Optimization method and application of sclareol production strain based on various metabolic models

Abstract

The invention discloses an optimization method and application of sclareol production strains based on various metabolic models. The method comprises the following steps: integrating phenotype data and transcriptome data of a chassis strain to construct an integration model, and predicting an endogenous transformation target within a genome range based on a metabolism model; the host strain is selected from saccharomyces cerevisiae. According to the method, an enzyme constraint model and a non-intuitive gene editing target mining strategy are adopted, and a new method and a new modified target are provided for constructing a microbial cell factory for efficiently synthesizing sclareol.

Description

Technical Field

[0001] This invention belongs to the fields of microbial metabolic engineering and synthetic biology. Specifically, it is an optimization method and application of perillaldehyde-producing strains based on multiple metabolic models. It utilizes a systems biology computing platform to integrate multiple advanced metabolic models and multi-omics data to rationally design and optimize perillaldehyde-producing strains, as well as the superior strains obtained through this method and their applications. Background Technology

[0002] Sclareol is a commercially valuable diterpenoid compound widely used in the fragrance, cosmetics, and pharmaceutical industries. Utilizing microorganisms as cell factories for biomanufacturing is a key approach to achieving sustainable and environmentally friendly production of sclareol. However, constructing efficient microbial cell factories remains a significant challenge. The core bottleneck lies in the fact that cellular metabolism is a complex and precisely regulated network; rationally and accurately identifying gene modification targets that can significantly enhance the yield of the desired product is a difficult problem in the field of metabolic engineering.

[0003] Traditional genome-scale metabolic models (GEMs) are primarily based on stoichiometry constraints, neglecting the limited intracellular enzyme resources and protein synthesis capacity, thus limiting their accuracy in predicting cell phenotypes. In recent years, advanced models combining enzyme kinetics and protein resource allocation principles, such as enzyme-constrained genome-scale models (ecGEMs) and models integrating transcription, translation, and metabolic processes (such as ETFL models), have demonstrated stronger predictive capabilities. These models can more realistically simulate the physiological state of cells under resource constraints.

[0004] Despite the immense potential of these advanced models, a standard, user-friendly computational tool and a systematic design process are currently lacking to translate their theoretical advantages into practical, experimentally guiding modification strategies. This is particularly true when re-optimizing a superior industrial strain that has already undergone multiple rounds of engineering modifications, requiring the discovery of deeper, non-intuitive gene targets, which places higher demands on computational design methods.

[0005] Therefore, there is an urgent need in this field for a systematic strain design method that can: 1) comprehensively utilize the advantages of different advanced models to obtain complementary prediction results; 2) integrate the strain's own physiological and transcriptomic data to achieve personalized and accurate predictions; 3) incorporate key biological principles such as the allocation of cell resources at the subcompartmental level into the constraints to improve prediction accuracy; and 4) verify its effectiveness in industrial strain optimization through practical applications. Summary of the Invention

[0006] To achieve the above objectives, the present invention provides an optimization method and application for optimizing the perillaldehyde production strain.

[0007] This invention adopts the following technical solution: an optimization method for perillaldehyde-producing strains based on multiple metabolic models, comprising the following steps:

[0008] a) Obtain the initial microbial strain for perillaldehyde production and collect its phenotypic parameters and transcriptome data under specific culture conditions;

[0009] b) Based on the genomic information of the microbial strain, construct at least two different frameworks of genome-scale metabolic models that include enzyme constraints; the genome-scale metabolic models include the enzyme-constrained genome-scale model ecGEM and the model EFL that integrates gene expression and metabolic networks;

[0010] c) Integrate the strain-specific phenotypic parameters and transcriptome data collected in step a) into ecGEM and EFL, and calibrate the two models;

[0011] d) For different physiological stages of the strain during fermentation, independent metabolic simulations were performed using calibrated ecGEM and EFL to predict the set of gene modification targets that can increase the yield of perillaldehyde.

[0012] e) Comprehensively analyze and screen gene modification targets predicted from different models and different physiological stages to obtain an optimized target list;

[0013] f) The optimized target list from step e) is screened for novelty and mechanism rationality to finally obtain gene modification targets that increase perillaldehyde production.

[0014] The phenotypic parameters include: maximum specific growth rate, specific glucose consumption rate, specific ethanol production rate, and specific product production rate.

[0015] The transcription data was obtained in the following manner:

[0016] Total RNA was extracted from samples taken during the glucose consumption and ethanol consumption phases, and transcriptome sequencing was performed to obtain the whole genome transcriptional expression profile of strain SCX42 at the two growth stages. The data represent gene expression levels.

[0017] The process of integrating the strain-specific phenotypic parameters and transcriptome data collected in step a) into ecGEM and EFL, and calibrating the two models, specifically involves performing the following steps for each genome-scale metabolic model:

[0018] The obtained phenotypic parameters are used as hard constraints for the genome-scale metabolic model: the values ​​of maximum specific growth rate and glucose specific consumption rate are set as constraints for the corresponding reactions in the genome-scale metabolic model, and the constraints are fixed flux values ​​or upper and lower limits.

[0019] The GIMME algorithm was used to integrate transcriptome data into a genome-scale metabolic model: the reaction flux associated with genes whose expression levels were below a threshold was preferentially set to zero, generating a metabolic model that could reflect the metabolic state of a specific growth stage.

[0020] Subcellular protein mass constraints were imposed: Based on the subcellular protein localization information in the Saccharomyces cerevisiae proteomics dataset and proteome annotation database, the range of total protein mass fraction of organelles under various culture conditions was statistically determined; then, in the genome-scale metabolic model, corresponding organelle constraints were added according to the annotation information of each gene product, that is, the range of total protein mass fraction of each organelle was used as the upper and lower limits of the corresponding protein pool in the genome-scale metabolic model.

[0021] For different physiological stages during the fermentation process of the strain, independent metabolic simulations were performed using calibrated ecGEM and EFL to predict the set of gene modification targets that could increase perillaldehyde production. A comprehensive analysis and screening of gene modification targets predicted from different models and different physiological stages yielded an optimized target list, including the following steps:

[0022] (1) Establish four simulation scenarios: Combine the calibrated ecGEM and EFL models and the two growth stages to establish four independent simulation scenarios. Each simulation scenario includes a model corresponding to a growth stage, which serves as the scenario model.

[0023] (2) Initial target set generation: The L1 prediction level of the ecFactory algorithm is applied to the four scenario models respectively, and the minimum total protein usage is set as the objective function to solve the problem. Finally, the potential gene modification target set under each scenario is obtained.

[0024] (3) Target scoring and ranking: For each potential gene modification target set and its corresponding scenario model, simulate the perturbation of the corresponding enzyme dosage for each preliminary target and calculate the impact on the perillol production throughput to obtain the perillol product throughput control coefficient value, i.e., FCCp value. Based on the FCCp value, the targets are ranked to obtain a ranked target list.

[0025] FCCp>0 indicates that upregulating the enzyme corresponding to this target site is beneficial to production, and is therefore an upregulated target site.

[0026] FCCp<0 indicates that downregulating or knocking out the target is beneficial to production, which is called downregulating / knocking out the target.

[0027] The absolute value of FCCp represents the strength of the target's influence; the larger the absolute value, the higher the ranking.

[0028] (4) Target list integration: The target lists after scoring and sorting under four different scenario simulations are integrated and deduplicated to generate an optimized list of potential gene targets.

[0029] The optimized target list from step e) is screened for novelty and mechanistic rationality to ultimately obtain gene modification targets that increase perillaldehyde production, including the following steps:

[0030] Final target screening: The optimized list of potential gene targets is screened using both novelty and mechanism rationality criteria. Specifically: Novelty screening: Based on pathway enrichment analysis and scenario cross-analysis, targets in the optimized list located in preset, known central metabolic pathways and main product synthesis pathways are excluded to identify non-intuitive gene targets; Mechanism rationality screening: Targets that conform to specific metabolic engineering principles are selected, including those that enhance the supply of key precursors required for the synthesis of sagebrushol or inhibit competitive metabolic pathways that consume the key precursors.

[0031] An optimization method based on multiple metabolic models for perillaldehyde-producing strains was applied to improve the microbial fermentation production of perillaldehyde. A genetically engineered strain was obtained by regulating gene modification targets that enhance perillaldehyde yield in the SCX42H strain. The gene modification targets for enhancing perillaldehyde yield are one or more of TGL3, ALE1, SLC1, IDP1, ODC1, and ALD6. The regulation methods include one or more of the following gene modification targets: knockout, overexpression, and attenuation. The genetically engineered strain contains one or more gene modifications selected from: TGL3 gene knockout, ALE1 gene knockout, SLC1 gene knockout, IDP1 gene knockout, ODC1 gene knockout, and ALD6 gene overexpression.

[0032] The present invention has the following beneficial effects and advantages:

[0033] This invention comprehensively utilizes the advantages of different advanced models to obtain complementary prediction results; integrates the strain's own physiological and transcriptomic data to achieve personalized and accurate prediction; incorporates key biological principles such as the allocation of cell resources at the subcompartmental level into the constraints to improve prediction accuracy; and verifies its effectiveness in industrial strain optimization through practical application.

[0034] Using the above method, this invention was applied to a highly optimized *Saccharomyces cerevisiae* strain that produces perillaldehyde, and several novel and effective gene modification targets were discovered. Experimental verification showed that knocking out or downregulating lipid synthesis pathway genes (such as TGL3, ALE1, SLC1) and amino acid synthesis pathway genes (such as IDP1, ODC1) that compete with acetyl-CoA, as well as overexpressing upstream pathway genes that enhance acetyl-CoA supply (such as ALD6), can significantly increase the yield of perillaldehyde, with the highest increase reaching 40%. This invention also protects the genetically engineered strain obtained by the above method, and the application of this method and strain to the fermentation production of perillaldehyde. Attached Figure Description

[0035] Figure 1 illustrates the overall workflow for strain-optimized target prediction;

[0036] Figure 2 shows a schematic diagram of the predicted sources and screening process of gene modification targets in the embodiment, specifically demonstrating the distribution and overlap of the predicted targets through different metabolic models (ecGEM and EFL) and different fermentation stages (glucose consumption, ethanol consumption).

[0037] Figure 3 shows the yield of perillaldehyde after target modification;

[0038] Figure 4 illustrates the metabolic mechanism by which validated effective targets enhance perillaldehyde production. Detailed Implementation

[0039] The method of the present invention will be described in detail below with reference to a specific embodiment.

[0040] This invention provides an optimization method and application for strains that produce perillaldehyde. The core of this method lies in a systematic workflow, including:

[0041] 1. Data Acquisition and Model Construction: First, an initial strain with a certain capacity for perillaldehyde production (such as a preliminarily modified Saccharomyces cerevisiae) was selected and cultured under simulated fermentation conditions. Key phenotypic data (such as specific growth rate and substrate consumption rate) and whole-genome transcriptional data were collected. Simultaneously, based on the genomic information of this strain, two different but complementary advanced metabolic models were constructed or invoked: an enzyme-constrained genome-scale model (ecGEM) and a flux model integrating expression and thermodynamics (ETFL).

[0042] 2. Personalized Model Calibration and Constraint Integration: Collected strain-specific data are integrated into the two models described above, enabling the models to more accurately reflect the physiological state of the specific strain. A key innovation lies in the introduction of proteomic constraints at the subcompartmental level. Based on published proteomic data, the stable protein mass fraction ranges of major organelles (such as mitochondria and cytoplasm) under different conditions are determined and used as additional constraints for the models. This step ensures that the metabolic activities simulated by the models conform to the fundamental laws governing the spatial allocation of intracellular protein resources, effectively filtering out biologically unreasonable predictions.

[0043] 3. Multi-model, Multi-scenario Target Prediction: Another key innovation of this invention lies in utilizing the complementarity of models. Calibrated ecGEM and ETFL models are used to simulate two key physiological stages in the fermentation process: the glucose consumption stage and the ethanol consumption stage (for strains with biphasic growth, such as Saccharomyces cerevisiae). In the simulation, a protein-level objective function closer to biological reality, such as minimizing total protein usage (ppFBA), is employed to predict the impact of different genes (knockout, downregulation, or overexpression) on perillaldehyde production, thereby generating a series of potential gene modification targets.

[0044] Target screening and experimental validation: A comprehensive analysis was performed on targets predicted from two models and two physiological stages. This method pays particular attention to non-overlapping targets that can only be predicted by specific models or under specific conditions, as these targets are often non-intuitive but may play an important role in performance improvement. Several candidate genes were screened from the comprehensive target list, and the initial strain was modified using gene editing technologies (such as CRISPR / Cas9). The modification effect was verified through fermentation experiments.

[0045] Example 1: Optimization of a high-yield *Saccharomyces cerevisiae* strain using the method of the present invention

[0046] 1. Initial strain, data collection and model preparation

[0047] Initial strain: The high-yielding *Saccharomyces cerevisiae* strain SCX42, which was previously constructed in our laboratory and underwent multiple rounds of systematic modification, was selected. Metab Eng . 2023;75:19-28).

[0048] Data collection:

[0049] Phenotypic data acquisition: SCX42 strain was cultured in a standard laboratory fermenter, and parameters such as cell biomass (OD600), glucose concentration, ethanol concentration, and perillaldehyde titer were continuously monitored online. These time-series data were then fitted into a Gompertz growth kinetic model to accurately determine the key phenotypic parameters of the strain at different growth stages, specifically including: maximum specific growth rate, specific glucose consumption rate, specific ethanol production rate, and specific product production rate.

[0050] Transcriptome data acquisition: Total RNA was extracted from samples taken during mid-log growth (glucose consumption stage) and ethanol consumption stage, and then transcriptome sequencing (RNA-seq) was performed to obtain the whole genome transcriptional expression profile of strain SCX42 at the two key growth stages. The specific data are gene expression level data in units such as kilobase million transcripts (TPM).

[0051] Model preparation: The ecGEM model and EFL model (a simplified version of the ETFL model without thermodynamic constraints) of Saccharomyces cerevisiae were used to reduce computational complexity and make a fairer comparison with ecGEM.

[0052] 2. Model Calibration and Target Prediction

[0053] Data integration:

[0054] Phenotypic data integration: The phenotypic parameters obtained in step 1 are used as hard constraints for the model. For example, the maximum specific growth rate and the specific glucose consumption rate are set as fixed flux values ​​or upper and lower limits for the corresponding reactions in the model (such as biomass synthesis reaction and glucose uptake reaction).

[0055] Transcriptome data integration: The transcriptome data obtained in step 1 was integrated into the ecGEM and EFL models using the GIMME algorithm. This algorithm, while ensuring the model can simulate preset biological functions (such as achieving minimum growth rate and product production rate), contextually simplifies the metabolic network based on gene expression levels (TPM values). Specifically, the reaction flux associated with genes with low expression levels is preferentially set to zero, thereby generating a highly personalized metabolic model that reflects the metabolic state at a specific growth stage (glucose consumption or ethanol consumption).

[0056] Subcellular protein mass constraints were imposed: Based on published Saccharomyces cerevisiae proteomics datasets and protein subcellular localization information from databases such as UniProt, the total protein mass fraction range of major organelles such as cytoplasm, mitochondria, and nucleus was statistically calculated under various culture conditions. Then, corresponding organelle constraints were added to the model based on the annotation information of each gene product, that is, the total protein mass fraction range of each organelle was used as the upper and lower limits of the corresponding protein pool in the model.

[0057] Simulation and Prediction:

[0058] 1) Establish four simulation scenarios. By combining two models (ecGEM, EFL) and two growth stages (glucose consumption, ethanol consumption), a total of four independent and personalized simulation scenarios were established.

[0059] 2) Initial target set generation. Using the four calibrated and constrained personalized models obtained in the preceding steps, the L1 prediction level of the ecFactory algorithm (aimed at generating the broadest possible set of candidate targets) is applied to each of the four scenario models. During algorithm execution, "minimizing total protein usage" (ppFBA) is set as the objective function. The final set of potential genetic modification targets for each scenario is obtained.

[0060] 3) Target Scoring and Ranking. Using the four preliminary target sets generated in step two and their corresponding personalized models, flux control coefficient (FCC) analysis was employed to score and rank each preliminary target. FCC is a mature metabolic control analysis method; its calculated value can quantify the ultimate impact of small changes in the expression level of a single enzyme on the synthesis rate of the target product (perillyl alcohol). The calculated FCCp value of the perillyl alcohol product was used as the ranking criterion. FCCp > 0 indicates that upregulating the enzyme corresponding to the target is beneficial to production, classifying it as an upregulated target; FCCp < 0 indicates that downregulating or knocking out the target is beneficial to production, classifying it as a downregulated / knockout target. The absolute value of FCCp represents the strength of the target's influence; the larger the absolute value, the higher the ranking. Specifically, the perturbation of the corresponding enzyme dosage for each preliminary target is simulated and the impact on the perturbation of perturbation production flux is calculated to obtain the perturbation product flux control coefficient (FCCp) value. Based on the FCCp value, the targets are sorted to obtain a list of four sorted targets.

[0061] 4) Target List Integration. The four target lists after scoring and sorting in step three were integrated and deduplicated to generate an optimized list containing 72 potential gene targets (e.g., Figure 1 (As shown).

[0062] 5) Target screening.

[0063] Step 1: Global Analysis.

[0064] Pathway enrichment analysis: Pathway enrichment was performed on 72 targets, identifying the metabolic pathways with the highest concentration of targets. Results showed these pathways to be related to lipid metabolism, amino acid metabolism, terpene synthesis, and central metabolism (e.g., ...). Figure 2 (As shown).

[0065] Cross-scenario analysis: Through Venn diagrams (such as...) Figure 2 (As shown) Analyze the distribution of target points across different models and stages.

[0066] Step 2: Set the screening criteria.

[0067] Standard 1 (Novelty): To identify non-intuitive gene targets, we screen out known genes related to central metabolism and terpene synthesis pathways, while retaining genes related to lipid metabolism and amino acid metabolism.

[0068] Criterion 2 (Interpretability): The screening criteria focus on genes that can "enhance the supply of key precursors (acetyl-CoA)" or "inhibit precursor / energy competition".

[0069] Step 3: Focus on key pathways. Based on the enrichment analysis in Step 1 and the screening criteria in Step 2, the screening focus was placed on the two pathways with the largest number of predicted targets and the closest competition with precursors: lipid metabolism and amino acid metabolism. Additionally, genes from the ethanol metabolism pathway that were predicted as highly efficient targets in the "ethanol consumption phase" simulation were also included.

[0070] Step 4: Refinement and Validation. Under the above criteria, the selected targets undergo final refinement. Specifically, for predicted targets that are protein complexes composed of multiple genes, only one key subunit gene is selected for validation.

[0071] Final Results: Following the above screening criteria, six genes were ultimately selected for experimental validation, including:

[0072] Lipid metabolism (competitive pathways): TGL3, ALE1, SLC1

[0073] Amino acid metabolism (competitive pathways): IDP1, ODC1

[0074] Ethanol metabolism (precursor supply): ALD6.

[0075] Example 2

[0076] Experimental verification:

[0077] Unless otherwise specified, the experimental methods used in the following examples are all conventional methods, and the materials and reagents used can be purchased commercially. See Tables 1 to 3.

[0078] Table 1. Information related to overexpressed genes in the embodiments of this application.

[0079]

[0080] Table 2. Target sequences of the knockout genes involved in the embodiments of this application.

[0081]

[0082] Table 3. Composition of Delft culture medium used in the embodiments of this application.

[0083]

[0084] Unless otherwise specified, this application uses a culture medium supplemented with 20 g / L glucose and 100 mg / L histidine.

[0085]

[0086]

[0087] This application embodiment is based on SCX42H ( Metab Eng The strain was modified using CRISPR / Cas9 (2023;75:19-28). The genetic modification in this application was performed using the CRISPR / Cas9 system (see reference: Mans R, van Rossum HM, Wijsman M, Backx A, Kuijpers NG, van den Broek M, Daran-Lapujade P, Pronk JT, vanMaris AJ, Daran JM. CRISPR / Cas9: a molecular Swiss army knife for simultaneous introduction of multiple genetic modifications in...). Saccharomyces cerevisiae . FEMS Yeast Res . 2015 Mar;15(2):fov004.)

[0088] The specific implementation method of target knockout is as follows TGL3 Taking genes as an example: First, construct targeted... TGL3 The sgRNA expression vector containing the gene's ORF box was then amplified separately. TGL3The upstream and downstream sequences of the gene were extracted, and the complete donor DNA fragment was obtained by fusion PCR. Subsequently, the gRNA expression vector and gene expression cassette (500 ng each) were transformed into Saccharomyces cerevisiae by chemical transformation and plated onto selection plates at 30°C. o After static incubation for 3 days, transformants were cultured in liquid SD medium and verified as correct by colony PCR. The transformed strains were then plated on plates containing 5-fluoroorotic acid for plasmid loss. The plasmid-loss-free strains were stored for later use. Other 4 target sites ( IDP1, ALE1, SLC1, ODC1 The knockout process followed the above procedure, achieving seamless knockout. Engineered strains TG3, TG2, TG4, TG5, and TG6 were obtained respectively.

[0089] Target overexpression ALD6 Specific construction method: Constructing targeted chromosomal integration sites IX-2 The sgRNA expression vector was then used; subsequently, the sgRNA expression vector and donor DNA (500 ng each) were chemically transformed into SCX42H, and plated onto selection plates at 30°C. o After being incubated statically for 3 days, the transformants were cultured in liquid SD medium and verified to be correct by colony PCR. They were then plated on plates containing 5-fluoroorotic acid to remove plasmids. The strain after plasmid removal was named engineered strain TG1.

[0090] Furthermore, the six engineered strains and the control strain SCX42H were fermented in shake flasks using Delft medium (Table 3), and the yield of perillaldehyde and cell density (OD600) were measured at 72 and 120 hours. The results are as follows:

[0091] Knocking out TGL3 (TG3) increased the yield of salinomyol by 16%.

[0092] Knocking out ALE1 (TG4) increased the yield of sagelinol by 24%.

[0093] Knocking out SLC1 (TG5) increased the yield of salidomide by 22%.

[0094] Knocking out IDP1 (TG2) increased the yield of sagelinol by 12%.

[0095] Knocking out ODC1 (TG6) increased the yield of sagelinol by 22%.

[0096] Overexpression of ALD6 (TG1) increased the yield of perillyl alcohol by 12%.

[0097] Analysis of the target mechanism revealed that it mainly revolves around the supply and competition for the key precursor acetyl-CoA. Knocking out the TGL3, ALE1, and SLC1 genes weakened the lipid synthesis pathway that competes with acetyl-CoA. Knocking out the IDP1 and ODC1 genes weakened the glutamate synthesis pathway, which also consumes acetyl-CoA (via the TCA cycle). Overexpression of ALD6, however, enhanced the replenishment pathway from ethanol to acetyl-CoA (e.g., ...). Figure 4 (As shown). Therefore, the target site screened by the method of the present invention can still increase the yield of perillaldehyde based on the existing strain SCX42H (e.g. Figure 3 (As shown in the figure). This demonstrates the powerful ability of the method of the present invention to discover effective, non-intuitive gene targets.

[0098] This invention provides a systematic, multi-level computational design framework that successfully applies fundamental biological principles such as protein resource allocation to metabolic engineering practices. By combining complementary advanced models, integrating strain-specific omics data, and imposing subcellular constraints, this method can accurately and efficiently provide new and effective gene modification strategies for further optimization of industrial strains, solving a key bottleneck in the field of metabolic engineering and possessing significant scientific value and broad industrial application prospects.

[0099] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. An optimization method for perillaldehyde-producing strains based on multiple metabolic models, characterized in that, Includes the following steps: a) Obtain the initial microbial strain for perillaldehyde production and collect its phenotypic parameters and transcriptome data under specific culture conditions; b) Based on the genomic information of the microbial strain, construct at least two different frameworks of genome-scale metabolic models that include enzyme constraints; the genome-scale metabolic models include the enzyme-constrained genome-scale model ecGEM and the model EFL that integrates gene expression and metabolic networks; c) Integrate the strain-specific phenotypic parameters and transcriptome data collected in step a) into ecGEM and EFL, and calibrate the two models; d) For different physiological stages of the strain during fermentation, independent metabolic simulations were performed using calibrated ecGEM and EFL to predict the set of gene modification targets that can increase the yield of perillaldehyde. e) Comprehensively analyze and screen gene modification targets predicted from different models and different physiological stages to obtain an optimized target list; f) The optimized target list from step e) is screened for novelty and mechanism rationality to finally obtain gene modification targets that increase perillaldehyde production.

2. The method for optimizing the perillaldehyde-producing strain based on multiple metabolic models according to claim 1, characterized in that, The phenotypic parameters include: maximum specific growth rate, specific glucose consumption rate, specific ethanol production rate, and specific product production rate.

3. The method for optimizing the perillaldehyde-producing strain based on multiple metabolic models according to claim 1, characterized in that, The transcription data was obtained in the following manner: Total RNA was extracted from samples taken during the glucose consumption and ethanol consumption phases, and transcriptome sequencing was performed to obtain the whole genome transcriptional expression profile of strain SCX42 at the two growth stages. The data represent gene expression levels.

4. The method for optimizing the perillaldehyde-producing strain based on multiple metabolic models according to claim 1, characterized in that, The process of integrating the strain-specific phenotypic parameters and transcriptome data collected in step a) into ecGEM and EFL, and calibrating the two models, specifically involves performing the following steps for each genome-scale metabolic model: The obtained phenotypic parameters are used as hard constraints for the genome-scale metabolic model: the values ​​of maximum specific growth rate and glucose specific consumption rate are set as constraints for the corresponding reactions in the genome-scale metabolic model, and the constraints are fixed flux values ​​or upper and lower limits. The GIMME algorithm was used to integrate transcriptome data into a genome-scale metabolic model: the reaction flux associated with genes whose expression levels were below a threshold was preferentially set to zero, generating a metabolic model that could reflect the metabolic state of a specific growth stage. Subcellular protein mass constraints were imposed: Based on the subcellular protein localization information in the Saccharomyces cerevisiae proteomics dataset and proteome annotation database, the range of total protein mass fraction of organelles under various culture conditions was statistically determined; then, in the genome-scale metabolic model, corresponding organelle constraints were added according to the annotation information of each gene product, that is, the range of total protein mass fraction of each organelle was used as the upper and lower limits of the corresponding protein pool in the genome-scale metabolic model.

5. The method for optimizing the perillaldehyde-producing strain based on multiple metabolic models according to claim 1, characterized in that, For different physiological stages during the fermentation process of the strain, independent metabolic simulations were performed using calibrated ecGEM and EFL to predict the set of gene modification targets that could increase perillaldehyde production. A comprehensive analysis and screening of gene modification targets predicted from different models and different physiological stages yielded an optimized target list, including the following steps: (1) Establish four simulation scenarios: Combine the calibrated ecGEM and EFL models and the two growth stages to establish four independent simulation scenarios. Each simulation scenario includes a model corresponding to a growth stage, which serves as the scenario model. (2) Initial target set generation: The L1 prediction level of the ecFactory algorithm is applied to the four scenario models respectively, and the minimum total protein usage is set as the objective function to solve the problem. Finally, the potential gene modification target set under each scenario is obtained. (3) Target scoring and ranking: For each potential gene modification target set and its corresponding scenario model, simulate the perturbation of the corresponding enzyme dosage for each preliminary target and calculate the impact on the perillol production throughput to obtain the perillol product throughput control coefficient value, i.e., FCCp value. Based on the FCCp value, the targets are ranked to obtain a ranked target list. FCCp > 0 indicates that upregulating the enzyme corresponding to this target site is beneficial to production, and is therefore an upregulated target site. FCCp < 0 indicates that downregulating or knocking out the target is beneficial to production, which is called downregulating / knocking out the target. The absolute value of FCCp represents the strength of the target's influence; the larger the absolute value, the higher the ranking. (4) Target list integration: The target lists after scoring and sorting under four different scenario simulations are integrated and deduplicated to generate an optimized list of potential gene targets.

6. The method for optimizing the perillaldehyde-producing strain based on multiple metabolic models according to claim 1, characterized in that, The optimized target list from step e) is screened for novelty and mechanistic rationality to ultimately obtain gene modification targets that increase perillaldehyde production, including the following steps: Final target screening: The optimized list of potential gene targets is screened using both novelty and mechanism rationality criteria. Specifically: Novelty screening: Based on pathway enrichment analysis and scenario cross-analysis, targets in the optimized list located in preset, known central metabolic pathways and main product synthesis pathways are excluded to identify non-intuitive gene targets; Mechanism rationality screening: Targets that conform to specific metabolic engineering principles are selected, including those that enhance the supply of key precursors required for the synthesis of sagebrushol or inhibit competitive metabolic pathways that consume the key precursors.

7. The application of the optimization method for perillaldehyde-producing strains based on multiple metabolic models as described in claim 1, characterized in that, The method is used to improve the production of perillol by microbial fermentation.

8. The application according to claim 7, characterized in that, Genetically engineered strains were obtained by regulating gene modification targets that increase perillaldehyde production in the SCX42H strain; the gene modification targets that increase perillaldehyde production are one or more of TGL3, ALE1, SLC1, IDP1, ODC1, and ALD6.

9. The application according to claim 7, characterized in that, The regulatory mechanism involves one or more of the following: knockout, overexpression, or attenuation of gene modification targets that increase perillaldehyde production.

10. The application according to claim 9, characterized in that, The genetically engineered strain contains one or more gene modifications, wherein the gene modifications are selected from: TGL3 gene knockout, ALE1 gene knockout, SLC1 gene knockout, IDP1 gene knockout, ODC1 gene knockout, and ALD6 gene overexpression.

Images