Protein crystal rapid identification method and device

Abstract

The invention discloses a method and a device for quickly identifying protein crystals, which are used for solving the technical problem that the production state of the protein crystals cannot be quickly judged and analyzed in the prior art. The method comprises the following steps: acquiring a crystal microscopic image of a to-be-detected protein crystal; inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology identification, and obtaining category information and contour data of the protein crystal to be detected; performing post-processing analysis based on the category information and the contour data to obtain a kinetic analysis result of the to-be-detected protein crystal, so that through high-throughput analysis based on artificial intelligence and in combination with a post-processing program set for quantitative analysis of the protein crystal, the same protein crystals with different morphologies can be quickly identified, and the detection accuracy is improved. And subsequent kinetic analysis can be carried out on the protein crystal.

Description

Technical Field

[0001] The present invention relates to the technical field of protein crystallization research, and in particular to a protein crystal rapid identification method, a protein crystal rapid identification device, an electronic device and a storage medium. Background Art

[0002] Protein crystals have a wide range of applications. In practical applications, protein crystals are often used in the field of structural biology for X-ray diffraction to determine the three-dimensional structure of proteins or protein complexes. Protein crystals also have potential advantages in the way they are used in the biopharmaceutical industry. For example, they can stabilize the properties of drugs and control the release of drugs. Protein crystals can also play a role in materials science as templates for the in situ growth of nanomaterials.

[0003] In recent years, high-throughput screening instruments and technologies for screening protein crystallization conditions have made great progress. The main implementation platforms are crystallization screening robots and microfluidic chips. However, the large amount of crystallization test data generated by these two methods still requires long-term, labor-intensive inspections, which is not conducive to rapid judgment and analysis of the state of the resulting protein crystals. Therefore, how to construct a rapid processing of protein crystallization test results is an important issue that needs to be solved for high-throughput screening of protein crystallization conditions. Summary of the invention

[0004] The present invention provides a protein crystal rapid identification method, a protein crystal rapid identification device, an electronic device and a storage medium, which are used to solve or partially solve the technical problem that the existing related technology cannot quickly judge and analyze the generation state of protein crystals.

[0005] The present invention provides a method for rapid identification of protein crystals, the method comprising:

[0006] Acquire crystal microscopic images of the protein crystals to be tested;

[0007] Inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition to obtain category information and contour data of the protein crystal to be tested;

[0008] Post-processing analysis is performed based on the category information and profile data to obtain the dynamic analysis results of the protein crystal to be tested.

[0009] Optionally, the step of inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition to obtain category information and contour data of the protein crystal to be tested comprises:

[0010] Inputting the crystal microscopic image into a pre-trained instance segmentation model to identify the crystal morphology of the protein crystal to be tested;

[0011] Based on the crystal morphology recognition result, performing crystal morphology classification on the crystal microscopic image;

[0012] According to the crystal morphology classification result, the crystal microscopic image is segmented to obtain the category information and contour data of the protein crystal to be tested.

[0013] Optionally, the construction and training process of the instance segmentation model includes:

[0014] Obtain microscopic images of protein crystals under different concentration conditions of protein and precipitant solutions;

[0015] Marking the morphology of protein crystals with different morphologies in each of the protein crystal microscopic images, and further integrating them into a protein crystal data set;

[0016] An initial segmentation model is constructed based on the YOLOv5 model framework, and the initial segmentation model is trained and verified according to the protein crystal dataset to obtain an instance segmentation model.

[0017] Optionally, the performing model training and verification on the initial segmentation model according to the protein crystal dataset to obtain an instance segmentation model comprises:

[0018] Dividing the protein crystal data set into an identification training set and an identification verification set according to a preset ratio;

[0019] The initial segmentation model is trained according to the recognition training set, and during the training process, the protein crystal microscopic images under different concentration conditions of protein and precipitant solution are recognized and segmented based on the morphological markers;

[0020] The recognition validation set is used to perform model validation on the trained initial segmentation model, obtain model performance evaluation, and output the final instance segmentation model.

[0021] Optionally, the method further comprises:

[0022] The prepared protein solution and precipitant solutions of different concentrations are mixed according to a preset volume ratio to form solutions of different supersaturation degrees;

[0023] The protein crystals produced in the solutions with different supersaturation degrees are placed on an inverted microscope stage, and corresponding protein crystal microscopic images are collected.

[0024] Optionally, performing post-processing analysis based on the category information and the profile data to obtain the kinetic analysis result of the protein crystal to be tested includes:

[0025] Combining the category information with the contour data, performing crystal contour processing on the protein crystal to be tested to obtain crystal contour parameters;

[0026] Counting the number of crystals produced of the protein crystal to be tested;

[0027] The kinetic analysis results of the protein crystal to be tested are obtained by analyzing the changes of the crystal contour parameters and the number of crystals produced over time.

[0028] Optionally, the protein crystal to be tested corresponds to a plurality of groups of different morphology categories, each group of the morphology categories corresponds to a plurality of segmented crystals, and the combining of the category information and the contour data to perform crystal contour processing on the protein crystal to be tested to obtain crystal contour parameters includes:

[0029] Determining the number of segmented crystals corresponding to each morphological category in the protein crystal to be tested according to the category information;

[0030] For each of the split crystals in each group of morphology categories, the outline of the split crystal is rotated until the long side of the outline of the split crystal is rotated to a vertical direction;

[0031] Extracting the length value and the width value of the segmented crystal from the outline data, calculating the number of all pixels within the segmented crystal outline, and obtaining the crystal size of the segmented crystal based on the length value, the width value and the number of all pixels in combination with a scale;

[0032] The length value, the width value and the crystal size of each of the segmented crystals are integrated together as the crystal contour parameters of the protein crystal to be tested.

[0033] The present invention also provides a protein crystal rapid identification device, comprising:

[0034] A crystal microscopic image acquisition module, used to acquire a crystal microscopic image of the protein crystal to be tested;

[0035] A crystal morphology recognition module, used for inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition, and obtaining the category information and contour data of the protein crystal to be tested;

[0036] A post-processing analysis module is used to perform post-processing analysis based on the category information and the profile data to obtain a kinetic analysis result of the protein crystal to be tested.

[0037] The present invention also provides an electronic device, the device comprising a processor and a memory:

[0038] The memory is used to store program code and transmit the program code to the processor;

[0039] The processor is used to execute any one of the above methods for rapid protein crystal identification according to the instructions in the program code.

[0040] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium is used to store program codes, and the program codes are used to execute the method for rapid protein crystal identification as described in any one of the above items.

[0041] It can be seen from the above technical solutions that the present invention has the following advantages:

[0042] A method for rapid protein crystal identification is provided. First, a crystal microscopic image of the protein crystal to be tested is obtained; then the crystal microscopic image is input into a pre-trained instance segmentation model for morphology recognition to obtain the category information and contour data of the protein crystal to be tested; finally, post-processing analysis is performed based on the category information and contour data to obtain the kinetic analysis results of the protein crystal to be tested.

[0043] Therefore, through high-throughput analysis based on artificial intelligence, combined with a post-processing program set up for quantitative analysis of protein crystals, it is not only possible to quickly identify protein crystals that are identical but have different morphologies, but also to perform subsequent kinetic analysis of the protein crystals. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A flowchart of the steps of a method for rapid identification of protein crystals;

[0046] Figure 2 This is a schematic diagram of the overall process of a method for rapid identification of protein crystals;

[0047] Figure 3 Representative microscopic images of a protein under different crystallization conditions in a crystallization screening experiment;

[0048] Figure 4 A confusion matrix for evaluating the recognition performance of protein crystals by the YOLOv5-based instance segmentation model;

[0049] Figure 5A schematic diagram of contour rotation in parameter post-processing for quantitative analysis of protein crystals;

[0050] Figure 6 This is a schematic diagram of the curve of the number of a protein crystal changing with time;

[0051] Figure 7 A schematic diagram of a curve showing the change of the aspect ratio of a protein crystal over time;

[0052] Figure 8 This is a schematic diagram of the curve of the area of ​​a protein crystal changing with time;

[0053] Fig. 9 This is a structural block diagram of a device for rapid protein crystal identification. DETAILED DESCRIPTION

[0054] The embodiments of the present invention provide a method for rapid identification of protein crystals, a device for rapid identification of protein crystals, an electronic device and a storage medium, which are used to solve or partially solve the technical problem in the existing related technologies that the production state of protein crystals cannot be quickly determined and analyzed.

[0055] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] As an example, in recent years, high-throughput screening instruments and technologies for screening protein crystallization conditions have made great progress. The main implementation platforms are crystallization screening robots and microfluidic chips. However, the large amount of crystallization test data generated by these two methods still requires long-term, labor-intensive inspections, which is not conducive to rapid judgment and analysis of the state of the resulting protein crystals. Therefore, how to construct a method for rapidly processing the results of protein crystallization tests is an important issue that needs to be solved for high-throughput screening of protein crystallization conditions.

[0057] Further analysis shows that the protein crystallization process is affected by many factors including the pH value of the solution, the type and concentration of the precipitant, and the protein concentration. Different crystallization conditions may form protein crystals with different properties such as morphology and size. In addition, the different morphologies and sizes of crystals affect the application of protein crystals. For example, the most difficult bottleneck problem of the X-ray diffraction method is that it is difficult to obtain single crystals that are large enough and of high quality. Protein crystals are usually required to have a longest dimension of at least 0.1 mm (unit: millimeter) to provide a lattice with sufficient volume to obtain good diffraction fringes.

[0058] Therefore, one of the core invention points of the embodiment of the present invention is: in view of the deficiencies in the prior art, a method for identifying and analyzing the crystallization dynamics of high-throughput protein crystals based on artificial intelligence is proposed. First, a crystal microscopic image of the protein crystal to be tested is obtained; then the crystal microscopic image is input into the pre-trained instance segmentation model for morphology recognition, and the category information and contour data of the protein crystal to be tested are obtained; finally, post-processing analysis is performed based on the category information and contour data to obtain the dynamic analysis results of the protein crystal to be tested. Thus, through high-throughput analysis based on artificial intelligence, combined with the post-processing program set for quantitative analysis of protein crystals, it is possible to not only quickly judge the results of protein crystallization tests, but also quickly identify protein crystals with the same but different morphologies; it is also possible to quantitatively analyze protein crystals with different morphologies by segmenting protein crystals and quantitatively analyzing the growth parameters of protein crystals such as length, width, size and number, so as to achieve the analysis of protein crystallization dynamics. Further, through the technical solution of the present invention, it is not only helpful to prepare protein crystals of suitable size, morphology and application value, but also to provide new ideas for protein crystallization research, which is of great significance and value for studying new protein structures and important biological applications.

[0059] Reference Figure 1 , shows a flowchart of a method for rapid identification of protein crystals provided by an embodiment of the present invention, which may specifically include the following steps:

[0060] Step 101, obtaining a crystal microscopic image of a protein crystal to be tested;

[0061] In practical applications, it is first necessary to obtain a crystal microscopic image of the protein crystal to be identified. The crystal microscopic image can be obtained by mixing a prepared protein solution and a precipitant solution of a preset concentration in a certain volume ratio to form a corresponding supersaturated solution, then placing the protein crystals produced in the supersaturated solution on an inverted microscope stage for observation, and collecting the corresponding crystal microscopic image by photographing or the like.

[0062] Step 102, inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition to obtain category information and contour data of the protein crystal to be tested;

[0063] The purpose of the technical solution of the present invention is to propose a method for high-throughput analysis based on artificial intelligence that can be used for the cultivation, identification and kinetic analysis of protein crystals.

[0064] As an important tool for analyzing images, artificial intelligence can achieve pixel-level segmentation of different targets and distinction between different categories. This can not only make up for the shortcomings of traditional crystallization test results, which require long processing time and only judge whether the entire image is crystallized, but also solve the current problem of difficulty in quickly identifying protein crystals with different morphologies and sizes.

[0065] Through further analysis, the present invention believes that instance segmentation, as a computer vision task that can predict the exact pixel boundaries of each independent object in each image, can meet the above-mentioned protein crystal recognition and segmentation requirements. Among the many neural network learning models, the YOLOv5 (You Only Look Once version 5, an open source deep learning model) instance segmentation network performs well, and compared with other networks, it has faster and more accurate prediction capabilities on the same large data set in the same operating environment.

[0066] To achieve the above goals, it is necessary to build and train an instance segmentation model based on a certain scale of protein crystal dataset combined with the YOLOv5 network framework.

[0067] Based on this, the present invention firstly carried out a protein crystallization screening experiment. The specific process is: the prepared protein solution and the precipitant solution of different concentrations are mixed according to a preset volume ratio (a certain volume ratio) to form solutions of different supersaturation; the protein crystals produced in the solutions of different supersaturation are placed on the stage of an inverted microscope, and the corresponding protein crystal microscopic images are collected.

[0068] Next is the construction and training process of the instance segmentation model, which mainly includes: obtaining protein crystal microscopic images under different protein and precipitant solution concentration conditions (i.e., images obtained through protein crystal screening experiments); morphologically labeling protein crystals of different morphologies in each protein crystal microscopic image (such as setting labels according to the morphologies of different protein crystals, such as the currently common rod-shaped, square-shaped, and sheet-shaped morphologies), and further integrating them into a protein crystal dataset; constructing an initial segmentation model based on the YOLOv5 model framework, and training and verifying the initial segmentation model based on the protein crystal dataset to obtain an instance segmentation model.

[0069] Furthermore, the initial segmentation model is trained and verified according to the protein crystal dataset to obtain an instance segmentation model, which can be specifically: the protein crystal dataset is divided into a recognition training set and a recognition verification set according to a preset ratio; the initial segmentation model is trained according to the recognition training set, and during the training process, the protein crystal microscopic images under different protein and precipitant solution concentration conditions are recognized and segmented based on morphological markers; the trained initial segmentation model is verified using the recognition verification set to obtain a model performance evaluation, and the final instance segmentation model is output.

[0070] Based on the construction and training of the instance segmentation model, in practical applications, artificial intelligence-based protein crystal recognition and classification can be quickly achieved through the trained instance segmentation model.

[0071] Specifically, the process of inputting the crystal microscopic image into the pre-trained instance segmentation model for morphology recognition to obtain the category information and contour data of the protein crystal to be tested can be: inputting the crystal microscopic image into the pre-trained instance segmentation model to identify the crystal morphology of the protein crystal to be tested; based on the crystal morphology recognition result, performing crystal morphology classification on the crystal microscopic image; according to the crystal morphology classification result, segmenting the crystal microscopic image to obtain the category information and contour data of the protein crystal to be tested.

[0072] Step 103: Perform post-processing analysis based on the category information and the profile data to obtain a kinetic analysis result of the protein crystal to be tested.

[0073] Finally, based on the recognition and analysis results of the crystal microscopy images of the protein crystals to be tested by the instance segmentation model, a post-processing program for quantitative analysis of protein crystals is set up.

[0074] Specifically, the process of performing post-processing analysis based on the category information and the profile data to obtain the kinetic analysis results of the protein crystal to be tested can be achieved by executing the following steps S01 to S03:

[0075] Step S01: combining the category information and the contour data, performing crystal contour processing on the protein crystal to be tested to obtain crystal contour parameters;

[0076] After the crystal microscopic image of the protein crystal to be tested is identified and segmented by the instance segmentation model, it can be concluded that the protein crystals to be identified and tested can correspond to multiple groups of different morphological categories (such as rods, squares, sheets, etc.), and each group of morphological categories can correspond to multiple segmented crystals. Based on this, the specific process of step S01 can be implemented by executing the following sub-steps S011 to S014:

[0077] Step S011: determining the number of segmented crystals corresponding to each morphological category in the protein crystal to be tested according to the category information;

[0078] Step S012: for each segmented crystal in each group of morphology categories, rotating the outline of the segmented crystal until the long side of the outline of the segmented crystal is rotated to a vertical direction;

[0079] Step S013: extracting the length value and width value of the segmented crystal from the contour data, calculating the number of all pixels in the segmented crystal contour, and obtaining the crystal size of the segmented crystal based on the length value, width value and all pixel numbers combined with the scale;

[0080] Step S014: Integrate the length value, width value and crystal size of each segmented crystal into the crystal contour parameters of the protein crystal to be tested.

[0081] Step S02: Counting the number of crystals produced of the protein crystal to be tested;

[0082] The number of crystal contours identified by the instance segmentation model is counted as the number of protein crystals grown (i.e., the number of crystals generated).

[0083] Step S03: By analyzing the changes of crystal contour parameters (length, width, and size of each segmented crystal) and the number of crystals produced over time, the kinetic analysis results of the protein crystal to be tested are obtained.

[0084] In an embodiment of the present invention, a method for identifying and analyzing the crystallization kinetics of high-throughput protein crystals based on artificial intelligence is proposed. Through high-throughput analysis based on artificial intelligence, combined with a post-processing program set for quantitative analysis of protein crystals, it is possible not only to quickly determine the results of protein crystallization experiments, but also to quickly identify protein crystals that are identical but have different morphologies; it is also possible to quantitatively analyze protein crystals with different morphologies by segmenting protein crystals and quantitatively analyzing growth parameters such as the length, width, size, and number of protein crystals, thereby achieving a kinetic analysis of protein crystallization. Furthermore, the technical solution of the present invention will not only help prepare protein crystals of suitable size, morphology, and application value, but will also provide new ideas for protein crystallization research, which is of great significance and value to the study of new protein structures and important biological applications.

[0085] For better explanation, the following briefly describes the general process of rapid protein crystal identification introduced in the previous content, combined with specific protein crystal screening experiments and subsequent analysis and processing. Figure 2 , showing the overall flow chart of a method for rapid protein crystal identification provided by an embodiment of the present invention.

[0086] (I) Protein crystallization screening experiment. Protein crystals are generated through protein crystallization condition screening experiment and crystal microscopic images are collected. The specific steps include the following:

[0087] Step 1) First, lysozyme stock solution (60 mg / ml (unit: mg / ml)) and sodium chloride solution of different concentrations (0.5M-5M (unit: volume molar concentration, expressed as mol / L, moles / liter)) were mixed at a certain volume ratio (varying from 1:9 to 9:1). Then, a screening experiment was performed in a 96-well plate: 10μl (volume unit: microliter) of the mixed crystallization solution of a specific volume ratio was added to each well, and then covered with 50μl of liquid paraffin. Ten parallel experiments were performed under each concentration condition to form droplets of different supersaturation.

[0088] Step 2) Place the well plate in a 4°C (unit: degrees Celsius) environment for incubation. Continuous observations were performed at 0h (unit: hours), 24h, 48h, 72h, 96h, 120h, 144h, 168h, 192h and 216h. The crystal microscopic image of each well was acquired by shooting with a microscope camera. The image resolution of the crystal microscopic image is 1280×640, and the photo file name is named in chronological order starting from 1 according to the sodium chloride concentration and the volume ratio of lysozyme to sodium chloride. Each crystal microscopic image contains protein crystals with different morphologies produced by experiments with different concentrations and volume ratios.

[0089] Taking 100 μm (unit: micrometer) as the scale unit, the protein crystallization experiment results collected in this example for crystallization condition screening are as follows: Figure 3 As shown, representative microscopic images under different crystallization conditions in a protein crystallization screening experiment are specifically shown.

[0090] (II) Protein crystal identification and classification based on artificial intelligence. The specific steps include:

[0091] Step 1) In combination with the previous protein crystal screening experiment, a total of 200 crystal microscopic images containing protein crystals of different morphologies were collected in this example to create a protein crystal dataset for artificial intelligence.

[0092] Step 2) Use the 4-fold cross-validation method to train and evaluate the instance segmentation network built based on YOLOv5. The dataset is divided into a recognition training set and a recognition validation set at a ratio of 4:1. Among them, the recognition training set is used to train the YOLOv5 network, and the categories of all crystal microscopy images in each sub-dataset used for cross-validation contain six labels as morphological markers: "rod", "cubic", "plate", "sheet", "tail" and "end".

[0093] Step 3) Download the weight file for training large datasets and load it onto the YOLOv5 instance segmentation network to initialize the network parameters. Then, based on the recognition training set, transfer learning (TL) is used to train the YOLOv5 instance segmentation network. During the training process, the optimization method used by the model is stochastic gradient descent (SGD). Set the learning rate to 0.01, the batch size to 4, and the number of training iterations to 300. For example, the software environment is Ubuntu 16.0, CUDA 11.1 and the corresponding cudnn 8.0, and the framework selects Pytorch under python 3.8.

[0094] Step 4) Use the recognition validation set to verify the instance segmentation model obtained after training. Through experiments, the recognition and segmentation speed of each protein crystal microscopy image on the recognition validation set reached 15ms (unit: milliseconds), and the type accuracy of "rod", "cubic", "plate", "sheet", "tail" and "end" reached 98%, 99%, 98%, 99%, 98% and 99% respectively. Thus, the protein crystal type and location information is output through the YOLOv5 instance segmentation neural network, and the average accuracy can reach 98.5%. Exemplarily, the confusion matrix used to evaluate the recognition performance of the YOLOv5-based instance segmentation model for protein crystals is as follows: Figure 4 shown.

[0095] Step 5) Referring to the protein crystal screening experimental method, obtain the crystal microscopic image of the protein crystal that needs to be detected and identified (usually, in order to save experimental time and cost, the protein crystals that need to be detected and identified are multiple groups of protein crystals obtained by crystallization under different concentrations or experimental conditions. For the sake of convenience, in the embodiment of the present invention, these crystals that need to be detected and identified are collectively referred to as protein crystals to be tested), and input the crystal microscopic image into the instance segmentation model for morphology recognition to obtain the category information and contour data of the protein crystal.

[0096] (III) Post-processing procedures for quantitative analysis of protein crystals. Specifically, the following steps are included:

[0097] Step 1) According to the category information, determine the number of segmented crystals corresponding to each morphological category in the protein crystal. For each segmented crystal, use the computer vision software Opencv to obtain the minimum circumscribed rectangle and ellipse of the segmented crystal outline and the angle with the horizontal direction according to the coordinates of the segmented crystal contour points and the functions cv2.minAreaRect and cv2.ellipse.

[0098] Step 2) Based on the type of the split crystal, the outline of the crystal is rotated at an angle between the minimum circumscribed rectangle and ellipse of the outline and the horizontal direction until the long side of the outline of the split crystal is rotated to the vertical direction.

[0099] Step 3) Define the line connecting the top and bottom points of the rotated crystal contour data points as the length of the crystal. Similarly, define the line connecting the leftmost point and the rightmost point as the width of the crystal. Then, extract the length and width values ​​of the split crystal from the contour data through function calculation.

[0100] Step 4) Use Opencv's function cv2.contourArea to calculate the crystal contour area of ​​each segmented crystal according to Green's formula.

[0101] Step 5) Convert the crystal length value, width value and contour area calculated by the function into actual length value, width value and area value according to the ruler.

[0102] Thus, by using the instance segmentation model to identify and segment the protein crystal information, the crystal length, width and other parameters are calculated. For example, the rotation angle of the crystal outline is obtained based on the minimum circumscribed rectangle and ellipse of the crystal outline, and then the length and width of the crystal are obtained by rotating the outline and combining the outline data. At the same time, the size of the crystal is obtained by counting the total number of pixels occupied by the crystal outline. For example, the schematic diagram of the outline rotation in the parameter post-processing for quantitative analysis of protein crystals is shown in FIG. Figure 5 As shown. Wherein, L represents the length of the split crystal, width represents the width of the split crystal, and AR represents the aspect ratio of the split crystal.

[0103] Step 6) Count the number of crystal contours identified by the instance segmentation model under each experimental condition as the number of protein crystals grown under this experimental condition (i.e., the number of crystals produced under this concentration condition).

[0104] Step 7) Perform a dynamic analysis of protein crystals based on the time-varying curves of the length, width, area, and number of each segmented crystal in the protein crystal. For example, the concentration of the protein solution affects the supersaturation, and supersaturation, as the main driving force for crystallization, affects the formation of crystals, which is reflected in the number, aspect ratio, and area of ​​the crystals. Figures 6 to 8 As shown. Among them, Figure 6 This is a schematic diagram of the curve showing the change of the number of protein crystals over time; Figure 7 Schematic diagram of the curve of the aspect ratio of protein crystals changing with time; Figure 8 Schematic diagram of the curve showing the change of protein crystal area over time.

[0105] Combination Figures 6 to 8 Analysis shows that at the same precipitant concentration, the lower the protein concentration (e.g. Figures 6 to 8 Under the condition of 12 mg / ml protein solution), the smaller the supersaturation driving crystallization, the longer it takes to push the solute in the solution to the solid crystal. From this, it can be seen that the number, aspect ratio and number of the 12 mg / ml protein solution concentration appear slower than those of the other solution concentrations.

[0106] Therefore, by adopting the technical solution of the present invention, through high-throughput analysis based on artificial intelligence, combined with a post-processing program set for quantitative analysis of protein crystals, it is not only possible to quickly identify protein crystals that are identical but have different morphologies, but also to perform dynamic analysis of protein crystals by segmenting protein crystals and quantitatively analyzing growth parameters such as the length, width, size, and number of protein crystals. Therefore, the high-throughput analysis method based on artificial intelligence provided by the embodiment of the present invention has broad application prospects in the fields of protein crystal dynamics, structural biology, and the pharmaceutical industry.

[0107] Reference Fig. 9 , shows a structural block diagram of a protein crystal rapid identification device provided by an embodiment of the present invention, which may specifically include:

[0108] A crystal microscopic image acquisition module 901 is used to acquire a crystal microscopic image of the protein crystal to be tested;

[0109] A crystal morphology recognition module 902 is used to input the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition, and obtain the category information and contour data of the protein crystal to be tested;

[0110] The post-processing analysis module 903 is used to perform post-processing analysis based on the category information and the profile data to obtain the kinetic analysis results of the protein crystal to be tested.

[0111] In an optional embodiment, the crystal morphology recognition module 902 includes:

[0112] A crystal morphology recognition submodule, used for inputting the crystal microscopic image into a pre-trained instance segmentation model to recognize the crystal morphology of the protein crystal to be tested;

[0113] A crystal morphology classification module, used for classifying the crystal morphology of the crystal microscopic image based on the crystal morphology recognition result;

[0114] The crystal microscopic image segmentation module is used to segment the crystal microscopic image according to the crystal morphology classification result to obtain the category information and contour data of the protein crystal to be tested.

[0115] In an optional embodiment, the device further includes:

[0116] A protein crystal microscopic image acquisition module, used to acquire protein crystal microscopic images under different protein and precipitant solution concentration conditions;

[0117] A morphology labeling module, used for labeling protein crystals of different morphologies in each of the protein crystal microscopic images, and further integrating them into a protein crystal data set;

[0118] The model training and verification module is used to build an initial segmentation model based on the YOLOv5 model framework, and to perform model training and verification on the initial segmentation model according to the protein crystal dataset to obtain an instance segmentation model.

[0119] In an optional embodiment, the model training and verification module includes:

[0120] A data set division module, used for dividing the protein crystal data set into an identification training set and an identification verification set according to a preset ratio;

[0121] A model training module, used for training the initial segmentation model according to the recognition training set, and in the training process, identifying and segmenting protein crystal microscopic images under different concentration conditions of protein and precipitant solution based on morphological markers;

[0122] The model verification module is used to use the recognition verification set to perform model verification on the trained initial segmentation model, obtain model performance evaluation, and output a final instance segmentation model.

[0123] In an optional embodiment, the device further includes:

[0124] A solution preparation module is used to mix the prepared protein solution and precipitant solutions of different concentrations according to a preset volume ratio to form solutions of different supersaturation degrees;

[0125] The microscopic image acquisition module is used to place the protein crystals produced in the solutions with different supersaturation degrees on the stage of an inverted microscope and to acquire corresponding protein crystal microscopic images.

[0126] In an optional embodiment, the post-processing analysis module 903 includes:

[0127] A crystal contour processing module, used to combine the category information and contour data to perform crystal contour processing on the protein crystal to be tested to obtain crystal contour parameters;

[0128] A crystal quantity statistics module, used for counting the number of crystals produced of the protein crystal to be tested;

[0129] The kinetic analysis module is used to obtain the kinetic analysis results of the protein crystal to be tested by analyzing the changes of the crystal contour parameters and the number of crystals produced over time.

[0130] In an optional embodiment, the protein crystal to be tested corresponds to a plurality of groups of different morphological categories, each group of the morphological categories corresponds to a plurality of segmented crystals, and the crystal contour processing module includes:

[0131] A segmented crystal quantity determination module, used to determine the quantity of segmented crystals corresponding to each morphological category in the protein crystal to be tested according to the category information;

[0132] A crystal contour rotation module, used for rotating the contour of the split crystal until the long side of the contour of the split crystal is rotated to a vertical direction;

[0133] A crystal size calculation module, used to extract the length value and width value of the segmented crystal from the outline data, calculate the number of all pixels in the segmented crystal outline, and obtain the crystal size of the segmented crystal based on the length value, the width value and the number of all pixels in combination with a scale;

[0134] The crystal contour parameter integration module is used to integrate the length value, the width value and the crystal size of each of the segmented crystals into the crystal contour parameters of the protein crystal to be tested.

[0135] As for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the aforementioned method embodiment.

[0136] An embodiment of the present invention further provides an electronic device, the device comprising a processor and a memory:

[0137] The memory is used to store the program code and transmit the program code to the processor;

[0138] The processor is used to execute the protein crystal rapid identification method of any embodiment of the present invention according to the instructions in the program code.

[0139] An embodiment of the present invention further provides a computer-readable storage medium, which is used to store program codes, and the program codes are used to execute the protein crystal rapid identification method of any embodiment of the present invention.

[0140] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0141] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0142] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0143] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0144] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0145] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for rapid identification of protein crystals, characterized in that: include: Acquire crystal microscopic images of the protein crystals to be tested; Inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition to obtain category information and contour data of the protein crystal to be tested; Post-processing analysis is performed based on the category information and profile data to obtain the dynamic analysis results of the protein crystal to be tested.

2. The method for rapid protein crystal identification according to claim 1, characterized in that: The step of inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition to obtain category information and contour data of the protein crystal to be tested includes: Inputting the crystal microscopic image into a pre-trained instance segmentation model to identify the crystal morphology of the protein crystal to be tested; Based on the crystal morphology recognition result, performing crystal morphology classification on the crystal microscopic image; According to the crystal morphology classification result, the crystal microscopic image is segmented to obtain the category information and contour data of the protein crystal to be tested.

3. The method for rapid protein crystal identification according to claim 2, characterized in that: The construction and training process of the instance segmentation model includes: Obtain microscopic images of protein crystals under different concentration conditions of protein and precipitant solutions; Marking the morphology of protein crystals with different morphologies in each of the protein crystal microscopic images, and further integrating them into a protein crystal data set; An initial segmentation model is constructed based on the YOLOv5 model framework, and the initial segmentation model is trained and verified according to the protein crystal dataset to obtain an instance segmentation model.

4. The method for rapid protein crystal identification according to claim 3, characterized in that: The performing model training and verification on the initial segmentation model according to the protein crystal data set to obtain an instance segmentation model comprises: Dividing the protein crystal data set into an identification training set and an identification verification set according to a preset ratio; The initial segmentation model is trained according to the recognition training set, and during the training process, the protein crystal microscopic images under different concentration conditions of protein and precipitant solution are recognized and segmented based on the morphological markers; The recognition validation set is used to perform model validation on the trained initial segmentation model, obtain model performance evaluation, and output the final instance segmentation model.

5. The method for rapid protein crystal identification according to claim 3 or 4, characterized in that: Also includes: The prepared protein solution and precipitant solutions of different concentrations are mixed according to a preset volume ratio to form solutions of different supersaturation degrees; The protein crystals produced in the solutions with different supersaturation degrees are placed on an inverted microscope stage, and corresponding protein crystal microscopic images are collected.

6. The method for rapid protein crystal identification according to claim 1, characterized in that: The post-processing analysis based on the category information and the profile data to obtain the kinetic analysis results of the protein crystal to be tested includes: Combining the category information with the contour data, performing crystal contour processing on the protein crystal to be tested to obtain crystal contour parameters; Counting the number of crystals produced of the protein crystal to be tested; The kinetic analysis results of the protein crystal to be tested are obtained by analyzing the changes of the crystal contour parameters and the number of crystals produced over time.

7. The method for rapid protein crystal identification according to claim 6, characterized in that: The protein crystal to be tested corresponds to a plurality of groups of different morphological categories, each group of the morphological categories corresponds to a plurality of segmented crystals, and the crystal contour processing is performed on the protein crystal to be tested by combining the category information with the contour data to obtain crystal contour parameters, including: Determining the number of segmented crystals corresponding to each morphological category in the protein crystal to be tested according to the category information; For each of the split crystals in each group of morphology categories, the outline of the split crystal is rotated until the long side of the outline of the split crystal is rotated to a vertical direction; Extracting the length value and the width value of the segmented crystal from the outline data, calculating the number of all pixels within the segmented crystal outline, and obtaining the crystal size of the segmented crystal based on the length value, the width value and the number of all pixels in combination with a scale; The length value, the width value and the crystal size of each of the segmented crystals are integrated together as the crystal contour parameters of the protein crystal to be tested.

8. A protein crystal rapid identification device, characterized in that: include: A crystal microscopic image acquisition module, used to acquire a crystal microscopic image of the protein crystal to be tested; A crystal morphology recognition module, used for inputting the crystal microscopic image into a pre-trained instance segmentation model for morphology recognition, and obtaining the category information and contour data of the protein crystal to be tested; A post-processing analysis module is used to perform post-processing analysis based on the category information and the profile data to obtain a kinetic analysis result of the protein crystal to be tested.

9. An electronic device, characterized in that: The device comprises a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the protein crystal rapid identification method according to any one of claims 1 to 7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium is used to store program codes, and the program codes are used to execute the method for rapid protein crystal identification according to any one of claims 1 to 7.

Images