Plant tissue culture scheme generation system and method based on multi-source knowledge base and AI

Abstract

The invention discloses a plant tissue culture scheme generation system and method based on a multi-source knowledge base and AI. The system comprises: (1) a data acquisition unit used for gathering related data; (2) the knowledge base construction unit is used for carrying out cleaning fusion and ontology modeling on multi-source heterogeneous data; (3) an AI model training unit constructs a multi-dimensional parameter prediction function group through deep learning; (4) the scheme generation and evaluation module outputs optimal culture conditions of each stage according to variety characteristics and performs effect scoring; and (5) a feedback updating module which writes an experiment result and a score back to the knowledge base and triggers model incremental learning and scheme iteration. According to the method, data access and knowledge base updating are completed firstly, then a trained model is called to predict a culture medium formula and culture conditions, and self-adaptive optimization is carried out according to a feedback result. Compared with a traditional trial and error mode depending on experience, the formula optimization period is remarkably shortened, the tissue culture success rate of a new variety is increased, and the method is suitable for plant industrialized production and breeding research and development.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of plant tissue culture and artificial intelligence technology, in particular to a plant tissue culture scheme generation system and method based on a multi-source knowledge base and AI. BACKGROUND

[0002] Plant tissue culture technology, as a core technology of modern biological engineering, plays an irreplaceable role in germplasm preservation, detoxification seedling, new variety breeding, etc. However, traditional tissue culture technology has some prominent defects in practical application: first, it is highly dependent on experience, and the existing technical scheme is seriously dependent on expert experience, and there is a lack of quantitative standard for key parameters such as culture medium component ratio and culture condition, resulting in large differences in culture efficiency obtained by different operators; second, multi-factor coupling analysis is missing, and the tissue culture effect is influenced by multiple parameters such as explant physiological state, culture medium composition, and culture environment, and it is difficult to establish an effective parameter correlation model by traditional test method, and the single formula optimization period is as long as 3-6 months, and the production cost is high. Although artificial intelligence technology has been gradually applied in the field of agriculture, the integration of knowledge is insufficient, and there is a lack of structured knowledge base specially designed for tissue culture scene, resulting in low standardization of existing tissue culture system, unstable output rate of high-quality seedlings, and other industrialization obstacles, which is in sharp contradiction with the parameter control precision required by modern agricultural factory production.

[0003] Traditional plant tissue culture technology is highly dependent on expert experience and repeated trial and error, especially in the scene of new variety breeding or large-scale production, the culture medium formula and key parameters (hormone, carbon source, light, temperature, etc.) often need to be determined through dozens to hundreds of experiments, resulting in long research and development cycle, high cost, and large fluctuation of success rate. The existing optimization scheme based on rules or single data source is difficult to fully utilize the growing tissue culture literature, historical experiments and production monitoring data. SUMMARY

[0004] The purpose of the present application is to provide a plant tissue culture scheme generation system and method based on a multi-source knowledge base and AI, to solve the problems of strong experience dependence, long culture scheme optimization period, and high trial and error cost in traditional plant tissue culture method.

[0005] In order to achieve the above-mentioned purpose, the technical scheme adopted by the present application is as follows:

[0006] The plant tissue culture scheme generation method based on a multi-source knowledge base and AI comprises the following steps:

[0007] 1) Data collection: collecting plant tissue culture related data, and performing denoising processing, field standardization processing and metadata tagging processing on the collected data to form a structured raw data set;

[0008] 2) Knowledge base construction: After data cleaning, entity recognition, relationship extraction and semantic mapping of the data in the original data set, a multi-source heterogeneous plant tissue culture knowledge base with a clear hierarchical structure is formed;

[0009] 3) AI model training: Call the deep learning model to use the standardized data in the knowledge base as training corpus for model training to obtain a multi-dimensional function group for predicting culture parameters. The input data during model training includes One-hot encoding of plant families and genera, numerical characteristics of explant types and historical parameter information. The culture parameters predicted by the model include key parameters such as disinfection methods, culture medium formula and culture conditions in each culture stage;

[0010] 4) Scheme generation and evaluation: According to the target plant variety, plant characteristics and culture target input by the user, the multi-dimensional function group trained in step 3) is used to predict the initial culture scheme parameters. Then, the Bayesian optimization algorithm combined with the genetic algorithm is used to optimize and search the prediction results, and finally the complete culture scheme is output, and the comprehensive score and scheme confidence are output;

[0011] 5) Feedback update: According to the culture scheme output in step 4), the actual experiment is carried out, the experimental process data is recorded, the comprehensive score is calculated, and the knowledge base is written back; if the comprehensive score or the scheme confidence is lower than the preset threshold, or the amount of newly added data in the knowledge base reaches the preset threshold, the AI model in step 3) is automatically triggered for fine tuning and incremental learning update, and enters the next round of scheme generation iteration.

[0012] Further, the data collected in step 1) includes historical experimental record data, real-time monitoring data in the production process, patent literature data and non-patent literature data, etc.

[0013] Further, the knowledge base in step 2) includes structured data and unstructured data. The structured data refers to plant classification, organization, fixed field name, clear data type / unit, and can be directly retrieved and calculated by program, which covers high-frequency numerical values and fields such as hormone concentration, care intensity and culture stage, as well as semantic connections for expressing multi-relationship networks such as "variety-stage-parameter-result". The unstructured data includes culture medium formula, literature abstract, picture, etc. The structured data is stored in a database, and the unstructured data is stored in a vector database after semantic vector coding. The structured data and unstructured data are one-to-one corresponding through unique relationship identifier (ID).

[0014] Further, the hierarchical structure in step 2) includes plant varieties, explant types, culture stages, parameter types, experimental results, etc.

[0015] Further, the deep learning model in step 3) is a multi-task prediction model based on the Transformer architecture.

[0016] Further, the model training of step 3) adopts a transfer learning and incremental learning mode, and when the knowledge base data is newly added to reach a threshold, model fine-tuning and parameter updating are automatically triggered.

[0017] Further, the culture conditions of step 3) include hormone concentration, carbon source concentration, light intensity, culture period, etc.

[0018] Further, the proxy model of step 4) is preferably a deep Gaussian process (Deep Gaussian Process, DGP), and an expected improvement (EI) function is used to select the next round of parameter combinations to be evaluated, and a genetic algorithm is used to process the fast search of discrete parameters such as culture bottle types.

[0019] Further, the complete culture scheme of step 4) includes disinfection methods, induction culture stage parameters, proliferation culture stage parameters, rooting culture stage parameters, etc.

[0020] Further, the experimental process data of step 5) include survival rate, induction rate, proliferation fold, rooting rate, etc.

[0021] The application also provides a plant tissue culture scheme generation system based on multi-source knowledge base and AI, the system comprises:

[0022] (1) a data acquisition unit for acquiring and storing relevant raw data from historical experimental record data, real-time monitoring data of production process, patent literature data and non-patent literature data;

[0023] (2) a knowledge base construction unit for data cleaning, entity recognition, relationship extraction, semantic mapping and ontology modeling of the raw data, and construction of a multi-source heterogeneous knowledge base for the field of plant tissue culture;

[0024] (3) an AI model training unit for training a deep learning model based on the knowledge base, obtaining a multi-dimensional function group for predicting culture parameters, and regularly updating according to newly added data;

[0025] (4) a scheme generation and evaluation module for receiving input characteristics such as target plant varieties and explant types, generating a complete culture scheme according to the multi-dimensional function group, and outputting a comprehensive score and scheme confidence;

[0026] (5) a feedback updating module for receiving actual experimental results, writing experimental feedback data back to the knowledge base, and triggering the AI model training unit for parameter fine-tuning and model iteration.

[0027] The application has the following beneficial effects:

[0028] 1. The method of the present application first completes data access and knowledge base updating, then calls the trained model to predict the medium formula and culture conditions, and optimizes them adaptively according to the feedback results. Compared with the traditional trial-and-error method relying on experience, the present application significantly shortens the formula optimization period, improves the success rate of new variety tissue culture, and is suitable for plant factory production and breeding research and development.

[0029] 2. The present application realizes rapid prediction, automatic evaluation and continuous iterative optimization of medium formula and culture conditions by unified integration of multi-source heterogeneous data, construction of knowledge base and the aid of deep learning and Bayesian optimization algorithms, thereby significantly shortening the development period of new variety tissue culture scheme and improving the success rate.

[0030] 3. The method of the present application realizes the automatic and intelligent generation and continuous optimization of plant tissue culture scheme, which can significantly reduce the number of experiments, shorten the optimization period of the scheme, improve the survival rate, effectively reduce the test cost, improve the breeding efficiency, meet the actual needs of factory production and plant breeding research and development. BRIEF DESCRIPTION OF DRAWINGS

[0031] Figure 1 The figure is a flowchart of the method of the present application. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly, completely and in detail in combination with the embodiments below, but the implementation of the present application is not limited to this embodiment. All other embodiments obtained by those skilled in the art without creative labor on the basis of the embodiments of the present application also belong to the protection scope of the present application.

[0033] As shown in Figure 1 The present embodiment provides a plant tissue culture scheme generation system and method based on multi-source knowledge base and AI. The present example takes Ficus hirta Vahl of Moraceae Ficus as the target plant. The plant tissue culture scheme generation system and method based on multi-source knowledge base and AI are used to realize the tissue culture scheme formulation of Ficus hirta Vahl, including the following steps:

[0034] The plant tissue culture scheme generation method based on multi-source knowledge base and AI includes the following steps:

[0035] S1: Data collection: collect relevant historical experimental record data, production process real-time monitoring data, patent literature data and non-patent literature data of plant tissue culture (Moraceae Ficus plants such as Ficus microcarpa, Ficus pellucida, Ficus virens, Ficus carica, Ficus taiwaniana, Ficus elastica etc.), and perform denoising processing, field standardization processing and metadata tagging processing on the collected data to form a structured raw data set;

[0036] S2: Knowledge base construction: After data cleaning, entity recognition, relationship extraction and semantic mapping of the data of the original data set, a multi-source heterogeneous plant tissue culture knowledge base with a clear hierarchical structure (plant variety, explant type, culture stage, parameter type, experimental result, etc.) is formed; specifically, the knowledge base includes structured data and unstructured data, the structured data refers to plant classification, organization, with fixed field name, clear data type / unit, can be directly retrieved and calculated by program, covers high frequency numerical value and field such as hormone concentration, care intensity, culture stage, and semantic connection for expressing "variety-stage-parameter-result" and other multi-relationship networks, the unstructured data includes culture medium formula, literature abstract, picture, etc., the structured data is stored by using a database, the unstructured data is stored in a vector database after being encoded by a semantic vector, and the structured data and the unstructured data are one-to-one corresponding through a unique relationship identifier (ID).

[0037] S3: AI model training: calling a deep learning model to train the model by taking the data standardized in the knowledge base as training corpus to obtain a multi-dimensional function group for predicting culture parameters, preferably training a multi-task prediction model based on a Transformer architecture, the input data during model training includes One-hot encoding of plant families, numerical characteristics of explant types and historical parameter information, and the culture parameters predicted by the model include key parameters such as sterilization methods, culture medium formula, culture conditions (hormone concentration, carbon source concentration, light intensity, culture period, etc.) of each culture stage;

[0038] In addition, the model training adopts a transfer learning and incremental learning mode, and when the knowledge base data is newly added to reach a threshold value, model fine-tuning and parameter updating are automatically triggered;

[0039] S4: scheme generation and evaluation: according to the target plant variety (rough leaf fig), plant characteristics and culture target (sterilization survival rate 50%, induction rate 60%, proliferation multiple 3.0, rooting rate 90%, while taking into account production cost), the multi-dimensional function group obtained by training in step S3 is used to predict the initial culture scheme parameters, and then a Bayesian optimization algorithm combined with a genetic algorithm is used to optimize and search the prediction results, and finally a complete culture scheme including sterilization methods, induction culture stage parameters, proliferation culture stage parameters and rooting culture stage parameters is output, and a comprehensive score and scheme confidence are output;

[0040] Specifically, the proxy model of Bayesian optimization is preferably a deep Gaussian process (Deep Gaussian Process, DGP), and an expected improvement (EI) function is used to select the next round of parameter combinations to be evaluated, and the genetic algorithm is used to process the fast search of discrete parameters such as culture bottle types;

[0041] S5: Feedback update: according to the culture scheme output in step S4, actual experiments are carried out, and experimental process data (survival rate, induction rate, proliferation fold, rooting rate, etc.) are recorded, and the comprehensive score is calculated to write back to the knowledge base; if the comprehensive score or the scheme confidence is lower than the preset threshold, or the amount of newly added data in the knowledge base reaches the preset threshold, the AI model in step S3 is automatically triggered for fine tuning and incremental learning update, and the next round of scheme generation iteration is entered.

[0042] The data in step S1 described in the embodiment include, but are not limited to, the classification of plants, the type of explants, the pretreatment of selected materials, the disinfection method of explants, the induction medium formula, the induction conditions, the induction rate, the growth condition of adventitious buds, the proliferation medium formula, the proliferation fold, the growth of seedlings, the medium formula for strong seedlings, the rooting medium formula, the rooting rate, the rooting length, the rooting days and the number of roots, the transplanting substrate, and the transplanting survival rate.

[0043] In step S4 described in the embodiment, the target plant is Ficus hirta Vahl of Ficus subgenus Ficus, also known as Ficus hirta Vahl, which is a commonly used raw material for medicinal food in South China. It was first recorded in the Qing Dynasty in “Shengcao Yingshi Beiyao”. It is sweet and pungent, and belongs to the lung, spleen, stomach, large intestine and liver meridians. According to the “Chinese Pharmacopoeia”, Ficus hirta Vahl is the dried root of Ficus hirta Vahl of Moraceae. It can be harvested all year round. Remove the root hairs, wash them, cut them into slices, and dry them. It has the functions of invigorating the spleen and tonifying qi, and removing dampness and promoting qi. Therefore, Ficus hirta Vahl has both medicinal and edible values.

[0044] The plant characteristics are that the young branches of Ficus hirta Vahl are hollow, the old branches are lignified, the leaves are alternate, the young branches, leaves and fruits are covered with short and hard villi, the leaf shape is like a palm, the fruit shape is like a peach, and the leaf shape is diverse, which can be divided into unsplit, three-lobed, five-lobed, seven-lobed, and nine-lobed.

[0045] The culture target specifically provides the plant tissue culture scheme of Ficus hirta Vahl, including the selection and disinfection method of explants, and the disinfection survival rate is at least 50%; the induction medium formula and culture conditions are provided, and the induction rate is required to reach 60%; the proliferation medium formula and culture conditions are provided, and the proliferation fold is required to reach 3.0; the rooting medium formula and culture conditions are provided, and the rooting rate is required to reach 90%, while the production cost is considered.

[0046] The complete culture scheme generated in step S4 in the present example is as follows: crude Ficus microcarpa L. var. crassifolia Cheng stem segments are used as explants, and 0.1% HgCl2 is used for sterilization for 15 min; the induction culture medium is 1 / 2MS medium + 6-BA (0.25 mg / L) + NAA (0.05 mg / L) + IBA (0.05 mg / L); the proliferation culture medium is MS medium + 6-BA (0.5 mg / L) + NAA (0.05 mg / L); the rooting culture medium is 1 / 2MS medium + 6-BA (0.25 mg / L) + IBA (0.05 mg / L) + AC (0.05 g / L); the culture temperature is 24 ± 2°C, the light time is 12 h / d, and the light intensity is 2000-2500 lx.

[0047] Further, the actual experiment is carried out according to the complete culture scheme generated in step S4, and specifically includes the following steps:

[0048] S41: Pretreatment and selection of explants

[0049] According to the culture scheme generated by the system, the healthy and disease-free Ficus microcarpa L. var. crassifolia Cheng is selected, and after spraying with 50% carbendazim 1000 times liquid for 5 days, it is carefully dug up, the roots are washed, the roots are soaked in 70% thiophanate-methyl solution diluted 1200 times for 20 min for sterilization, and then transplanted into the substrate which has been irrigated with 0.01% potassium permanganate solution, and should be transplanted into the greenhouse. After planting, spray with chlorothalonil every 7 days for sterilization. When the newly grown Ficus microcarpa L. var. crassifolia Cheng sprouts a tender bud segment of more than 10 cm, it is cut off as an explant for rapid propagation of sterile stems, the mature leaves are removed, the new tender leaves are retained, and the stems with axillary buds are cut into 3-4 cm long segments.

[0050] S42: Selection of sterilization time

[0051] According to the culture scheme generated by the system, the stem segments with leaves removed are soaked in washing powder solution for 15 min, brushed with a soft brush to remove surface impurities, then washed with running water for 120 min, and then the washed stem segments are trimmed into 2-3 cm long small segments with one axillary bud, and then transferred to a clean bench for standby use. The bud-containing stem segments are soaked in 75% alcohol for 30 s, rinsed with sterile water once, then soaked in 0.1% HgCl2 solution for 10 min, 15 min and 20 min respectively, rinsed with sterile water for 5 times, then dried with sterile filter paper, and then inoculated into the culture medium, with 1 segment of explant per bottle, and 3 repeats. After 15 days, the contamination rate, mortality rate and survival rate are counted respectively.

[0052] The results in Table 1 show that different disinfection times significantly affected the contamination rate, mortality rate, and survival rate during the induction process of *Ficus macrophylla*. When the disinfection time with 0.1% HgCl2 was 10 min, the disinfection was insufficient, with an average contamination rate as high as 45.33%. As the disinfection time increased, the contamination rate gradually decreased, but prolonged disinfection may cause irreversible damage to the stem segments, leading to an increased mortality rate. Therefore, disinfection with 0.1% HgCl2 for 15 min reduced both the contamination rate and mortality rate while ensuring a survival rate of 61%. This indicates that the culture protocol generated according to the system prediction can achieve the target survival rate of over 50%.

[0053] Table 1. Effects of different disinfection times on explants

[0054]

[0055] Note: Different uppercase letters in the same column indicate significant differences at the 0.05 level.

[0056] S43: Screening of induction culture media

[0057] Based on the culture protocol generated by the system, using 1 / 2 MS medium as the basal medium, and employing a 3-factor, 3-level L9(3) culture medium. 4 Nine treatments were designed using an orthogonal design with different combinations of plant hormones. Sterilized explants were inoculated into different induction media, with three replicates. Observations were performed every 5 days, and the growth of sterile stem segments under different treatments was recorded. After 30 days of inoculation, the induction rate of explants in different treatments was calculated, and the optimal induction medium was selected.

[0058] Table 2. L9(3) explant induction medium 4 Orthogonal experimental design

[0059]

[0060] From the results of Table 3, it can be seen that the different kinds of different concentrations of plant hormones have different effects on the induction of Ficus ingluroides. Among them, according to the system generated culture scheme for treatment group 1, the induction rate is 61.3%; while the expanded culture scheme, the average induction rate of treatment 6 is the highest, which is 69.11%. The R value of each factor is different, which shows that different plant hormones have different effects on the induction of Ficus ingluroides, among which 6-BA and IBA have more obvious effect, and NAA has little effect. According to the results of k value, the optimal induction medium combination is A2B1C1, that is, 1 / 2MS medium + 6-BA (0.5 mg / L) + NAA (0.05 mg / L) + IBA (0.05 mg / L), and the induction effect of Ficus ingluroides is better. The results of variance analysis (Table 4) show that the effects of 6-BA and IBA on the induction rate of Ficus ingluroides explants reached a very significant level, while NAA had no significant effect, which is consistent with the results of range analysis.

[0061] From the above experimental results, it can be seen that the culture scheme generated according to the system can achieve the target requirement of induction rate of 60%, and the production cost is lower.

[0062] Table 3 Range analysis of different plant hormones on the induction of Ficus ingluroides explants

[0063]

[0064] Table 4 Orthogonal test analysis table of different plant hormones on the induction of Ficus ingluroides explants

[0065]

[0066]

[0067] Note: ** indicates significant difference at P = 0.01 level, the same below.

[0068] S44: Screening of proliferation medium

[0069] According to the culture scheme generated by the system, MS medium is used as the basic medium, and L9(3 4 ) orthogonal design of different plant hormone combinations with 3 factors and 3 levels is used, a total of 9 treatments. The tissue culture seedlings induced by primary culture are cut into sterile stem segments with axillary buds, inoculated into the proliferation medium of each treatment, and repeated 3 times. Proliferation culture is carried out under the culture conditions, and the effects of different exogenous plant hormone combinations on the state of adventitious buds and proliferation coefficient are compared every 5d, and the results are counted after 25d to screen out the most suitable proliferation medium.

[0070] Table 5 L9(3 4 ) orthogonal experimental design of proliferation medium

[0071]

[0072] From the results of Table 6, it can be seen that the effects of different kinds and concentrations of plant hormones on the proliferation of Ficus ingluroides are different. The proliferation multiple of the treatment group 1 according to the culture scheme generated by the system is 3.17, and the average proliferation multiple of the treatment 7 in the expanded culture scheme is the highest, which is 3.78. The R values of each factor are different, indicating that different plant hormones have different effects on the proliferation multiple of Ficus ingluroides, among which the effects of 6-BA and NAA are more obvious, and KT hardly affects it. According to the results of k value, the optimal proliferation culture medium combination is A3B1C3, that is, MS medium + 6-BA (1.5 mg / L) + NAA (0.05 mg / L) + KT (1.0 mg / L), and the proliferation effect of Ficus ingluroides is better. The results of variance analysis (Table 7) show that the effects of 6-BA and NAA on the proliferation multiple of Ficus ingluroides reach a very significant level, while the effect of KT is not significant, which is consistent with the results of range analysis.

[0073] From the above experimental results, it can be seen that the culture scheme generated according to the system prediction can achieve the target requirement of a proliferation multiple of 3.0, and the production cost is low.

[0074] Table 6 Range analysis of different plant hormones on the proliferation of Ficus ingluroides

[0075]

[0076] Table 7 Orthogonal test variance analysis table of different plant hormones on the proliferation of Ficus ingluroides

[0077]

[0078] S45: Screening of rooting medium

[0079] According to the culture scheme generated by the system, 1 / 2MS medium is used as the basic medium, and L9(3 4 ) orthogonal design with 3 factors and 3 levels is used to design different hormone combinations, a total of 9 treatments. After the proliferation culture for 30 days, the single plant is cut when the adventitious buds grow to 3-5 cm high, and the rooting induction culture is carried out, repeated 3 times. The rooting situation is observed and counted every 5 days, the effects of different concentrations and types of plant hormones and activated carbon on the rooting of Ficus ingluroides are studied and analyzed, and the results are counted after 20 days, so as to screen out the suitable rooting induction medium.

[0080] Table 8 L9(3 4 ) orthogonal experiment design of rooting medium

[0081]

[0082] The rough Ficus microcarpa Linn.f. var.sublanceolata Becc seedlings obtained from subculture were inoculated into rooting medium containing different plant hormones and activated carbon concentrations. After 5 days of inoculation, white root points were first observed, some seedlings began to root, and the root system showed yellow-white radiation. As can be seen from the results in Table 9, different types and concentrations of plant hormones and activated carbon have different effects on the rooting of rough Ficus microcarpa Linn.f. var.sublanceolata Becc in vitro. Among them, according to the culture scheme generated by the system, the treatment group 1, the rooting rate reached 93.23%; while in the expanded culture scheme, the rooting rate of treatment 4 was as high as 100%. The R values of each factor are also different, indicating that different plant hormones and activated carbon concentrations have different degrees of influence on the multiplication of rough Ficus microcarpa Linn.f. var.sublanceolata Becc, among which the effect of IBA is the most obvious. According to the results of k value, the optimal rooting medium combination can be screened as A2B1C3, that is, MS medium + 6-BA(0.5mg / L) + IBA(0.05mg / L) + AC(0.15g / L). But from the perspective of production cost, combined with the test results, the concentration of AC is selected as 0.1g / L, which can ensure that the rooting rate reaches 100% and reduce the production cost. The results of variance analysis( Table 10) show that the effect of three different factors on the multiplication of rough Ficus microcarpa Linn.f. var.sublanceolata Becc reached a very significant level, which is consistent with the results of range analysis.

[0083] From the above experimental results, it can be seen that the culture scheme generated according to the system can achieve the goal of a rooting rate of 90% and has a low production cost.

[0084] Table 9 Range analysis of different plant hormones and activated carbon concentrations on the rooting of rough Ficus microcarpa Linn.f. var.sublanceolata Becc in vitro

[0085]

[0086]

[0087] Table 10 Orthogonal test analysis of variance table of different plant hormones and activated carbon concentrations on the rooting of rough Ficus microcarpa Linn.f. var.sublanceolata Becc in vitro

[0088]

[0089] Through the cooperation of the above modules and the embodiment, the application can collect patent literature data, non-patent literature data, historical experiment record data and production process real-time monitoring data related to plant tissue culture, form a structured original data set, solve the problem of experience-based and lagging update of traditional tissue culture, adopt a multi-source heterogeneous plant tissue culture knowledge base with a clear hierarchical structure (including but not limited to plant varieties, explant types, culture stages, parameter types and experimental results), enhance the systematicness and integrity of the tissue culture knowledge, call a deep learning framework, use the standardized data in the knowledge base as training corpus, train an AI model, predict and generate a culture scheme, reduce the trial and error cost, and realize precise knowledge service.

Claims

1. A method for generating plant tissue culture protocols based on multi-source knowledge base and AI, characterized in that, Comprise the following steps: 1) Data collection: collect plant tissue culture related data, and perform denoising processing, field standardization processing and metadata tagging processing on the collected data to form a structured original data set; 2) Knowledge base construction: after data cleaning, entity recognition, relationship extraction and semantic mapping of the data in the original data set, a multi-source heterogeneous plant tissue culture knowledge base with a clear hierarchical structure is formed; 3) AI model training: call the deep learning model to train the model with the standardized data in the knowledge base as the training corpus to obtain a multi-dimensional function group for predicting culture parameters. The input data during model training includes One-hot encoding of plant families and genera, numerical characteristics of explant types, and historical parameter information. The culture parameters predicted by the model include disinfection methods, culture medium formulations, and culture conditions at each culture stage; 4) Scheme generation and evaluation: based on the user input of target plant species, plant characteristics and culture goals, the multi-dimensional function group trained in step 3) is used to predict the initial culture scheme parameters. Then, the Bayesian optimization algorithm combined with the genetic algorithm is used to optimize and search the prediction results, and finally the complete culture scheme is output, along with the comprehensive score and scheme confidence; 5) Feedback update: perform actual experiments according to the culture scheme output in step 4), record the experimental process data, calculate the comprehensive score, and write back to the knowledge base; if the comprehensive score or scheme confidence is lower than the preset threshold, or the amount of new data in the knowledge base reaches the preset threshold, automatically trigger the AI model in step 3) for fine-tuning and incremental learning update, and enter the next round of scheme generation iteration.

2. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The data collected in step 1) includes historical experimental record data, real-time monitoring data during production process, patent literature data, and non-patent literature data.

3. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The knowledge base in step 2) includes structured data and unstructured data. Structured data refers to plant classification, organization, fixed field name, clear data type / unit, and direct retrieval and calculation by program. Unstructured data includes culture medium formulation, literature abstract, and picture. Structured data is stored in a database, and unstructured data is stored in a vector database after semantic vector encoding. Structured data and unstructured data are one-to-one corresponding through unique relationship identifiers.

4. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The hierarchical structure in step 2) includes plant species, explant type, culture stage, parameter type, and experimental result.

5. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The deep learning model in step 3) is a multi-task prediction model based on the Transformer architecture. The model training uses transfer learning and incremental learning methods. When the knowledge base data reaches the threshold, the model is automatically fine-tuned and the parameters are updated.

6. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The culture conditions in step 3) include hormone concentration, carbon source concentration, light intensity, and culture period.

7. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The surrogate model of Bayesian optimization in step 4) is a deep Gaussian process, and the expected improvement function is used to select the next round of parameter combinations to be evaluated. Genetic algorithm is used to quickly search discrete parameters.

8. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, The complete culture scheme output in step 4) includes disinfection method, induction culture stage parameters, proliferation culture stage parameters, and rooting culture stage parameters.

9. The multi-source knowledge base and AI based plant tissue culture protocol generation method as claimed in claim 1 wherein, Step 5) the experimental process data includes survival rate, induction rate, proliferation fold, and rooting rate.

10. A multi-source knowledge base and AI based plant tissue culture protocol generation system, characterized in that, The system comprises: (1) a data acquisition unit for acquiring and storing relevant raw data from historical experimental record data, real-time monitoring data of the production process, patent literature data, and non-patent literature data; (2) a knowledge base construction unit for data cleaning, entity recognition, relationship extraction, semantic mapping, and ontology modeling of the raw data, and for constructing a multi-source heterogeneous knowledge base oriented to the field of plant tissue culture; (3) an AI model training unit for training a deep learning model based on the knowledge base, obtaining a multi-dimensional function group for predicting culture parameters, and regularly updating the incremental learning according to the newly added data; (4) a scheme generation and evaluation module for receiving input features including target plant varieties and explant types, generating a complete culture scheme according to the multi-dimensional function group, and outputting a comprehensive score and a scheme confidence; (5) a feedback updating module for receiving actual experimental results, writing back the experimental feedback data to the knowledge base, and triggering the AI model training unit to perform parameter fine-tuning and model iteration.

Images