the chemical rule knowledge base. If new inconsistencies or risks are found, further corrections are made. This verification and correction cycle continues until the parameter combination no longer needs to be corrected after a complete verification, indicating that it has reached a stable state, or the preset maximum number of iterations has been reached. The parameter combination finally output is the result of multiple iterative verification and corrections, which reduces the risk of introducing new problems due to a single correction and improves the reliability of the parameter combination.

[0099] In a specific example, suppose the target peptide sequence contains an amino acid residue prone to racemization. Step S21 obtains the quantitative additive parameter combination output by the data-driven model, which may include a higher reaction temperature and a common coupling agent. Step S22 performs an initial validation against the chemical rule knowledge base, which contains a rule indicating "racemization-prone residue + high temperature + common coupling agent -> high racemization risk." The initial validation results identify a high racemization risk for this parameter combination. Based on this result, step S23 makes an initial correction to the parameter combination, for example, lowering the reaction temperature to 10°C and recommending the use of a coupling agent known to inhibit racemization. Step S24 initiates an iterative validation. The revised parameter combination (low temperature, specific coupling agent) is used as input for revalidation. In the first iterative revalidation, this new combination is checked against the chemical rule knowledge base. It may be found that low temperature significantly prolongs the reaction time, conflicting with another rule: "Excessive reaction time + certain protecting groups -> risk of protecting group shedding." The revalidation results identify new risks. Based on this result, further corrections are made, such as fine-tuning the temperature to 15°C and adding a small amount of catalyst to the reaction system to shorten the reaction time. The second iteration begins, and the revised parameter combination (15°C, specific coupling agent, and a small amount of catalyst) is recalibrated. If this verification does not reveal any new discrepancies with chemical rules or the induction of side reactions, a stable state is determined and the iteration terminates. The final output is this parameter combination that has undergone multiple corrections and verifications.

[0100] like Figure 4 As shown, step S22 in this embodiment includes the following steps:

[0101] S221, obtain one or more parameter subsets in which the parameters in the current quantitative addition parameter combination have synergistic or antagonistic effects, and for each identified parameter subset, retrieve parameter interaction rules corresponding to the parameter subset from a chemical rule knowledge base or a parameter interaction knowledge database, the parameter interaction rules indicating the effects on the progress, efficiency of the target chemical reaction, or the occurrence probability, degree of side reactions caused by the combined action of the member parameters in the parameter subset;

[0102] S222, based on each retrieved parameter interaction rule, perform adjustment on the compliance checking conclusion of the single rule verification result for the individual parameter or the entire parameter combination in the parameter subset or the risk level or occurrence probability of the side reactions caused by the combined action of the member parameters in the parameter subset to obtain parameter interaction adjustment information;

[0103] S223, according to the chemical rule knowledge base, perform single rule verification on the obtained quantitative addition parameter combination to identify parameters that do not comply with the chemical rules in the chemical rule knowledge base or may cause side reactions, and obtain a single rule verification result;

[0104] S224, combine the single rule verification result and the parameter interaction adjustment information to obtain a preliminary verification result.

[0105] Specifically, in the process of medical peptide synthesis, the accuracy of the quantitative addition parameter combination directly affects the synthesis result. The traditional verification method may only check whether each parameter complies with the independent rule, ignoring the additional risks or effects that may be caused by the interaction between parameters. In step S22, the analysis of the interaction between parameters is introduced to improve the comprehensiveness of the verification.

[0106] First, identify parameter groups in the quantitative addition parameter combination that may have synergistic or antagonistic effects. For example, the concentration of the coupling agent and the reaction temperature may have mutual influence. Then, query the knowledge base to obtain rules describing how these parameter combinations affect the reaction or side reactions. These rules may indicate that even if the coupling agent concentration and temperature are within the acceptable range individually, their specific combination may significantly increase the occurrence probability of a certain side reaction.

[0107] At the same time, perform basic single rule verification to check whether each parameter or parameter combination complies with independent chemical rules. Finally, combine the results of single rule verification with the adjustment information obtained from parameter interaction analysis. If the parameter interaction rules indicate that a certain parameter combination significantly increases the risk of side reactions, even if the single rule verification does not find any problems, this risk information will be integrated into the final preliminary verification result.

[0108] Therefore, the initial verification results not only reflect the compliance of the parameters with the independent rules, but also consider the overall effect of the parameter combination, thereby more accurately identifying potential problems and reducing the synthetic risks caused by parameter interactions.

[0109] In a specific example, for a target peptide structure, the system obtains its quantitative addition parameter combination, such as the molar equivalent of coupling agent A is 1.5, the molar equivalent of base B is 2.0, and the reaction temperature is 25°C. The system recognizes that the molar equivalent of coupling agent A and the molar equivalent of base B may have a synergistic effect, and retrieves a rule from the knowledge base: when the molar equivalent of coupling agent A is higher than 1.2 and the molar equivalent of base B is higher than 1.8, the probability of occurrence of a specific side reaction C increases significantly. At the same time, the system performs a single rule check and finds that the 1.5 molar equivalents of coupling agent A comply with its independent use rules (for example, the recommended range 0.8-2.0), the 2.0 molar equivalents of base B also comply with its independent use rules (for example, the recommended range 1.0-2.5), and the reaction temperature of 25°C complies with its independent use rules (for example, the recommended range 20-30°C). The single rule check results show that these parameters comply with the rules when viewed individually. However, based on the retrieved parameter interaction rules, the system assessed that the combination of 1.5 molar equivalents of coupling agent A and 2.0 molar equivalents of base B significantly increased the risk of side reaction C, generating parameter interaction adjustment information, indicating a high risk for side reaction C. Ultimately, the system combined the single rule verification results with the parameter interaction adjustment information to produce an initial verification result, indicating that although this parameter combination individually complies with the rules, there is a high risk of side reaction C due to the interaction between the parameters, and may generate a corresponding warning message.

[0110] like Figure 5 As shown, step S223 in this embodiment includes the following steps:

[0111] S2231. Retrieving chemical rules related to the current quantitative addition parameter combination and the structural characteristics of the target peptide from the chemical rule knowledge base, and identifying rules with their own confidence scores or probabilistic descriptions of their applicable conditions among the related chemical rules, to obtain chemical rules with accompanying uncertainty information;

[0112] S2232. For each chemical rule identified with accompanying uncertainty information, calculate the probability that the current quantitative addition parameter combination complies with the chemical rule with accompanying uncertainty information based on the chemical rule's confidence score or probabilistic description of its applicable conditions and the matching of the quantitative addition parameter combination with the preconditions of the chemical rule;

[0113] S2233. For one or more preset side reactions, based on one or more chemical rules related to the side reaction, combined with the probability of compliance of the current quantitative addition parameter combination with the chemical rule with accompanying uncertainty information, and using a probability synthesis method when multiple related chemical rules contribute to the risk of the same side reaction, calculate the comprehensive risk probability of the current quantitative addition parameter combination causing the side reaction;

[0114] S2234. Using the calculated probability of the current quantitative addition parameter combination complying with each chemical rule and the calculated comprehensive risk probability of the current quantitative addition parameter combination triggering each preset side reaction, identify parameters that do not comply with the chemical rules in the chemical rule knowledge base or may trigger side reactions, and obtain a single rule verification result.

[0115] Among them, chemical rules related to the current quantitative addition parameter combination and the structural characteristics of the target peptide are retrieved from the chemical rule knowledge base, and the rules with their own confidence scores or probabilistic descriptions of their applicable conditions in the relevant chemical rules are identified to obtain chemical rules with accompanying uncertainty information; for each chemical rule with accompanying uncertainty information obtained, based on the confidence score of the chemical rule itself or the probabilistic description of its applicable conditions, and the matching of the quantitative addition parameter combination with the premise conditions of the chemical rule, the probability of the current quantitative addition parameter combination meeting the chemical rule with accompanying uncertainty information is calculated; for one or more preset secondary reactions, Based on one or more chemical rules related to the side reaction, combined with the compliance probability of the current quantitative addition parameter combination with the chemical rule with the accompanying uncertainty information, and using a probability synthesis method when multiple related chemical rules contribute to the risk of the same side reaction, the comprehensive risk probability of the current quantitative addition parameter combination triggering the side reaction is calculated; using the calculated compliance probability of the current quantitative addition parameter combination with each chemical rule and the calculated comprehensive risk probability of the current quantitative addition parameter combination triggering each preset side reaction, parameters that do not comply with the chemical rules in the chemical rule knowledge base or may trigger side reactions are identified, and a single rule verification result is obtained.

[0116] In addition, chemical rules with uncertainty information can be identified by means of metadata tags or fields pre-set for each rule in the chemical rule knowledge base, such as a field representing a confidence value or a field describing an applicable probability distribution. The retrieval process distinguishes deterministic rules from uncertain rules according to the presence or absence of these fields. The calculation of the compliance probability of the current quantitative parameter combination with a chemical rule with uncertainty information can be a combination operation of the rule's own confidence or probability information with the degree to which the parameter combination satisfies the rule's premise. For example, if the rule's premise is fully satisfied, the compliance probability equals the rule's own confidence; if partially satisfied, the confidence is weighted according to the degree of matching. When calculating the comprehensive risk probability of inducing a side reaction, the individual risk contributions of multiple rules related to a specific side reaction are taken as inputs, and a probability synthesis method is used for integration. The probability synthesis method can be implemented using Bayesian networks, fuzzy logic reasoning, or a simple probability aggregation model, thereby obtaining a quantitative probability of side reaction occurrence.

[0117] In a specific example, consider the synthesis of a peptide containing an Asp-Gly sequence, which is known to be susceptible to imidization under specific conditions. A data-driven model recommends a set of parameters, including the use of HATU as a coupling agent, DIPEA as a base, DMF as a solvent, and a reaction temperature of 25°C. The system performs a single rule validation. First, it retrieves rules related to the Asp-Gly sequence, HATU, DIPEA, DMF, and a temperature of 25°C from the chemical rule knowledge base. The system identifies the rule: "Using HATU / DIPEA to couple Asp-Gly sequences in DMF increases the risk of imidization at 25°C (confidence level: 0.7)." This rule is identified as having associated uncertainty information. The system then calculates the match between the current parameter combination (HATU, DIPEA, DMF, 25°C) and the rule's preconditions and finds a perfect match. Based on the rule's confidence level of 0.7, the calculated probability of compliance with the rule for the current parameter combination is 0.7. The system also identified another rule: "The Asp-Gly sequence itself is prone to imidization (probability under typical conditions: 0.8)". Both rules are related to imidization side reactions. The system uses a probabilistic synthesis method, such as considering the independent contributions of the two rules for probabilistic aggregation, and calculates the comprehensive risk probability of the current parameter combination triggering an imidization side reaction, for example, the calculated result is 0.65. Based on the calculated rule compliance probability of 0.7 and the imidization risk probability of 0.65, the system identifies that parameters such as HATU, DIPEA, DMF, and 25°C are related to potential imidization risks, and outputs this information as a single rule verification result, such as marking these parameters as "high imidization risk-associated parameters" with risk probability values. As a result, R&D personnel obtain a quantitative risk assessment, rather than just a simple rule violation prompt.

[0118] like Figure 6 As shown, in this embodiment, step S2233 includes the following steps:

[0119] S22331. Retrieving and obtaining, for a plurality of chemical rules related to the side reaction, a dependency description between the chemical rules from a rule dependency data source, wherein the dependency description indicates an interaction pattern of the chemical rules in jointly influencing the probability of occurrence of the side reaction;

[0120] S22332, before performing the probability synthesis operation, modifying, according to the obtained dependency description, the individual risk contribution of at least one rule of the plurality of chemical rules to the side reaction, to obtain a modified individual risk contribution, and / or constructing a probability synthesis structure integrating the dependency description, and obtaining, as inputs of the probability synthesis structure, the individual risk contributions of the plurality of chemical rules related to the side reaction, each of which is generated by a corresponding rule to the side reaction, wherein the individual risk contribution indicates a compliance probability of the current quantitative addition parameter combination to the corresponding rule calculated for the corresponding rule, or is derived from probability information directly included in the corresponding rule for describing the side reaction;

[0121] S22333, performing the probability synthesis operation by using the obtained modified individual risk contribution, and / or performing the probability synthesis operation by using the probability synthesis structure constructed in the foregoing step and the obtained individual risk contributions as inputs of the probability synthesis structure, so as to calculate, through the probability synthesis operation, a comprehensive risk probability of the side reaction induced by the current quantitative addition parameter combination in combination with the dependency relationship among the chemical rules.

[0122] wherein, for the plurality of chemical rules related to the side reaction, a dependency description among the chemical rules is retrieved and obtained from a rule dependency data source, the dependency description indicating an interaction mode of the chemical rules in jointly affecting the probability of the side reaction; before performing the probability synthesis operation, modifying, according to the obtained dependency description, the individual risk contribution of at least one rule of the plurality of chemical rules to the side reaction, to obtain a modified individual risk contribution, and / or constructing a probability synthesis structure integrating the dependency description, and obtaining, as inputs of the probability synthesis structure, the individual risk contributions of the plurality of chemical rules related to the side reaction, each of which is generated by a corresponding rule to the side reaction, wherein the individual risk contribution indicates a compliance probability of the current quantitative addition parameter combination to the corresponding rule calculated for the corresponding rule, or is derived from probability information directly included in the corresponding rule for describing the side reaction; performing the probability synthesis operation by using the obtained modified individual risk contribution, and / or performing the probability synthesis operation by using the probability synthesis structure constructed in the foregoing step and the obtained individual risk contributions as inputs of the probability synthesis structure, so as to calculate, through the probability synthesis operation, a comprehensive risk probability of the side reaction induced by the current quantitative addition parameter combination in combination with the dependency relationship among the chemical rules, and the rule dependency data source can store the interaction mode among the chemical rules, for example, indicating that there is an enhancing, inhibiting or mutually exclusive relationship among the rules.

[0123] In one specific example, consider a target peptide that can undergo a beta-elimination side reaction under certain conditions. Assume there are two chemical rules associated with this: rule R1 indicates that a certain sequence fragment is prone to beta-elimination under basic conditions, with an individual risk contribution of P1; and rule R2 indicates that a certain protecting group is prone to induce beta-elimination under the action of a certain deprotection reagent, with an individual risk contribution of P2. From the rule dependency data source, it is discovered that there is a synergistic enhancement relationship between rules R1 and R2, i.e. when the sequence fragment and the protecting group both satisfy the conditions, the risk of beta-elimination is significantly higher than the simple addition of each alone. Based on the acquired synergistic enhancement dependency description, before performing the probability synthesis operation, P1 and P2 can be modified, for example, a nonlinear combination function f(P1, P2) = P1 + P2 + a*P1*P2 (where a > 0 represents an enhancement factor) is used to calculate the modified joint contribution. Alternatively, a simple Bayesian network is constructed, in which there are two parent nodes representing the satisfaction of the conditions of rules R1 and R2, respectively, and a child node representing the occurrence of the beta-elimination side reaction, and the conditional probability table of the child node is defined according to the dependency relationship of R1 and R2. P1 and P2 are input probabilities of the parent nodes, and the comprehensive risk probability of the beta-elimination side reaction is calculated by network inference. For example, if P1 = 0.4, P2 = 0.3, and the dependency relationship indicates synergistic enhancement, simple addition may result in 0.7, but considering the enhancement effect, the modified joint contribution or the Bayesian network inference result may be 0.85, which more accurately reflects the actual risk.

[0124] As Figure 7 shown, in the present embodiment, step S24 includes the following steps:

[0125] S241, after obtaining the modified or warning-attached quantitative addition parameter combination after the initial verification, an iterative verification process is started, and in each iteration of the iterative verification process, re-verification is performed on the quantitative addition parameter combination of the current iteration round according to the chemical rule knowledge base;

[0126] S242, it is judged whether the iterative verification process meets a preset iteration termination condition, the iteration termination condition indicating that the updated quantitative addition parameter combination reaches a stable state without further modification after one complete re-verification or the number of iterations of the iterative process reaches a preset upper limit; if the iteration termination condition is met, the updated quantitative addition parameter combination is taken as the modified or warning-attached quantitative addition parameter combination; if the preset iteration termination condition is not met, the updated quantitative addition parameter combination is taken as the input quantitative addition parameter combination of the next iteration round, and the re-verification in each iteration of the iterative process is returned to be performed, and finally a re-verification result is obtained;

[0127] S242. Based on the re-verification result, the quantitative addition parameter combination of the current iteration round is re-corrected to obtain a final quantitative addition parameter combination after correction or with a warning.

[0128] Among them, after the revised or warning quantitative addition parameter combination obtained after the initial verification is proposed, the iterative verification process is started, and in each iteration of the iterative verification process, the quantitative addition parameter combination of the current iteration round is re-verified according to the chemical rule knowledge base; it is judged whether the iterative verification process meets the preset iteration termination condition, and the iteration termination condition indicates that the updated quantitative addition parameter combination reaches a stable state after a complete re-verification without further correction or the number of iterations of the iterative process reaches a preset upper limit; if the iterative termination condition is met, the updated quantitative addition parameter combination is used as the revised or warning quantitative addition parameter combination; if the preset iterative termination condition is not met, the updated quantitative addition parameter combination is used as the input quantitative addition parameter combination of the next iteration round, and the re-verification in each iteration of the iterative process is returned to finally obtain the re-verification result; based on the re-verification result, the quantitative addition parameter combination of the current iteration round is re-corrected to obtain the final revised or warning quantitative addition parameter combination.

[0129] In a specific example, an initial parameter combination P = {p1: 1.0, p2: 2.0} is initially verified and corrected to P' = {p1: 0.8, p2: 2.2}, accompanied by a warning message W1 indicating that the value of p1 may not comply with rule R_A. An iterative verification process is initiated. In the first iteration, P' is reverified (S241). This reverification reveals that while p1 = 0.8 no longer violates R_A, p2 = 2.2 does not comply with rule R_B, and the combination of p1 and p2 may trigger a side effect F1. The termination condition is determined to indicate that the parameter combination has changed and the upper limit of the number of iterations has not been reached. The termination condition is not met. Based on the reverification results, P' is re-corrected. The system corrects p2 to 2.0 and adds a warning W2 indicating the risk of the combination of p1 and p2. This results in a new parameter combination P" = {p1: 0.8, p2: 2.0}, accompanied by warnings W1 and W2. P" is used as input for the next iteration. In the second iteration, P" is rechecked. The recheck found that P" meets all rules and there are no new risks. By judging the termination condition (S242), the parameter combination P" does not need further correction after this recheck and reaches a stable state. The termination condition is met. The iteration ends and outputs P" = {p1: 0.8, p2: 2.0} with warnings W1 and W2 as the final quantitative addition parameter combination after correction or with warnings.

[0130] like Figure 8 As shown, in this embodiment, step S3 includes the following steps:

[0131] S31, obtaining the relevance result between the modified or warning-attached quantitative addition parameter combination and the chemical rule knowledge base;

[0132] S32, evaluating the influence of the modified or warning-attached quantitative addition parameter combination on one or more preset optimization targets based on the relevance result, and obtaining a decision result.

[0133] In step S31, the modified or warning-attached quantitative addition parameter combination is obtained by querying the chemical rule knowledge base. By comparing each parameter in the parameter combination with the chemical rules stored in the knowledge base, it is identified whether the parameter combination conforms to the rules, whether there is a potential conflict or risk, and the relevance result is presented in the form of a relevance report or a score.

[0134] In step S32, the influence of the parameter combination on the preset optimization targets is evaluated based on the relevance result obtained in step S32. The preset optimization targets can include yield, purity, impurity content, cost, reaction time, etc. The evaluation process receives the parameter combination, the relevance result, and the preset optimization targets as inputs, outputs the predicted influence on each target or a comprehensive evaluation result, and finally generates a decision result. The result reflects the comprehensive performance of the parameter combination on multiple optimization targets.

[0135] As shown in FIG. 8, in this embodiment, step S321 includes the following steps: Figure 9

[0136] S321, obtaining the selected modified or warning-attached quantitative addition parameter combination and the user's preference information for multiple conflicting medical peptide optimization targets;

[0137] S322, for each of the multiple conflicting medical peptide optimization targets, calculating the quantitative achievement degree of the medical peptide optimization target based on the obtained modified or warning-attached quantitative addition parameter combination; identifying the conflict relationship between the multiple conflicting medical peptide optimization targets based on the obtained modified or warning-attached quantitative addition parameter combination and the multiple conflicting medical peptide optimization targets, and quantifying the conflict relationship to obtain a conflict relationship quantification result;

[0138] S323, according to the calculated quantitative achievement degree of each medical peptide optimization target, the obtained conflict relationship quantification result, and the obtained user preference information, using a comprehensive evaluation algorithm to calculate the comprehensive influence evaluation of the selected parameter combination on the multiple conflicting medical peptide optimization targets;

[0139] S324, obtaining a decision result based on the comprehensive influence evaluation.

[0140] ​The quantitative addition parameter combination to be evaluated and the user-set optimization target weight or priority information are obtained. For each optimization target, the predicted performance value of the current parameter combination on the target is calculated. At the same time, the mutual influence mode between different optimization targets is analyzed, for example, the improvement of one target may lead to the decline of another target, and the degree of mutual influence is converted into a numerical value. The individual target performance value, the mutual influence value between targets, and the obtained user preference information are taken as inputs, and a preset algorithm model is used for calculation, and a value or level reflecting the overall performance of the parameter combination is output. Step S324 performs decision generation operation, and determines the final decision output according to the obtained overall performance value or level.

[0141] This step is used to solve the problem of how to evaluate the parameter combination and make a decision when there are multiple interrelated and possibly interdependent optimization targets (such as yield, purity, cost, and time) in the optimization process of the quantitative addition parameters of medical peptides. First, a parameter combination that has been preliminarily corrected or with a warning is obtained, and user input about the importance of different optimization targets is received. Then, the system calculates the expected performance value of the parameter combination on each single optimization target, and analyzes the degree of mutual influence between these targets, and quantifies this influence. Subsequently, these information is combined with the user's preferences, and an algorithm is used to calculate the overall evaluation of the parameter combination considering all targets, target relationships and user requirements. Finally, based on this overall evaluation, a decision result about the parameter combination is generated.

[0142] In summary, in one aspect of the embodiment described above, a machine learning-based medical peptide quantitative addition parameter optimization method is provided, which identifies and corrects the situation where the chemical rules in the chemical rule knowledge base do not conform or cause side reactions by constructing a chemical rule knowledge base and verifying the parameter combination output by the data-driven model, and generates a warning to obtain a decision result. According to the decision result, feedback is obtained and the knowledge base is updated or the model parameters are adjusted, and thus a chemical rule-based parameter verification and correction decision explanation is constructed in the medical peptide quantitative addition parameter optimization decision-making scenario, which can update the chemical rule knowledge base or adjust the data-driven model parameters based on the decision result and with the participation of experts in decision-making, improve the reliability of the parameter optimization result, and support the iterative improvement of system knowledge and model.

[0143] As the second aspect of the embodiment of the present application, as shown in Figure 10 A machine learning-based medical peptide quantitative addition parameter optimization system 100 is provided, which comprises:

[0144] The chemical rule knowledge base construction module 101 is used for obtaining original data associated with the field of medical peptide synthesis from multiple specific channels, fusing the obtained original data using a data fusion algorithm to obtain a chemical rule knowledge base;

[0145] The data acquisition and parameter correction warning module 102 is used for obtaining the structural characteristics of the target peptide, obtaining a quantitative addition parameter combination based on the structural characteristics of the target peptide and the parameters input by the data-driven model, verifying the obtained quantitative addition parameter combination according to the chemical rule knowledge base to identify a verification result that does not conform to the chemical rules in the chemical rule knowledge base or that may cause a side reaction, and correcting or generating a warning for the quantitative addition parameter combination according to the verification result, to obtain a quantitative addition parameter combination that is corrected or accompanied by a warning;

[0146] The decision basis generation module 103 is used for generating a decision result based on the quantitative addition parameter combination that is corrected or accompanied by a warning.

[0147] The feedback and calibration module 104 is used for obtaining feedback information based on the decision result and updating the chemical rule knowledge base or adjusting the preset parameters in the data-driven model based on the obtained feedback information, wherein the feedback information includes user-suggested parameter adjustments or supporting chemical principles.

[0148] In the two aspects of the embodiment, a medical peptide quantitative addition parameter optimization system based on machine learning is constructed based on an optimization method for medical peptide quantitative parameters, and the system framework is completed by configuring a chemical rule knowledge base construction module, a data acquisition and parameter correction warning module, a decision basis generation module, and a feedback and calibration module, to complete the verification of the chemical rule knowledge base and the parameter combination output by the data-driven model, identify the chemical rule inconsistency or side reaction in the chemical rule knowledge base and correct or generate a warning, obtain a decision result, and obtain feedback according to the decision result and update the knowledge base or adjust the model parameters, thereby constructing a medical peptide quantitative addition parameter optimization decision-making scenario that can integrate chemical rules for parameter correction and provide chemical logic for decision explanation, update the chemical rule knowledge base or adjust the data-driven model parameters based on the decision result and with the participation of experts in the decision-making process in the feedback process of the decision result, improve the reliability of the parameter optimization result, and support the iterative improvement of the system knowledge and the model.

[0149] Those skilled in the art can make various modifications and changes to the above embodiments according to the disclosure and teachings herein. Therefore, the application is not limited to the specific embodiments disclosed and described above, and some modifications and changes to the application shall fall within the protection scope of the claims of the application. In addition, although some specific terms are used in the specification, these terms are only for convenience of description and do not constitute any limitation on the application.

Claims

1. A method for optimizing parameters for quantitative addition of pharmaceutical peptides based on machine learning, characterized in that: The steps include: S1. Obtain raw data related to the field of pharmaceutical peptide synthesis from multiple specific channels, and use rule extraction technology to fuse the obtained raw data to obtain a chemical rule knowledge base; S2. Obtaining structural features of the target peptide, obtaining a quantitative addition parameter combination based on the structural features of the target peptide in combination with parameters input by the neural network model, verifying the obtained quantitative addition parameter combination according to a chemical rule knowledge base to identify verification results that are inconsistent with chemical rules in the chemical rule knowledge base or may induce side reactions, and modifying the quantitative addition parameter combination or generating a warning based on the verification results to obtain a modified quantitative addition parameter combination or a quantitative addition parameter combination with a warning; S3. Generate a decision result based on the revised or additionally warned quantitative addition parameter combination; S4. Obtaining feedback information based on the decision result and updating the chemical rule knowledge base or adjusting preset parameters in the neural network model based on the obtained feedback information, wherein the feedback information includes the parameter adjustment suggested by the user or the chemical principle supporting the suggestion; Wherein, the step S2 specifically includes: S21, obtaining a quantitative addition parameter combination based on the structural characteristics of the target peptide and the parameters output by the neural network model; S22. Performing a primary verification on the obtained quantitative addition parameter combination according to a chemical rule knowledge base to identify a primary verification result that is inconsistent with the chemical rule in the chemical rule knowledge base or may cause a side reaction; S23. Based on the initial verification result, the current quantitative addition parameter combination is initially corrected or a warning is generated to obtain a corrected quantitative addition parameter combination or a quantitative addition parameter combination with a warning obtained after the initial verification; S24. After obtaining the corrected or warned quantitative addition parameter combination obtained after the initial verification, iterative verification processing is performed on the target peptide information set, and the current quantitative addition parameter combination is verified to obtain a final verification result associated with the current quantitative addition parameter combination, and a corrected or warned quantitative addition parameter combination is obtained based on the final verification result.

2. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 1, characterized in that: The step S1 comprises the following steps: S11. Obtaining chemical application knowledge information related to amino acid sequences, protecting groups, and coupling reagents in the field of pharmaceutical peptide synthesis; S12. Based on the chemical application knowledge information, rules are formulated according to the peptide structural characteristics and process conditions, as well as the potential risks or recommended operations corresponding to the peptide structural characteristics and process conditions, and structured storage is performed according to the data set. The obtained data sets are aggregated to obtain a chemical rule knowledge base.

3. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 1, characterized in that: The step S22 specifically includes: S221. Obtain one or more parameter subsets in which synergistic or antagonistic effects exist between member parameters in the current quantitative addition parameter combination, and for each identified parameter subset, retrieve a parameter interaction rule corresponding to the parameter subset from a chemical rule knowledge base or a parameter interaction knowledge database, wherein the parameter interaction rule indicates the effect of the member parameters in the parameter subset acting together on the progress and efficiency of the target chemical reaction or on the probability and degree of occurrence of a side reaction; S222. Based on each retrieved parameter interaction rule, adjust the compliance verification conclusion of a single parameter or the entire parameter combination in the parameter subset based on the single rule verification result, or evaluate the risk level or occurrence probability of a side effect caused by the joint action of the member parameters in the parameter subset, to obtain parameter interaction adjustment information; S223. Performing a single rule verification on the obtained quantitative addition parameter combination based on the chemical rule knowledge base to identify parameters that are inconsistent with the chemical rules in the chemical rule knowledge base or that may cause side reactions, and obtaining a single rule verification result; S224: Combine the single rule verification result and the parameter interaction adjustment information to obtain an initial verification result.

4. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 3, characterized in that: The step S223 specifically includes: S2231. Retrieving chemical rules related to the current quantitative addition parameter combination and the structural characteristics of the target peptide from the chemical rule knowledge base, and identifying rules with their own confidence scores or probabilistic descriptions of their applicable conditions among the related chemical rules, to obtain chemical rules with accompanying uncertainty information; S2232. For each chemical rule identified with accompanying uncertainty information, calculate the probability that the current quantitative addition parameter combination complies with the chemical rule with accompanying uncertainty information based on the chemical rule's confidence score or probabilistic description of its applicable conditions and the matching of the quantitative addition parameter combination with the preconditions of the chemical rule; S2233. For one or more preset side reactions, based on one or more chemical rules related to the side reaction, combined with the probability of compliance of the current quantitative addition parameter combination with the chemical rule with accompanying uncertainty information, and using a probability synthesis method when multiple related chemical rules contribute to the risk of the same side reaction, calculate the comprehensive risk probability of the current quantitative addition parameter combination causing the side reaction; S2234. Using the calculated probability of the current quantitative addition parameter combination complying with each chemical rule and the calculated comprehensive risk probability of the current quantitative addition parameter combination triggering each preset side reaction, identify parameters that do not comply with the chemical rules in the chemical rule knowledge base or may trigger side reactions, and obtain a single rule verification result.

5. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 4, characterized in that: The step S2233 specifically includes: S22331. Retrieving and obtaining, for a plurality of chemical rules related to the side reaction, a dependency description between the chemical rules from a rule dependency data source, wherein the dependency description indicates an interaction pattern of the chemical rules in jointly influencing the probability of occurrence of the side reaction; S22332. Based on the obtained dependency description, before executing the probability synthesis operation, modify the individual risk contribution of at least one of the multiple chemical rules to the side reaction caused by the rule to obtain a modified individual risk contribution, and / or construct a probability synthesis structure that integrates the dependency description, and obtain the individual risk contribution of each of the multiple chemical rules related to the side reaction to the side reaction as an input to the probability synthesis structure, wherein the individual risk contribution is indicated by a probability of compliance of the current quantitative addition parameter combination with the chemical rule calculated for the corresponding rule, or is derived from probability information directly contained in the corresponding rule that describes the triggering of the side reaction; S22333. Perform a probability synthesis operation using the corrected individual risk contribution obtained, and / or perform a probability synthesis operation using the probability synthesis structure constructed in the previous step and the individual risk contribution obtained as its input, so as to calculate the comprehensive risk probability of causing the side reaction by combining the dependency relationship between chemical rules and the current quantitative addition parameter combination through the probability synthesis operation.

6. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 1, characterized in that: The step S24 specifically includes: S241. After obtaining the corrected or warned quantitative addition parameter combination obtained after the initial verification, initiating an iterative verification process, wherein in each iteration of the iterative verification process, re-verifying the quantitative addition parameter combination of the current iteration round according to the chemical rule knowledge base; S242. Determine whether the iterative verification process satisfies a preset iteration termination condition, where the iteration termination condition includes that the updated quantitative addition parameter combination reaches a stable state without further correction after a complete re-verification or the number of iterations of the iterative process reaches a preset upper limit; if the iteration termination condition is satisfied, use the updated quantitative addition parameter combination as the corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning; if the preset iteration termination condition is not satisfied, use the updated quantitative addition parameter combination as the input quantitative addition parameter combination for the next iteration, return to perform the re-verification in each iteration of the iterative process, and ultimately obtain a re-verification result; S243. Based on the re-verification result, the quantitative addition parameter combination of the current iteration round is re-corrected to obtain a final quantitative addition parameter combination after correction or with a warning.

7. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 1, characterized in that: The step S3 specifically includes: S31. Obtaining a correlation result between the corrected or warned quantitative addition parameter combination and the chemical rule knowledge base; S32. Evaluate the impact of the modified or warned quantitative addition parameter combination on one or more preset optimization targets based on the correlation results, and obtain a decision result.

8. The method for optimizing parameters of quantitative addition of pharmaceutical peptides based on machine learning according to claim 7, characterized in that: The step S32 specifically includes: S321, obtaining a selected modified or warned quantitative addition parameter combination and user preference information for multiple conflicting pharmaceutical peptide optimization goals; S322. For each of the multiple conflicting pharmaceutical peptide optimization targets, based on the obtained corrected or warned quantitative addition parameter combination, calculate a quantitative achievement degree of the pharmaceutical peptide optimization target; based on the obtained corrected or warned quantitative addition parameter combination and the multiple conflicting pharmaceutical peptide optimization targets, identify conflict relationships between the multiple conflicting pharmaceutical peptide optimization targets, and quantify the conflict relationships to obtain conflict relationship quantification results; S323. Based on the calculated quantitative achievement degree of each pharmaceutical peptide optimization goal, the obtained quantitative results of the conflict relationships, and the obtained user preference information, a comprehensive evaluation algorithm is used to calculate a comprehensive impact evaluation of the selected modified or warned quantitative addition parameter combination on the multiple conflicting pharmaceutical peptide optimization goals. S324, obtain a decision result based on the comprehensive impact assessment.

9. A machine learning-based pharmaceutical peptide quantitative addition parameter optimization system, characterized in that: include: A chemical rule knowledge base construction module is used to obtain raw data related to the field of pharmaceutical peptide synthesis from multiple specific channels, and to fuse the obtained raw data using rule extraction technology to obtain a chemical rule knowledge base; A data acquisition and parameter correction and warning module, which is used to obtain the structural characteristics of the target peptide, obtain a quantitative addition parameter combination based on the structural characteristics of the target peptide in combination with the parameters input by the neural network model, verify the obtained quantitative addition parameter combination according to the chemical rule knowledge base to identify verification results that are inconsistent with the chemical rules in the chemical rule knowledge base or may cause side reactions, and correct the quantitative addition parameter combination or generate a warning based on the verification results to obtain a corrected quantitative addition parameter combination or a warning-attached quantitative addition parameter combination; A decision basis generation module, the decision basis generation module is used to generate a decision result based on the modified or warned quantitative addition parameter combination; a feedback and calibration module, the feedback and calibration module being configured to obtain feedback information based on the decision results and, based on the obtained feedback information, update the chemical rule knowledge base or adjust preset parameters in the neural network model, wherein the feedback information includes parameter adjustments suggested by the user or chemical principles supporting the suggestion; The data acquisition and parameter correction warning module includes: a structural feature and parameter combination acquisition unit, the structural feature and parameter combination acquisition unit being used to acquire a quantitative addition parameter combination based on the structural features of the target peptide and the parameters output by the neural network model; a parameter verification unit, configured to perform an initial verification on the obtained quantitative addition parameter combination based on a chemical rule knowledge base, to identify initial verification results that are inconsistent with the chemical rules in the chemical rule knowledge base or that may induce side reactions; A correction and warning unit, wherein the correction and warning unit is used to perform an initial correction on the current quantitative addition parameter combination or generate a warning based on the initial verification result, and obtain a corrected quantitative addition parameter combination or a quantitative addition parameter combination with a warning obtained after the initial verification; an iterative verification unit, wherein the iterative verification unit is used to perform an iterative verification process on the target peptide information set based on the corrected quantitative addition parameter combination or the quantitative addition parameter combination with a warning obtained after the initial verification, and verify the current quantitative addition parameter combination to obtain a final verification result associated with the current quantitative addition parameter combination, and obtain a corrected quantitative addition parameter combination or a quantitative addition parameter combination with a warning based on the final verification result.

Images