Machine learning-based method for quantitatively analyzing β-sheet structure in silk fibroin material

Abstract

Provided is a machine learning-based method for quantitative analysis and comparison of beta-sheet structures in silk fibroin materials, relating to the fields of material characterization technology and biomedical materials. The method includes small-angle X-ray scattering detection experimentation and big data-based machine learning model analysis. The SAXS experimentation should be performed on a device that meets a detection sensitivity requirement to obtain scattered light intensity I data. After the machine learning model inputs SAXS experimentation q-I data of an unknown silk fibroin material, a shape and a feature dimension of a beta-sheet structure thereof are output. This method has higher convenience. The method can visually reflect the statistical results of the structural features in the sample without sample pretreatment, and can reflect nanostructure information in a scale range of <500 nm. Analyzing light scattering results by means of a machine learning method more accurately infers detailed structural information of a material, thereby optimizing material design.

Description

A method for quantitatively analyzing beta-sheet structure in silk fibroin material based on machine learning TECHNICAL FIELD

[0001] The present application relates to the field of material characterization and biomedical material technology, and particularly relates to a method for quantitatively analyzing microstructure of silk fibroin material based on machine learning. BACKGROUND

[0002] The molecular assembly structure of proteins in vivo and in vitro has a significant impact on their performance and function. In the biomedical field, structural proteins such as collagen and silk fibroin have been developed as biomedical materials, which exhibit excellent biocompatibility. However, there are multiple complex molecular interactions such as hydrogen bonds, van der Waals forces, and hydrophobic interactions in the topological structure of proteins, and at the same time, the protein structure is sensitive to heat, force and other factors, and the accurate and quantitative analysis of the microstructure is still a bottleneck restricting its application.

[0003] Silk fibroin is a protein extracted from natural silk, which has typical beta-sheet, helix and random coil conformation structures. The films, solutions, gels and scaffolds made of silk fibroin materials have become an important category in biomedical materials. Most of the amino acid sequences in silk fibroin are GAGAGS repeat modules, with a repeat number of 432 times. When two segments of GAGAGS are arranged in parallel and locked by hydrogen bonds, an anti-parallel beta-sheet structure is formed, which can be considered as the smallest repeat unit of beta-sheet structure (Figure 1). In silk fibroin materials, such as solutions / gels, the size of the beta-sheet structure depends on the preparation conditions, and it will continuously evolve over time. Different silk fibroin materials will contain different sizes of beta-sheet structures, which can be considered as sheet or cuboid structures stacked by the smallest repeat units. As shown in Figure 2, the long side of the cuboid is the a direction of the hydrogen bond formed between the molecular chains, the wide side is the b direction of the sheet along the stacking, and the height is the c direction of the chain extension. Currently, the chain spacing and layer spacing information have been obtained by wide-angle X-ray diffraction (WAXD), but there is still a lack of quantitative analysis method for sheet and cuboid beta-sheet structures with sizes ranging from 10 nm to 500 nm.

[0004] Small-angle X-ray scattering (SAXS) involves exposing a sample to high-energy, hard X-rays with very small wavelengths, i.e. 0.07-0.2 nm, and photon energies higher than ~10 keV. Such high energies enable a strong enough scattered light signal to be obtained at a distance of 3-4 m from the X-ray source, i.e. at very small scattering angles. Therefore, the SAXS method has become an important tool for analyzing protein structures. This method is suitable for solution, gas and solid samples and is a non-invasive, non-destructive characterization technique. The same sample produces a unique scattering signal determined by its internal structure, so the microstructure of the sample can be accurately and effectively inferred from the scattering results. Small-angle scattering includes SAXS and small-angle neutron scattering (SANS) and is commonly used to study the shape, structure and interaction of proteins and polymers in solution.

[0005] Machine learning is a data-driven method that has a significant advantage in solving multivariate nonlinear problems. Supervised learning in machine learning covers two major algorithms, regression and classification, which are used for predicting continuous and discrete results, respectively, and can effectively assist in research topics in the field of natural sciences. These mainstream algorithms include decision tree (Decision Tree), random forest (Random Forest), support vector machine (SVM), naïve Bayes (Naïve Bayes), ensemble learning (Ensemble Learning), and multilayer perceptron (MLP). Machine learning, as a key technology in the field of artificial intelligence, has had a very significant impact in the fields of computer image, natural language processing, content generation, etc. In the field of materials science, machine learning models based on big data are also increasingly widely used. By constructing a machine learning model to predict the size of the characteristic beta-sheet structure in silk fibroin, experimental costs can be effectively reduced, and the material development and optimization process can be accelerated. SUMMARY

[0006] The present application aims to protect a method for quantitatively analyzing the beta-sheet structure in silk fibroin materials. This method uses SAXS as an experimental technique, combined with machine learning methods and models, to obtain shape and size information of the beta-sheet structure in silk fibroin solutions, gels and other materials. This quantitative method has important reference value for constructing the structure-activity relationship of silk fibroin materials, optimizing the preparation process of silk fibroin materials, and expanding their application scenarios.

[0007] To achieve the above-mentioned application purposes, the present application provides the following technical solutions:

[0008] The specific steps are as follows:

[0009] (1) Data collection and processing. The present application measures a large number of scattering curve data (i.e. data of scattering intensity I changing with scattering vector q) of silk fibroin solution by SAXS experiment, and preliminarily processes the scattering data. Then the β-sheet structure size of the sample is characterized based on an independent method. Assuming that the β-sheet structure is spherical particles, in the small scattering angle region, the scattering signal satisfies Guinier law. Therefore, the present application uses Guinier law to calculate the β-sheet structure size of the sample, and uses the calculated numerical value as the reference size of the β-sheet structure. Based on this method, the data set required for training the machine learning model is obtained.

[0010] (2) Feature extraction and size calibration. The present application extracts the feature indicators of SAXS scattering curve by various numerical analysis methods. These features are related to the physical and chemical processes of protein β-sheet in silk fibroin solution, and also reflect the morphological characteristics of β-sheet structure to some extent. The relevant features include but are not limited to:

[0011] a) Scattering curve features: such as scattering curve slope, scattering intensity maximum / minimum, etc.

[0012] b) Radial distribution function: the characteristic peak value is obtained by Fourier transform calculation.

[0013] c) Structure factor S(q): the structure factor reflects the short-range order of atoms or molecules in the system, and can provide information about the correlation between atoms.

[0014] Through various methods, all the features involved in modeling are standardized to ensure the quality and consistency of the features.

[0015] (3) AI model training. Select a suitable traditional machine learning model for training. The optional models include:

[0016] a) Random Forest: Random Forest is an ensemble learning algorithm composed of multiple decision trees, which can construct different decision trees by multiple random sampling and take their average value for prediction. Random Forest has the advantages of anti-overfitting, processing high-dimensional data and processing nonlinear relationship, and is suitable for β-sheet structure size prediction problem.

[0017] b) Support Vector Machine (SVM): Support Vector Machine is a machine learning model for classification and regression. Support Vector Machine can achieve efficient prediction by finding the optimal hyperplane, especially when using kernel functions (such as RBF kernel) to handle nonlinear problems, Support Vector Machine can capture the complex relationship between β-sheet structure size and structure.

[0018] During model training, first, all collected data is divided into training set and test set according to a certain proportion (such as 8:2); then, the training set data is used for model fitting, and preliminary evaluation and parameter adjustment are carried out through the test set. In the model training process, the cross-validation or Bayesian optimization method is used to search for the optimal combination of model hyperparameters. The model hyperparameters include: the number and depth of decision trees in the random forest, and the kernel function selection and penalty parameter C in SVM.

[0019] The evaluation of the model is carried out by mean square error (MSE) and goodness of fit (R2) for the trained machine learning model:

[0020]

[0021] Where n represents the total number of protein sample data sets, and are the model predicted value and the theoretically calculated value of the i-th data, is the average value of all predicted values .

[0022] (4) AI model evaluation and optimization. The machine learning model obtained by training can be used to predict the size of the beta-sheet structure in the silk fibroin solution. To evaluate the model effect, new solution samples need to be subjected to SAXS experiments, and the scattering curve data of the new samples are processed in the same way as the training data set, and the extracted features and the trained model are used to predict the size of the beta-sheet structure of the new samples. Compare the model prediction result with the result calibrated by Guinier's law. If the result reaches a certain accuracy, it means that the AI model has reached the level of industrial application, and the next step is to deploy the trained model in the production environment for application, effectively assisting the structure design and synthesis experiment of silk fibroin, reducing the cost and time of experimental measurement. If the result accuracy does not reach the available level, the model needs to be continuously improved by modifying the model training parameters, increasing the data volume, adjusting the feature extraction, etc.

[0023] The method for quantitatively analyzing the shape and size information of the beta-sheet structure in the silk fibroin material based on machine learning provided by the application gives the shape and size information of the beta-sheet structure based on SAXS experimental data and machine learning model.

[0024] The silk fibroin material includes but is not limited to silk fibroin solution, gel, film, powder, sponge, block material, fiber and other macroscopic forms, and a suitable SAXS sample stage / sample injection system is selected according to the material form.

[0025] The SAXS experiment has a scattering angle range of 0.1- , and the scattering vector is The beta-sheet structure in the silk fibroin material is used as a scatterer.

[0026] The machine learning model takes the SAXS experimental result, i.e., the I-q data curve, as input, extracts the mathematical features of the curve, and predicts the shape and size of the beta-sheet structure in the silk fibroin material.

[0027] The machine learning model is based on a training set containing >100 groups of data, and the machine learning model with the highest prediction accuracy is selected from the AI model.

[0028] The >100 groups of data sets include I-q curve data obtained by performing SAXS experimental tests on >100 samples, the revolution radius of the scatterer calculated based on the Guinier law, and the power law of q and I in a specific interval.

[0029] The machine learning model extracts feature data for model training from the scattering curve characterized by the SAXS experiment, and the revolution radius fitted according to the Guinier law as the target of model prediction.

[0030] The machine learning model uses k-fold cross-validation to improve the generalization ability of the model, and uses Bayesian optimization to search for the optimal model hyperparameters.

[0031] The machine learning model uses mean square error and R 2 The machine learning model uses mean square error and R

[0032] The application of the present application is very wide, and its uses include: obtaining the mechanical structure-activity relationship of the material or the relationship between the biological performance and the structure of the material or assisting in optimizing the process.

[0033] For example, the silk fibroin material prepared under specific process conditions can be further determined for mechanical performance indicators such as modulus after completing the quantitative analysis of the beta-sheet structure, so as to obtain the mechanical structure-activity relationship of the material.

[0034] The silk fibroin material prepared under specific process conditions can be further determined for biological performance indicators such as in vivo degradation / absorption rate based on a rat subcutaneous animal model after completing the quantitative analysis of the beta-sheet structure, so as to obtain the relationship between the biological performance and the structure of the material.

[0035] The present application has the following beneficial effects:

[0036] (1) The present application proposes a measurement method on the sub-micron size level, which can quantitatively measure the size of the beta-sheet structure in the silk fibroin solution / gel. The present application uses SAXS as the detection method, and finally obtains the stacking condition and corresponding size of the beta-sheet in the material under the condition of obtaining the SAXS signal.

[0037] (2) After accumulating a large amount of SAXS detection data of various silk fibroin solutions / gels, the size of the beta-sheet corresponding to the silk fibroin solution / gel is obtained according to the corresponding law, and a deep learning model can be established according to the accumulated data, and the size of the beta-sheet of the silk fibroin solution / gel to be measured can be calculated according to the model.

[0038] (3) After simulating and calculating the size of the beta-sheet structure of the silk fibroin solution / gel system according to the machine learning model, the physical and chemical properties of the related material can be inferred according to the information, which provides strong assistance for the efficient design of silk protein materials.

[0039] (4) Compared with the traditional experimental method, the machine learning algorithm can complete the prediction of the size of the beta-sheet structure of a large amount of protein sequences in a few minutes, greatly shortening the research period and reducing the overall cost of research investment. BRIEF DESCRIPTION OF DRAWINGS

[0040] Fig. 1 is the minimum repeating unit of the beta-sheet structure of the repeating sequence GAGAGS in silk fibroin;

[0041] Fig. 2 is a schematic diagram of the sheet and cuboid beta-sheet structure formed by stacking the minimum repeating unit in silk fibroin;

[0042] Fig. 3 is a flow chart of the application of the machine learning assisted beta-sheet structure size prediction model;

[0043] Fig. 4 is an I-q signal diagram of a simple mixed solvent system for SAXS test;

[0044] Fig. 5 is an I-q signal diagram of a simple mixed solvent system for SAXS test;

[0045] Fig. 6 is an I-q signal diagram of two scaffolds for SAXS test;

[0046] Fig. 7 is a picture of a silk fibroin sample;

[0047] Fig. 8 is an I-q signal diagram of a silk fibroin aqueous solution for SAXS test. DETAILED DESCRIPTION

[0048] The present application will be further described below in conjunction with specific embodiments, but the protection scope of the present application is not limited to this:

[0049] Example 1:

[0050] Six hours after the physical gelation, the solution was transferred to a glass cuvette and SAXS measurement was performed on the Xeuss 3.0 platform of Xenocs company. The sample was placed in the liquid window, and two detectors were set at the distances of 1798 mm and 300 mm to cover a wide range of q values. The SAXS curves of the heterogeneous solution system containing β-sheet structure and the simple mixed solvent system were obtained (Figure 4). Assuming that the β-sheet structure is a spherical particle, in the low q value (small scattering angle) region, the SAXS curve is in the Guinier region (q <1.3) follows the Guinier law:

[0051]

[0052] The slope of the linear fitting of The slope of the linear fitting of was obtained, and the gyration radius of the particle was ~ 0.5 nm, which was converted into a particle diameter D ~ 1.3 nm, which was consistent with the size of the β-sheet structure of the smallest repeating structure predicted by the amino acid sequence.

[0053] This example proves by testing that the size of the β-sheet structure can be obtained from the structure information of the chemical bond in theory, and at the same time, the Guinier law is followed.

[0054] Example 2:

[0055] To explore the effect of concentration, different concentrations of silk fibroin HFIP solution were prepared (preparation method as shown in Table 1), and samples 2-1, 2-2, and 2-3 were obtained. Relying on the Xeuss 3.0 platform of Xenocs company, the sample was placed in the liquid window, and two detectors were set at the distances of 1798 mm and 300 mm to cover a wide range of q values. The SAXS curves of the heterogeneous solution system and the simple mixed solvent system were obtained (Figure 5). Assuming that the β-sheet structure is a spherical particle, the Guinier law of the SAXS curve in the small scattering angle region was used to calculate the size of the β-sheet structure in each sample. The appropriate low q value space < 1.3 was selected, and it was ensured that the data did not contain noise or a strong background scattering signal; then, the relationship between I and q was calculated by data processing of the I-q curve; thirdly, the least square method was used to calculate the relationship between and and​ Linear fitting is performed to calculate the slope m and intercept c of the fitted straight line; finally, according to the formula The radius of gyration is calculated from the slope of the fitting .

[0056] Table 1: The size of β-sheet corresponding to different concentrations of silk fibroin solution

[0057]

[0058] This example demonstrates that by SAXS characterization of silk fibroin solutions of different concentrations, the size of the β-sheet data is obtained by Guinier formula calculation, and the size of the β-sheet corresponding to different concentrations of silk fibroin solution is different.

[0059] Example 3:

[0060] A small molecule crosslinking agent is used to prepare a silk fibroin scaffold with a physical and chemical double crosslinking structure by freeze gelation. Compared with pure silk fibroin scaffold samples. Relying on the Beijing Synchrotron Radiation Facility 1W2A line station, SAXS test is performed on the two scaffolds to obtain I-q signal (Figure 6). Combined with the theory of SAXS, i.e. Guinier law, the size of the β-sheet structure is calculated as shown in Table 2.

[0061] Table 2: The size of the β-sheet of the silk fibroin scaffold prepared by different crosslinking methods

[0062]

[0063] This example demonstrates that by SAXS characterization and Guinier formula calculation, the size of the β-sheet of the silk fibroin scaffold prepared by different crosslinking methods is different.

[0064] Example 4: This example establishes a machine learning method and model, and by SAXS characterization of different silk fibroin materials, the scattering curve of SAXS is input into the trained machine learning model to obtain the size information of the β-sheet structure.

[0065] (1) 120 silk fibroin samples are prepared, including solutions, gels, membranes, powders, sponges, blocks, and fibers (as shown in Figure 7).

[0066] (2) The scattering curve data (including scattering intensity I and scattering vector q) of all silk fibroin samples are measured by SAXS experiment. These I-q curves are used to characterize the internal structure of the sample and provide microscopic information about particle size, shape, distribution, etc.

[0067] (3) The I-q curve data obtained needs to be preprocessed before being input into the machine learning model. The specific preprocessing steps include:

[0068] - Normalization: Normalize all I-q curve data to make the I values of different samples within the same order of magnitude range, preventing unstable model training caused by too large numerical range.

[0069] - Since the number and interval of I-q curve data points may vary for different samples, consistent processing of q values is needed to unify the q values of each sample within the same range and step size through interpolation method.

[0070] (4) According to Guinier's law, select I-q curve data in the low q region, and obtain the radius of gyration , which is an important physical quantity representing the geometric size of the β-sheet scattering body.

[0071] (5) Based on the pre-processed I-q curve, the radius of gyration , the features for machine learning modeling are constructed, and the features of each sample include:

[0072] - I-q curve feature: a set of I values corresponding to standardized q values.

[0073] - Scattering curve feature: maximum and minimum values of scattering intensity.

[0074] - Radial distribution function: obtain the position of all characteristic peaks by Fourier transform of I-q curve.

[0075] - Structure factor: separate the structure factor S(q) of the sample from the experimental data, which can provide information on the correlation between atoms.

[0076] (6) Use the obtained silk fibroin sample database to split according to 80% training set and 20% test set, and establish machine learning models through the following 5 ways: random forest, support vector machine, naive Bayes, multilayer perception, and XGBoost. In the model training process, use cross-validation method to update the training set and test set to improve the generalization ability of the model. Adjust the model hyperparameters to ensure the model obtains the optimal effect.

[0077] (7) Calculate the performance indicators of each model on the test set, including:

[0078] - Mean Squared Error (MSE), used to measure the difference between the predicted value of the model and the calculated value of Guinier formula.

[0079] - R 2 value: reflects the goodness of fit of the model to the data, R 2The closer the value is to 1, the better the fitting effect of the model.

[0080] (8) Among the 5 machine learning models, the model with better performance indicators is selected as the model applied in the prediction stage. In this embodiment, the random forest model shows the highest prediction accuracy, the lowest MSE, and the R² value close to 1, so the random forest model is selected as the final model.

[0081] Table 3. The size range of β-sheet of silk fibroin materials in solution, gel, film, powder, sponge, block, and fiber forms using machine learning models

[0082]

[0083] This embodiment proves that the size of the β-sheet of silk fibroin materials in solution, gel, film, powder, sponge, block, and fiber forms can be effectively measured by machine learning models.

[0084] Example 5:

[0085] This embodiment verifies the machine learning method and model. A silk fibroin aqueous solution with a concentration of 2.4% is prepared, the I-q curve of SAXS is obtained, and theoretical calculation is performed according to the Guinier law to obtain R g = 4.8 nm (Figure 8). The I-q curve is input into the trained machine learning model, and the size of the β-sheet structure is 4.9 nm, with a prediction accuracy of 98%.

Claims

1. A method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin materials based on machine learning, which gives the shape and size information of β-sheet structure based on SAXS experimental data and machine learning model, wherein SAXS represents small-angle X-ray scattering.

2. The method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin materials based on machine learning according to claim 1, wherein the silk fibroin materials include but are not limited to silk fibroin solutions, gels, membranes, powders, sponges, blocks, fibers and other macroscopic forms, and a suitable SAXS sample stage / injection system is selected according to the material form.

3. The method for quantitatively analyzing the shape and size information of the β-sheet structure in silk fibroin material based on machine learning according to claim 1, wherein the scattering angle θ in the SAXS experiment ranges from 0.1- , , the scattering vector is , λ is the x-ray wavelength, and the β-sheet structure in the silk fibroin material acts as a scatterer.

4. The method of claim 1, wherein the machine learning model is trained to predict the shape and size of the beta-sheet structure in the silk fibroin material using SAXS data, I-q, I-q 2 and other data structure information as input, by extracting mathematical features of the curve.

5. The method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin materials based on machine learning according to claim 4, wherein the machine learning model is selected from 5 machine learning models based on a training set containing >100 groups of data, and the machine learning model with the highest prediction accuracy is selected. The >100 groups of data sets include I-q curve data obtained by SAXS experimental test on >100 samples, the gyration radius of the scatterer calculated based on Guinier's law, and the power law of q and I in a certain interval.

6. The method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin materials based on machine learning according to claim 1, wherein the machine learning model extracts feature data for model training from the scattering curve characterized by SAXS experiment, and the gyration radius fitted according to Guinier's law as the target of model prediction.

7. The method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin materials based on machine learning according to claim 1, wherein the machine learning model uses k-fold cross-validation to improve the generalization ability of the model, and uses Bayesian optimization to search for the optimal model hyperparameters.

8. The method of claim 1, wherein the machine learning model is evaluated using mean squared error and R 2 values.

9. The method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin materials based on machine learning according to claim 1, comprising the following steps: (1) Data collection and processing, scattering curve data of a large number of silk fibroin solutions, i.e. data of scattering intensity I changing with scattering vector q, are measured by SAXS experiment, and the scattering data are preliminarily processed; then the β-sheet structure size of the measured sample is characterized by an independent method; (2) Feature extraction and size calibration, the characteristic indexes of SAXS scattering curve are extracted by various numerical analysis means, and the related features include but are not limited to: a) Scattering curve features: including scattering curve slope, scattering intensity maximum / minimum; b) Radial distribution function: the characteristic peak value is obtained by Fourier transform calculation; c) Structure factor S(q): the structure factor reflects the short-range order of atoms or molecules in the system, and can provide information about the correlation between atoms; All the features involved in modeling are standardized to ensure the quality and consistency of the features; (3) AI model training, including selecting a suitable traditional machine learning model for training, and the optional models include: a) Random Forest: Random Forest is an ensemble learning algorithm composed of multiple decision trees, which can construct different decision trees through multiple random sampling and take their average to make predictions; b) Support Vector Machine (SVM): Support Vector Machine is a machine learning model for classification and regression. Support Vector Machine achieves efficient prediction by finding the optimal hyperplane, especially when dealing with nonlinear problems using kernel functions (such as RBF kernel), Support Vector Machine can capture the complex relationship between β-sheet structure size and structure; During model training, first, all collected data is divided into training set and test set according to certain proportion (such as 8:2); then, the training set data is used to fit the model, and the test set is used for preliminary evaluation and parameter adjustment. In the model training process, cross-validation or Bayesian optimization method is used to search for the optimal combination of model hyperparameters; Model hyperparameters include: the number of decision trees and their depth in Random Forest, and the selection of kernel function and penalty parameter C in SVM; The evaluation of the model is done by the mean squared error, MSE, and the goodness of fit, R 2 Evaluation of the trained machine learning model: ； where n represents the total number of protein sample data sets, and are the model predicted and theoretical calculated values of the i-th data, respectively, is all predicted values The average value of the two is taken as the final prediction result of the model. (4) AI model evaluation and optimization; the machine learning model obtained by training can be used to predict the size of β-sheet structure in silk fibroin solution; to evaluate the model effect, new solution samples need to be subjected to SAXS experiment, and the scattering curve data of new samples are processed in the same way as the training data set, and the extracted features and trained model are used to predict the β-sheet structure size of new samples. Compare the model prediction result with the result calibrated by Guinier law. If the result reaches a certain accuracy, it means that the AI model has reached the level of industrial application, and the next step is to deploy the trained model in the production environment for application, effectively assisting the structure design and synthesis experiment of silk fibroin, reducing the cost and time of experimental measurement. If the result precision does not reach the available level, the model needs to be continuously improved by modifying the model training parameters, increasing the data volume, adjusting the feature extraction, etc.

10. A method for quantitatively analyzing the shape and size information of β-sheet structure in silk fibroin material based on machine learning according to any one of claims 1 to 9, the use comprising: Obtain the relationship between the mechanical structure of the material and the biological performance of the material and the structure, or assist in optimizing the process.

Images