A method, device, electronic device and storage medium for screening anti-hypoxia drugs

Abstract

A method, device, electronic device and storage medium for screening anti-hypoxic drugs provided by the present invention relate to the technical field of drug screening. The method obtains the whole-field images of each cell group cultured in a non-hypoxic environment, cultured in a hypoxic environment without drug treatment, and cultured in a hypoxic environment with a candidate drug treatment. Using a pre-trained cell portrait segmentation network, subcellular structure images of multiple cells in each cell group are extracted from the whole-field images of each cell group and input into a pre-trained cell hypoxia scoring network to obtain the hypoxia scores of multiple cells in each cell group. According to the hypoxia scores of multiple cells in each cell group, it is determined whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions, thereby improving the screening speed and the accuracy of the screening results.

Description

Technical Field

[0001] The present invention relates to the technical field of drug screening, and more particularly, to a method, apparatus, electronic device, and storage medium for screening anti-hypoxia drugs. Background Art

[0002] High altitude pulmonary edema is a potentially fatal high altitude disease caused by biological and morphological changes in vascular endothelial and alveolar epithelial cells due to hypoxia.

[0003] Existing methods for screening drugs against high altitude pulmonary edema rely on specific biomarkers and staining processes to qualitatively or semi-quantitatively analyze changes in cell morphology, suffering from problems such as slow speed and strong subjectivity of detection results. Summary of the Invention

[0004] The present invention provides a method, apparatus, electronic device, and storage medium for screening anti-hypoxia drugs, which can improve the screening speed and the accuracy of screening results.

[0005] Embodiments of the present invention may be implemented as follows:

[0006] In a first aspect, the present invention provides a method for screening anti-hypoxia drugs, the method comprising:

[0007] Obtaining full-field images of a first cell group, a second cell group, and a third cell group, wherein each cell in the first cell group is cultured in a non-hypoxic environment, each cell in the second cell group is cultured in a hypoxic environment and not treated with a drug, and each cell in the third cell group is cultured in a hypoxic environment and treated with a candidate drug;

[0008] Using a pre-trained cell portrait segmentation network to extract subcellular structure images of multiple cells in the first cell group from the full-field image of the first cell group, extract subcellular structure images of multiple cells in the second cell group from the full-field image of the second cell group, and extract subcellular structure images of multiple cells in the third cell group from the full-field image of the third cell group;

[0009] Inputting the subcellular structure images of multiple cells in the first cell group, the subcellular structure images of multiple cells in the second cell group, and the subcellular structure images of multiple cells in the third cell group into a pre-trained cell hypoxia scoring network respectively to obtain hypoxia scores of multiple cells in the first cell group, hypoxia scores of multiple cells in the second cell group, and hypoxia scores of multiple cells in the third cell group;

[0010] Determine whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells in the first cell group, the hypoxia scores of multiple cells in the second cell group, and the hypoxia scores of multiple cells in the third cell group.

[0011] Optionally, the hypoxia score is positively correlated with the degree of cell hypoxia. The step of determining whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells in the first cell group, the hypoxia scores of multiple cells in the second cell group, and the hypoxia scores of multiple cells in the third cell group includes:

[0012] Take the mean of the hypoxia scores of multiple cells in the first cell group as the hypoxia score of the first cell group, take the mean of the hypoxia scores of multiple cells in the second cell group as the hypoxia score of the second cell group, and take the mean of the hypoxia scores of multiple cells in the third cell group as the hypoxia score of the third cell group;

[0013] If the hypoxia score of the third cell group is greater than the hypoxia score of the first cell group and less than the hypoxia score of the second cell group, determine that the candidate drug is a drug that can improve the cell survival state under hypoxic conditions.

[0014] Optionally, the step of training the cell portrait segmentation network includes:

[0015] Obtain a first training dataset and a second training dataset. The first training dataset includes first pseudocolor images of multiple test cell groups, and the second training dataset includes second pseudocolor images of multiple test cell groups. The oxygen concentration in the culture environment of each test cell group is different. The first pseudocolor image is the nuclear pseudocolor image of the test cell group and a nuclear mask is outlined on the first pseudocolor image. The second pseudocolor image is synthesized from the nuclear pseudocolor image, the actin / Golgi pseudocolor image, and the plasma membrane pseudocolor image of the test cell group, and a cell membrane mask is outlined on the second pseudocolor image;

[0016] Input the first training dataset and the second training dataset into a pre-constructed cell portrait segmentation network for training to obtain a trained cell portrait segmentation network.

[0017] Optionally, the test cell group includes multiple test cells. The step of training the cell hypoxia scoring network includes:

[0018] Obtain multiple first sample images and multiple second sample images. One of the first sample images corresponds to one test cell group, and one test cell group corresponds to at least one of the first sample images. Each of the first sample images contains the cell nuclei of multiple test cells in the corresponding test cell group. One of the second sample images corresponds to one test cell group, and one test cell group corresponds to at least one of the second sample images. Each of the second sample images contains the cell nuclei, actin / Golgi apparatus, and plasma membrane of multiple test cells in the corresponding test cell group;

[0019] Using the multiple first sample images, the multiple second sample images, and the trained cell portrait segmentation network, obtain the subcellular structure images of each test cell in each test cell group;

[0020] Input the subcellular structure images of each test cell in each test cell group into a pre-constructed cell hypoxia scoring network for training to obtain a trained cell hypoxia scoring network.

[0021] Optionally, the step of using the first sample images, the second sample images, and the trained cell portrait segmentation network to obtain the subcellular structure images of each test cell in each test cell group includes:

[0022] Input the multiple first sample images and the multiple second sample images into the trained cell portrait segmentation network, obtain multiple nuclear masks from each of the first sample images, and obtain multiple cell membrane masks from each of the second sample images;

[0023] Match each of the nuclear masks and each of the cell membrane masks to determine the nuclear mask and the cell membrane mask belonging to the same single test cell, and obtain the nuclear masks and cell membrane masks of each test cell;

[0024] Using the nuclear masks and cell membrane masks of each test cell, based on the nuclear pseudo-color images, endoplasmic reticulum pseudo-color images, actin / Golgi apparatus pseudo-color images, mitochondrial pseudo-color images, plasma membrane pseudo-color images, and RNA pseudo-color images of each test cell group, obtain the subcellular structure images of each test cell.

[0025] Optionally, the nuclear mask and the cell membrane mask belonging to the same test cell satisfy the following formula:

[0026]

[0027] Where is the jth nuclear mask, is the kth cell membrane mask.

[0028] Optionally, the subcellular structure image includes a nuclear grayscale image, an endoplasmic reticulum grayscale image, an actin / Golgi grayscale image, a mitochondrial grayscale image, a plasma membrane grayscale image, and an RNA grayscale image. The subcellular structure image of any test cell satisfies the following formula:

[0029]

[0030] Wherein, is the nuclear grayscale image of the test cell, is the endoplasmic reticulum grayscale image of the test cell, is the actin / Golgi grayscale image of the test cell, is the mitochondrial grayscale image of the test cell, is the plasma membrane grayscale image of the test cell, and is the RNA grayscale image of the test cell, is the nuclear mask of the test cell, is the cell membrane mask of the test cell, I D is the nuclear pseudocolor image of the test cell group where the test cell is located, I E is the endoplasmic reticulum pseudocolor image of the test cell group where the test cell is located, is the actin / Golgi pseudocolor image of the test cell group where the test cell is located, i M is the mitochondrial pseudocolor image of the test cell group where the test cell is located, I P is the plasma membrane pseudocolor image of the test cell group where the test cell is located, is the RNA pseudocolor image of the test cell group where the test cell is located.

[0031] In a second aspect, the present invention provides an anti-hypoxia drug screening device, which includes:

[0032] An acquisition module, configured to acquire the full-field images of the first cell group, the full-field images of the second cell group, and the full-field images of the third cell group. Among them, the cells in the first cell group are cultured in a non-hypoxic environment, the cells in the second cell group are cultured in a hypoxic environment and not treated with drugs, and the cells in the third cell group are cultured in a hypoxic environment and treated with a candidate drug;

[0033] An extraction module, configured to use a pre-trained cell image segmentation network to extract subcellular structure images of multiple cells within the first cell group from the whole-field image of the first cell group, extract subcellular structure images of multiple cells within the second cell group from the whole-field image of the second cell group, and extract subcellular structure images of multiple cells within the third cell group from the whole-field image of the third cell group;

[0034] A processing module, configured to respectively input the subcellular structure images of multiple cells within the first cell group, the subcellular structure images of multiple cells within the second cell group, and the subcellular structure images of multiple cells within the third cell group into a pre-trained cell hypoxia scoring network, to obtain hypoxia scores of multiple cells within the first cell group, hypoxia scores of multiple cells within the second cell group, and hypoxia scores of multiple cells within the third cell group;

[0035] A screening module, configured to determine whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells within the first cell group, the hypoxia scores of multiple cells within the second cell group, and the hypoxia scores of multiple cells within the third cell group.

[0036] In a third aspect, the present invention provides an electronic device, which includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, it implements the anti-hypoxia drug screening method as described in the first aspect above.

[0037] In a fourth aspect, the present invention provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, it implements the anti-hypoxia drug screening method as described in the first aspect above.

[0038] Other features and advantages disclosed in the present invention will be described in the subsequent specification, or, some features and advantages can be inferred from the specification or determined without doubt, or can be obtained by implementing the above technologies disclosed in the present invention.

[0039] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. Description of the Drawings

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0041] Figure 1 A flowchart showing the process of a method for screening anti - hypoxia drugs provided by an embodiment of the present invention;

[0042] Figure 2 An example diagram of a full - field image provided by an embodiment of the present invention;

[0043] Figure 3 An example diagram of a sub - cellular structure image provided by an embodiment of the present invention;

[0044] Figure 4 A flowchart showing the implementation process of step S104 provided by an embodiment of the present invention;

[0045] Figure 5 A flowchart showing the training process of a cell portrait segmentation network provided by an embodiment of the present invention;

[0046] Figure 6 A flowchart showing the training process of a cell hypoxia scoring network provided by an embodiment of the present invention;

[0047] Figure 7 A flowchart showing the implementation process of step S302 provided by an embodiment of the present invention;

[0048] Figure 8 A functional unit block diagram of an anti - hypoxia drug screening device provided by an embodiment of the present invention;

[0049] Figure 9 A structural schematic block diagram of an electronic device provided by an embodiment of the present invention.

[0050] Icons: 100 - anti - hypoxia drug screening device; 101 - acquisition module; 102 - extraction module; 103 - processing module; 104 - screening module; 200 - electronic device; 210 - memory; 220 - processor. Detailed implementation manners

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Usually, the components of the embodiments of the present invention described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations.

[0052] Accordingly, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0053] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0054] In addition, terms such as "first" and "second" are used for descriptive purposes only and cannot be construed as indicating or implying relative importance.

[0055] It should be noted that the features in the embodiments of the present invention can be combined with each other without conflict.

[0056] The existing technologies for screening anti-high altitude pulmonary edema drugs mainly have the following deficiencies:

[0057] (1) Relying on experts' visual assessment or semi-automated image analysis techniques, it takes a long time to process a large amount of image data, the analysis speed is slow, and it is difficult to meet the requirements of high-throughput screening;

[0058] (2) Relying on experts' assessment, it is highly subjective, and different assessors may obtain inconsistent results, affecting the accuracy and repeatability of the analysis;

[0059] (3) It is difficult to accurately quantify the subtle changes in cell morphology, especially in the early stage of hypoxic injury, and these subtle changes may be key indicators in drug screening;

[0060] (4) Limited ability to process large-scale data and unable to effectively process thousands of high-content image data;

[0061] (5) Lack of verification methods and over-reliance on experimental animals.

[0062] In order to address the above deficiencies of the existing technologies, the embodiments of the present invention provide an anti-hypoxia drug screening method, which will be introduced in detail below.

[0063] Please refer to Figure 1 , this anti-hypoxia drug screening method includes steps S101 to S104.

[0064] S101, obtaining the full-field images of the first cell group, the full-field images of the second cell group, and the full-field images of the third cell group.

[0065] Among them, the cells in the first cell group are cultured in a non-hypoxic environment, the cells in the second cell group are cultured in a hypoxic environment without drug treatment, and the cells in the third cell group are cultured in a hypoxic environment and treated with a candidate drug.

[0066] In a possible implementation, 10 candidate drugs with potential anti-hypoxic effects can be selected from a natural product library and a traditional Chinese medicine formula library. One non-hypoxic group, one hypoxic group (5% O2), and 10 hypoxic treatment groups are set up. In each group, 6-well A549 and HPMEC cells are seeded at a density of 3000 cells / well in a black plate.

[0067] After cell attachment, the 10 hypoxic treatment groups have their culture media changed and different candidate drugs are added, that is, the 10 hypoxic treatment groups correspond one-to-one with the 10 drugs. The non-hypoxic group and the hypoxic group have their conventional culture media changed.

[0068] After 24 hours of incubation, the hypoxic group and the 10 hypoxic treatment groups are all transferred to a three-gas hypoxic incubator for an additional 24 hours of incubation, while the non-hypoxic group continues to be cultured under normal oxygen concentration during this period. Then, the cells in the non-hypoxic group, the hypoxic group, and the 10 hypoxic treatment groups are stained and photographed.

[0069] Cell staining refers to staining the cells using six specific dyes included in a cell imaging kit (different specific dyes correspond to different cell parts). After the cells are fixed and permeabilized, 50 μl of staining solution is added to each well and incubated in the dark at 37°C and 5% CO2 for 30 minutes, and then the cells are washed three times with HBSS solution.

[0070] Cell photography refers to placing the microplate in a high-content imaging system, configuring the microscope acquisition device, imaging at a magnification of 60× using a water immersion objective, capturing multiple whole-field images in each well, and performing this in six different channels.

[0071] As Figure 2 shown, the whole-field images include six channels, namely the pseudocolor image of the cell nucleus (DNA), the pseudocolor image of the endoplasmic reticulum (ER), the pseudocolor image of actin / golgi apparatus (AG), the pseudocolor image of mitochondria (Mito), the pseudocolor image of the plasma membrane (PM), and the pseudocolor image of RNA.

[0072] The pseudocolor image of the cell nucleus contains the cell nuclei of multiple cells, the pseudocolor image of the endoplasmic reticulum contains the endoplasmic reticulum of multiple cells, the pseudocolor image of actin / golgi apparatus contains the actin / golgi apparatus of multiple cells, the pseudocolor image of mitochondria contains the mitochondria of multiple cells, the pseudocolor image of the plasma membrane contains the plasma membranes of multiple cells, and the pseudocolor image of RNA contains the RNA of multiple cells.

[0073] Understandably, the first cell group is the non-hypoxic group, the second cell group is the hypoxic group, and the third cell group is any hypoxic treatment group.

[0074] S102. Using a pre-trained cell image segmentation network, extract the subcellular structure images of multiple cells within the first cell group from the whole-field image of the first cell group, extract the subcellular structure images of multiple cells within the second cell group from the whole-field image of the second cell group, and extract the subcellular structure images of multiple cells within the third cell group from the whole-field image of the third cell group.

[0075] Among them, the subcellular structures include the nucleus, endoplasmic reticulum, actin / Golgi apparatus, mitochondria, plasma membrane, and RNA.

[0076] As Figure 3 shown, the subcellular structure image of a single cell includes the grayscale image of the nucleus of the cell, the grayscale image of the endoplasmic reticulum of the cell, the grayscale image of the actin / Golgi apparatus of the cell, the grayscale image of the mitochondria of the cell, the grayscale image of the plasma membrane of the cell, and the grayscale image of the RNA of the cell.

[0077] S103. Input the subcellular structure images of multiple cells within the first cell group, the subcellular structure images of multiple cells within the second cell group, and the subcellular structure images of multiple cells within the third cell group into the pre-trained cell hypoxia scoring network respectively, to obtain the hypoxia scores of multiple cells within the first cell group, the hypoxia scores of multiple cells within the second cell group, and the hypoxia scores of multiple cells within the third cell group.

[0078] Understandably, for each cell, input the grayscale image of the nucleus of the cell, the grayscale image of the endoplasmic reticulum of the cell, the grayscale image of the actin / Golgi apparatus of the cell, the grayscale image of the mitochondria of the cell, the grayscale image of the plasma membrane of the cell, and the grayscale image of the RNA of the cell into the pre-trained cell hypoxia scoring network, and the hypoxia score of the cell can be obtained.

[0079] The hypoxia score is positively correlated with the hypoxia degree of the cell, that is, for any cell, if the hypoxia score of the cell is higher, the hypoxia degree of the cell is more severe.

[0080] S104. Determine whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells within the first cell group, the hypoxia scores of multiple cells within the second cell group, and the hypoxia scores of multiple cells within the third cell group.

[0081] In a possible implementation, as Figure 4 shown, step S104 includes sub-steps S104-1 to S104-2.

[0082] S104-1. Take the mean of the hypoxia scores of multiple cells in the first cell group as the hypoxia score of the first cell group, take the mean of the hypoxia scores of multiple cells in the second cell group as the hypoxia score of the second cell group, and take the mean of the hypoxia scores of multiple cells in the third cell group as the hypoxia score of the third cell group.

[0083] S104-2. If the hypoxia score of the third cell group is greater than the hypoxia score of the first cell group and less than the hypoxia score of the second cell group, then determine that the candidate drug is a drug that can improve the cell survival state under hypoxic conditions.

[0084] It can be understood that for the 10 hypoxia treatment groups in the foregoing example, by determining whether the hypoxia score of each hypoxia treatment group is greater than the hypoxia score of the non-hypoxic group and less than the hypoxia score of the hypoxic group, a drug that can improve the cell survival state under hypoxic conditions is screened out from the 10 candidate drugs corresponding one-to-one to the 10 hypoxia treatment groups.

[0085] Generally, for any drug, the degree of hypoxia of the cells under its action is more severe than that of the cells in a non-hypoxic environment, that is, the hypoxia scores of the hypoxia treatment groups corresponding to different candidate drugs are often greater than the hypoxia score of the non-hypoxic group. For a candidate drug that can improve the cell survival state under hypoxic conditions, the hypoxia score of the corresponding hypoxia treatment group is less than the hypoxia score of the hypoxic group. For a candidate drug that cannot improve the cell survival state under hypoxic conditions, the hypoxia score of the corresponding hypoxia treatment group will not be less than the hypoxia score of the hypoxic group. Therefore, in order to improve the screening efficiency, the hypoxia scores of the hypoxia treatment group and the hypoxic group can be directly compared to determine whether the corresponding candidate drug can improve the cell survival state under hypoxic conditions.

[0086] The training process of the cell portrait segmentation network is introduced below.

[0087] Please refer to Figure 5 , the training process of the cell portrait segmentation network includes steps S201 to S202.

[0088] S201. Obtain a first training data set and a second training data set.

[0089] Among them, the first training data set includes first pseudo-color images of multiple test cell groups, the second training data set includes second pseudo-color images of multiple test cell groups, the oxygen concentration in the culture environment of each test cell group is different, the first pseudo-color image is the nuclear pseudo-color image of the test cell group and a nuclear mask is outlined on the first pseudo-color image, and the second pseudo-color image is synthesized from the nuclear pseudo-color image, actin / golgi pseudo-color image and plasma membrane pseudo-color image of the test cell group and a cell membrane mask is outlined on the second pseudo-color image.

[0090] In a possible implementation, A549 and HPMEC cell lines (obtained from the ATCC cell bank) can be cultured in DMEM containing 10% FBS, L-glutamine (0.292 mg / ml), penicillin (100 U / ml), and streptomycin (100 mg / ml). The cells are seeded in a 96-well black plate at a density of 8000 cells / well.

[0091] Four parallel plates were prepared for each cell line and incubated overnight in a cell culture incubator at 37 °C and 5% CO2. After the cells adhered, three plates of each cell line were transferred to hypoxic culture chambers with different oxygen concentrations (1% O2, 3% O2, and 5% O2 respectively), and maintained at 37 °C and 5% CO2 for 24 hours. During this period, the fourth plate of each cell line was continuously cultured under standard oxygen concentration. Then the cells on each plate were stained and photographed.

[0092] The cells on each plate were stained using six specific dyes included in the cell imaging kit (different specific dyes correspond to different cell parts). After the cells were fixed and permeabilized, 50 μl of staining solution was added to each well and incubated in the dark at 37 °C and 5% CO2 for 30 minutes, and then the cells were washed three times with HBSS solution.

[0093] The microplate was placed in a high-content imaging system, and the microscope acquisition device was configured. Images were taken at a magnification of 60× using a water immersion objective, capturing multiple full-field images in each well, and performed in six different channels.

[0094] The original full-field images were converted into pseudo-color images with six channels using ImageJ, and the full-field images were denoised, corrected, and normalized, etc., to ensure the data quality of the obtained cell portraits, meeting the requirements of deep learning model analysis.

[0095] The cells on each black plate placed in the same hypoxic culture chamber were regarded as a test cell group, and the cells on each black plate placed under standard oxygen concentration were regarded as a test cell group, for a total of four test cell groups. It can be understood that multiple full-field images can be obtained for each test cell group.

[0096] For any full-field image, the pseudo-color image of the cell nucleus in this full-field image was used with the LabelMe annotation software, and a mask covering the cell nucleus region was manually outlined by selecting vertices to generate a polygon, obtaining a first pseudo-color image.

[0097] Combine the nuclear pseudo-color image, actin / Golgi pseudo-color image, and plasma membrane pseudo-color image in the full-field image into one image, and then use the LabelMe annotation software on the combined image to manually draw a mask covering the cell membrane area by selecting vertices to generate a polygon, obtaining a second pseudo-color image.

[0098] Traverse each full-field image to obtain multiple first pseudo-color images and multiple second pseudo-color images. The multiple first pseudo-color images form a first training dataset, and the multiple second pseudo-color images form a second training dataset.

[0099] S202, Input the first training dataset and the second training dataset into a pre-constructed cell portrait segmentation network for training to obtain a trained cell portrait segmentation network.

[0100] In the embodiments of the present invention, a cell portrait segmentation network is constructed using transfer learning based on the YOLO v8 framework, and at the same time, the segmentation of the cell nucleus (the segmentation object is the first pseudo-color image) and the segmentation of the cell membrane (the segmentation object is the second pseudo-color image) are performed.

[0101] During the training process, various data augmentation techniques can also be used to enhance the diversity and robustness of the first training dataset and the second training dataset, including adjustments to the HSV (hue, saturation, and value) color space, geometric transformations such as translation, scaling, and horizontal flipping, and mosaic augmentation techniques, etc. Metrics such as the mean average precision (mAP) can also be used to evaluate the effectiveness of object detection and segmentation.

[0102] The training process of the cell hypoxia scoring network is introduced below.

[0103] Please refer to Figure 6 , The training process of the cell hypoxia scoring network includes steps S301 to S303.

[0104] S301, Obtain multiple first sample images and multiple second sample images.

[0105] Among them, one first sample image corresponds to a test cell group, a test cell group corresponds to at least one first sample image, and each first sample image contains the cell nuclei of multiple test cells in the corresponding test cell group.

[0106] One second sample image corresponds to a test cell group, a test cell group corresponds to at least one second sample image, and each second sample image contains the cell nuclei, actin / Golgi, and plasma membrane of multiple test cells in the corresponding test cell group.

[0107] In the embodiments of the present invention, multiple whole-field images can be obtained for each test cell group. The nuclear pseudo-color image in each whole-field image can be used as a first sample image, and the nuclear pseudo-color image, actin / Golgi pseudo-color image, and plasma membrane pseudo-color image in each whole-field image can be synthesized into a second sample image.

[0108] S302. Using multiple first sample images, multiple second sample images, and the trained cell portrait segmentation network, obtain the subcellular structure images of each test cell in each test cell group.

[0109] Among them, the subcellular structure diagram of a single test cell includes the nuclear gray-scale image of the test cell, the endoplasmic reticulum gray-scale image of the test cell, the actin / Golgi gray-scale image of the test cell, the mitochondrial gray-scale image of the test cell, the plasma membrane gray-scale image of the test cell, and the RNA gray-scale image of the test cell.

[0110] In a possible implementation, please refer to Figure 7 , step S302 includes sub-steps S302-1 to S302-3.

[0111] S302-1. Input multiple first sample images and multiple second sample images into the trained cell portrait segmentation network, and obtain multiple nuclear masks from each first sample image and multiple cell membrane masks from each second sample image.

[0112] Among them, a first sample image includes the nuclei of multiple test cells. Using the trained cell portrait segmentation network, draw masks covering the nuclear regions of each test cell on the first sample image to obtain multiple nuclear masks.

[0113] Similarly, a second sample image includes the nuclei, actin / Golgi, and plasma membranes of multiple test cells. Using the trained cell portrait segmentation network, draw masks covering the cell membrane regions of each test cell on the second sample image to obtain multiple cell membrane masks.

[0114] S302-2. Match each nuclear mask and each cell membrane mask to determine the nuclear mask and cell membrane mask belonging to the same test cell, and obtain the nuclear mask and cell membrane mask of each test cell.

[0115] Among them, the nuclear mask and cell membrane mask belonging to the same test cell satisfy the following formula:

[0116]

[0117] In the formula, is the j-th nuclear mask, is the k-th cell membrane mask.

[0118] S302-3: Using the nuclear masks and cell membrane masks of each test cell, based on the nuclear pseudo-color images, endoplasmic reticulum pseudo-color images, actin / Golgi pseudo-color images, mitochondrial pseudo-color images, plasma membrane pseudo-color images, and RNA pseudo-color images of each test cell group, obtain the subcellular structure images of each test cell.

[0119] Among them, the subcellular structure image of any test cell satisfies the following formula:

[0120]

[0121] Among them, is the grayscale image of the nucleus of this test cell, is the grayscale image of the endoplasmic reticulum of this test cell, is the grayscale image of actin / Golgi of this test cell, is the grayscale image of mitochondria of this test cell, is the grayscale image of the plasma membrane of this test cell, and is the grayscale image of RNA of this test cell, is the nuclear mask of this test cell, is the cell membrane mask of this test cell, I D is the nuclear pseudo-color image of the test cell group where this test cell is located, I E is the endoplasmic reticulum pseudo-color image of the test cell group where this test cell is located, is the actin / Golgi pseudo-color image of the test cell group where this test cell is located, I M is the mitochondrial pseudo-color image of the test cell group where this test cell is located, I P is the plasma membrane pseudo-color image of the test cell group where this test cell is located, is the RNA pseudo-color image of the test cell group where this test cell is located.

[0122] S303: Input the subcellular structure images of each test cell in each test cell group into a pre-constructed cell hypoxia scoring network for training to obtain a trained cell hypoxia scoring network.

[0123] In the embodiments of the present invention, a neural network for classifying whether a cell is in a hypoxic state is designed based on ResNet-50. ResNet-50 is improved to process six-channel input data (including the grayscale image of the nucleus, the grayscale image of the endoplasmic reticulum, the grayscale image of actin / Golgi, the grayscale image of mitochondria, the grayscale image of the plasma membrane, and the grayscale image of RNA). The output layer is modified to two-dimensional, the sigmoid function is used as the activation function, and the hypoxia probability output by the sigmoid function is used as the hypoxia score.

[0124] When training the network, the size of the subcellular structure images was adjusted to 224×224 pixels using the OpenCV library. In addition, the training (80%) and validation (20%) datasets were created by random splitting, and the binary cross-entropy loss function was selected as the optimization objective, and the stochastic gradient descent method was used to optimize the network.

[0125] The performance of the cell hypoxia scoring network was evaluated using accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic (ROC) curve (AUC).

[0126] Before training the cell hypoxia scoring network, for each test cell, morphological features of the cell portrait need to be extracted and analyzed to elucidate the law of cell hypoxia response.

[0127] Among them, the morphological features of the cell portrait include features such as nuclear area, perimeter, shape factor, aspect ratio, and contour calculated using the nuclear mask, features such as cell area, perimeter, shape factor, aspect ratio, and contour calculated using the cell membrane mask, and gray-level co-occurrence matrix attributes (contrast, dissimilarity, homogeneity, energy, correlation, ASM), granularity, and average intensity of each subcellular structure.

[0128] For the extracted morphological features of the cell portrait, methods such as analysis of variance, heat map, clustering, volcano plot, and dimensionality reduction analysis were used to explore the morphological features of the cell portrait under different hypoxia intensities and elucidate the law of cell hypoxia response.

[0129] To execute the corresponding steps in the above method embodiments and each possible implementation manner, an implementation manner of an anti-hypoxia drug screening device 100 is given below.

[0130] Please refer to Figure 8 As shown in the figure, the anti-hypoxia drug screening device 100 includes an acquisition module 101, an extraction module 102, a processing module 103, and a screening module 104.

[0131] The acquisition module 101 is configured to acquire full-field images of a first cell group, full-field images of a second cell group, and full-field images of a third cell group, wherein each cell in the first cell group is cultured in a non-hypoxic environment, each cell in the second cell group is cultured in a hypoxic environment and not treated with a drug, and each cell in the third cell group is cultured in a hypoxic environment and treated with a candidate drug.

[0132] The extraction module 102 is used to extract the subcellular structure images of multiple cells in the first cell group from the whole-field image of the first cell group, extract the subcellular structure images of multiple cells in the second cell group from the whole-field image of the second cell group, and extract the subcellular structure images of multiple cells in the third cell group from the whole-field image of the third cell group by using a pre-trained cell image segmentation network.

[0133] The processing module 103 is used to input the subcellular structure images of multiple cells in the first cell group, the subcellular structure images of multiple cells in the second cell group, and the subcellular structure images of multiple cells in the third cell group into a pre-trained cell hypoxia scoring network respectively, and obtain the hypoxia scores of multiple cells in the first cell group, the hypoxia scores of multiple cells in the second cell group, and the hypoxia scores of multiple cells in the third cell group.

[0134] The screening module 104 is used to determine whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells in the first cell group, the hypoxia scores of multiple cells in the second cell group, and the hypoxia scores of multiple cells in the third cell group.

[0135] The anti-hypoxia drug screening device 100 further includes a segmentation network training module and a scoring network training module.

[0136] The segmentation network training module is used to obtain a first training data set and a second training data set. The first training data set includes the first pseudo-color images of multiple test cell groups, and the second training data set includes the second pseudo-color images of multiple test cell groups. The oxygen concentration in the culture environment of each test cell group is different. The first pseudo-color image is the nucleus pseudo-color image of the test cell group and a nucleus mask is outlined on the first pseudo-color image. The second pseudo-color image is synthesized from the nucleus pseudo-color image, the actin / Golgi pseudo-color image, and the plasma membrane pseudo-color image of the test cell group, and a cell membrane mask is outlined on the second pseudo-color image. The first training data set and the second training data set are input into a pre-constructed cell image segmentation network for training to obtain a trained cell image segmentation network.

[0137] A scoring network training module is configured to obtain multiple first sample images and multiple second sample images. One first sample image corresponds to one test cell group, and one test cell group corresponds to at least one first sample image. Each first sample image contains the cell nuclei of multiple test cells in the corresponding test cell group. One second sample image corresponds to one test cell group, and one test cell group corresponds to at least one second sample image. Each second sample image contains the cell nuclei, actin / Golgi apparatus, and plasma membrane of multiple test cells in the corresponding test cell group. By using the multiple first sample images, the multiple second sample images, and the trained cell portrait segmentation network, subcellular structure images of each test cell in each test cell group are obtained. The subcellular structure images of each test cell in each test cell group are input into a pre-constructed cell hypoxia scoring network for training to obtain a trained cell hypoxia scoring network.

[0138] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working process of the anti-hypoxia drug screening device 100 described above can refer to the corresponding process in the foregoing method embodiment, and will not be elaborated herein.

[0139] Furthermore, an embodiment of the present invention also provides an electronic device 200. Please refer to Figure 9 , the electronic device 200 may include a memory 210 and a processor 220.

[0140] Among them, the processor 220 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits for controlling the execution of a program for the anti-hypoxia drug screening method provided in the foregoing method embodiment.

[0141] The memory 210 may be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or may also be an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 210 may exist independently and be connected to the processor 220 through a communication bus. The memory 210 may also be integrated with the processor 220. Among them, the memory 210 is used to store machine-executable instructions for implementing the solution of this application. The processor 220 is used to execute the machine-executable instructions stored in the memory 210 to implement the above method embodiments.

[0142] An embodiment of the present invention also provides a computer-readable storage medium containing a computer program, and the computer program can be used to perform related operations in a method for screening anti-hypoxia drugs provided by the above method embodiment when executed.

[0143] The above is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for screening anti-hypoxia drugs, characterized in that The method includes: Obtaining the whole-field images of the first cell group, the second cell group, and the third cell group, where each cell in the first cell group is cultured in a non-hypoxic environment, each cell in the second cell group is cultured in a hypoxic environment and not treated with a drug, and each cell in the third cell group is cultured in a hypoxic environment and treated with a candidate drug; Using a pre-trained cell portrait segmentation network to extract the subcellular structure images of multiple cells in each of the first cell group, the second cell group, and the third cell group from their respective whole-field images; Inputting the subcellular structure images of multiple cells in the first cell group, the second cell group, and the third cell group into a pre-trained cell hypoxia scoring network respectively to obtain the hypoxia scores of multiple cells in each cell group; Determining whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells in the first cell group, the second cell group, and the third cell group; The steps of training the cell hypoxia scoring network include: Obtaining multiple first sample images and multiple second sample images. One of the first sample images corresponds to a test cell group, and one test cell group corresponds to at least one of the first sample images. Each of the first sample images contains the cell nuclei of multiple test cells in the corresponding test cell group. One of the second sample images corresponds to a test cell group, and one test cell group corresponds to at least one of the second sample images. Each of the second sample images contains the cell nuclei, actin / Golgi apparatus, and plasma membrane of multiple test cells in the corresponding test cell group; Inputting the multiple first sample images and the multiple second sample images into the trained cell portrait segmentation network to obtain multiple nuclear masks from each of the first sample images and multiple cell membrane masks from each of the second sample images; Matching each of the nuclear masks and each of the cell membrane masks to determine the nuclear masks and cell membrane masks that belong to the same test cell, and obtaining the nuclear masks and cell membrane masks of each test cell; Using the nuclear masks and cell membrane masks of each test cell, based on the nuclear pseudo-color images, endoplasmic reticulum pseudo-color images, actin / Golgi apparatus pseudo-color images, mitochondrial pseudo-color images, plasma membrane pseudo-color images, and RNA pseudo-color images of each test cell group, obtaining the subcellular structure images of each test cell; Inputting the subcellular structure images of each test cell in each test cell group into a pre-constructed cell hypoxia scoring network for training to obtain the trained cell hypoxia scoring network.

2. The method according to claim 1, characterized in that The hypoxia score is positively correlated with the degree of cell hypoxia. The step of determining whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells in the first cell group, the second cell group, and the third cell group includes: The mean of the hypoxia scores of multiple cells in the first cell group is used as the hypoxia score of the first cell group, the mean of the hypoxia scores of multiple cells in the second cell group is used as the hypoxia score of the second cell group, and the mean of the hypoxia scores of multiple cells in the third cell group is used as the hypoxia score of the third cell group; If the hypoxia score of the third cell group is greater than the hypoxia score of the first cell group and less than the hypoxia score of the second cell group, it is determined that the candidate drug is a drug that can improve the cell survival state under hypoxic conditions.

3. The method according to claim 1, wherein The steps of training the cell portrait segmentation network include: Obtain a first training dataset and a second training dataset. The first training dataset includes first pseudo-color images of multiple test cell groups. The second training dataset includes second pseudo-color images of multiple test cell groups. The oxygen concentration in the culture environment of each test cell group is different. The first pseudo-color image is the nucleus pseudo-color image of the test cell group and a nucleus mask is outlined on the first pseudo-color image. The second pseudo-color image is synthesized from the nucleus pseudo-color image, the actin / Golgi pseudo-color image, and the plasma membrane pseudo-color image of the test cell group, and a cell membrane mask is outlined on the second pseudo-color image; Input the first training dataset and the second training dataset into a pre-constructed cell portrait segmentation network for training to obtain a trained cell portrait segmentation network.

4. The method according to claim 1, characterized in that, The nucleus mask and the cell membrane mask belonging to the same test cell satisfy the following formula: Among them, is the th cell nucleus mask, is the th cell membrane mask.

5. The method according to claim 1, characterized in that The subcellular structure image includes a nucleus grayscale image, an endoplasmic reticulum grayscale image, an actin / Golgi grayscale image, a mitochondrion grayscale image, a plasma membrane grayscale image, and an RNA grayscale image. The subcellular structure image of any test cell satisfies the following formula: Among them, is the grayscale image of the nucleus of the test cell, is the grayscale image of the endoplasmic reticulum of the test cell, is the grayscale image of actin / Golgi of the test cell, is the grayscale image of the mitochondria of the test cell, is the grayscale image of the plasma membrane of the test cell, and is the grayscale image of the RNA of the test cell, is the nuclear mask of the test cell, is the cell membrane mask of the test cell, is the pseudo-color image of the nucleus of the test cell group where the test cell is located, is the pseudo-color image of the endoplasmic reticulum of the test cell group where the test cell is located, is the pseudo-color image of actin / Golgi of the test cell group where the test cell is located, is the pseudo-color image of the mitochondria of the test cell group where the test cell is located, is the pseudo-color image of the plasma membrane of the test cell group where the test cell is located, is the pseudo-color image of the RNA of the test cell group where the test cell is located.

6. An anti-hypoxia drug screening device, characterized in that, The device includes: An acquisition module for acquiring the whole-field images of the first cell group, the second cell group, and the third cell group. Among them, the cells in the first cell group are cultured in a non-hypoxic environment, the cells in the second cell group are cultured in a hypoxic environment and not treated with drugs, and the cells in the third cell group are cultured in a hypoxic environment and treated with a candidate drug; An extraction module for using a pre-trained cell portrait segmentation network to extract the subcellular structure images of multiple cells in each cell group from the whole-field images of the first cell group, the second cell group, and the third cell group respectively; A processing module, configured to input subcellular structure images of multiple cells in the first cell group, the second cell group, and the third cell group into a pre-trained cell hypoxia scoring network respectively, so as to obtain hypoxia scores of multiple cells in each cell group. Wherein, the steps of training the cell hypoxia scoring network include: obtaining a plurality of first sample images and a plurality of second sample images, one of the first sample images corresponding to a test cell group, one test cell group corresponding to at least one of the first sample images, each of the first sample images containing cell nuclei of multiple test cells in the corresponding test cell group, one of the second sample images corresponding to a test cell group, one test cell group corresponding to at least one of the second sample images, each of the second sample images containing cell nuclei, actin / golgi apparatus, and plasma membranes of multiple test cells in the corresponding test cell group; inputting the plurality of first sample images and the plurality of second sample images into a trained cell portrait segmentation network, obtaining a plurality of nucleus masks from each of the first sample images, and obtaining a plurality of cell membrane masks from each of the second sample images; matching each of the nucleus masks and each of the cell membrane masks to determine the nucleus mask and the cell membrane mask belonging to the same test cell, so as to obtain the nucleus mask and the cell membrane mask of each test cell; using the nucleus mask and the cell membrane mask of each test cell, based on the nucleus pseudo-color image, endoplasmic reticulum pseudo-color image, actin / golgi apparatus pseudo-color image, mitochondrial pseudo-color image, plasma membrane pseudo-color image, and RNA pseudo-color image of each test cell group, obtaining subcellular structure images of each test cell; inputting the subcellular structure images of each test cell in each test cell group into a pre-constructed cell hypoxia scoring network for training, so as to obtain a trained cell hypoxia scoring network; A screening module, configured to determine whether the candidate drug is a drug that can improve the cell survival state under hypoxic conditions according to the hypoxia scores of multiple cells in the first cell group, the second cell group, and the third cell group.

7. An electronic device, characterized in that, It includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, it implements the anti-hypoxia drug screening method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a computer program, and when the computer program is executed by a processor, it implements the anti-hypoxia drug screening method according to any one of claims 1 to 5.

Images