Method for evaluating and grading quality of thesium chinense or finished thesium chinense

Abstract

The invention relates to the technical field of chemical detection, and particularly discloses a thesium chinense or a thesium chinense finished product quality evaluation and grading method. According to the application, a high-content imaging technology system is applied to thesium chinense and finished product quality evaluation for the first time, and the quality control concept transformation from chemical indexes to biological values is realized; a quantitative relation between chemical components and biological activity is established through a machine learning algorithm, and rapid biological value prediction based on chemical analysis is realized; a complete technical system from wild resource evaluation to preparation quality control is established, and a scientific basis is provided for thesium chinense high-quality resource screening and artificial cultivation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of chemical detection, in particular to a quality evaluation and grading method for bai-rui-grass or its finished product. BACKGROUND

[0002] As a traditional anti-infective Chinese medicine, bai-rui-grass and its main preparation bai-rui granules have remarkable curative effect on treating upper respiratory tract infection and other diseases. However, the quality control of the medicinal material and its preparation is facing severe challenges: the current quality control method mainly relies on the chemical content determination of single or a few index components such as kaempferol. This "single component, single index" mode cannot reflect the overall biological potency of the synergistic effect of multiple active components in the medicinal material, resulting in weak correlation between product quality and clinical efficacy.

[0003] More prominently, bai-rui-grass currently mainly relies on wild resources. Due to differences in growth environment and genetic background, medicinal materials from different producing areas and harvest periods have great differences in quality. Traditional chemical analysis methods cannot accurately evaluate the influence of such variations on actual efficacy, which restricts the standardization of products and the stability of curative effect. SUMMARY

[0004] In view of the core problems existing in the quality control of bai-rui-grass and its preparation, such as disconnection between chemical index and clinical efficacy, inaccurate quality evaluation of wild resources, and low efficiency of traditional biological evaluation methods, the purpose of the application is to overcome the deficiencies of the prior art and provide a quality evaluation and grading method for bai-rui-grass or its finished product based on high-content imaging and machine learning.

[0005] To achieve the above purpose, the technical scheme adopted by the application is as follows: The application provides a quality evaluation and grading method for bai-rui-grass or its finished product, which comprises the following steps: S1, establishing an inflammatory cell model and verifying the inflammatory cell model; S2, performing multi-target point immunofluorescence staining on the verified inflammatory cell model in step S1, and then performing feature extraction and screening on the stained cells by using a high-content image acquisition and analysis method; the screened features include NF-κB nuclear-plasma ratio, TNF-α expression intensity, cell area and mitochondrial membrane potential, and a comprehensive efficacy index formula is constructed; Comprehensive efficacy index=(IRR_NF-κB*0.4)+(IRR_TNF-α*0.3)+(IRR_cell area*0.2)+(IRR_mitochondria*0.1), IRR is inhibition rate / recovery rate; S3, pretreating bai-rui-grass or its finished product, and then treating the pretreated bai-rui-grass or its finished product by using ultra-high performance liquid chromatography coupled with quadrupole time-of-flight mass spectrometry to obtain monomeric compounds and their chemical data; S4, testing the dose effect of the monomer compound of step S3 on the inflammation cell model obtained after the treatment of step S2, and determining the key quality markers verified by activity; S5, performing machine learning on the data of the key quality markers verified by activity of step S4, constructing a prediction model, and dividing the Bai Li Cao or its finished product into four grades of special grade, first grade, second grade and qualified grade based on the prediction model; Among them, the special grade: the comprehensive pharmacodynamic index is greater than or equal to P75; the first grade: P75 is greater than the comprehensive pharmacodynamic index and is greater than or equal to P50; The second grade: P50 is greater than the comprehensive pharmacodynamic index and is greater than or equal to P25; the qualified grade: the comprehensive pharmacodynamic index is less than P25.

[0006] The quality evaluation and grading method provided in the application uses high-content imaging technology to perform multi-parameter and quantitative biological effect characterization of Bai Li Cao or its finished product at the cell level, establishes a pharmacodynamic evaluation system, and provides a direct basis for quality grading of wild medicinal material resources; then, a machine learning algorithm is used to deeply analyze the complex correlation between chemical components and pharmacodynamic data, and an intelligent prediction model of “chemical composition-biological effect” is constructed.

[0007] The application can establish a new quality control method directly linked to clinical efficacy, realize rapid quality control by quickly predicting biological titer through chemical components, and ultimately complete the whole-process quality monitoring from intelligent grading of wild resources to efficient testing of finished products.

[0008] As a preferred embodiment of the quality evaluation and grading method of Bai Li Cao or its finished product described in the application, the construction method of the inflammation cell model in step S1 comprises the following steps: The human lung cancer epithelial cells and mouse macrophages are co-cultured, and LPS is used for stimulation to obtain an inflammation cell model.

[0009] Preferably, the human lung cancer epithelial cells include A549 human lung cancer epithelial cells. The mouse macrophages include RAW264.7 mouse macrophages.

[0010] The stimulation concentration of LPS is 0.01-1.0 μg / mL.

[0011] As a preferred embodiment of the quality evaluation and grading method of Bai Li Cao or its finished product described in the application, the method for verifying the inflammation cell model in step S1 comprises: detecting cell viability by CCK-8 method, detecting TNF-α release amount by ELISA, and observing cell morphological changes by high-content bright field imaging.

[0012] As a preferred embodiment of the quality evaluation and grading method of Bai Li Cao or its finished product described in the application, the multi-target immunofluorescence staining in step S2 comprises the following steps: The inflammation cell model verified in step S1 is fixed, permeabilized, blocked, then rabbit anti-NF-κB p65 and mouse anti-TNF-α are added for incubation, then Cy3-goat anti-rabbit, AF647-goat anti-mouse and dye are added for incubation, after incubation, the sample is mounted and stored.

[0013] As a preferred embodiment of the quality evaluation and grading method of the hundred-petal grass or its finished product, in step S2, the features with p-value <0.001 and effect size Cohen's d >2.0 are screened out.

[0014] In the technical solution of the present application, the high-content imaging technology system is applied to the quality evaluation of the hundred-petal grass and its finished product for the first time, realizing the change of the quality control concept from "chemical index" to "biological titer".

[0015] As a preferred embodiment of the quality evaluation and grading method of the hundred-petal grass or its finished product, in step S3, the pretreatment includes the following steps: The hundred-petal grass or its finished product is treated with reflux in methanol, and the supernatant is taken and filtered; and / or, In step S3, the chromatographic column used by the ultra-high performance liquid chromatography tandem quadrupole time-of-flight mass spectrometry is Waters HSS T3, 2.1x100mm, 1.8μm; the mobile phase is 0.1% formic acid water-acetonitrile, and the mass spectrometry condition is ESI positive and negative ion mode, and the scanning range is m / z 50-1000.

[0016] As a preferred embodiment of the quality evaluation and grading method of the hundred-petal grass or its finished product, in step S4, before the dose-effect is performed, the chemical data obtained in step S3 is divided into a training set and a test set, and the key ingredients with a pharmacodynamic correlation degree >0.70 are screened out by gray correlation analysis.

[0017] As a preferred embodiment of the quality evaluation and grading method of the hundred-petal grass or its finished product, in step S4, the active ingredients with EC50 <50 μM and Emax >0.45 are selected as key quality markers; The active ingredients include hundred-petal grassin I, sophorin, eriodictyol, 5-O-feruloyl quinic acid, luteolin, isosophorin, kaempferol-3-O-β-D-glucoside, and abscisic acid.

[0018] As a preferred embodiment of the quality evaluation and grading method of the hundred-petal grass or its finished product, in step S4, the dose of the monomer compound is 1-50 μM.

[0019] As a preferred embodiment of the quality evaluation and grading method of the Herba Centellae or its finished product, in the step S5, the data of the key quality markers verified in the step S4 are divided into a training set and a test set, standardized, and then subjected to model training by using a vector regression SVR algorithm, grid search optimization parameters, to obtain an SVR model.

[0020] The present application establishes a quantitative relationship between chemical composition and biological activity by a machine learning algorithm, and realizes rapid prediction of biological potency based on chemical analysis.

[0021] Compared with the prior art, the present application has the following beneficial effects: The present application provides a quality evaluation and grading method of Herba Centellae or its finished product based on high-content imaging and machine learning. The present application first applies high-content imaging technology system to the quality evaluation of Herba Centellae and its finished product, realizes the change of quality control concept from "chemical indicators" to "biological potency", establishes a quantitative relationship between chemical composition and biological activity by a machine learning algorithm, realizes rapid prediction of biological potency based on chemical analysis, and establishes a complete technical system from wild resource evaluation to finished product quality control, thereby providing a scientific basis for high-quality resource screening and artificial cultivation of Herba Centellae. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 The flow chart of the quality evaluation and grading method of Herba Centellae or its finished product based on high-content imaging and machine learning; Figure 2 The total ion chromatogram of Herba Centellae extract; Figure 3 The possible cleavage pathway of Rosthopsin. DETAILED DESCRIPTION

[0023] For better illustrating the purpose, technical scheme and advantages of the present application, the present application will be further described in combination with the drawings and specific embodiments.

[0024] In the following examples, the experimental methods used are conventional methods unless otherwise specified, and the materials, reagents, etc. used are commercially available unless otherwise specified, and the components of the raw materials used in each parallel experiment are the same.

[0025] In the following examples: Cell lines, reagents and antibodies: Human lung cancer epithelial cells A549, mouse monocyte macrophage RAW264.7 were purchased from American Type Culture Collection; DMEM / F12 (1:1) medium, special grade fetal bovine serum, penicillin-streptomycin solution (100X), trypsin: 0.25% Trypsin-EDTA (1X), phosphate buffer, purchased from GIBCO; dimethyl sulfoxide: cell culture grade DMSO, lipopolysaccharide, 4% paraformaldehyde, Triton X-100, bovine serum albumin, rabbit anti-NF-κB p65 monoclonal antibody, mouse anti-TNF-α monoclonal antibody, Cy3-labeled goat anti-rabbit IgG, Alexa Fluor 647-labeled goat anti-mouse IgG were purchased from Sigma. Positive drug: dexamethasone, purchased from MCE, purity: 99.98%, prepared into 10 mM stock solution with DMSO, stored at -20°C. CCK-8 cell viability assay kit, TNF-α ELISA kit, purchased from Nanjing Jiancheng Bio.

[0026] Medicinal materials and finished products: Wild Saururus chinensis: a total of 50 batches were collected, collected from 10 main producing areas in China (S1: Anhui; S2: Sichuan; S3: Gansu; S4: Shaanxi; S5: Hubei; S6: Hunan; S7: Guizhou; S8: Shanxi; S9: Hebei; S10: Inner Mongolia) from January to March 2025, 5 batches from each producing area. The numbers are BRC-Herb-2022-001 to BRC-Herb-2022-050. Saururus chinensis granules: corresponding to the above 50 batches of medicinal materials, produced and prepared by Jiuhua Huayuan Pharmaceutical Co., Ltd., a total of 50 batches, numbered C1-C50. The correspondence between the finished product and the raw material is recorded in the sample tracking table.

[0027] Instruments and equipment: CO2 incubator (Thermo Scientific), multifunctional enzyme label instrument (Berthold), desktop high-speed refrigerated centrifuge (Eppendorf), automatic cell counter (Countstar), inverted fluorescence microscope (Shanghai Meikai Optoelectronics), high-content imaging system (PerkinElmer, Model Operetta CLS), image analysis software PerkinElmer, Harmony®4.9, ultra-high performance liquid chromatography-quadrupole-time-of-flight mass spectrometry (Waters, ACQUITY UPLC I-Class / Xevo G2-XS QTOF), analytical balance (Mettler Toledo).

[0028] Example 1, a high-content imaging and machine learning-based evaluation and grading method for 100-ribs grass and its finished product quality Figure 1 The present embodiment provides a Saururus chinensis and its finished product quality evaluation and grading method based on high-content imaging and machine learning (such as Figure 2as shown), comprising the following steps: S1, establishment and verification of cell model: (1) Cell recovery and culture: Recovery: Take the cryopreservation tube (A549: cryopreservation number TC-2022-011-P5; RAW264.7: TC-2022-012-P5) from the liquid nitrogen tank immediately and place it in a 37°C constant temperature water bath, continuously gently shake to ensure complete thawing within 60 seconds; after wiping the outer wall of the cryopreservation tube with 75% ethanol, transfer the cell suspension to a 15 mL centrifuge tube containing 5 mL of pre-warmed complete culture medium (preheated in a 15 mL centrifuge tube) in a clean bench, centrifuge at 1000 rpm for 5 minutes.

[0029] Resuspension and culture: carefully discard the supernatant (containing DMSO), add 1 mL of fresh complete culture medium, gently blow the cell pellet with a 1 mL pipette to resuspend the cells, avoiding the generation of air bubbles. Transfer the cell suspension to a T25 cell culture flask, add culture medium to a total volume of 5 mL; gently mix by cross shaking; place in a 37°C, 5% CO2 saturated humidity incubator for static culture.

[0030] Medium change and subculture: the first full medium change is performed after 24 hours to completely remove residual DMSO and dead cells. Change the medium every 2-3 days thereafter. When the cell confluence reaches 85%-90%, subculture. Discard the old medium, rinse once with pre-warmed PBS, add 1 mL of 0.25% trypsin, and digest at 37°C for about 1-2 minutes. After observing the increase in cell gap and rounding under an inverted microscope, immediately add 2 mL of complete culture medium to terminate digestion. Gently blow the bottom of the bottle with a pipette to ensure that all cells are detached to form a single cell suspension. Transfer the cell suspension to a 15 mL centrifuge tube and centrifuge at 1000 rpm for 5 minutes. Discard the supernatant, resuspend the cells with an appropriate amount of complete culture medium, subculture at a ratio of 1:3 to 1:5, and distribute to new T75 culture flasks.

[0031] (2) Inflammatory cell model construction and condition optimization: Co-culture system establishment: well-grown A549 and RAW264.7 cells in the logarithmic growth phase were digested and counted. Using Countstar counting, the A549 cell density was 9.5×10 5 cells / mL, and the RAW264.7 cell density was 1.1×10 6 cells / mL. Mix according to the cell number ratio of A549:RAW264.7=10:1 (for example, take 1 mL of A549 suspension and 90.9 μL of RAW264.7 suspension), and adjust the mixed cell density to 8.0×10 4cells / mL. Cells were seeded in a volume of 100 μL per well in PerkinElmer dedicated 96-well black clear bottom cell culture plates, i.e. 8.0 × 103cells per well. Liquid was added slowly along the well wall to avoid bubble formation and cell washout. Cell plates were incubated in the incubator for 24 hours to allow cells to adhere and establish a stable co-culture system.

[0032] LPS stimulation condition optimization: concentration gradient: LPS concentration gradient was set as: 0 (blank control), 0.01 μg / mL, 0.1 μg / mL, 0.5 μg / mL, 1.0 μg / mL. 6 replicates were set for each concentration. LPS was diluted with serum-free DMEM / F12.

[0033] Time gradient: 1.0 μg / mL LPS was used, and stimulation was performed for 0 hour, 2 hours, 4 hours, 6 hours, 8 hours, and 12 hours, respectively. 6 replicates were set for each time point.

[0034] Evaluation method: cell viability: CCK-8 method was used. After stimulation, 10 μL CCK-8 solution was added to each well, and incubation was continued for 2 hours. The absorbance at 450 nm was measured using a multifunctional enzyme label instrument. Inflammatory factor release: after stimulation, the supernatant of each well was carefully collected, and cell debris was removed by centrifugation at 3000 rpm for 10 minutes. The concentration of TNF-α in the supernatant was detected using a mouse TNF-α ELISA kit according to the instructions. Cell morphology observation (high-content imaging method): in an independent experimental plate, immediately after stimulation, the high-content imaging system was used to collect bright field images of living cells, and the morphological changes of the cells were qualitatively evaluated.

[0035] S2, establishment of high-content analysis method: (1) Standardized process of immunofluorescence staining: Fixation: after the end of cell intervention, the culture medium was aspirated, 100 μL of pre-warmed PBS was added to each well, and gentle washing was performed twice; the PBS was aspirated, and 100 μL of 4% paraformaldehyde was added to each well, and the cells were fixed at room temperature for 15 minutes in the dark.

[0036] Permeabilization: the fixing solution was aspirated, 100 μL of 0.1% Triton X-100 (freshly prepared with PBS) was added to each well, and permeabilization was performed at room temperature for 10 minutes.

[0037] Blocking: the permeabilization solution was aspirated, 100 μL of 5% BSA (dissolved in PBS) blocking solution was added to each well, and blocking was performed at room temperature for 1 hour.

[0038] Primary antibody incubation: Prepare primary antibody working solution (rabbit anti-NF-κB p65 1:500 + mouse anti-TNF-α 1:250, dissolved in 1% BSA / PBS); aspirate the blocking solution, and add 50 μL of the primary antibody working solution to each well; seal the cell plate with sealing film and incubate at 4°C overnight (about 16 hours).

[0039] Secondary antibody and dye incubation: recover the primary antibody (can be reused 2-3 times), and rinse with PBST for 3 times. Prepare a mixed working solution of secondary antibody and dye (Cy3-goat anti-rabbit 1:1000 + AF647-goat anti-mouse 1:1000 + MitoTracker Green 200 nM + Hoechst 33342 5 μg / mL, dissolved in 1% BSA / PBS). Add 50 μL of the mixed working solution to each well, and incubate at room temperature for 1 hour in the dark.

[0040] Mounting and storage: after incubation, rinse with PBST for 3 times. Add 100 μL of PBS to each well to prevent drying. Immediately perform image acquisition, or wrap the cell plate with aluminum foil and store in a 4°C refrigerator, and complete the acquisition within 24 hours.

[0041] (2) High-content image acquisition and analysis method establishment: Image acquisition parameters: objective lens: 20x dry lens (numerical aperture NA=0.4) camera: sCMOS, imaging mode: fluorescence; Channel and exposure time (determined by pre-experiment optimization): Channel 1 (Hoechst): Ex 355-385nm, Em 430-500nm, exposure time 50 ms, laser power 25%.

[0042] Channel 2 (MitoTracker Green): Ex 460-490nm, Em 500-550nm, exposure time 200 ms, laser power 30%.

[0043] Channel 3 (Cy3): Ex 520-550nm, Em 560-630nm, exposure time 400 ms, laser power 35%.

[0044] Channel 4 (Alexa Fluor 647): Ex 600-640nm, Em 650-760nm, exposure time 300 ms, laser power 40%.

[0045] Number of fields of view collected per well: 9 non-overlapping fields of view to ensure statistical significance.

[0046] Autofocus: Laser-based fast focusing using Hoechst channel (focusing height: 110 μm).

[0047] (3) Harmony software analysis pipeline: “Find Nuclei” (identify nuclei): Method: Method B (based on intensity threshold and shape); Channel: Channel 1 (Hoechst); Parameters: Approximate Diameter: 10 μm, Splitting Coefficient: 0.4, Background Correction: ON, Threshold: 0.2. Output: All nuclei are identified as the base regions for subsequent analysis. The software will mark the nuclei that are poorly segmented or overlapped, and manual review is needed to confirm the segmentation effect.

[0048] “Find Cytoplasm” (identify cytoplasm): Method: Method B Channel: Channel 2 (MitoTracker Green) / / use mitochondrial signal to outline the cytoplasm profile; Parameters: Individual Threshold: 0.3, Cell Individualization: 0.5, Cytoplasm Padding: 2 μm. Output: The complete single cell region is expanded based on the nucleus as the seed.

[0049] “Calculate Intensity Properties” (calculate intensity properties): For each nucleus region and cytoplasm region, the following intensity characteristics are calculated: Intensity Mean (average fluorescence intensity); Intensity Max (maximum fluorescence intensity); Intensity Min (minimum fluorescence intensity); Intensity StdDev (standard deviation of fluorescence intensity); Intensity Total (fluorescence intensity integral) “Calculate Morphology Properties” (calculate morphological properties): For each nucleus and cell region, calculate: Area (area, μm²); Roundness (roundness, 1 for perfect circle); Eccentricity (eccentricity, 0 for circular, 1 for linear); Width and Height (bounding box width and height) “Add Column” (add custom column): Create the key derived feature NFkB_Nuc_Cyt_Ratio: Formula: $im.Cy3.Mean.Nuclei / $im.Cy3.Mean.Cytoplasm; This ratio quantifies the degree of translocation of NF-κB p65 from the cytoplasm to the nucleus and is a core indicator of inflammatory activation.

[0050] “Filter Populations”: Set a filter to exclude abnormal cells with Cell_Area < 100 μm² or > 3000 μm², as well as dying / dead cells with Nucleus_Roundness < 0.5, to ensure that the analysis is performed on a healthy and intact cell population.

[0051] (4) Feature extraction and screening: Imaging analysis was performed on the blank control group and the model control group (n=18 wells each), totaling over 250,000 single cells. Twenty-eight raw feature values ​​were extracted from each cell. Intergroup comparisons were performed using a T-test, and features with p-value < 0.001 and effect size (Cohen's d) > 2.0 were selected.

[0052] (5) Construct the formula for the comprehensive efficacy index: Comprehensive efficacy index = (IRR_F1 * W1) + (IRR_F2 * W2) + (IRR_F3 * W3) + (IRR_F4 * W4).

[0053] Wherein, IRR is the inhibition rate / recovery rate, and W is the weight. W1=0.4 (NF-κB pathway is the core hub); W2=0.3 (TNF-α is the core effector); W3=0.2 (cell morphology is an important auxiliary indicator); W4=0.1 (mitochondrial function is an indicator of cell state).

[0054] S3. Chemical fingerprinting of raw medicinal materials and finished products: (1) Sample pretreatment: Accurately weigh 1.000 g of Bai Rui herbal powder (passed through a No. 3 sieve) or the contents of Bai Rui granules and place it in a 50 mL stoppered conical flask. Accurately add 50.00 mL of methanol solution, weigh it (record as W1), and reflux for 30 minutes; remove it, cool it to room temperature, weigh it again (record as W2), and replenish the lost weight (W1-W2) with 70% methanol, shake well, and let it stand. Take about 1.5 mL of the supernatant, filter it through a 0.22 μm microporous membrane, and take the filtrate as the test solution, and transfer it to a 2 mL LC vial. At the same time, prepare a sample blank (methanol) and a quality control sample (mix all samples in equal amounts).

[0055] (2) UPLC-QTOF-MS analysis conditions: Chromatographic conditions: Column: Waters ACQUITY UPLC HSS T3 (2.1 mm x 100 mm, 1.8 μm).

[0056] Mobile phase: A: 0.1% formic acid in water; B: acetonitrile; Column temperature: 45 °C; Flow rate: 0.40 mL / min; Injection volume: 2.0 μL.

[0057] The gradient elution program is shown in Table 1.

[0058] Table 1 Mass spectrometry conditions: Ion source: ESI (electrospray ion source); Scan mode: positive and negative ion modes were collected respectively; Data acquisition mode: MS^E (Centroid), low collision energy 6 eV, high collision energy ramp 20 - 40 eV. Scan range: m / z 50 - 1000; Scan time: 0.2 s; Capillary voltage: positive ion 2.5 kV, negative ion 2.2 kV; Cone voltage: 40 V; Ion source temperature: 150 °C.

[0059] (3) Data processing and compound identification: Data processing: The raw data (.raw file) was imported into Waters' Progenesis QI software (v3.0). The quality control sample was selected for peak identification and alignment. The parameters were set: retention time tolerance 0.10 min, mass accuracy tolerance 5.0 ppm. The software automatically performed peak identification (minimum peak intensity 500 counts), retention time correction (maximum correction 1.0 min) and peak alignment. All identified peaks were normalized by the total abundance of all compounds, and then a two-dimensional data matrix containing sample name, compound identification (retention time _ mass-to-charge ratio) and peak area was exported (.csv format).

[0060] Compound identification: Primary identification: The accurate molecular weight of the compound (error <5.0 ppm) was compared with the self-built database of chemical components of Pogostemonis Herba, HMDB and PubChem database.

[0061] Secondary identification: The secondary mass spectrum fragment ions under high energy scanning were compared with the fragment information in the database, literature reported data or standard spectrum to confirm.

[0062] Identification level: Follow Schymanski standard: Level 1 (confirmed by standard), Level 2 (confirmed by literature or database spectra), Level 3 (speculated candidate structure by mass fragmentation). Most of this study is Level 2 and Level 3.

[0063] S4, Discovery and validation of key quality markers: (1) Data preparation and division: Integrate the chemical data (X, normalized peak area of 53 chromatographic peaks) and corresponding comprehensive efficacy index (Y) of 50 batches of Pterocephalus heterophyllus into a Pandas DataFrame. Use the sklearn.model_selection.train_test_split function of Python to randomly divide it into a training set (35 batches) and a test set (15 batches) in a ratio of 7:3, and set random_state=42 to ensure that the results are reproducible.

[0064] (2) Grey correlation analysis: Use Python to write a grey correlation analysis algorithm.

[0065] The specific steps include: a. Dimensionless processing (initial value transformation); b. Calculate the absolute difference between each chemical sequence and the efficacy sequence; c. Find the minimum and maximum difference; d. Calculate the correlation coefficient (resolution coefficient p is 0.5) of each sample point; e. Take the average of the correlation coefficient of each chromatographic peak, i.e. the grey correlation degree of the peak. Set the correlation degree threshold > 0.70, and screen out the key variables for subsequent activity verification.

[0066] (3) Activity verification: Standard purchase: Purchase commercial standards such as astragalin, luteolin, kaempferol-3-O-β-D-glucoside, ferulic acid, vanillic acid, abscisic acid, isolariciresinol, and ethyl trans-p-hydroxycinnamate from Shanghai Yuanye Biotechnology Co., Ltd. The purity was detected by HPLC to be ≥98%.

[0067] Isolation and purification of monomeric compounds: For compounds such as Pterocephalus I, Cimicifugan, Isoastragalin, and 5-O-Feruloylquinic Acid, which are difficult to purchase commercially, directional separation was performed from the total extract of Pterocephalus heterophyllus by semi-preparative high-performance liquid chromatography.

[0068] Purification condition: Agilent 1260 Infinity II semi-preparative system, XBridge Prep C18 column (10 mm x 250 mm, 5 μm), mobile phase of methanol-water, gradient elution, flow rate 3.0 mL / min, UV detector. The target fraction was collected, the solvent was removed by rotary evaporation, and white or off-white powder was obtained by freeze-drying.

[0069] Structure confirmation: The isolated pure compound was analyzed by nuclear magnetic resonance spectrum (NMR, Bruker AVANCE NEO500 MHz spectrometer) for 1H-NMR and 13C-NMR analysis, and the chemical structure was confirmed by comparison with literature data.

[0070] Cellular pharmacodynamic verification experiment: Cell model: A549 / RAW264.7 (10:1) co-culture inflammation model was established.

[0071] Grouping and administration: blank control group: no LPS and drug; model control group: containing LPS (1 μg / mL), no drug; positive control group: dexamethasone (10 μM) + LPS; monomer compound group: the key ingredients obtained in the previous step were respectively prepared into 100 mM stock solution with DMSO, and when used, they were diluted with serum-free medium. Each compound was set at 5 concentration gradients: 1 μM, 5 μM, 10 μM, 25 μM, 50 μM. Pre-protection was performed 2 hours before the addition of LPS stimulation.

[0072] Detection and analysis: After the intervention, the established high-content analysis method was strictly followed to operate, to quantify the effect of each compound on each core feature, and to calculate the comprehensive pharmacodynamic index at each concentration.

[0073] Data processing: GraphPad Prism 9.0 software was used to draw the dose-effect curve, and the half effective concentration (EC50) and maximum effect (Emax) of each compound were calculated.

[0074] S5, machine learning prediction model construction: (1) Data preparation: The verified active ingredient content data and pharmacodynamic index of 50 batches of medicinal materials were divided into training set and test set according to 7:3. Support vector regression modeling.

[0075] (2) Model training: Library: from sklearn.svm import SVR; Kernel function: Radial Basis Function (kernel='rbf'); Data standardization: Z-score standardization (mean=0, standard deviation=1) on X data of training set and test set using sklearn.preprocessing.StandardScaler; Parameter optimization: 5-fold cross-validation grid search using GridSearch CV, parameter range: 'C': [0.1, 1, 10, 100], 'gamma': [0.001, 0.01, 0.1, 1]; Model training: Train the final SVR model on the standardized training set using the optimal parameters; Model evaluation: Predict the standardized test set using the trained model, calculate R² and RMSE (3) Grade division: Based on the distribution of the predicted values of the training set, calculate its quartiles.

[0076] Division criteria: Super (excellent): Comprehensive efficacy index ≥ P75; First (good): P75 > Comprehensive efficacy index ≥ P50; Second (medium): P50 > Comprehensive efficacy index ≥ P25 Qualified (poor): Comprehensive efficacy index < P25.

[0077] (4) Finished product inspection application Input the chemical data (only containing key variables) of 50 batches of BR granular finished products into the trained SVR model to predict their efficacy indexes and grades. Compare the predicted results with the "expected grade" based on the actual efficacy of their raw medicinal materials, and calculate the coincidence rate.

[0078] Experimental results: S1, establishment and verification results of cell model: (1) Cell recovery and culture maintenance: After A549 and RAW264.7 cells were recovered, they all adhered within 24 hours. A549 cells showed typical epithelial-like, pavement stone-like morphology with clear boundaries; RAW264.7 cells were round, some adhered, some grew in suspension, and had strong refractive properties. After two passages, the growth of both cells was stable, the morphology was uniform, and there was no sign of microbial contamination. The survival rate of cells before freezing was higher than 95%, and after recovery and subculture, the cell state met the experimental requirements.

[0079] (2) Inflammatory cell model construction and condition optimization: After 24 hours of cell inoculation, under the inverted microscope, it was observed that the cells completely adhered and distributed uniformly. A549 cells formed a monolayer, and RAW264.7 cells adhered to the A549 cell layer or in the interstitial space, and the coexistence of the two cells was good without mutual inhibition.

[0080] The effect of LPS concentration gradient (stimulation for 6h) on cell viability (CCK-8 method) is shown in Table 2.

[0081] The effect of LPS concentration gradient (stimulation for 6h) on TNF-α release (ELISA method) is shown in Table 3.

[0082] The effect of LPS time gradient (1.0 μg / mL) on TNF-α release (ELISA method) is shown in Table 4.

[0083] Table 2 After 6 hours of stimulation by 1.0 μg / mL LPS, the relative viability of the cells remained above 90%, with no significant difference from the blank control group (p>0.05), indicating that LPS at this concentration had no obvious cytotoxicity on the co-culture system.

[0084] Table 3 Table 4 After 6 hours of stimulation by LPS, the amount of TNF-α release reached a plateau, with a concentration of 1850 pg / mL, which was 21.8 times that of the blank control group, with a highly significant difference (p<0.001). This condition could induce a strong and stable inflammatory response.

[0085] High-content brightfield images clearly showed that, compared with the blank control group, in the model group stimulated by 1.0 μg / mL LPS for 6 hours, the volume of RAW264.7 cells significantly increased, the cell body stretched, and the pseudopodia increased and lengthened, showing a typical activated state; the morphological changes of A549 cells were relatively small, but the intercellular connections became slightly loose.

[0086] Conclusion: Through the comprehensive analysis of cell viability, inflammatory factor release, and cell morphology, the optimal modeling condition was successfully determined as LPS 1.0 μg / mL, stimulated for 6 hours. The model under this condition is stable and reliable, and is suitable for the subsequent anti-inflammatory efficacy evaluation of P. centipedae.

[0087] S2, Results of the establishment of high-content analysis method: (1) Immunofluorescence staining: All samples completed the standard immunofluorescence staining process. After staining, the cell plate was previewed under the Operetta CLS system, and it could be seen that: Channel 1 (Hoechst): The nucleus exhibits clear and bright blue fluorescence with a complete outline, uniform distribution, and no excessive aggregation or fragmentation. Channel 2 (MitoTracker Green): The mitochondrial network in the cytoplasm shows clear green fluorescence, successfully outlining the cytoplasm. Channel 3 (Cy3): NF-κB p65 protein exhibits red fluorescence. In the control group, fluorescence is mainly distributed in the cytoplasm; in the model group, fluorescence is strongly concentrated in the nucleus. Channel 4 (Alexa Fluor 647): TNF-α protein exhibits far-red fluorescence. The signal in the control group is weak; the signal in the model group is significantly enhanced. Crosstalk between channels is negligible. The staining background is clean, and the signal-to-noise ratio is high. Multi-target immunofluorescence staining was successful, meeting the requirements for high-content imaging.

[0088] (2) Establishment of high-content image acquisition and analysis methodology: Image Acquisition: Following the set parameters, the system successfully acquired four-channel images from nine fields of view for all samples. Images were accurately focused, with moderate fluorescence signal intensity, and no overexposure or weak signal. Approximately 43,200 high-quality fluorescence images were obtained (100 samples * 9 fields of view / sample * 4 channels / field of view + control).

[0089] Image analysis workflow results: "Find Nuclei": The software successfully identified the vast majority of cell nuclei with accurate segmentation. After manual review, approximately 95% of the cell nuclei were correctly segmented, and the few overlapping nuclei were improved by adjusting the Splitting Coefficient parameter.

[0090] “Find Cytoplasm”: Based on the MitoTracker Green signal, the cytoplasmic region was successfully identified and expanded, and it matched well with the nuclear region.

[0091] Feature extraction: The process successfully extracted 28 preset morphological and intensity feature values ​​from each cell and correctly calculated the derived feature NFkB_Nuc_Cyt_Ratio.

[0092] Cell screening: The filter successfully excluded approximately 3% of abnormal cells (too small or too large in area, or with poor nuclear roundness), ensuring the quality of subsequent analysis data.

[0093] (3) Screening of significant features and formulation of comprehensive efficacy index: Feature extraction and screening results: A total of 258,642 valid single-cell data points from the blank control group and the model control group were analyzed. After T-test comparison, the p-values ​​of all four features were less than 0.0001, and the effect size (Cohen's d) far exceeded 2.0.

[0094] The specific data for the four core features finally determined between the two groups are shown in Table 5.

[0095] Table 5 (4) Calculation of comprehensive efficacy index: Based on the established formula, the overall efficacy index of all administered samples was calculated. The overall efficacy index ranged from 0.12 to 0.81 for 50 batches of BR herb extract. The overall efficacy index ranged from 0.18 to 0.78 for 50 batches of BR granules.

[0096] Example of representative sample calculation (BR-03, 200 μg / mL): IRR_F1 = 1 - (0.62 / 2.65) = 0.766; IRR_F2 = 1 - (2800 / 12900) =0.783; IRR_F3 = 1 - (520 / 945) = 0.450; IRR_F4 = (4800 / 3380) - 1 = 0.420; Overall efficacy index = (0.766 * 0.4) + (0.783 * 0.3) + (0.450 * 0.2) + (0.420 * 0.1) = 0.306 + 0.235 + 0.090 + 0.042 = 0.673.

[0097] S3. Results of chemical fingerprinting of raw medicinal materials and finished products: (1) Sample pretreatment and UPLC-QTOF-MS analysis: All 100 samples (50 batches of herb + 50 batches of finished product) underwent pretreatment and were analyzed. The system showed good adaptability, with high chromatographic overlap in the quality control samples. The relative standard deviations (RSDs) of retention time and peak intensity were less than 0.2% and 8.0%, respectively, indicating that the entire analytical process was stable and reliable.

[0098] (2) Data processing and compound identification: Peak identification and alignment results: Progenesis QI software successfully identified and aligned 53 common chromatographic peaks from all samples. Compound identification results: Through the comparison of primary and secondary mass spectrometry data, the structures of 39 compounds were identified or presumed (including flavonoids, phenolic acids, lignans, terpenes, etc.). The error between the accurate mass number of all identified compounds and the theoretical value was less than 5.0 ppm. The secondary mass spectrometry fragments matched the structural formula reasonably.

[0099] As shown in Figure 3 , taking peak 20 (eriodictyol) as an example, the molecular formula is C 15 H 12 O6, and the quasi-molecular ion peak in the negative ion mode is m / z 287.0559 [M-H] - . The obvious fragment ions in the secondary mass spectrometry diagram are m / z 135.0439, 151.0025, and 269.0454. According to their elemental composition and related literature, it is presumed that the eriodictyol cleavage is through the removal of a molecule of H2O to obtain the fragment ion of m / z 269.0454 [M-H-H2O]-, and the flavone retro-Diels-Alder fragmentation (RDA) occurs to obtain the fragment ions of m / z 135.0439 [C7H4O3] - and m / z 151.0025 [C8H8O3] - , and the possible cleavage pathway is shown in ​ .

[0100] S4, discovery and verification of key mass markers: (1) Data preparation and division: The data of 50 batches of P. chinensis were randomly divided into a training set (35 batches) and a test set (15 batches). The pharmacodynamic index range of the training set was 0.15 ~ 0.79, with an average of 0.49; the pharmacodynamic index range of the test set was 0.18 ~ 0.81, with an average of 0.47. The data distribution of the two groups was balanced.

[0101] (2) Gray correlation analysis results: The grey correlation degrees of 50 chromatographic peaks and comprehensive pharmacodynamic indexes were calculated. Twelve key components (correlation degree > 0.70) were screened out: centaureidin I (peak 8): 0.92; isohydangin (peak X): 0.77; apigenin (peak 6): 0.79; 5-O-feruloylquinic acid (peak 14): 0.82; kaempferol-3-O-β-D-glucoside (peak 21): 0.75; abscisic acid (peak 9): 0.71; trans-p-hydroxycinnamic acid ethyl ester (peak 12): 0.70; (3) Anti-inflammatory activity verification of monomer compounds: The dose-effect curves of 12 monomer compounds were successfully fitted.

[0102] The activity verification results are shown in Table 6.

[0103] Table 6 Activity confirmation and cause-effect relationship establishment: Among the 12 candidate components, 8 compounds (centaureidin I, isohydangin, apigenin, 5-O-feruloylquinic acid, kaempferol-3-O-β-D-glucoside, abscisic acid) showed clear concentration-dependent anti-inflammatory activity (EC50<50 μM). Their pure products reproduced the pharmacodynamic effect of high-quality centaury extract at the cellular level, confirming that they are the effective material basis for the anti-inflammatory effect of centaury and establishing a solid cause-effect relationship.

[0104] Final determination of key quality markers: Compounds with EC50<50 μM and Emax>0.45 were selected as the final key quality markers. Accordingly, the following 8 components were finally determined: centaureidin I, isohydangin, apigenin, 5-O-feruloylquinic acid, kaempferol-3-O-β-D-glucoside, abscisic acid.

[0105] Exclusion of ineffective components: Abietic acid, isolariciresinol, vanillic acid, and trans-p-hydroxycinnamic acid ethyl ester have weak activity at the tested concentrations. Although they have statistical correlation with pharmacodynamic effects in medicinal materials (possibly as biosynthetic precursors or with a symbiotic relationship with other components), they are not direct main active contributors, and therefore are excluded when building the final quality control model.

[0106] S6, machine learning prediction model construction: (1) Data preparation and preprocessing: The content data of 8 key quality markers (Bai'ercao I, Hibiscus syriacus, St. John's wort, 5-O-feruloyl quinic acid, luteolin, isohispidin, kaempferol-3-O-β-D-glucoside, abscisic acid) in 50 batches of Bai'ercao medicinal materials were used as characteristic variables (X), and the corresponding comprehensive pharmacodynamic index was used as the target variable (Y) to construct the original data set. The pandas library of Python was used for data integration to form a 50x9 two-dimensional data matrix (8 characteristics + 1 target variable).

[0107] The train-test-split function of the scikit-learn library was used to set the random seed random-state=42, and the data set was divided into a training set (35 batches) and a test set (15 batches) at a ratio of 7:3. The characteristic variables were subjected to Z-score standardization to eliminate the influence of dimension and ensure the stability of model training.

[0108] (2) Model training and optimization: The support vector regression (SVR) algorithm was selected, and the radial basis function (RBF) was used as the kernel function. The GridSearchCV function of scikit-learn was used for 5-fold cross-validation grid search, and the parameter search range included: regularization parameter C: [0.1, 1, 10, 100]; Kernel function coefficient gamma: [0.001, 0.01, 0.1, 1]; The optimal parameter combination determined by grid search is C=10, gamma=0.1. The optimal parameters were used to train the final SVR model on the standardized training set, and a nonlinear mapping relationship between the content of chemical components and the pharmacodynamic index was established.

[0109] (3) Model performance evaluation: The coefficient of determination (R²) and the root mean square error (RMSE) were used as model performance evaluation indexes. On the training set, the model performed excellently, with R² reaching 0.94 and RMSE being 0.032; on the test set, the model showed good generalization ability, with R² being 0.90 and RMSE being 0.041. These indicators show that the constructed SVR model can accurately predict the biological potency of Bai'ercao.

[0110] (4) Establishment of grade division standard: Based on the distribution characteristics of the predicted values of the training set, the quartiles (the data in the training set are arranged from low to high, and the value corresponding to 25% is P25, and the median is P50) were calculated as the basis for grade division: Superior (excellent): predicted pharmacodynamic index ≥ P75 (≥ 0.62); First class (good): P75 > predicted pharmacodynamic index ≥ P50 (0.47-0.62); Secondary (medium): P50 > predicted efficacy index ≥ P25 (0.32-0.47); Pass (poor): predicted efficacy index < P25 (<0.32).

[0111] The classification criteria ensure the distinction between each level, while considering the feasibility of practical application.

[0112] (5) Model application and verification: Intelligent grading results of BaiRuiCao resources: Superior: 11 batches (22%), such as BR-03, BR-13, BR-17 (all from S3, S7 producing areas). First-class: 17 batches (34%); Secondary: 15 batches (30%); Pass: 7 batches (14%), such as BR-29, BR-44 (all from S9, S10 producing areas).

[0113] The coincidence rate with the measured classification is 96% (48 / 50). Two batches of boundary samples (BR-19, BR-38) have a first-class difference.

[0114] Fast quality control results of BaiRui granules: The predicted coincidence rate is 94% (47 / 50) of the product predicted grade consistent with the expected grade based on its raw materials.

[0115] Abnormal batch analysis: 3 batches of non-conforming products are all predicted by the model to be a lower grade than expected, and the production records are found to have process parameter fluctuations.

[0116] Conclusion: The model is successfully applied to product quality control and can effectively reflect the impact of process fluctuations on product biological activity.

[0117] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and do not limit the scope of protection of the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the essence and scope of the technical solutions of the present application.

Claims

1. A method for quality evaluation and grading of Centella asiatica or its finished product, characterized in that, Includes the following steps: S1. Establish and validate inflammatory cell models; S2. The inflammatory cell model validated in step S1 was subjected to multi-target immunofluorescence staining, and then the stained cells were subjected to feature extraction and screening using high-content image acquisition and analysis methodology. The screened features included NF-κB nuclear-cytoplasmic ratio, TNF-α expression intensity, cell area, and mitochondrial membrane potential. A comprehensive drug efficacy index formula was constructed. The overall efficacy index is calculated as follows: (IRR_NF-κB × 0.4) + (IRR_TNF-α × 0.3) + (IRR_cell area × 0.2) + (IRR_mitochondria × 0.1), where IRR is the inhibition rate / recovery rate. S3. Take Centella asiatica or its finished product for pretreatment, and then use ultra-high performance liquid chromatography-tandem quadrupole time-of-flight mass spectrometry to process the pretreated Centella asiatica or its finished product to obtain monomeric compounds and their chemical data. S4. Test the dose-effect of the monomeric compound from step S3 on the inflammatory cell model obtained after treatment in step S2 to identify key quality markers that have been validated for activity. S5. Perform machine learning on the data of key quality markers that have undergone activity verification in step S4, build a prediction model, and classify Bai Rui Cao or its finished products into four grades: premium, first grade, second grade, and qualified based on the prediction model. Among them, the special grade is defined as having a comprehensive efficacy index ≥ P75; the first grade is defined as having a comprehensive efficacy index ≥ P50. Level 2: P50 > Overall efficacy index ≥ P25; Pass: Overall efficacy index <P25。 2. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S1, the method for constructing the inflammatory cell model includes the following steps: Human lung cancer epithelial cells and mouse macrophages were co-cultured and stimulated with LPS to obtain an inflammatory cell model.

3. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S1, the methods for verifying the inflammatory cell model include: detecting cell viability using the CCK-8 assay, detecting TNF-α release using ELISA, and observing cell morphological changes using high-content bright-field imaging.

4. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S2, multi-target immunofluorescence staining includes the following steps: The inflammatory cell model validated in step S1 was fixed, permeabilized, and blocked. Then, rabbit anti-NF-κB p65 and mouse anti-TNF-α were added for incubation. Cy3-goat anti-rabbit, AF647-goat anti-mouse and dye were added for further incubation. After incubation, the slides were mounted and stored.

5. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S2, features with p-value < 0.001 and effect size Cohen's d > 2.0 are selected.

6. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S3, the preprocessing includes the following steps: The herb or its finished product is refluxed in methanol, and the supernatant is collected and filtered. And / or, In step S3, the ultra-high performance liquid chromatography-tandem quadrupole time-of-flight mass spectrometry uses a Waters HSST3 column, 2.1×100mm, 1.8μm; the mobile phase is 0.1% formic acid-acetonitrile; the mass spectrometry conditions are ESI positive and negative ion mode, and the scan range is m / z 50-1000.

7. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S4, before the dose-effect analysis, the chemical data obtained in step S3 is divided into a training set and a test set, and key components with a correlation degree greater than 0.70 with the efficacy are screened by grey relational analysis.

8. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S4, active ingredients with EC50 < 50 μM and Emax > 0.45 are selected as key quality markers; The active ingredients include gentianin I, gentianin, gentianin, 5-O-feruloylquinic acid, luteolin, isorientin, kaempferol-3-O-β-D-glucoside, and abscisic acid.

9. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S4, the dosage of the monomeric compound is 1~50 μM.

10. The method for quality evaluation and grading of Centella asiatica or its finished product as described in claim 1, characterized in that, In step S5, the data of the key quality markers that have undergone activity verification in step S4 are divided into training set and test set, and the data is standardized. Then, the vector regression (SVR) algorithm is used to optimize the parameters and train the model to obtain the SVR model.

Images