A method and device for automatically designing and optimizing biological and chemical experiments

Abstract

The present invention belongs to the field of information data processing, and specifically relates to a method and device for automatically designing and optimizing biological and chemical experiments. The technical solution includes: a method for designing or optimizing biological and chemical experiments, including: retrieving literature; extracting parts related to experimental steps and details; obtaining a set of alternative core objects in the experiment; constructing an experimental process; extracting the real core objects of the experiment; combing the changing laws of the core objects, and verifying and optimizing the experimental process based on the changing laws; extracting experimental parameters, counting and summarizing the optional range of experimental parameters, and filling in the experimental process to form a complete experimental process. The present invention takes the core objects in the experiment as the key and index, and the natural changing laws of the core objects as the core, designs and optimizes the experimental methods, shortens the experimental design time, ensures that the designed experiments conform to objective laws, overcomes the shortcomings of existing designed biological and chemical experiments, and improves the operability and scientificity of the experiments.

Description

Technical Field

[0001] The invention belongs to the field of information data processing, and specifically relates to a method and a device for automatically designing and optimizing biological and chemical experiments. Background Art

[0002] Biological experiments and chemical experiments are essential and important components of scientific research. With the rise and development of molecular biology, chemical biology and other disciplines, designing more research-oriented and open experiments to explore the basic mechanisms of life processes, discover new biomarkers and drug targets, etc., has become the focus and hotspot of experimental design.

[0003] However, biological and chemical experiments are complex, involve many interdisciplinary subjects, have fast technology updates, have inconsistent terminology and standards, and even have poor quality scientific literature, which means that researchers often need to go through a long period of exploration, repetition, failure, and summary when designing experiments before they can gradually form an experimental method that is highly operational and reproducible. This greatly wastes the researchers' energy and time, and also greatly increases the difficulty and cost of frontier research.

[0004] Large Language Model (LLM) is an artificial intelligence technology that can learn the structure, rules, and semantics of a language by training on massive amounts of text data, and can generate text with a natural language style or answer natural language questions.

[0005] Currently known large language models cannot provide high-quality and fast analysis and summarization of experimental processes. The main reason is that existing large language models simply assemble words into sentences through semantic analysis and cannot understand the principles and logic of experimental processes like humans. Therefore, the summaries often contain a lot of errors.

[0006] If we can use large language models to provide a method for automatically and intelligently designing and optimizing biochemical experiments, it will be expected to reduce the difficulty of experimental design, shorten the experimental design time, overcome the shortcomings of the existing biological and chemical experimental design process, and accelerate the development of biochemical related sciences. Summary of the invention

[0007] The purpose of the present invention is to provide a method and device for automatically designing and optimizing biological and chemical experiments.

[0008] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a method for designing or optimizing biological and chemical experiments, comprising the following steps:

[0009] S1. Search and collect literature containing the scheme and steps of the experiment; for example, call the API interface E-utilities provided by the PubMed database through the Python programming language to realize batch search;

[0010] S2, using a large language model (such as Llama3.0) to perform conventional semantic learning and segmentation on the documents collected in step S1, and extracting the steps and details related to the experiment in the documents;

[0011] S3. Set a large language model, adjust the semantic classification principle, and obtain a set of candidate core objects in the experiment; the core object is the main processing object or key target product of the experiment. For example: adjust the Llama3.0 parameters, the main parameters are temperature adjustment in the range of 0.3-1.5, preferably 0.6, Top-k sampling in the range of 1-100, preferably 35, and Top-p sampling in the range of 0.80-0.95, preferably 0.90.

[0012] S4. Set the parameters of the large language model (for example, the temperature is adjusted to 1.0-1.2, the Top-k sampling is adjusted to 60-80, and the Top-p sampling range is 0.80-0.85), extract the experimental steps, and construct the experimental process;

[0013] S5, setting the parameters of the large language model (preferably adjusting the parameters to be consistent with S3), combining the experimental process obtained in step S4, analyzing each core object in the core object set obtained in step S3, and extracting the real core objects of the experiment according to the experimental purpose of the experiment;

[0014] S6, sorting out the change rules followed by the real core objects in the experimental changes, and based on the change rules, verifying, adjusting and optimizing the experimental process of step S4;

[0015] S7. Extract all experimental parameters that appear in the experimental process adjusted in step S6, count and summarize the optional range of the experimental parameters in the literature retrieved in step S1, and add the parameters and parameter ranges into step S6 to form a complete, detailed and operational experimental process.

[0016] Preferably, the method further comprises:

[0017] S8. Iterate steps S6 and S7 until there is no significant change in the experimental process and experimental parameters (the difference in experimental parameters is less than 10%), and the experimental design or optimization is completed.

[0018] Preferably, the method of obtaining the set of candidate core objects in the experiment in step S3 comprises the following steps:

[0019] (3-1) Extract the nouns and noun phrases in all experimental steps in step S2 as the core object set A;

[0020] (3-2) Set the lower limit n of the frequency of each noun and noun phrase in set A in the experimental steps, remove the nouns and noun phrases with a frequency lower than n in the experimental steps from set A, and obtain the streamlined core object set B;

[0021] (3-3) Extract the nouns and noun phrases from the experiment name given by the user and compare them with the core object set B. If the core object set B lacks the nouns and noun phrases extracted from the experiment name in this step, add these nouns and noun phrases to the core object set B to obtain the core object set C;

[0022] (3-4) Nouns following the verbs in the screening experimental steps are compared with the core object set C. If the noun following the verb is missing in the core object set C, it is added to the core object set C to obtain the core object set D.

[0023] Preferably, the method for obtaining the experimental process in step S4 comprises the following steps:

[0024] (4-1) searching for a core object set D in the documents retrieved in step S1. After the core object in the core object set D is retrieved, extracting the entire sentence of the core object in the document and the three entire sentences closest to the sentence, thereby obtaining a language set F.

[0025] (4-2) Perform semantic analysis and cross-comparison on the sentences extracted from different documents, merge the same or similar experimental operations for the same core object, delete duplicate descriptions / operations, and for similar operations, retain the operations with the highest frequency, and obtain the language set G;

[0026] (4-3) Performing logical analysis on each experimental operation in set G to generate an experimental process. Preferably, performing logical analysis on each experimental operation in set G to generate an experimental process comprises the following steps: arranging the experimental steps according to the order in which different steps appear in the original document, and if the order in which the same experimental step appears in different documents is inconsistent, the order in which the step appears most frequently in the document shall prevail, and experimental process A shall be obtained.

[0027] Preferably, after obtaining Experimental Flow A, the sequence of the experimental steps of Experimental Flow A is revised according to the recognized and public biological principles or chemical principles corresponding to the experiment to obtain Experimental Flow B.

[0028] Preferably, the method of extracting the real core object of the experiment in step S5 comprises the following steps:

[0029] (5-1) Using a large language model such as Llama3 through relationship extraction technology, identify the relationship between the candidate core objects in the core object set D, as well as the relationship between the candidate core objects and other nouns / noun phrases in the experimental steps that are not in the core object set D, and combine the natural language understanding of the large language model to identify the experimental purpose. Based on the relationship identification and the experimental purpose, select the target substance in the experimental purpose as the core object, and add, delete and revise the set D to obtain a single core object E1 or a core object set E2 composed of multiple core objects;

[0030] (5-2) Use the large language model to understand and identify the meaning of each substance in the core object E1 or set E2, and classify the core object according to the substance category in natural science;

[0031] (5-3) Analyze the meaning and context of each experimental step through natural language understanding, and combine the classification and meaning analysis of the core objects in step (5-2) to determine whether the core objects are appropriate and reasonable in each experimental step and the experimental purpose, thereby determining whether the core objects selected in step (5-1) are reasonable;

[0032] (5-4) Judge the overall experimental purpose and experimental steps one by one; if it is judged that there are no unreasonable items in this step, continue to the subsequent steps; if it is judged that there is at least one unreasonable item in this step, the verification is unqualified, return to step S4, replace the core object set D in step S4 with the core object E1 or set E2, re-arrange the steps of the experimental process, and then continue to step S5; repeat S4 and S5 until there are no unreasonable items.

[0033] Accordingly, a computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.

[0034] Accordingly, a computer-readable storage medium stores a computer program / instruction thereon, which implements the steps of the method when executed by a processor.

[0035] Accordingly, a computer program product comprises a computer program / instruction, which implements the steps of the method when executed by a processor.

[0036] The present invention has the following beneficial effects: Based on a large language model, the present invention provides a method and device for automatically and intelligently designing and optimizing biological and chemical experiments. The present invention takes the core object in the experiment as the key and index, and the natural change law of the core object as the basis and core, to design and optimize the experimental method, thereby, on the one hand, shortening the time for the system to select more scientific processing steps from the complicated experimental process step information, and on the other hand, ensuring that the designed experiment conforms to objective laws, thereby improving the operability and scientificity of the experiment. In the design process, the method of the present invention improves the reliability of the final generated experiment through multiple verifications, proofreading and optimization such as comparison of biological / chemical principles, semantic understanding, logical analysis, and natural change laws of core objects, and achieves the purpose of automatically designing and optimizing experimental methods without human intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the operation flow of the present invention;

[0038] Figure 2 Screenshot of the initial interface of the software designed according to the method of the present invention;

[0039] Figure 3 To use Figure 2 Screenshot of the experimental process B obtained by the software;

[0040] Figure 4 To use Figure 2 Screenshot of the experimental process after the software optimized the experimental process B;

[0041] Figure 5 To use Figure 2 The software ultimately produces a complete experimental method. DETAILED DESCRIPTION

[0042] The present invention provides a method and device for designing and optimizing biological and chemical experiments. The experiments to be designed or optimized are experiments in the field of biology or organic chemistry, including but not limited to polymerase chain reaction (PCR), DNA fragment ligation, plasmid extraction, cell counting, enzyme catalysis reaction, etc. The process of the method is as follows Figure 1 As shown, the specific steps include:

[0043] S1. Automatically search and collect documents containing the schemes and steps of the experiments to be designed or optimized through public scientific papers and book databases. At the same time, search for textbooks and reference books in the technical field corresponding to the experiments to be designed or optimized to form a common knowledge database in this field. In subsequent operations, when it is necessary to confirm and compare common knowledge information such as the laws of material changes, the order of changes, and the basic properties of materials that may be involved in the experimental design, priority is given to searching and confirming from the common knowledge database. If the corresponding information is missing from the common knowledge database, supplementary search and confirmation will be carried out from public channels on the Internet.

[0044] The preferred scheme is: when searching to form a common knowledge database, according to the disclosure of common knowledge, the basic principles of Tables 1 and 2 are entered and formed (this table is only a part of the recognized principle examples, as an example, not all known principles; the software will continuously supplement and improve Tables 1 and 2 during operation), and all retrieved biochemical experiments involving relevant principles or corresponding descriptions of the principles are added under the principle. The basic biological principles are shown in Table 1, and the basic chemical principles are shown in Table 2. The method of improving Tables 1 and 2 includes: if new principles are retrieved when using the design software provided by the present invention, additional retrieval and comparison confirmation can be performed for the new principles. If after comparison, they are different from the principles in Tables 1 and 2, and the new principles are retrieved in more than c articles (c is the number of documents, which can be set as needed) of core journals and above, or in textbooks / reference books / national standards / industry standards, the new principles are added to the basic principle table.

[0045] Table 1 Basic biological principles

[0046] in principle describe Application Examples Cell theory All living things are made up of cells, which are the basic units of life. Cell culture, tissue culture Central Dogma Genetic information is transcribed from DNA into RNA and then translated into protein PCR, RT-PCR, Western Blot Enzyme reaction Enzymes are biological catalysts that are highly specific and sensitive to temperature and pH. Enzyme reaction experiment, enzyme activity determination Membrane transport The selective permeability of the cell membrane determines the entry and exit of substances Drug permeability test, cell uptake test Genetic variation Gene mutation, recombination and selection lead to biological diversity Genetic experiments, mutant screening Signal transduction Cells transmit information through signaling molecules to regulate biological processes Cell signaling pathway research, drug action mechanism research Ecological relationship Interactions between organisms and their environment and other organisms Microbial ecology research, plant-pathogen interaction research Developmental Biology The process of development of an organism from embryo to mature individual Developmental stage studies, morphogenesis experiments Enzyme kinetics The relationship between the reaction rate of an enzyme and the concentration of its substrate can be described by the Michaelis-Menten equation. Enzyme activity assay, enzyme kinetics experiment DNA replication DNA undergoes semiconservative replication during cell division, with each new strand containing a copy of the original strand. DNA extraction, PCR, DNA sequencing DNA Structure Eukaryotes have introns, prokaryotes do not PCR, cDNA analysis RNA transcription The synthesis of RNA under the guidance of DNA template plays an important role in gene expression RNA extraction, RT-PCR, and Northern Blot experiments Protein Translation The sequence of amino acids under the guidance of mRNA forms proteins Protein expression, protein purification, Western Blot

[0047] Table 2 Basic chemical principles

[0048] in principle describe Application Examples Law of Conservation of Mass In a chemical reaction, the total mass of the reactants and products remains constant. Stoichiometry, reaction equilibrium calculations The law of conservation of energy Energy does not appear or disappear during a reaction; it is simply converted from one form to another. Thermodynamic experiments, energy conversion research Principle of Chemical Equilibrium In a state of dynamic equilibrium, chemical reactions have the same reaction rates. Equilibrium constant determination and reaction condition optimization Acid-base balance The pH value affects the concentration of hydrogen ions in the solution, affecting the reaction rate and the properties of the products. Titration experiment, buffer solution preparation Redox reaction Redox reactions involve the transfer of electrons, the interaction of oxidants and reductants Electrochemical reaction, redox titration Reaction rate The reaction rate is affected by factors such as temperature, concentration, and catalyst. Reaction kinetics research, catalyst performance testing Solubility equilibrium Solubility is the maximum amount of a substance that can dissolve in a solvent at a specific temperature. Solubility determination, precipitation reaction Intermolecular forces Intermolecular forces include van der Waals forces, hydrogen bonds, etc., which affect the physical and chemical properties of substances. Solution preparation, material property research Atomic Conservation In a chemical reaction, the total number of atoms remains the same and elements do not change into other elements. Balancing reaction equations and calculating the amount of substances Chemical reactions A chemical reaction is a process in which a substance undergoes chemical changes to produce new substances. Study of reaction types (e.g., displacement reactions, acid-base reactions) Reaction heat The heat released or absorbed during a chemical reaction affects the direction and extent of the reaction. Thermochemical experiments, reaction heat calculations coordination chemistry Metal ions and ligands form complexes with specific structures and properties. Coordination chemistry experiments, research on the properties of complexes

[0049] In order to ensure the accuracy of the experimental design in the subsequent steps, enough data should be collected in the search stage. The preferred solution is: in this step, all relevant documents are retrieved and collected, or 10,000 or more publicly published documents are retrieved and collected. If the experimental study is small, all the documents retrieved are collected. If the experimental study is large, exceeding 10,000, the collection quantity is set to a, a ≥ 10,000 according to the processing power of the processor of the experimental design device, etc.; collect the most relevant and most recent a documents retrieved (relevance is better than time).

[0050] The retrieval is performed with the name of the experiment to be designed or optimized as the first keyword. For example, the name of the experiment to be designed is "a method for catalyzing the degradation of chitosan using chitosanase", and the keyword extraction is "chitosanase catalyzing degradation of chitosan". However, this extraction method may cause a large number of missed detections due to problems such as the keywords being too long, the order of word arrangement in the literature being restricted, and the different expressions. Therefore, the preferred scheme is: the experimental name is split into nouns and verbs and used as keywords, and then searched in a way of connecting "and". For example: "Chitosanase catalyzes the degradation of chitosan" is split into "chitosan", "enzyme catalysis" and "degradation", and the search formula is: "chitosan and enzyme catalysis and degradation", and the order of word arrangement is not limited. Considering that there may be synonyms, a more preferred scheme is: after splitting the experimental name, further introduce synonyms and near-synonyms, and search in the way of "or". For example, synonyms / near synonyms of "degradation" may include "hydrolysis", "enzymolysis", and "decomposition", so the search formula is changed to: "Chitosan and enzyme catalysis and (degradation or hydrolysis or enzymolysis or decomposition)".

[0051] The preferred solution is: match the experimental name with Tables 1 and 2. If it can match one or several specific principles in Tables 1 and 2, then use the key noun in the description of the matched principle as the second keyword, and perform a secondary search in the results obtained after searching with the first keyword. Perform subsequent steps based on the literature after the second search. If the extracted second keyword already exists in the previous keyword, you can also introduce the antonym of the second keyword and search in the "not" way. For example: Chitosanase-catalyzed chitosan degradation belongs to an enzymatic reaction, and the keyword "enzyme catalysis" is extracted. This keyword already exists in the first search. The search formula in the second search with enzyme catalysis is changed to: "Chitosan and enzyme catalysis and (degradation or hydrolysis or enzymatic hydrolysis or decomposition) not non-enzyme catalysis".

[0052] S2. Use a large language model (such as Llama 3) to perform semantic learning and segmentation on the literature collected in step S1, and extract the parts of the literature related to the experimental steps and details. Infer and summarize the experimental steps according to the experimental logic (i.e., the experimental process framework with a preliminary order), and obtain b experimental processes containing different experimental steps, where b is the number of literature collected in step S1.

[0053] Among them, the experimental logic is deduced from the literature through advanced natural language processing (NLP) technology, and the experimental steps, conditions and parameters are systematically extracted and analyzed. Specifically, it includes: keyword extraction, named entity recognition, dependency parsing and context understanding to ensure accurate information extraction and clear identification of logical relationships.

[0054] S3. Set up a large language model, adjust the principles of semantic classification, and obtain a set of candidate core objects in the experiment.

[0055] The core object is the main treatment object or key target product of the experiment. In biochemical experiments, the core object usually runs through the entire experiment. It is generally reflected in the name of the experiment and will appear frequently during the experiment. For example: in the polymerase chain reaction (PCR) experiment, the core experimental object is the target DNA fragment, the core experimental object in the DNA fragment connection experiment and the plasmid extraction experiment is the plasmid, the core experimental object in the cell counting is the cell, and the core experimental object in the enzyme catalytic reaction is the enzyme. The existence, transfer and change of the core objects in the biochemical experiment in different experimental steps obey the laws of nature. Therefore, as long as the extraction, transfer, transformation and purification processes of these core objects can be analyzed and inferred, the logical essence of the experiment can be inferred more accurately. Therefore, find out the core object in the experiment, and design the experimental process around the change law and change logic of the core object according to the purpose of the experiment, so as to obtain a logically clear, universal and scientifically reasonable experimental process.

[0056] The steps for obtaining the candidate core object set include:

[0057] (3-1) Extract the nouns and noun phrases in all experimental steps in step S2 as the core object set A.

[0058] (3-2) As needed, set the lower limit n of the frequency of nouns and noun phrases in the experimental steps, remove the nouns and noun phrases with a frequency lower than n from the set A, and obtain the streamlined core object set B.

[0059] (3-3) Extract the nouns and noun phrases from the experiment name and compare them with the core object set B. If the core object set B lacks the nouns and noun phrases extracted from the experiment name, add them to the core object set B to obtain the core object set C.

[0060] (3-4) Nouns after verbs in the screening experimental steps are compared with the core object set C. If the noun after the verb is missing in the core object set C, it is added to the core object set C to obtain the core object set D. For example, "purify protein", purification is a verb, protein is a noun, if "protein" is not in the core object set C, it is added.

[0061] The preferred solution is: further comprising steps (3-5), wherein while screening the literature and obtaining the core object set, the extracted core objects are constructed into a core object database.

[0062] The database is a collection of keywords (all nouns) and phrases (all noun phrases) related to various experimental methods. In the process of continuous training and using the design device (software) provided by the present invention to perform experimental design, the newly inferred core objects are continuously added to the database. When the device provided by the present invention is used to perform experimental design next time, after obtaining the experimental steps in step S2 and before step (3-1) of step S3, the database can be used to perform a preliminary screening of the experimental steps. If the experimental steps (preferably the necessary steps in the experimental steps) are screened to contain keywords and phrases in the database, these keywords and phrases are used as candidate core objects and as part of the core object set A, and then the core object extraction steps (3-1)-(3-4) are performed; if they are not screened, the core object extraction steps (3-1)-(3-4) are performed normally. Considering that in the field of biochemistry, the same substance may not be completely unified in the writing method and may have multiple names, by constructing a database and performing a preliminary screening, the core object set can be supplemented based on the previous experimental design data, reducing the possibility of the true core object being extracted incorrectly due to reasons such as name expression.

[0063] More preferably, the data in the database can also be prioritized. After entering the database, keywords and phrases that have not been used within x times (x refers to the number of times the software and device of the present invention are used to design experiments) have their priorities lowered. Even if they are screened from the database during the next experimental design, they will not be added to the candidate core object set (but those that are normally screened according to the above steps (3-1)-(3-4) will still be normally included in the candidate core object set). After the keywords and phrases are downgraded, y experimental designs are performed again. If the downgraded keywords and phrases are screened z times (z < y) within y times, the level of the keywords and phrases is restored. If they are screened in the next experimental design, they will be restored as candidate core objects.

[0064] When the device is continuously used to design experiments, the database data will be continuously improved and gradually trained to form a targeted and perfect database.

[0065] S4. Set up a large language model, extract experimental steps, and build an experimental framework. The specific steps include:

[0066] (4-1) Search for the core object set D in document b. After retrieving the core object in the core object set D, extract the entire sentence in the document and the three closest entire sentences to the sentence to obtain the language set F.

[0067] (4-2) Perform semantic analysis and cross-comparison on the sentences extracted from different documents, merge the same or similar experimental operations for the same core object, delete duplicate descriptions / operations, and for similar operations, retain the operations with the highest frequency of occurrence to obtain the language set G.

[0068] (4-3) Perform logical analysis on each experimental operation in set G and generate an experimental process, including:

[0069] (4-3-1) Arrange the experimental steps according to the order in which the different steps appear in the original literature. If the same experimental step appears in different orders in different literature, the order with the highest frequency of appearance in the literature shall prevail, and a preliminary experimental process A is obtained.

[0070] The preferred solution is: further including step (4-3-2), understanding all the experimental steps of Experimental Process A, determining or revising the sequence of the experimental steps (experimental logic) according to general biological or chemical principles, and obtaining a revised Experimental Process B.

[0071] In this step, compare and confirm which principle in Tables 1 and 2 the experiment corresponding to Experimental Process A belongs to. If the comparison fails, skip this step. After the match is successful, compare and adjust the order of the experimental steps based on the description of the principle and the law of material change to obtain Experimental Process B.

[0072] The preferred solution is: in order to improve the accuracy of principle matching, after initially selecting the matching principle through text matching, the present invention also constructs a new formula to confirm whether the experimental process A is indeed in line with the selected biological principle or chemical principle. The formula is:

[0073]

[0074] Among them: Score: comprehensive score, used to judge the matching degree between principles and processes, with a value of 0-1; you can set the score higher than the required value to determine whether the selected principle is reasonable, and generally set the score higher than 0.75 to determine whether the principle is reasonable. i : The automatically learned weights of the ith experimental step are automatically determined by a large language model such as Llama 3 based on the training data. PM i : The matching degree of the ith experimental step with the selected biological or chemical principle is automatically calculated by a large language model such as Llama 3. Each experimental step may correspond to one biological or chemical principle, so the entire experiment may correspond to multiple different biological or chemical principles. i The comprehensive score calculated by this formula can reflect the overall matching of all selection principles in the experimental process. min PM: The minimum matching value among all experimental steps.max : The maximum value of the matching degree in all experimental steps. mean(w i ): The mean of all step weights. SmoothingFactor: A constant value, automatically calculated by large language models such as the Llama 3 model.

[0075] This formula is used to judge the matching degree of biological or chemical principles. The calculation and adjustment of all parameters are automatically completed by the large language model, and can be adaptively adjusted according to different experimental conditions to improve the intelligence level of the system. At the same time, the adaptive weighting and normalization processing enables the system to dynamically adjust according to the characteristics of the actual experimental steps and the data situation, enhancing the adaptability and flexibility of the system. Moreover, all necessary calculation steps and processing methods are integrated into one formula, which simplifies the matching scoring process and saves computing resources. This formula can handle data with different ranges and distributions, and is particularly suitable for scenarios that require processing a large amount of experimental data and steps, improving the processing capability and scoring accuracy in complex data environments.

[0076] The preferred solution is: further comprising step (4-3-3) verification. Specifically, it comprises step (4-3-3-1), verifying and comparing experimental process B with d% of documents (20%≤d%<100%, the value of d is set as required) randomly selected from documents b to determine whether it is reasonable.

[0077] The verification and comparison method includes: generating experimental processes (named "document processes") for randomly selected documents respectively. Experimental process B is compared with each document process one by one. The Jaccard similarity coefficient is used to quantify the similarity between experimental processes of different documents. First, define two sets that need to be compared, such as experimental process B and a certain document process. Then, calculate the intersection of the two sets, that is, the number of elements that exist in the two sets. Next, calculate the union of the two sets, that is, the total number of all elements (no repetitions). Finally, divide the size of the intersection by the size of the union to get the Jaccard similarity coefficient. If the Jaccard similarity coefficient of the verification result is ≥ e (e can be set as needed, generally set above 0.8), it is judged that there is no deviation in the verification; if the Jaccard similarity coefficient in the verification result is less than e, it is considered that there is a deviation in experimental process B.

[0078] If there is a deviation in the verification, return to step (4-3-2) of this step, revise and generate experimental process B, and verify and compare again, repeating the cycle until there is no deviation in the verification.

[0079] The preferred solution is: further comprising step (4-3-3-2). Step (4-3-3-2) can be performed after completing step (4-3-3-1), or after completing step (4-3-2) step (4-3-3-1) can be skipped, or only (4-3-3-2) can be performed, or (4-3-3-2) can be performed first and then (4-3-3-1).

[0080] Step (4-3-3-2), the present invention constructs a new formula to verify whether the experimental process B is reasonable. The formula is:

[0081]

[0082] Among them: Score: comprehensive score, used to measure the quality and consistency of the experimental process, with a value of 0-1; you can set the score higher than the required value to consider the process reasonable, generally set the score> 0.75 to consider the experimental process reasonable. n: the number of features in the experimental process, for example: select the three features of experimental logic, core object matching, and the number of references, then n=3; you can select features as needed to determine the number of features. i : The input of the i-th feature comes from the feature extraction result of the text by a large language model such as the Llama 3 model. f(x i ):Feature x i The square root function of θ. i :Feature x i The weight of is determined by a large language model such as the Llama3 model through learning and adaptation. α: adjustment factor. A large language model such as the Llama3 model automatically determines the optimal value through an optimization process. Z: normalization factor, the sum of all feature scores, used to standardize the score.

[0083] If the comprehensive score of the verification is lower than the set value, return to step (4-3-2), revise and generate experimental process B, and verify and compare again, repeating the cycle until there is no deviation in the verification.

[0084] The weights and adjustment factors in the formula are automatically learned and determined by the model without manual setting, so that the system can automatically optimize according to the data and adapt to different experimental processes and conditions. Moreover, the formula can integrate multiple features, such as experimental logic, core object matching, reference contribution, etc., and flexibly reflect the multi-dimensional information of different experimental processes; thus effectively solving the defects of the existing technology that the experimental process induction process requires manual intervention or cannot understand the experimental process in multiple dimensions.

[0085] The preferred solution is: further comprising step (4-3-4), performing weight analysis on the experimental process B that has been verified to be free of deviation. Specifically comprising:

[0086] From the three aspects of experimental logic, experimental content and experimental parameters, weight analysis was performed at a ratio of 30%, 50% and 20% respectively, and each experimental step of experimental process B was analyzed and reasoned. Steps with a weight exceeding 90% were set as necessary steps, and the remaining steps were non-essential steps.

[0087] Specifically, the specific parameters and experimental operation details (reflected in adjectives and adverbs) in experimental process B are deleted to obtain a brief version of the experimental process, named machine process. D% of the documents are randomly selected from b documents (20%≤d%<100%, d can be set as needed), and an experimental process (document process) is generated for each of the d% documents. The machine process is compared with these document processes one by one, and the frequency of occurrence of each step in the calculation process in the experimental content is simulated by the random forest calculation method (experimental content, range 0-1, 0 means completely mismatched or completely dissimilar, 1 means completely matched or completely similar. The matching accuracy is set as needed, the same below), as well as the matching rate of the sequence of the upper and lower steps of the step (experimental logic, range 0-1), and the matching rate of the experimental content and experimental logic is obtained respectively. Then the experimental parameters of each step in experimental process B are extracted, and the random forest calculation method is also used to calculate the matching degree of the machine parameters in each step with the parameters in each document process (experimental parameters, range 0-1), and the matching rate of the experimental parameters is obtained. The comprehensive matching value is calculated according to the weight ratio of 30% for experimental logic, 50% for experimental content, and 20% for experimental parameters (the comprehensive matching value of a complete match is 100%, and each weight ratio can be set as needed. This is an example. Generally, the proportion of experimental content is greater than that of experimental logic, which is greater than that of experimental parameters).

[0088] The steps with a comprehensive matching value calculation score ≥ n% are identified as necessary steps, the steps with a score < m% are deleted, and the remaining steps are non-essential steps. m and n are the comprehensive matching values ​​calculated by the above method for a step in experimental process B, which can be set as needed, for example, n is set to 90 and m is set to 30. The necessary steps are built into the core experimental framework, and then according to the matching frequency of each non-essential step in the experimental logic, each non-essential step is inserted between each necessary step at the position with the highest matching rate in the experimental logic, forming an experimental process C with rich experimental details.

[0089] S5: Set up a large language model, combine the necessary steps obtained in step S4, analyze the physical meaning of each core object in the core object set D obtained in step S3, and extract and discover the real core objects of the experiment. Specifically, it includes the following steps:

[0090] (5-1) Through relationship extraction technology, identify the relationship between the candidate core objects in the core object set D, as well as the relationship between the candidate core objects and other nouns / noun phrases in the experimental steps that are not in the core object set D. Through natural language understanding, identify the purpose of the experiment. Based on the relationship identification and the experimental purpose, select the experimental object as the core object, and add, delete and revise the set D.

[0091] For example, "add HCl to adjust the pH of the protein solution". HCl and protein are both nouns. Through relationship identification and experimental purpose analysis, protein is the treatment object, and the experimental purpose is to adjust the pH of the protein solution. Therefore, the target is protein. If set D includes both HCl and protein, then HCl is deleted from set D and protein is retained; if set D only contains protein but not HCl, no treatment is performed; if set D only contains HCl but not protein, then protein is added and HCl is deleted.

[0092] In a single experiment, there is likely to be more than one core object, so this step may obtain a single core object E1 or a core object set E2 consisting of multiple core objects.

[0093] The preferred solution is: in this step, the key core objects and step core objects are extracted respectively. The key core objects of the whole experiment are inferred according to the overall experimental purpose, and the core objects of each step are inferred according to the specific experimental purpose of each step.

[0094] For example: a method for expressing purified protein. Through the overall analysis of the experimental name and the experimental process finally obtained in step S4, the overall experimental purpose is to obtain protein, so protein is determined to be the key core object of the experiment. The experiment includes multiple steps, among which the purpose of step 1 is to extract DNA, so DNA is selected as the core object of step 1 (RNA and protein are impurities in this step and are deleted from the core object set of step 1); in step 2, the purpose is to obtain protein, so protein is the core object of step 2 (DNA and RNA are impurities in this step and are deleted from the core object set of step 2).

[0095] The key core objects and the core objects of each step can be set up separately or put in the same set E2. When put in the same set E2, each substance in E2 is marked as belonging to the key core object or the core object of a specific step. For the sake of ease of explanation, the following text uniformly uses the method of marking in the same set E2.

[0096] (5-2) Understand and identify the meaning of each substance in the core object E1 or set E2, and classify the core objects.

[0097] The categories include: biological samples, chemical reagents / solutions, and others. Among them, biological samples include cells (primary cells, cell lines, etc.), microorganisms (bacteria, fungi, viruses, etc.), tissues and biomacromolecules (DNA, RNA, proteins, etc.), and others. Chemical reagents / solutions include inorganic reagents, organic reagents, biological / biochemical reagents, indicators, and others.

[0098] (5-3) Analyze the meaning and context of each necessary step through natural language understanding. Based on the contextual meaning of the necessary steps and the classification and meaning analysis of the core objects in step (5-2), determine whether the core objects selected in step (5-1) are reasonable.

[0099] For example, the purpose of a certain step is to adjust the pH, and it is judged that chemical reagents should be added in this step. However, the substance actually added in this step is protein, which belongs to the category of biological sample substances, and the step is judged to be unreasonable.

[0100] According to the operation of this step, the overall experimental purpose and all necessary steps are judged one by one. If the overall core object is judged to be consistent with the overall experimental purpose in this step, and there are no unreasonable items in all necessary steps, then continue to the subsequent links. If it is judged that there is at least one unreasonable item in this step, the verification fails, and return to step S4, replace the core object set D in step S4 with the core object E1 or set E2, and re-sort the steps of the experimental process. Then continue to step S5.

[0101] Repeat S4 and S5 until there are no unreasonable items in the overall experimental purpose and all necessary steps.

[0102] S6. According to the experimental steps sorted out in step S4 and / or the experimental purpose identified in step S5, determine the type of change (physical change, chemical change or biological change) of the core object determined in step S5 during the experiment. According to the experimental purpose, experimental steps, and the common knowledge database, sort out the changes and order of the substances when the core object undergoes physical changes, chemical changes or biological changes.

[0103] Based on the order of the physical, chemical or biological changes of the core object determined in step S5, experimental process C is verified in combination with the change process of the core object undergoing physical, chemical or biological changes in b documents and online searches.

[0104] During the verification process, if one or several experimental steps are found to be inconsistent with the change law of the core object, the verification will be judged as unqualified. This may be caused by errors in the content of the retrieved literature, inconsistent expressions of biochemical terms, inaccurate term identification / extraction, etc.

[0105] For example: a method of using chitosanase to catalyze the degradation of chitosan. Step S5 determines that "chitosan" is one of the core objects, and the enzymatic hydrolysis of chitosan is a biological change. In the process of gradual enzymatic hydrolysis of chitosan, the degree of polymerization gradually decreases, and it is gradually enzymatically hydrolyzed into lower molecular chitosan, and even becomes chitosan oligosaccharides. If the verification finds that the change of chitosan in step S4 does not conform to this rule, the verification is judged to be unqualified.

[0106] For experimental processes that fail the verification, if the order of the overall process steps or the order of operations within a certain step is adjusted to meet the change law of the corresponding core object, then the corresponding order is adjusted to obtain the optimized experimental process D, and then the verification of this step is re-performed. After the verification is qualified, proceed to step S7.

[0107] For example, in the experimental process C of "Method of using chitosanase to catalyze the degradation of chitosan", chitosan was first hydrolyzed into chitosan oligosaccharides with a degree of polymerization of 10, and then continued to hydrolyze to become chitosan with a high degree of polymerization. The overall experimental steps were adjusted according to the order in which the degree of polymerization of chitosan gradually changed from large to small, and the optimized experimental process was obtained, and then re-verified.

[0108] For the experimental process that fails the verification, if it is impossible to adjust the order of the process steps to conform to the change law of the core object, then return to step (3-4) of step S3.

[0109] It should be noted that: after the screening and review in the aforementioned steps, the probability of selecting the wrong core object is small, and it is generally an error in the experimental process. However, if this step fails the verification and directly returns to step S4 to re-sort the experimental process steps, the amount of calculation is large and time-consuming. Therefore, using the core object as the index, return to step S3, and use the core object F corresponding to the experimental process that failed the verification as the keyword, and re-perform a semantic association search for the core object F in the experimental process of document b to obtain the search set G. Then delete the core object F in the core object set D of step S3, and revise the core object set D according to the situation of the search set G according to the following steps:

[0110] If the search set G contains the core object F, the frequency of occurrence of the core object F in the experimental process of the b article is analyzed. If the frequency of occurrence of the core object F is higher than the set value (the frequency is set as needed, and the frequency is generally required to be higher than 50%), the search set G is added to the core object set D in step S3 while retaining the core object F, and the subsequent steps are repeated until the verification is qualified. If the frequency of occurrence of the core object F is less than the set value, the core object F is deleted from the search set G, and the revised search set G is added to the core object set D as a whole, and the subsequent steps are repeated until the verification is qualified.

[0111] If the search set G does not include the core object F, the entire search set G is added to the core object set D, and the subsequent steps are repeated until the verification is qualified.

[0112] A more preferred solution is: if the core object F is the core substance of the step, you only need to re-perform steps S3 to S6 for the steps corresponding to the core object F, and fix the overall experimental process and other steps not involving the core object F. After the steps corresponding to the core object F are revised and verified to be qualified, replace the corresponding steps in the overall experimental process.

[0113] S7, extract all experimental parameters (such as temperature, concentration, time, gradient, etc.) that appear in the experimental process of step S6 to form an experimental parameter range. Specifically including:

[0114] (7-1) extracting all experimental parameters appearing in the experimental process of step S6;

[0115] (7-2) The extracted experimental parameters are counted and summarized through the large language model, and the parameters with a frequency of occurrence above f are selected (f is the frequency of occurrence of the parameter in the experimental process of article b, which can be set as needed) to obtain the parameters required for the experiment. Then, the values ​​of these parameters with a frequency of occurrence above g (g is the frequency of occurrence of the parameter value in the experimental process of article b, which can be set as needed) are selected to obtain the parameter selection range.

[0116] The parameters and parameter ranges are added into step S6 to form a complete and operable experimental process.

[0117] The preferred solution includes: (7-3) after obtaining the parameters and parameter selection ranges and before adding the parameters and selection ranges to step S6, the key parameters are verified using the core object to determine whether the parameters (and ranges) conform to the variation law of the core object and the purpose of the experiment. If they do not conform to the variation law or the purpose of the experiment, the verification fails.

[0118] For example, in a method for extracting humic acid, the key parameter is the pH of the extraction environment, and the key parameter selection range is pH=1-2. However, it is well known in the art that humic acid is insoluble in a strong acidic environment. Therefore, the purpose of extracting humic acid cannot be achieved in this environment. Therefore, the parameter value range verification fails.

[0119] If the verification fails, you can choose any of the following methods:

[0120] (7-3-1) Parameters (or ranges) that fail the verification will be retained, highlighted and prompted, and the user will be reminded to review the parameters when outputting the plan.

[0121] (7-3-2) Return to step (7-2) of step S7, extract all parameter conditions and ranges that appear in the corresponding experiment of the experimental parameters (or ranges) that failed the verification in the common knowledge database, select the experimental parameters with the highest frequency of occurrence, and replace the experimental parameters that failed the verification. Then repeat step (7-3) for verification. If the most frequently occurring parameter range is the experimental parameter that failed the verification, or after replacing the most frequently occurring parameter range, the verification still fails, or the experimental parameters (or ranges) that meet the requirements are not retrieved in the common knowledge database, the parameters that failed the verification are retained, highlighted and prompted, and the user is reminded to focus on the parameter when outputting the plan.

[0122] S8. Iterate steps S6 and S7 until there is no significant change in the experimental process and experimental parameters (repetition rate > 80%), and complete the experimental process design or optimization.

[0123] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Unless otherwise specified, the technical means used in the embodiments are conventional means well known to those skilled in the art. The data obtained are all average values ​​obtained after at least 3 repetitions, and all the data obtained in each repetition are valid data.

[0124] Example

[0125] Using the steps provided by the present invention, a computer program product (software) for designing and optimizing biochemical experiments is designed through the large language model Llama 3. The initial interface of the software is as follows: Figure 2 shown.

[0126] The software is used to design an experimental method for amplifying the DNA encoding the Monascus aminotransferase QT5. The method specifically comprises the following steps:

[0127] S1. Translate the experiment name into English “Amplification of transaminase QT5 encodingDNA from Monascus spp.” to facilitate subsequent retrieval.

[0128] The extracted search keywords were: "coding DNA" and "DNA amplification"; the search formula was: "codingDNA" and "DNA amplification". By searching public scientific papers and the largest scientific literature database (Scopus database), a total of 2,270 relevant documents were collected.

[0129] Monascus involved in this experiment is a type of eukaryotic organism, and the DNA sequence on its genome may contain introns (non-coding sequences). According to Table 1, eukaryotes was used as the second keyword, and a secondary search was performed among the 2270 retrieved documents, resulting in 1150 documents. The subsequent steps were based on the documents after the second search.

[0130] S2. Perform semantic learning and segmentation on the literature collected in step S1 through Llama 3, and separate the parts of the literature related to the experimental steps and details.

[0131] Specifically, we set Batch_size to control the balance between the speed and accuracy of Llama 3, and prioritize the speed within the limits of the server's video memory. We derived the experimental logic (basic experimental framework) from the literature through keyword extraction, named entity recognition, dependency parsing, and context understanding: grind Monascus → extract total RNA → reverse transcribe into cDNA → PCR reaction → design transaminase QT5 primers → obtain transaminase QT5 encoding DNA.

[0132] By reasoning and summarizing the experimental steps according to the experimental logic, we obtained 1150 experimental processes with different experimental steps, which mainly include 6 steps.

[0133] S3. Set up a large language model, adjust the principles of semantic classification, and obtain a set of candidate core objects in the experiment.

[0134] (3-1) Extract nouns and noun phrases from all experimental steps in 1150 papers as the core object set A.

[0135] (3-2) As needed, the lower limit of the frequency of nouns and noun phrases in the experimental steps is set to 10%, and nouns and noun phrases with an appearance frequency of ≤10% in the experimental steps are removed from set A to obtain a streamlined core object set B.

[0136] (3-3) Extract the nouns and noun phrases in the experiment names, specifically: Amplification and coding DNA, and compare them with the core object set B. The core object set B already has these two nouns and does not need to be changed, thus obtaining the core object set C.

[0137] (3-4) Screen the text content of the experimental methods section in the literature, compare the nouns after the verbs with the core object set C, and find new nouns "cDNA", "mRNA" and "primer". These nouns are missing in the core object set C, so they are added to the core object set C to obtain the core object set D.

[0138] S4, set Llama 3, compare the differences in the experimental steps in the six experimental processes obtained in step S3, and sort out the experimental steps according to the core object set D obtained in S2. It includes the following steps:

[0139] (4-1) Search for the core object set D in 1150 documents. After the core object in the core object set D is retrieved, the entire sentence of the core object in the document and the three closest entire sentences to the sentence are extracted together to obtain the language set F.

[0140] (4-2) Perform semantic analysis and cross-comparison on the sentences extracted from different documents, merge the same or similar experimental operations for the same core object, delete duplicate descriptions / operations, and for similar operations, retain the operations with the highest frequency of occurrence to obtain the language set G.

[0141] (4-3) Perform logical analysis on each experimental operation in set G and generate the experimental process:

[0142] (4-3-1) Arrange the experimental steps according to the order in which the different steps appear in the original literature. If the same experimental step appears in different orders in different literature, the order with the highest frequency of appearance in the literature shall prevail, and a preliminary experimental process A is obtained.

[0143] (4-3-2) Understand all the experimental steps of Experimental Process A, determine and revise the sequence of experimental steps (experimental logic) based on the three experimental principles of general microbial mRNA extraction, mRNA reverse transcription into cDNA, and PCR experiment, and obtain a revised Experimental Process B. The specific generated Experimental Process B is as follows Figure 3 At this time, the comprehensive score is 0.83, indicating that the experimental principles and the designed experimental process are well matched.

[0144] (4-3-3) 230 documents were randomly selected from 1,150 documents, and the machine process was compared with the experimental contents in these documents one by one. The frequency of occurrence of the experimental content in each step was simulated and calculated by the random forest calculation method: the frequencies of occurrence of experimental steps 1-6 were 0.98, 0.95, 0.94, 0.98, 0.99 and 1.0 respectively. For the corresponding experimental steps, the matching frequencies of the order in the previous and next steps were 1.0, 1.0, 1.0, 1.0, 1.0 and 1.0 respectively. The calculation method of this order matching frequency is: let the verified step be A, the previous step of A be B, and the next step of A be C. If the order is BAC, the order of the previous and next steps is correct, and the frequency is 1.0. For each wrong order, subtract 0.5. Then, the random forest calculation method was used to calculate the matching degrees between the experimental parameters calculated by artificial intelligence and the parameters reported in the literature, which were 0.84, 0.92, 0.87, 0.90, 0.91 and 0.88. The calculation method here is: the total number of parameters is set to n, the number of matching parameters is set to m, and the matching degree is m / n.

[0145] A weighted analysis was performed based on the experimental logic, experimental content, and experimental parameters with a ratio of 30%, 50%, and 20% respectively. The scores of the six experimental steps were all over 0.9 and were all identified as necessary steps. There were no unnecessary steps or steps that needed to be deleted.

[0146] The necessary steps are built into a core experimental framework, and then according to the experimental logic deduced from step S2, non-essential steps are inserted between the necessary steps to form an experimental process framework with rich experimental details. Specifically, between "extracting total RNA" and "reverse transcribing to cDNA", artificial intelligence inserts a total RNA purification step to avoid other impurities (such as proteins, DNA, and polysaccharides, etc.) from interfering with the subsequent step of "reverse transcription to cDNA". The specific experimental content is: precipitate the RNA by adding isopropanol or ethanol, and then wash with 75% ethanol to remove residual salts and other impurities 2-3 times. ‌‌The washed RNA is naturally dried in a fume hood, and then dissolved with an appropriate amount of DEPC water. ‌‌The concentration and purity of RNA are determined using a UV spectrophotometer, and the integrity of RNA is detected by electrophoresis. ‌

[0147] S5, set Llama 3, combine the necessary steps obtained in step S4, analyze the physical meaning of each core object in the core object set D obtained in step S3, extract and discover the real core objects of the experiment. Specifically include the following steps:

[0148] (5-1) Use Llama 3 to identify the relationship between candidate experimental objects and other experimental objects through relationship extraction technology. Identify the experimental purpose of the experimental steps through natural language understanding. Select the target object as the core object according to the experimental purpose. Extract the key core objects and step core objects respectively.

[0149] The key core objects identified and extracted include: cDNA and DNA.

[0150] The core objects of the identification and extraction steps include: mRNA (step 2), cDNA (step 3) and DNA (step 4).

[0151] The above core objects together form a core object set E2, and the core objects in the set E2 are marked as "key" and "step" respectively.

[0152] (5-2) Understand and identify the meaning of each substance in the core object set E2, and classify the core objects. The specific classification is as follows: The core objects of this experiment are all genetic information substances, excluding proteins, lipids, inorganic substances and organic compounds.

[0153] (5-3) Analyze the meaning and context of each necessary step through natural language understanding. Based on the contextual meaning of the necessary steps and the classification and meaning analysis of the core objects in step (5-2), determine whether it is reasonable for the core objects selected in step (5-1) to exist in this step. The judgment is: the overall experimental purpose and all necessary steps are judged one by one, and there are no unreasonable items in all necessary steps.

[0154] S6. According to the experimental steps sorted out in step S4, determine whether the core object determined in step S5 is physically transferred and biologically transformed in the experiment. According to the experimental purpose and experimental steps, combined with the common knowledge retrieved, sort out the changes in the core object when it changes, and the order of changes is as follows: physically, the total RNA is dissolved from the cell (nucleus), and then enriched and extracted from the cell by an organic solvent, and physically separated from substances such as proteins through the purification step; biologically, the mRNA in the total RNA is catalyzed by reverse transcriptase to form cDNA with a consistent base sequence, and the cDNA is then catalyzed by DNA polymerase to form a double-stranded DNA molecule with a consistent base sequence.

[0155] Based on the change order of the core objects determined in step S5, the experimental process of step S4 is adjusted and optimized to obtain the optimized experimental process. Specifically, in the correct experimental logic, PCR primers should be designed first, and then the primers should be used to implement PCR. However, the experimental logic given before is the opposite here ( Figure 4 ), so it is necessary to swap the experimental logic and corresponding experimental content of these two steps.

[0156] S7, extracting all experimental parameters appearing in the experimental process of step S6 to form an experimental parameter range. Specifically including:

[0157] (7-1) Extract all experimental parameters that appear in the experimental process of step S6.

[0158] (7-2) The extracted experimental parameters are counted and summarized through the large language model, and the substrate step time in step 3 is selected as the key parameter, the primer length and Tm value in step 4 are selected as the key parameters, and the number of cycles in step 5 is selected as the key parameter.

[0159] (7-3) Use core objects to verify key parameters: reversely summarize the key parameters into several experimental contents, and then match the change patterns of the core objects according to the experimental contents to determine whether the key parameters of this experiment match the core objects correctly.

[0160] The reference and parameter ranges that have passed the verification are added to the experimental process of step S6.

[0161] S8. Iterate steps S6 and S7 until the experimental process and experimental parameters do not change significantly (repetition rate > 80%), and complete the experimental process design. Figure 5 shown.

[0162] After expert review and experimental verification of this experimental process, the coding DNA sequence of transaminase QT5 was successfully obtained, proving that the obtained experimental process is scientific and reliable.

[0163] The embodiments described above are only descriptions of the preferred modes of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations, modifications, and substitutions made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for designing or optimizing biological or chemical experiments, characterized in that: The steps include: S1. Search and collect literature containing the protocols and steps of the described experiment; S2, using a large language model to perform semantic learning and segmentation on the documents collected in step S1, and extracting the parts of the documents related to the experimental steps and details; S3, setting a large language model, adjusting the principle of semantic classification, and obtaining a set of candidate core objects in the experiment; the core objects are the main processing objects or key target products of the experiment; S4. Set up a large language model, extract experimental steps, and construct an experimental process; S5, setting a large language model, combining the experimental process obtained in step S4, analyzing each core object in the core object set obtained in step S3, and extracting the real core objects of the experiment according to the experimental purpose of the experiment; S6, sorting out the changing rules of the core objects when the real core objects change, and verifying and optimizing the experimental process of step S4 based on the changing rules; S7. Extract all experimental parameters that appear in the experimental process adjusted in step S6, count and summarize the optional ranges of the experimental parameters in the literature retrieved in step S1, and add the parameters and parameter ranges into step S6 to form a complete and operational experimental process.

2. The method according to claim 1, characterized in that: Also includes: S8. Iterate steps S6 and S7 until there is no significant change in the experimental process and experimental parameters, and the experimental design or optimization is completed.

3. The method according to claim 1 or 2, characterized in that: Step S3: The method for obtaining a set of candidate core objects in the experiment comprises the following steps: (3-1) Extract the nouns and noun phrases in all experimental steps in step S2 as the core object set A; (3-2) Set the lower limit n of the frequency of nouns and noun phrases in the experimental steps, remove the nouns and noun phrases with a frequency lower than n from the set A, and obtain the streamlined core object set B; (3-3) Extract the nouns and noun phrases from the experiment name and compare them with the core object set B. If the core object set B lacks the nouns and noun phrases extracted from the experiment name, add them to the core object set B to obtain the core object set C; (3-4) Nouns following the verbs in the screening experimental steps are compared with the core object set C. If the noun following the verb is missing in the core object set C, it is added to the core object set C to obtain the core object set D.

4. The method according to claim 3, characterized in that: Step S4: The method for obtaining the experimental process comprises the following steps: (4-1) searching for a core object set D in the documents retrieved in step S1. After the core object in the core object set D is retrieved, extracting the entire sentence of the core object in the document and the three entire sentences closest to the sentence, thereby obtaining a language set F. (4-2) Perform semantic analysis and cross-comparison on the sentences extracted from different documents, merge the same or similar experimental operations for the same core object, delete duplicate descriptions / operations, and for similar operations, retain the operations with the highest frequency, and obtain the language set G; (4-3) Perform logical analysis on each experimental operation in set G and generate the experimental process.

5. The method according to claim 4, characterized in that: A method for generating an experimental process by performing logical analysis on each experimental operation in set G includes the following steps: arranging the experimental steps according to the order in which different steps appear in the original document; if the order in which the same experimental step appears in different documents is inconsistent, the order in which the step appears most frequently in the document shall prevail, thereby obtaining experimental process A.

6. The method according to claim 5, characterized in that: After obtaining the experimental process A, the sequence of the experimental steps of the experimental process A is revised according to the biological principle or chemical principle corresponding to the experiment to obtain the experimental process B; Or; after obtaining experimental process A, match the corresponding biological principle or chemical principle according to the text of experimental process A, and confirm whether experimental process A is indeed in line with the biological principle or chemical principle according to the following formula: ， Among them: Score: comprehensive score, used to judge the matching degree between principles and processes, with a value of 0-1; i : The automatically learned weight of the ith experimental step; PM i PM: the degree of compatibility of the ith experimental step with the selected biological or chemical principle; min : The minimum matching degree among all experimental steps; PM max : The maximum value of the matching degree in all experimental steps; mean(w i ):The mean of all step weights; SmoothingFactor: constant value; If the calculated value of Score is higher than the set value, it is judged that the experimental process A complies with the corresponding biological principle or chemical principle. The sequence of the experimental steps of the experimental process A is revised according to the principle to obtain the experimental process B.

7. The method according to claim 3, characterized in that: Step S5: The method of extracting the real core object of the experiment comprises the following steps: (5-1) Using the relationship extraction technology, identify the relationship between the candidate core objects in the core object set D, as well as the relationship between the candidate core objects and other nouns / noun phrases in the experimental steps that are not in the core object set D. Through natural language understanding, identify the purpose of the experiment. Based on the relationship identification and the experimental purpose, select the experimental object as the core object, and add, delete and revise the set D to obtain a single core object E1 or a core object set E2 consisting of multiple core objects. (5-2) Understand and identify the meaning of each substance in the core object E1 or set E2, and classify the core objects; (5-3) Analyze the meaning and context of each experimental step through natural language understanding, and combine the classification and meaning analysis of the core objects in step (5-2) to determine whether the core objects selected in step (5-1) are reasonable; (5-4) Judge the overall experimental purpose and experimental steps one by one; if it is judged that there are no unreasonable items in this step, continue to the subsequent steps; if it is judged that there is at least one unreasonable item in this step, the verification is unqualified, return to step S4, replace the core object set D in step S4 with the core object E1 or set E2, re-arrange the steps of the experimental process, and then continue to step S5; repeat S4 and S5 until there are no unreasonable items.

8. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images