Method for rapidly identifying PBAT degrading bacteria
Through the combination of hyperspectral imaging technology and deep learning models, a prediction model based on LSTM was established, which solved the problem of time-consuming and cost-effective traditional microbial screening methods, and achieved rapid and accurate screening of PBAT degraded bacteria, improving screening efficiency and comprehensiveness. .
Patent Information
- Application Number
- CN202411260237.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2024-12-03
AI Technical Summary
Traditional microbial screening methods are time-consuming, cost-effective, and are susceptible to human factors. It is difficult to screen out all potentially valuable strains and cannot effectively deal with the problem of microplastic pollution.
Hyperspectral imaging technology combined with deep learning models is used to quickly capture the multi-dimensional spectral information of the sample through PBAT-glucose co-metabolizing solid culture medium, and establish a prediction model based on LSTM to achieve rapid screening of PBAT-degraded bacteria.
It significantly shortens the screening time, reduces the cost of experimental materials and labor, improves the comprehensiveness and accuracy of screening, discovers more potential microorganisms, and overcomes the limitations of traditional methods.
Smart Images

Figure CN119067958A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the fields of chemometrics and microbiology, and in particular relates to a method for rapid identification of PBAT (polybutylene adipate terephthalate) degrading bacteria. Background Art
[0002] Microplastic pollution has become a global environmental problem. Traditional physical or chemical treatment methods usually require high energy consumption, high equipment and operating costs, and may also produce secondary pollution or harmful byproducts. To address this problem, many fields have begun to replace non-degradable plastics with biodegradable plastics in order to reduce long-term environmental burdens and alleviate microplastic pollution through microbial decomposition in the environment. Among them, polybutylene adipate terephthalate (PBAT) is a commonly used biodegradable plastic. Biological treatment methods use natural processes for degradation, which is in line with the self-repair mechanism of the ecosystem and helps to achieve sustainable environmental management. Therefore, identifying and screening microorganisms with plastic degradation capabilities is a key step in addressing this problem. Traditional microbial screening methods usually rely on long-term cultivation and observation, which takes several months to complete, resulting in low screening efficiency. In addition, they rely on manual operations, consume a lot of culture media and reagents, and have a high overall cost. Traditional methods have limitations and are easily affected by human factors. Efficiency and accuracy may be limited; all strains with potential value may not be screened. In addition, the presence of abundant carbon and energy sources in the natural environment makes cometabolic degradation a promising solution. During co-metabolism, microorganisms speed up the biodegradation process by using additional nutrients to help break down the carbon or energy needed for the process. Summary of the invention
[0003] To solve the above problems, the present invention proposes a method for quickly screening PBAT-degrading bacteria from a PBAT-glucose co-metabolism solid culture medium. Combining hyperspectral imaging technology, chemometrics and deep learning models, this method can quickly and non-destructively capture the multidimensional spectral information of the sample, thereby efficiently and accurately screening strains. This not only significantly shortens the screening time, reduces experimental materials and labor costs, and is environmentally friendly, but also overcomes the limitations of traditional methods in dealing with difficult-to-culture strains, and improves the comprehensiveness and accuracy of the screening.
[0004] The technical solution of the present invention is as follows:
[0005] A method for rapid identification of PBAT-degrading bacteria, the steps are as follows:
[0006] Step 1. Modeling sample preparation: Prepare PBAT-glucose solid culture medium modeling samples with different concentrations and collect hyperspectral images of the samples. Separate the solid culture medium and the culture dish before collection, place the solid culture medium on a polytetrafluoroethylene white plate, keep the bottom fitted, and place it in the center of the stage, keeping the placement consistent.
[0007] Step 2: Collect average spectrum: Select the region of interest (ROI), divide the ROI into 10×10 pixel points, and extract the average spectrum of the modeled samples in each small area. In order to improve the performance of the prediction model and suppress the spectral fluctuations affected by noise, background interference, and instrument errors, the ROI area is marked in spatial order so that the spatial distribution of the light source is associated with the ROI position.
[0008] Step 3: Establish a prediction model: Divide the extracted average spectrum into a training set and a validation set, and use a deep learning prediction model based on sequence (LSTM) features, including an input layer, multiple LSTM units, a fully connected layer, and an output layer. Use the average spectrum and position sequence as input variables at the same time, and the PBAT concentration on the culture medium as the output variable to establish a prediction model between the average spectrum and the PBAT concentration; input the input variables into the input layer, and then pass them to multiple LSTM units in turn to capture long-term dependencies, and then connect to the fully connected layer, and finally obtain the final prediction results through the output layer.
[0009] Step 4: Select the optimal prediction model: Validate the trained prediction model, and use the correlation coefficient (R) and root mean square error (RMSE) of the validation set as key performance indicators to comprehensively evaluate the accuracy and reliability of the model. The larger the R of the prediction model, the closer its value is to 1; the smaller the RMSE, the closer its value is to 0, indicating that the model has good prediction ability and robustness. The prediction model with the largest R is selected for subsequent actual sample prediction.
[0010] Step 5: Pretreatment of the actual sample to be predicted: dilute and spread the microorganisms enriched and cultured from the test environment on a solid culture medium for culture; after the culture is completed, gently remove the colonies on the surface of the solid culture medium, collect a hyperspectral image of the removal position, and preserve the removed colonies for subsequent screening and identification.
[0011] Step 6. Predict the actual sample: In order to suppress the influence of spatial fluctuation of the spectrum on the prediction model, select a reference position, which is located in the blank position below the colony, within 0-5 pixels from the colony position, and the area size of this area is consistent with the removal of the colony; calculate the predicted difference of PBAT concentration between the spectrum of the sterilization position and the reference position through the prediction model, and indirectly infer whether the unknown colony on the solid culture medium has a degradation effect on PBAT based on the positive and negative and size of the difference; if the difference is less than 0, it means that the colony corresponding to the area is PBAT-degrading bacteria.
[0012] Beneficial effects of the present invention: Compared with traditional methods, the present invention uses hyperspectral technology combined with deep learning model methods to shorten the identification time in strain identification, reduce experimental materials and labor costs, and improve the comprehensiveness and accuracy of screening to discover more potential microorganisms. BRIEF DESCRIPTION OF THE DRAWINGS
[0013] Figure 1 It is a schematic diagram of the ROI position sequence.
[0014] Figure 2 It is the correlation coefficient of AttentionLSTM, BiLSTM, and CNNLSTM.
[0015] Figure 3 It is the Euclidean distance of the four positions above, below, left and right from the center position.
[0016] Figure 4 It is the PBAT degradation value (PBAT0, PBAT1, PBAT2, PBAT3).
[0017] Figure 5 This is the predicted difference graph of the three bacteria. DETAILED DESCRIPTION
[0018] The specific implementation of the present invention is described in detail below in combination with the technical scheme and the accompanying drawings.
[0019] A method for rapid identification of PBAT-degrading bacteria of the present invention comprises the following specific steps:
[0020] Step 1: Preparation of PBAT emulsion: Dissolve 2g PBAT in 50mL dichloromethane and add 2mL sodium N-lauroylamino acid. Add deionized water to the mixture and stir for two hours at 80°C in a fume hood to ensure that the dichloromethane is completely evaporated. Add deionized water to 100mL to prepare a 2% PBAT emulsion for use.
[0021] Step 2: Preparation of solid culture medium samples: Prepare a batch of solid culture medium with different concentrations of PBAT and glucose as the main nutrient sources. The PBAT concentration levels in the culture medium are 50, 100, 150, 200, 250, 300, 400 and 500 mg / L, and the glucose concentration level ranges from 100 to 900 mg / L. The culture medium contains the following ingredients (per liter): 30 g of artificial sea salt, an appropriate amount of glucose and PBAT emulsion, and 20 g of agar powder.
[0022] Step 3: Collect the average spectrum. The HSI-NIR system was used to collect the average spectrum data in the range of 900-1700nm (172 spectral bands). The solid culture medium was gently separated from the culture dish, placed on a polytetrafluoroethylene plate and placed in the center of the instrument stage and kept in the same position to avoid the influence of uneven instrument light during the collection process. In order to reduce the influence of the dark current of the camera lens and the light source intensity on the original image, the original image was corrected for black and white according to formula (1):
[0023]
[0024] Among them, I cal represents the corrected image; I raw Represents the original image; I white Represents a full white reflective image; I dark Represents a completely black reflected image.
[0025] Step 4: ROI division. In order to improve the performance of the prediction model and suppress the spectral fluctuations affected by noise, background interference and instrument errors, the ROI is divided into 10×10 pixel points. The average spectrum in each small area is extracted to obtain the modeling samples and the position sequence is marked in sequence according to the size of the divided area. The marking method is as follows: Figure 1 shown.
[0026] Step 5: Model establishment and evaluation. The average spectral data and position sequence are used as input variables of the deep learning model, and the PBAT concentration is used as the output variable to establish a concentration prediction model between them. Although there are many types of deep learning models, this embodiment uses the following three deep learning models (combining attention mechanism, bidirectional neural network and convolutional neural network with long short-term memory network (AttentionLSTM, BiLSTM and CNNLSTM)), considering the spatial position information to establish a quantitative model between the NIR average spectrum and the PBAT concentration. Figure 2 It can be seen that the correlation coefficient of the CNNLSTM model is 0.7092, which is better than the other two models. Therefore, this model will be used for prediction next.
[0027] Step 6: Use the trained model to predict the actual sample. Dilute and spread the microorganisms enriched and cultured from the test environment on a solid culture medium for culture to prepare the actual sample. The present invention uses three strains (PBAT1, PBAT2, PBAT3) and a blank position (PBAT0) to verify the model. Figure 4 It can be seen that PBAT1 is a degrading bacteria, and PBAT2 and PBAT3 are non-degrading bacteria after verification by high performance liquid chromatography. After the culture is completed, the colonies on the surface of the solid culture medium are gently removed, and the hyperspectral image of the removed position is collected. At the same time, the removed colonies are preserved for subsequent screening and identification.
[0028] Step 7. In order to suppress the influence of spatial fluctuation of the spectrum on the prediction model, the spectrum of the colony growth (RP) and the reference position (BRP) is compared and the difference between the two is calculated; the positive and negative and the size of the predicted difference between the blank position (NP0) and the three strains of sterilization positions (RP1, RP2, RP3) and the reference position (BNP, BRP1, BRP2, BRP3) are considered to indirectly infer whether the unknown colony on the solid culture medium has a degradation effect on PBAT; if the difference is negative, it means that the colony corresponding to the area is a PBAT-degrading bacteria; the results are as follows Figure 5 As shown in the figure, the predicted difference between the sterilization position and the reference position of strain PBAT1 is less than 0, while the predicted differences of the other two strains (PBAT2, PBAT3) and the blank position (PBAT0) are 0 or greater than 0, indicating that the method of the present invention can quickly identify PBAT-degrading bacteria. Figure 4 The results are consistent.
[0029] Among them, for the selection of reference spectrum in step seven, considering the influence of space on the spectrum, an area without strain growth is selected as the central area (CR), and the Euclidean distance between the average spectrum (ACR, BCR, LCR, RCR) of the neighboring area (within the distance of 5 pixels in the four directions of up, down, left and right) and CR is calculated respectively, and the neighboring area with the smallest difference is selected as the comparison spectrum for difference operation to eliminate the spatial influence of space on the spectrum.
[0030] Step 8: According to the prediction results of step 7, the corresponding colonies are repeatedly purified and cultured to obtain pure cultures of PBAT-degrading bacteria.
Claims
1. A method for rapid identification of PBAT-degrading bacteria, characterized in that: Here are the steps: Step 1: Prepare the modeling samples: prepare the PBAT-glucose solid culture medium modeling samples with different concentrations, and collect the hyperspectral images of the samples. Separate the solid culture medium and the culture dish before collection, place the solid culture medium on the polytetrafluoroethylene white plate, keep the bottom fit, and place it in the center of the stage, keeping the placement position consistent; Step 2: Collect the average spectrum: select the region of interest (ROI), divide the ROI into 10×10 pixel points, and extract the average spectrum of the modeling samples in each small area; in order to improve the performance of the prediction model and suppress the spectral fluctuations affected by noise, background interference, and instrument errors, mark the ROI area in spatial order so that the spatial distribution of the light source can be associated with the ROI position; Step 3: Establish a prediction model: Divide the extracted average spectrum into a training set and a validation set, and adopt a deep learning prediction model based on sequence features, including an input layer, multiple LSTM units, a fully connected layer, and an output layer; use the average spectrum and position sequence as input variables at the same time, and the PBAT concentration on the culture medium as the output variable to establish a prediction model between the average spectrum and the PBAT concentration; input the input variables into the input layer, and then pass them to multiple LSTM units in turn to capture long-term dependencies, and then connect to the fully connected layer, and finally obtain the final prediction result through the output layer; Step 4: Select the optimal prediction model: Verify the trained prediction model and use the correlation coefficient R and root mean square error RMSE of the validation set as key performance indicators to comprehensively evaluate the accuracy and reliability of the model; the larger the R of the prediction model, the closer its value is to 1; The smaller the RMSE is, the closer its value is to 0, indicating that the model has good prediction ability and robustness. The prediction model with the largest R is selected for subsequent actual sample prediction; Step 5: Pretreatment of the actual sample to be predicted: dilute and spread the microorganisms enriched and cultured from the environment to be tested on a solid culture medium for culture; After the culture is completed, the colonies on the surface of the solid culture medium are gently removed, and a hyperspectral image of the removed position is collected. At the same time, the removed colonies are stored for subsequent screening and identification; Step 6. Predict the actual sample: In order to suppress the influence of spatial fluctuation of the average spectrum on the prediction model, select a reference position, which is located in the blank position below the colony, within 0-5 pixels from the colony position, and the area size of this area is consistent with the removal of the colony; calculate the predicted difference of PBAT concentration between the spectrum of the sterilization position and the reference position through the prediction model, and indirectly infer whether the unknown colony on the solid culture medium has a degradation effect on PBAT based on the positive and negative and size of the difference; if the difference is less than 0, it means that the corresponding colony in this area is PBAT-degrading bacteria.
Citation Information
Patent Citations
System and method for rapidly and autonomously detecting aerosol particles
CN114585903A
Rapid screening method of benzalkonium bromide co-metabolism degrading bacteria based on hyperspectral technology
CN115323029A
Citrobacter sp. BERC-2 for degrading PBAT and application thereof
CN118291305A
Method to identify microorganisms using spectroscopic technique
US20200354766A1
