A method for evaluating stability of polypeptide self-assembled microstructures

Abstract

This invention provides a method for assessing the stability of the microstructure of peptide self-assembly, relating to the field of biological image processing technology. The method includes: acquiring a real-time image sequence of the microstructure of the target peptide during self-assembly; preprocessing the images and extracting the contour features of the assembly to generate a boundary coordinate set; tracking the centroid displacement and boundary deformation of the same assembly between adjacent frames to construct a temporal feature vector, inputting it into a pre-trained stability discrimination model to output a structural perturbation score of the assembly at different time points; and classifying the assembly into one of three states—tending towards stability, reversible fluctuation, or irreversible instability—based on the trend of the score change. This invention can objectively distinguish between reversible fluctuation and irreversible instability, providing a quantitative technical means for real-time monitoring of the peptide self-assembly process.

Description

Technical Field

[0001] This invention relates to the field of biological image processing technology, specifically to a method for evaluating the stability of the microstructure of peptide self-assembly. Background Technology

[0002] Peptide self-assembly refers to the process by which peptide molecules spontaneously form ordered nano- or micro-scale structures through non-covalent interactions, and it has broad application prospects in fields such as biomaterials, drug delivery, and tissue engineering. The stability of peptide self-assembled structures directly affects their functionality and practical application value; therefore, assessing the stability of the microstructure of assemblies during self-assembly is of great significance. Currently, existing methods for assessing the stability of peptide self-assembled structures mainly rely on static image observation or human experience judgment. For example, static morphological images of assemblies are obtained through transmission electron microscopy or atomic force microscopy, and researchers then make subjective evaluations based on the morphological integrity of the assemblies in the images. These methods have significant shortcomings: first, static observation cannot capture the dynamic evolution of assemblies over time, making it difficult to distinguish between temporary reversible fluctuations and irreversible instability trends; second, subjective human evaluation lacks unified quantitative standards, suffers from poor consistency among different evaluators, and is difficult to handle large-scale image data.

[0003] Furthermore, some existing methods attempt to quantify structural changes by tracking individual geometric parameters of assemblies (such as particle size, aspect ratio, or area), but these single parameters are insufficient to fully reflect the complex geometric evolution of assemblies. Simultaneously, existing methods lack synergistic analysis of centroid migration and boundary deformation, failing to construct temporal characteristics describing geometric evolution. More critically, current technologies lack an end-to-end automated evaluation framework from microscopic image sequences to stable states, and lack technical solutions that combine dynamic geometric characteristics of assemblies with stability discrimination models. This makes it difficult to achieve objective and quantitative assessments of whether assemblies tend to stabilize, experience reversible fluctuations, or enter irreversible unstable states during real-time monitoring of peptide self-assembly.

[0004] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for evaluating the stability of the microstructure of peptide self-assembly, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for evaluating the stability of the microstructure of peptide self-assembly, comprising the following steps: Step 1: Obtain real-time microstructure image sequences of the target peptide during the self-assembly process as initial image data for evaluation; Step 2: Denoise and enhance the initial image data, extract the contour features of the assembly in each frame image, and generate structural contour data represented by the assembly boundary coordinate set; Step 3: Based on the structural contour data, track the centroid displacement and boundary deformation of the same assembly between adjacent time frames, and construct a temporal feature vector describing the geometric evolution of the assembly; Step 4: Input the time-series feature vector into the pre-trained stability discrimination model and output the structural perturbation score of each assembly at different time points; Step 5: Evaluate the stability of the assembly based on the trend of the structural disturbance score over time.

[0007] Furthermore, the logic for acquiring the initial image data is as follows: The target polypeptide sample is placed on a microscopic imaging platform, and raw images at multiple time points are continuously acquired at preset time intervals to form a raw image time series. Each frame of the original image time series is timestamped and spatially scaled to generate initial image data with physical size information and relative time order. Based on the initial image data, invalid frames that do not contain assemblies or whose imaging quality does not meet a preset threshold are removed, and the remaining valid frames are arranged in chronological order as the initial image data to be evaluated.

[0008] Furthermore, the imaging quality includes signal-to-noise ratio and edge sharpness; The image quality not meeting the preset threshold means that the signal-to-noise ratio of the image is lower than the first preset threshold, or the edge sharpness of the image is lower than the second preset threshold.

[0009] Furthermore, the specific execution process of step 2 is as follows: S21: Gaussian filtering for noise reduction and histogram equalization for enhancement are performed sequentially on each frame of the initial image data to obtain the preprocessed image; S22: An adaptive threshold segmentation algorithm is used on the preprocessed image to separate the assembly region from the background region in the image, resulting in a binarized image. S23: Perform morphological closing operation on the assembly region in the binarized image to fill the internal holes of the region and smooth the boundaries to obtain the optimized connected component of the assembly. S24: Extract the boundary pixel coordinates of the connected domain, arrange them in clockwise or counterclockwise order to form a boundary coordinate set, and output the boundary coordinate set as the structural contour data of the assembly in the frame image.

[0010] Furthermore, based on the structural contour data, the centroid displacement of the same assembly between adjacent time frames is tracked, and the specific logic underlying this is as follows: For adjacent time frames and frame Based on the two frames of images, the centroid coordinates of each assembly in each frame are calculated using their respective structural contour data. The calculation formula is as follows: ; In the formula, Indicates the centroid coordinates; is the total number of pixels contained in the structural contour data; i is the index of the boundary pixel. and Let x and y represent the x and y coordinates of the i-th boundary pixel, respectively. Based on the centroid coordinates, the nearest neighbor matching algorithm is used to match the frames. Assembly and frame in The assemblies are matched one-to-one to complete the tracking of the same assembly between adjacent time frames; For the same tracked assembly, calculate its centroid displacement using the following formula: ; In the formula, Indicates that the assembly is from the first Frame to the Centroid displacement of the frame; and These respectively indicate that the assembly is in the frame and frame The centroid coordinates in the equation.

[0011] Further, the boundary deformation of the assembly is calculated, specifically as follows: For frames Each pixel on the boundary of the assembly in the frame Find the boundary pixel with the closest Euclidean distance in the matrix and perform matching. Calculate the average distance between all matching point pairs as the boundary deformation. The calculation formula is as follows: ; In the formula, Indicates that the assembly is from the first Frame to the The boundary deformation of the frame; the larger the value, the more severe the deformation. Indicates the first Index of boundary pixels in the frame; and They represent the first The first frame The x and y coordinates of each boundary pixel; and They represent the first The first frame The x and y coordinates of each boundary pixel.

[0012] Furthermore, the calculated centroid displacement sequence and boundary deformation sequence The features are concatenated to construct a temporal feature vector, which takes the following form: ; in, This represents the total number of time frames. The temporal feature vector is used to describe the complete evolution trajectory of the geometry of the assembly during the self-assembly process.

[0013] Furthermore, the specific execution process of step 4 is as follows: S41: Obtain multiple sets of peptide self-assembly microstructure image sequences with known stability tags, extract the temporal feature vector of each assembly in each frame image according to steps 1-3, and label the assembly state at each time point with a structural perturbation truth value label according to the stability tag. The structural perturbation truth value label is a continuous value with a value range of 0 to 1, where 0 represents complete stability without perturbation and 1 represents complete instability and disintegration. S42: Construct a stability discrimination model. The model uses a bidirectional long short-term memory network as its basic architecture, comprising an input layer, two bidirectional LSTM hidden layers, a dropout layer, a fully connected layer, and a sigmoid output layer. The input layer receives the temporal feature vector. The bidirectional LSTM hidden layers capture the long-range dependencies of the temporal feature vector in both the forward and backward directions, with each hidden layer containing 64 memory units. The dropout layer has a dropout rate of 0.5 to prevent overfitting. The fully connected layer maps the high-dimensional features output by the LSTM to one-dimensional features. The sigmoid output layer converts the mapping result into a continuous value between 0 and 1, which serves as the structural perturbation score of the assembly at the current time point. S43: Take the temporal feature vector of each time point in the training dataset as the model input, take the structural perturbation ground value label corresponding to the time point as the supervision signal, use the mean squared error as the loss function, and use the Adam optimizer to perform parameter iterative optimization until the loss function converges to obtain the trained stability discrimination model. S44: Input the temporal feature vector of the assembly to be evaluated into the trained stability discrimination model in chronological order. The model outputs the structural perturbation score corresponding to each time point. The structural perturbation score is used to characterize the structural stability of the assembly at that time point. The closer the score is to 0, the more stable it is. The closer the score is to 1, the less stable it is.

[0014] Furthermore, the specific execution process of step 5 is as follows: S51: Obtain the structural perturbation score sequence of the assembly at all time points, denoted as... ,in Indicates the first The structural disturbance score at each time point ranges from 0 to 1. This represents the total number of time frames. S52: Perform trend analysis on the structural disturbance score sequence: calculate the sliding window mean sequence and local slope sequence of the sequence, where the width of the sliding window is set to... The average value of the sliding window is The local slope is the linear regression slope of the score within the window. This is used to characterize the direction and rate of local change in the score within the window; S53: Based on the mean and local slope of the sliding window, the stability state of the assembly is determined according to the following judgment logic: If the structural disturbance scores at a number of consecutive time points are all below a preset first score threshold, and the absolute value of the local slope within the corresponding time window is less than the slope threshold, then the assembly is judged to be stable. If there are at least two complete fluctuation cycles in the structural disturbance scoring sequence, and the continuous dwell time in each scoring interval is not less than the preset minimum number of dwell frames, and the arithmetic mean of the structural disturbance scores of the entire time series does not exceed the second scoring threshold, then it is determined that the assembly has reversible fluctuations. The fluctuation cycle is defined as the process of the score rising from below the first score threshold to above the third score threshold and then falling back below the first score threshold, or the process of the score falling from above the third score threshold to below the first score threshold and then rising back above the third score threshold. If the structural disturbance scores at a number of consecutive time points are all higher than the third score threshold, and the local slopes within the corresponding time window are all positive, and the score sequence within the time window shows a non-decreasing trend, then the assembly is judged to have entered an irreversible unstable state. If the conditions for both stability and reversible fluctuation are met simultaneously, but the complete criteria for either state are not fully met, then a tendency judgment is made based on the direction of the score change of the assembly within the most recent time window: when the local slope at multiple recent time points is negative or close to zero, it is judged as stabilizing; when the local slope alternates between positive and negative and the average score is lower than the second score threshold, it is judged as reversible fluctuation; when the local slope is positive and the score continues to rise, it is judged as irreversible instability.

[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention acquires real-time microstructural image sequences of target peptides during self-assembly, denoises and enhances the images, and extracts the contour features of the assemblies to generate structural contour data represented by boundary coordinate sets. Based on this, it tracks the centroid displacement and boundary deformation of the same assembly between adjacent time frames to construct a temporal feature vector describing the geometric evolution of the assembly. This invention overcomes the limitations of traditional static observation and single geometric parameter analysis, and for the first time incorporates centroid displacement and boundary deformation into the temporal feature expression. It can comprehensively capture the dynamic geometric evolution of the assembly over time, providing a rich, objective, and quantifiable feature foundation for subsequent stability assessment, and overcoming the problems of poor consistency and inability to handle large-scale data in subjective human evaluation.

[0016] This invention further inputs the temporal feature vector into a pre-trained stability discrimination model, outputting structural perturbation scores for each assembly at different time points. Based on the trend of the scores over time, the assembly is classified into one of three states: tending towards stability, reversible fluctuation, or irreversible instability. This invention establishes an end-to-end automated evaluation framework from microscopic image sequences to stability states. Through trend analysis of sliding window mean and local slope, as well as quantitative judgment logic, it can objectively distinguish between temporary reversible fluctuations and irreversible instability trends, realizing automated and quantitative evaluation of the stability of peptide self-assembly microstructures. This provides a reliable technical means for real-time monitoring and process optimization of peptide self-assembly processes. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall method flow of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly. Example

[0020] Please see Figure 1 The present invention provides a technical solution: A method for evaluating the stability of the microstructure of peptide self-assembly, comprising the following steps: Step 1: Obtain real-time microstructure image sequences of the target peptide during the self-assembly process as initial image data for evaluation; In this embodiment, the logic for obtaining the initial image data is as follows: The target polypeptide sample is placed on a microscopic imaging platform, and raw images at multiple time points are continuously acquired at preset time intervals to form a raw image time series. Each frame of the original image time series is timestamped and spatially scaled to generate initial image data with physical size information and relative time order. Based on the initial image data, invalid frames that do not contain assemblies or whose imaging quality does not meet a preset threshold are removed, and the remaining valid frames are arranged in chronological order as the initial image data to be evaluated.

[0021] The imaging quality includes signal-to-noise ratio and edge sharpness; The image quality not meeting the preset threshold means that the signal-to-noise ratio of the image is lower than the first preset threshold, or the edge sharpness of the image is lower than the second preset threshold.

[0022] In this embodiment, the target peptide sample is placed on a microscopic imaging platform. The microscopic imaging platform is preferably any one of a transmission electron microscope, atomic force microscope, or confocal fluorescence microscope. The imaging resolution should be sufficient to clearly distinguish the boundaries of individual assemblies, generally set to a spatial scale of 1-10 nanometers per pixel; the temperature of the imaging platform is controlled at... Humidity control at This is to maintain the stable progress of the peptide self-assembly process.

[0023] Original images at multiple time points are continuously acquired at preset time intervals to form an original image time series. The preset time interval is determined based on the peptide self-assembly speed: for rapidly assembling peptide systems, the time interval is set to 0.5-2 seconds; for slowly assembling peptide systems, the time interval is set to 10-60 seconds. The total acquisition time should cover the entire process from the start of assembly to the point where the structure tends to stabilize or completely destabilizes, generally set to 10 minutes to 2 hours. One frame of image is acquired at each time point, and the pixel size of each frame is uniformly set to 1024×1024 pixels, with a grayscale depth of 8 bits or 16 bits.

[0024] For each image frame, the following method is used to determine whether it contains an assembly: After binarizing the image, the number and area of ​​connected components are calculated. If there are no connected components in the image with an area greater than a preset area threshold (e.g., 50 pixels), the image frame is determined not to contain an assembly and is discarded. The preset area threshold is set according to the minimum resolvable size of the assembly, and is generally 10%-20% of the expected area of ​​the assembly.

[0025] The imaging quality includes signal-to-noise ratio (SNR) and edge sharpness; wherein, the SNR is calculated as follows: selecting the assembly region in the image as the signal region, and calculating the mean gray value of all pixels within that region. Select a background region far from the assembly (such as the four corners of the image, each region being 20×20 pixels), and calculate the standard deviation of the pixel grayscale values ​​within the background region. Then follow the formula Calculate the signal-to-noise ratio. Edge sharpness is calculated by performing a Sobel convolution operation on the image to obtain the gradient magnitude of each pixel. ,in and These are the first-order derivatives in the horizontal and vertical directions, respectively; then, pixels located on the boundaries of the assembly are extracted, and the mean gradient magnitude of these boundary pixels is calculated. , For the first The coordinates of each boundary pixel are used as the average value, which is the quantitative indicator of edge sharpness.

[0026] The image quality not meeting the preset threshold refers to the image's signal-to-noise ratio being lower than a first preset threshold, or the image's edge sharpness being lower than a second preset threshold. The first preset threshold is used to determine whether the assembly region in the image is excessively interfered with by noise. Its determination method is as follows: multiple sets of standard sample images with different signal-to-noise ratio levels are acquired, and multiple professional evaluators score the recognizability of the assembly boundaries in the images. The signal-to-noise ratio value corresponding to the inflection point of the recognizability score decrease is selected as the first preset threshold. In this embodiment, this threshold is preferably 5. The second preset threshold is used to determine whether the assembly boundaries are sharp enough for accurate tracking. Its determination method is as follows: a series of images of the same assembly under different focusing degrees are acquired, the edge sharpness index of each image is calculated, and professional evaluators determine whether the boundaries are clearly discernible. The edge sharpness value corresponding to the acceptable lower limit of sharpness is selected as the second preset threshold. In this embodiment, this threshold is preferably 30 (for 8-bit grayscale images, the gradient amplitude range is 0-255). and If the image quality is deemed satisfactory, it is considered to be acceptable; otherwise, if the image quality is deemed not to meet the preset threshold, the image frame is discarded.

[0027] Step 2: Denoise and enhance the initial image data, extract the contour features of the assembly in each frame image, and generate structural contour data represented by the assembly boundary coordinate set; The specific execution process of step 2 is as follows: S21: Gaussian filtering for noise reduction and histogram equalization for enhancement are performed sequentially on each frame of the initial image data to obtain the preprocessed image; S22: An adaptive threshold segmentation algorithm is used on the preprocessed image to separate the assembly region from the background region in the image, resulting in a binarized image. S23: Perform morphological closing operation on the assembly region in the binarized image to fill the internal holes of the region and smooth the boundaries to obtain the optimized connected component of the assembly. S24: Extract the boundary pixel coordinates of the connected domain, arrange them in clockwise or counterclockwise order to form a boundary coordinate set, and output the boundary coordinate set as the structural contour data of the assembly in the frame image.

[0028] In this embodiment, the specific execution process of step 2 is as follows: S21: Each frame of the initial image data is sequentially subjected to Gaussian filtering for noise reduction and histogram equalization for enhancement to obtain a preprocessed image. The Gaussian filtering uses a two-dimensional Gaussian kernel function to convolve the image; the preferred kernel size is 5×5 pixels, and the standard deviation is... Setting it to 1.0, this parameter combination can effectively suppress random noise generated during image acquisition while preserving the edge details of the assembly. Histogram equalization stretches the image gray-level histogram from a concentrated distribution to a uniform distribution. Specifically, it involves statistically analyzing the probability of each gray level in the image, calculating the cumulative distribution function, and then mapping the original gray values ​​to the full gray range of 0-255 based on the cumulative distribution function, thereby enhancing the contrast between the assembly and the background. S22: An adaptive threshold segmentation algorithm is applied to the preprocessed image to separate the assembly region from the background region, resulting in a binarized image. The adaptive threshold segmentation algorithm divides the image into multiple local windows (each window is preferably 101×101 pixels). A threshold is calculated independently within each window. Specifically, a Gaussian weighted average is used to calculate the weighted average of the pixels within the window as the threshold. For any pixel in the window, if its grayscale value is higher than the window threshold, it is marked as foreground (assembly region); otherwise, it is marked as background. This adaptive approach effectively overcomes the influence of uneven illumination on the segmentation effect. S23: Perform morphological closing operations on the assembly region in the binarized image to fill the internal holes and smooth the boundaries, obtaining the optimized connected components of the assembly. The morphological closing operation is a combination of dilation and erosion: the dilation operation involves sliding a structuring element (preferably a 3×3 pixel circular structuring element) on the image; if there are foreground pixels within the area covered by the structuring element, the pixel corresponding to the center of the structuring element is set as the foreground, thereby filling the small holes and cracks inside the assembly region; the erosion operation involves retaining the pixel corresponding to the center of the structuring element as the foreground if all pixels within the area covered by the structuring element are foreground, otherwise setting it as the background, thereby removing protruding burrs and smoothing the boundaries.

[0029] S24: Extract the boundary pixel coordinates of the connected components, arrange them in clockwise or counterclockwise order to form a boundary coordinate set, and output this boundary coordinate set as the structural contour data of the assembly in the frame image. The boundary extraction adopts an eight-neighbor boundary tracing algorithm: First, find the top left foreground pixel in the connected component as the starting point, and then search for the next boundary point in its eight neighbors in a clockwise direction from this point. Iterate until returning to the starting point, thus obtaining a closed sequence of boundary pixels. Each boundary point is stored in the form of its horizontal and vertical coordinates in the image coordinate system to form a boundary coordinate set.

[0030] Step 3: Based on the structural contour data, track the centroid displacement and boundary deformation of the same assembly between adjacent time frames, and construct a temporal feature vector describing the geometric evolution of the assembly; In this embodiment, the centroid displacement of the same assembly between adjacent time frames is tracked based on the structural contour data, and the specific logic is as follows: For adjacent time frames and frame Based on the two frames of images, the centroid coordinates of each assembly in each frame are calculated using their respective structural contour data. The calculation formula is as follows: ;

[0031] In the formula, Indicates the centroid coordinates; is the total number of pixels contained in the structural contour data; i is the index of the boundary pixel. and Let x and y represent the x and y coordinates of the i-th boundary pixel, respectively. Based on the centroid coordinates, the nearest neighbor matching algorithm is used to match the frames. Assembly and frame in The assemblies are matched one-to-one to complete the tracking of the same assembly between adjacent time frames; For the same tracked assembly, calculate its centroid displacement using the following formula: ;

[0032] In the formula, Indicates that the assembly is from the first Frame to the Centroid displacement of the frame; and These respectively indicate that the assembly is in the frame and frame The centroid coordinates in the equation.

[0033] In this embodiment, calculating the centroid displacement is significant in quantifying the overall migration degree of the assembly during the self-assembly process, thereby reflecting the positional stability of the assembly relative to the observation field of view. Specifically, the centroid coordinates... The centroid coordinates represent the geometric center of the region enclosed by the assembly's outline. Calculated based on the arithmetic mean of the coordinates of all pixels in the boundary coordinate set, it exhibits robustness to boundary noise and can stably characterize the spatial position of the assembly. When the assembly moves as a whole between adjacent time frames, its centroid coordinates change accordingly, resulting in a centroid displacement. The magnitude of this movement is quantified by calculating the straight-line distance between the centroids of two consecutive frames using the Euclidean distance formula. A larger value indicates more drastic movement of the assembly within that time interval, potentially suggesting external disturbances or dissociation / migration. Conversely, a smaller centroid displacement indicates a more stable position of the assembly in the field of view, typically reflecting a near-steady-state structure. It is important to note that centroid displacement only reflects the overall translational position and not shape changes; therefore, in practical applications, it needs to be used in conjunction with subsequent boundary deformation indices. In practice, the frame distance is first calculated using the formula... and frame The centroid coordinates of the same assembly are obtained, and then the difference between the two coordinates is calculated using the Euclidean distance formula to obtain the centroid displacement value. This value is a non-negative number, and the unit is related to the image spatial calibration coefficient (such as pixel or nanometer). By traversing all adjacent frame pairs, the centroid displacement sequence of the assembly throughout the self-assembly process can be obtained. This sequence describes the complete trajectory of the assembly's position evolution over time.

[0034] The boundary deformation of the assembly is calculated as follows: For frames Each pixel on the boundary of the assembly in the frame Find the boundary pixel with the closest Euclidean distance in the matrix and perform matching. Calculate the average distance between all matching point pairs as the boundary deformation. The calculation formula is as follows: ;

[0035] In the formula, Indicates that the assembly is from the first Frame to the The boundary deformation of the frame; the larger the value, the more severe the deformation. Indicates the first Index of boundary pixels in the frame; and They represent the first The first frame The x and y coordinates of each boundary pixel; and They represent the first The first frame The x and y coordinates of each boundary pixel.

[0036] In this embodiment, the significance of calculating boundary deformation lies in quantifying the degree of shape change of the assembly during the self-assembly process, thereby capturing the geometric evolution of non-translational properties such as local stretching, twisting, or fragmentation of the assembly boundary. These changes cannot be reflected by centroid displacement, but are crucial for determining the structural stability of the assembly. Specifically, the calculation logic for boundary deformation is as follows: for frame... Each pixel on the boundary of the assembled part, in the frame Traverse all pixels along the boundary of the assembly and calculate the distance between the current point and the frame using the Euclidean distance formula. The process calculates the straight-line distance between each boundary point in a frame and takes the minimum value as the best matching distance between that point and adjacent frames. This process is equivalent to calculating the straight-line distance between each boundary point in a frame. The boundary is a frame For each boundary point in the array, find its spatial nearest neighbor; for all... The minimum matching distances corresponding to each boundary point are summed and then divided by . The average matching distance is then obtained as the boundary deformation. This metric essentially measures the degree of shape difference between two boundary contours: if the assembly maintains its shape between adjacent frames (only overall translation or rotation occurs), then each boundary point can be matched within the frame. Finding the nearest neighbor matching point with a distance of zero or minimal on the boundary of the assembly results in a boundary deformation that approaches zero. Conversely, if the assembly undergoes significant local deformation, bulges, depressions, or partial disintegration, the boundary point cannot be precisely matched, the average minimum distance increases, and the boundary deformation increases accordingly. By traversing all adjacent frame pairs, the boundary deformation sequence of the assembly can be obtained. This sequence, together with the aforementioned centroid displacement sequence, describes the complete evolution trajectory of the assembly's geometry during self-assembly. Boundary deformation focuses on local shape changes, while centroid displacement focuses on overall positional migration. The two complement each other, providing comprehensive feature input for the subsequent stability discrimination model.

[0037] The calculated centroid displacement sequence and boundary deformation sequence The features are concatenated to construct a temporal feature vector, which takes the following form: ;

[0038] in, This represents the total number of time frames. The temporal feature vector is used to describe the complete evolution trajectory of the geometry of the assembly during the self-assembly process.

[0039] The significance of constructing a temporal feature vector by alternately splicing the centroid displacement sequence and the boundary deformation sequence is that it aligns and merges the centroid displacement describing the overall migration of the assembly and the boundary deformation describing the local shape changes in the time dimension to form a unified data structure, so that the subsequent stability discrimination model can use information from both the positional stability and morphological stability dimensions for comprehensive evaluation.

[0040] Step 4: Input the time-series feature vector into the pre-trained stability discrimination model and output the structural perturbation score of each assembly at different time points; In this embodiment, the specific execution process of step 4 is as follows: S41: Obtain multiple sets of peptide self-assembly microstructure image sequences with known stability tags, extract the temporal feature vector of each assembly in each frame image according to steps 1-3, and label the assembly state at each time point with a structural perturbation truth value label according to the stability tag. The structural perturbation truth value label is a continuous value with a value range of 0 to 1, where 0 represents complete stability without perturbation and 1 represents complete instability and disintegration. The known stability label refers to the true stability state known through experimental means or expert evaluation independent of this method. Specifically, it can be determined by measuring the instability time point of the assembly through synchronous in-situ characterization techniques (such as dynamic light scattering or circular dichroism spectroscopy), or by taking the average value after at least two professional researchers score each frame from 0 to 1 based on the morphological integrity of the assembly.

[0041] Based on this, the temporal feature vector of each assembly in each frame of the image is extracted according to steps 1-3, and the structural perturbation ground truth label is labeled for the assembly state at each time point according to the stability label mentioned above. The structural perturbation ground truth label is a continuous value between 0 and 1, where 0 represents complete stability without perturbation (clear and complete boundaries, centroid displacement and boundary deformation approach zero), 1 represents complete instability and disintegration (severely broken boundaries, drastically reduced area or disappearance from the field of view), and intermediate values ​​represent different degrees of perturbation state (e.g., 0.3 represents slight fluctuation, 0.7 represents severe deformation).

[0042] S42: Construct a stability discrimination model. The model uses a bidirectional long short-term memory network as its basic architecture, comprising an input layer, two bidirectional LSTM hidden layers, a dropout layer, a fully connected layer, and a sigmoid output layer. The input layer receives the temporal feature vector. The bidirectional LSTM hidden layers capture the long-range dependencies of the temporal feature vector in both the forward and backward directions, with each hidden layer containing 64 memory units. The dropout layer has a dropout rate of 0.5 to prevent overfitting. The fully connected layer maps the high-dimensional features output by the LSTM to one-dimensional features. The sigmoid output layer converts the mapping result into a continuous value between 0 and 1, which serves as the structural perturbation score of the assembly at the current time point. S43: Take the temporal feature vector of each time point in the training dataset as the model input, take the structural perturbation ground value label corresponding to the time point as the supervision signal, take the mean square error as the loss function, and take the Adam optimizer to perform parameter iterative optimization until the loss function converges. If the loss value is less than 0.01, the training is completed and a trained stability discrimination model is obtained. S44: Input the temporal feature vector of the assembly to be evaluated into the trained stability discrimination model in chronological order. The model outputs the structural perturbation score corresponding to each time point. The structural perturbation score is used to characterize the structural stability of the assembly at that time point. The closer the score is to 0, the more stable it is. The closer the score is to 1, the less stable it is.

[0043] Step 5: Evaluate the stability of the assembly based on the trend of the structural disturbance score over time; In this embodiment, the specific execution process of step 5 is as follows: S51: Obtain the structural perturbation score sequence of the assembly at all time points, denoted as... ,in Indicates the first The structural disturbance score at each time point ranges from 0 to 1. This represents the total number of time frames. S52: Perform trend analysis on the structural disturbance score sequence: calculate the sliding window mean sequence and local slope sequence of the sequence, where the width of the sliding window is set to... The average value of the sliding window is The local slope is the linear regression slope of the score within the window. This is used to characterize the direction and rate of local change in the score within the window; In this embodiment, S52 performs trend analysis on the structural disturbance scoring sequence. The purpose is to extract macroscopic indicators reflecting the stability evolution trend from the original scoring sequence, avoiding interference from single-point noise on the judgment results. Specifically, the width of the sliding window is first set. Its value is 5 and Rounding down to the smaller of the two values ​​ensures that the window width captures short-term fluctuations while adaptively expanding in long series to smooth long-term trends. For the th Take the point in time, and take the time before it. The point (i.e., the th point) To the The current window is composed of frames, and the arithmetic mean of all scores within the window is calculated as the sliding window mean. This mean reflects the overall level of recent ratings and can suppress jumps caused by random noise. Simultaneously, a linear regression analysis is performed on the ratings within the window, with time sequence as the independent variable and rating as the dependent variable, fitting a straight line whose slope... This refers to the local slope; a positive value indicates an upward trend in the score within that window (deteriorating stability), while a negative value indicates a downward trend (improving stability). The absolute value represents the rate of change. This is achieved by iterating through all time points (from the first...). Frame to the By analyzing the data in a single frame, a sliding window mean sequence and a local slope sequence of the same length as the original score sequence can be obtained, providing smooth and quantified trend features for subsequent stability assessment.

[0044] S53: Based on the mean and local slope of the sliding window, the stability state of the assembly is determined according to the following judgment logic: If the structural disturbance scores at a number of consecutive time points are all below a preset first score threshold, and the absolute value of the local slope within the corresponding time window is less than the slope threshold, then the assembly is judged to be stable. In this embodiment, the most recent consecutive preset number of time points is a fixed value between 3 and 5 time points, preferably 3, to avoid misjudgment caused by accidental low values ​​in a single frame; The first scoring threshold is used to define the upper limit of the "stable" score. It is determined by collecting multiple sets of scoring sequences of assemblies that have been confirmed to be in a stable state (with clear and complete boundaries, centroid displacement and boundary deformation approaching zero for more than 10 consecutive frames), and calculating the average score. The upper bound of the confidence interval is used as the first scoring threshold, which is 0.2 in this embodiment; The slope threshold is used to define "no obvious trend in the score". The method for determining it is as follows: perform linear regression on the score sequence of the stable state assembly, and multiply the standard deviation of the regression slope by 3 as the threshold. In this embodiment, the slope threshold is 0.02. If the structural disturbance scoring sequence contains at least two complete fluctuation cycles, and the continuous dwell time within each scoring interval is not less than the minimum number of dwell frames. In this embodiment If the arithmetic mean of the structural perturbation scores for the entire time series does not exceed the second scoring threshold, then the assembly is judged to have undergone reversible fluctuation. The fluctuation period is defined as the process of the score rising from below the first scoring threshold to above the third scoring threshold and then falling back below the first scoring threshold, or the process of the score falling from above the third scoring threshold to below the first scoring threshold and then rising back above the third scoring threshold, that is, completing a complete round trip of "low→high→low" or "high→low→high". The third scoring threshold is used to define the lower limit of the score for "significant instability". It is determined by collecting multiple sets of assembly scoring sequences that have been confirmed to be in the pre-disintegration state, and taking the value at the inflection point where the score begins to rise sharply as the third scoring threshold. In this embodiment, the third scoring threshold is 0.7. The second scoring threshold is used to define "the whole is still within a controllable range". It is determined by collecting multiple sets of assembly scoring sequences with known reversible fluctuation processes, and calculating the maximum value of their global average score as the second scoring threshold. In this embodiment, the second scoring threshold is 0.4. The requirement of at least two complete cycles eliminates single, accidental fluctuations, while the continuous dwell time within the scoring range of no less than two frames ensures that the score actually remains stable within the corresponding range rather than momentarily crossing it.

[0045] If the structural disturbance scores at a recent consecutive preset number of time points are all higher than the third scoring threshold, and the local slopes within the corresponding time windows are all positive, while the scoring sequence within the time window shows a non-decreasing trend, then the assembly is judged to have entered an irreversible unstable state. The recent consecutive preset number of time points is a fixed value between 3 and 5 time points, and is preferably 3 in this embodiment. The introduction of a non-decreasing trend ensures that the score is not only higher than the threshold but is also continuing to deteriorate, rather than falling back after oscillating at a high level.

[0046] If the conditions for both stability and reversible fluctuation are met simultaneously, but the complete criteria for either state are not fully met, then a tendency judgment is made based on the direction of the score change of the assembly within the most recent time window: when the local slope at multiple recent time points is negative or close to zero, it is judged as stabilizing; when the local slope alternates between positive and negative and the average score is lower than the second score threshold, it is judged as reversible fluctuation; when the local slope is positive and the score continues to rise, it is judged as irreversible instability.

[0047] Table 1: Experimental data on the self-assembly stability assessment of three groups of peptides

[0048] The table above shows the experimental data for evaluating the microstructural stability of three representative groups of peptide self-assembly, corresponding to three stability states: tending towards stability (Group A), reversible fluctuations (Group B), and irreversible instability (Group C). From the temporal changes in the structural perturbation score: Group A's score fluctuated between 0.11 and 0.14, consistently below the first score threshold of 0.2 for three consecutive time points, and the average value of the sliding window remained stable in the range of 0.12-0.13. The absolute value of the local slope was less than the slope threshold of 0.02, showing a negative or near-zero trend, indicating that the position and morphology of the assembly tended towards stability, meeting the criteria for "tending towards stability." The C group score rose continuously from 0.75 to 0.84, and was higher than the third scoring threshold of 0.7 for three consecutive time points. The mean of the sliding window increased from 0.75 to 0.793, and the local slopes were all positive (0.029-0.033). The scoring sequence showed a non-decreasing trend, indicating that the boundary deformation of the assembly continued to intensify and was irreversible, which met the judgment criteria of "irreversible instability".

[0049] Group B exhibited a typical reversible fluctuation process: the score jumped from 0.31 to 0.72 (above the third scoring threshold of 0.7), then fell back to 0.18 (below the first scoring threshold of 0.2), forming a complete "low→high→low" fluctuation cycle; the sliding window mean rose from 0.31 to 0.515 and then fell to 0.403, with the local slope changing from positive (0.096, 0.091) to negative (-0.11), indicating that the score first deteriorated sharply and then recovered on its own. If monitoring continues, a second complete fluctuation cycle is expected to be observed, and the global average score will not exceed the second scoring threshold (0.4), thus meeting the complete criteria for "reversible fluctuation". The above data verifies that the method proposed in this invention can objectively distinguish three stability states through trend analysis of structural perturbation scores, providing a reliable quantitative basis for real-time monitoring of peptide self-assembly processes.

[0050] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0051] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0052] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0053] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for evaluating the stability of the microstructure of peptide self-assembly, characterized in that, The specific steps include: Step 1: Obtain real-time microstructure image sequences of the target peptide during the self-assembly process as initial image data for evaluation; Step 2: Denoise and enhance the initial image data, extract the contour features of the assembly in each frame image, and generate structural contour data represented by the assembly boundary coordinate set; Step 3: Based on the structural contour data, track the centroid displacement and boundary deformation of the same assembly between adjacent time frames, and construct a temporal feature vector describing the geometric evolution of the assembly; Step 4: Input the time-series feature vector into the pre-trained stability discrimination model and output the structural perturbation score of each assembly at different time points; Step 5: Evaluate the stability of the assembly based on the trend of the structural disturbance score over time.

2. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 1, characterized in that: The logic for obtaining the initial image data is as follows: The target polypeptide sample is placed on a microscopic imaging platform, and raw images at multiple time points are continuously acquired at preset time intervals to form a raw image time series. Each frame of the original image time series is timestamped and spatially scaled to generate initial image data with physical size information and relative time order. Based on the initial image data, invalid frames that do not contain assemblies or whose imaging quality does not meet a preset threshold are removed, and the remaining valid frames are arranged in chronological order as the initial image data to be evaluated.

3. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 2, characterized in that: The imaging quality includes signal-to-noise ratio and edge sharpness; The image quality not meeting the preset threshold means that the signal-to-noise ratio of the image is lower than the first preset threshold, or the edge sharpness of the image is lower than the second preset threshold.

4. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 1, characterized in that: The specific execution process of step 2 is as follows: S21: Gaussian filtering for noise reduction and histogram equalization for enhancement are performed sequentially on each frame of the initial image data to obtain the preprocessed image; S22: An adaptive threshold segmentation algorithm is used on the preprocessed image to separate the assembly region from the background region in the image, resulting in a binarized image. S23: Perform morphological closing operation on the assembly region in the binarized image to fill the internal holes of the region and smooth the boundaries to obtain the optimized connected component of the assembly. S24: Extract the boundary pixel coordinates of the connected domain, arrange them in clockwise or counterclockwise order to form a boundary coordinate set, and output the boundary coordinate set as the structural contour data of the assembly in the frame image.

5. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 4, characterized in that: Based on the aforementioned structural contour data, the centroid displacement of the same assembly between adjacent time frames is tracked using the following specific logic: For adjacent time frames and frame Based on the two frames of images, the centroid coordinates of each assembly in each frame are calculated using their respective structural contour data. The calculation formula is as follows: ， In the formula, Indicates the centroid coordinates; is the total number of pixels contained in the structural contour data; i is the index of the boundary pixel. and Let x and y represent the x and y coordinates of the i-th boundary pixel, respectively. Based on the centroid coordinates, the nearest neighbor matching algorithm is used to match the frames. Assembly and frame in The assemblies are matched one-to-one to complete the tracking of the same assembly between adjacent time frames; For the same tracked assembly, calculate its centroid displacement using the following formula: ； In the formula, Indicates that the assembly is from the first Frame to the Centroid displacement of the frame; and These respectively indicate that the assembly is in the frame and frame The centroid coordinates in the equation.

6. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 5, characterized in that: The boundary deformation of the assembly is calculated as follows: For frames Each pixel on the boundary of the assembly in the frame Find the boundary pixel with the closest Euclidean distance in the matrix and perform matching. Calculate the average distance between all matching point pairs as the boundary deformation. The calculation formula is as follows: ； In the formula, Indicates that the assembly is from the first Frame to the The boundary deformation of the frame; the larger the value, the more severe the deformation. Indicates the first Index of boundary pixels in the frame; and They represent the first The first frame The x and y coordinates of each boundary pixel; and They represent the first The first frame The x and y coordinates of each boundary pixel.

7. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 6, characterized in that: The calculated centroid displacement sequence and boundary deformation sequence ; The features are concatenated to construct a temporal feature vector, which takes the following form: ； in, This represents the total number of time frames. The temporal feature vector is used to describe the complete evolution trajectory of the geometry of the assembly during the self-assembly process.

8. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 1, characterized in that: The specific execution process of step 4 is as follows: S41: Obtain multiple sets of peptide self-assembly microstructure image sequences with known stability tags, extract the temporal feature vector of each assembly in each frame image according to steps 1-3, and label the assembly state at each time point with a structural perturbation truth value label according to the stability tag. The structural perturbation truth value label is a continuous value with a value range of 0 to 1, where 0 represents complete stability without perturbation and 1 represents complete instability and disintegration. S42: Construct a stability discrimination model. The model uses a bidirectional long short-term memory network as its basic architecture, comprising an input layer, two bidirectional LSTM hidden layers, a dropout layer, a fully connected layer, and a sigmoid output layer. The input layer receives the temporal feature vector. The bidirectional LSTM hidden layers capture the long-range dependencies of the temporal feature vector in both the forward and backward directions, with each hidden layer containing 64 memory units. The dropout layer has a dropout rate of 0.5 to prevent overfitting. The fully connected layer maps the high-dimensional features output by the LSTM to one-dimensional features. The sigmoid output layer converts the mapping result into a continuous value between 0 and 1, which serves as the structural perturbation score of the assembly at the current time point. S43: Take the temporal feature vector of each time point in the training dataset as the model input, take the structural perturbation ground value label corresponding to the time point as the supervision signal, use the mean squared error as the loss function, and use the Adam optimizer to perform parameter iterative optimization until the loss function converges to obtain the trained stability discrimination model. S44: Input the temporal feature vector of the assembly to be evaluated into the trained stability discrimination model in chronological order. The model outputs the structural perturbation score corresponding to each time point. The structural perturbation score is used to characterize the structural stability of the assembly at that time point. The closer the score is to 0, the more stable it is. The closer the score is to 1, the less stable it is.

9. The method for evaluating the stability of the microstructure of polypeptide self-assembly according to claim 1, characterized in that: The specific execution process of step 5 is as follows: S51: Obtain the structural perturbation score sequence of the assembly at all time points, denoted as... ,in Indicates the first The structural disturbance score at each time point ranges from 0 to 1. This represents the total number of time frames. S52: Perform trend analysis on the structural disturbance score sequence: calculate the sliding window mean sequence and local slope sequence of the sequence, where the width of the sliding window is set to... The average value of the sliding window is The local slope is the linear regression slope of the score within the window. This is used to characterize the direction and rate of local change in the score within the window; S53: Based on the mean and local slope of the sliding window, the stability state of the assembly is determined according to the following judgment logic: If the structural disturbance scores at a number of consecutive time points are all below a preset first score threshold, and the absolute value of the local slope within the corresponding time window is less than the slope threshold, then the assembly is judged to be stable. If there are at least two complete fluctuation cycles in the structural disturbance scoring sequence, and the continuous dwell time in each scoring interval is not less than the preset minimum number of dwell frames, and the arithmetic mean of the structural disturbance scores of the entire time series does not exceed the second scoring threshold, then it is determined that the assembly has reversible fluctuations. The fluctuation cycle is defined as the process of the score rising from below the first score threshold to above the third score threshold and then falling back below the first score threshold, or the process of the score falling from above the third score threshold to below the first score threshold and then rising back above the third score threshold. If the structural disturbance scores at a number of consecutive time points are all higher than the third score threshold, and the local slopes within the corresponding time window are all positive, and the score sequence within the time window shows a non-decreasing trend, then the assembly is judged to have entered an irreversible unstable state. If the conditions for both stability and reversible fluctuation are met simultaneously, but the complete criteria for either state are not fully met, then a tendency judgment is made based on the direction of the score change of the assembly within the most recent time window: when the local slope at multiple recent time points is negative or close to zero, it is judged as stabilizing; when the local slope alternates between positive and negative and the average score is lower than the second score threshold, it is judged as reversible fluctuation; when the local slope is positive and the score continues to rise, it is judged as irreversible instability.

Images