A method for positioning and identifying ultra-bright light spots based on random forest classification algorithm
The identification and calibration of ultra-bright spots through the random forest classification algorithm was solved, and the problem of low quantitative ability in single-molecule positioning super-resolution microscopy was significantly improved, and the imaging ability was enhanced.
Patent Information
- Application Number
- CN202310359888.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-06
- Publication Date
- 2024-11-01
- Estimated Expiration
- 2043-04-06
AI Technical Summary
Single-molecule localized super-resolution microscopy faces great challenges in providing quantitative information on imaging biological structures, especially since the light switching of a specific molecule is essentially a random event, resulting in uncertainty in the single-molecule scintillation time, and in biological specimens The high molecular density leads to the appearance of ultra-bright spots and reduces the quantitative ability.
The ultra-bright spot positioning recognition method based on the random forest classification algorithm is adopted. By establishing a single-molecule library, simulating the ultra-bright spot data, training the random forest classification model, and applying this model to single-molecule positioning super-resolution microscopy of biological subdiffraction limit structure Imaging performs the identification of the positioning number and calibration of the ultra-bright spot, and then reconstructs the calibrated single-molecule positioning microscopy super-resolution image.
The classification accuracy of ultra-bright spots is improved, the quantitative ability of single-molecule positioning super-resolution microscopy is enhanced, the quantitative and comparative analysis ability of biological specimens is significantly improved, and the imaging ability of single-molecule positioning super-resolution microscopy is improved.
Smart Images

Figure CN116740705B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of single molecule positioning and identification, and in particular to a super bright spot positioning and identification method based on a random forest classification algorithm. Background Art
[0002] Single-molecule localization microscopy (SMLM), including STORM, PALM, DNA-PAINT, etc., uses the switching ability of certain fluorescent molecules to time-share multiple molecules within the diffraction limit. In the process of sequential imaging of the same field of view, each frame of the photo captures the signal of a single molecule, thus breaking the diffraction limit and achieving ultra-high resolution imaging (horizontal resolution of about 20nm, axial resolution of about 50nm); this technology also won the 2014 Nobel Prize. As the highest resolution super-resolution fluorescence imaging technology, SMLM has significant advantages and irreplaceable role when facing the size of most biological molecules of one to several nanometers, especially in visualizing sub-diffraction limit ultrastructures.
[0003] Recent advances in biological research have placed an urgent need for accurate quantitative assessment of biological structures at nanometer resolution. Despite its high resolution, single-molecule localization super-resolution microscopy still faces great challenges in providing quantitative information on imaged biological structures. This is because the photo-switching of a specific molecule is essentially a random event. The uncertainty of the blinking time of a single molecule during a defined image process poses a huge challenge to quantitative analysis based on the number of localizations. Despite this, there is still a lack of research on systematically evaluating or improving quantitative capabilities, and reports on quantification mainly focus on localization segmentation and counting.
[0004] According to the principle of single-molecule activation, the light spot in each camera frame is ideally contributed by the emission of a single molecule. This is achieved by utilizing excellent light-switchable fluorescent dyes such as AF647 and Cy5, which exhibit relatively low fluorescence-dark state duty cycles (0.05-0.1%). Nevertheless, molecules in biological specimens usually exhibit small sizes ranging from one nanometer to several nanometers, and usually have relatively high molecular abundance in the diffraction-limited cellular ultrastructure, such as approximately 325 tubulin dimers in a 200nm long microtubule segment. Therefore, although the single-molecule sample is sparse enough, there are still spots formed by the simultaneous emission of multiple molecules within the diffraction limit during single-molecule localization super-resolution microscopy imaging. This type of spot is called an ultra-bright spot. This special spot is mostly regarded as a single-molecule spot in image reconstruction, resulting in a decrease in the quantitative ability of single-molecule localization microscopy imaging.
[0005] Therefore, it is necessary to apply a method to identify and calibrate ultrabright localization for quantitative single-molecule localization super-resolution microscopy imaging to improve the quantitative capability of single-molecule localization super-resolution microscopy imaging and broaden its application in biomedical research. Summary of the invention
[0006] In order to solve the problems in the prior art, the present invention provides a super bright spot positioning and identification method based on a random forest classification algorithm.
[0007] The ultra-bright spot positioning and identification method based on the random forest classification algorithm of the present invention comprises the following steps:
[0008] Steps for establishing a single molecule library: used to establish a single molecule library for simulating single molecule localization super-resolution microscopy imaging;
[0009] The simulation steps of the ultra-bright spot data are as follows: randomly generating a plurality of position coordinates in the diffraction limit region, randomly extracting a corresponding number of single-molecule spots from the single-molecule library, and overlapping and placing them in the simulation region, so that the positioning coordinates of the single-molecule spots coincide with the generated position coordinates respectively;
[0010] Random forest classification model training step: using a random forest algorithm to train the single molecule spot and super bright spot data to obtain the random forest classification model;
[0011] Identification and calibration step: The random forest classification model is used to identify the positioning number and calibrate the positioning information of the ultra-bright spot of the biological sub-diffraction limit structure single-molecule localization super-resolution microscopy imaging.
[0012] Furthermore, the method also includes a super-resolution image reconstruction step: based on the identified and calibrated super-bright spot information, a calibrated single-molecule localization microscopy super-resolution image is reconstructed.
[0013] The present invention is further improved, and the step of establishing the single molecule library includes the following sub-steps:
[0014] S101: Preparation of single molecule samples, obtaining single molecule samples after fluorescence labeling;
[0015] S102: Single-molecule localization super-resolution microscopy imaging: Use imaging buffer to perform single-molecule localization super-resolution microscopy imaging of single-molecule samples to obtain image sequences of each single-molecule sample;
[0016] S103: Extraction of single-molecule data: Based on the image sequence, an imaging data set of each single-molecule localization microscopy imaging is obtained to form a single-molecule library, wherein the imaging data set includes an image sequence of a single molecule, and the image sequence includes an emission point and corresponding localization coordinates, frame number, and photon number.
[0017] Furthermore, the single-molecule localization super-resolution microscopy imaging method includes but is not limited to STORM, PALM, and DNA-PAINT single-molecule localization super-resolution microscopy imaging techniques.
[0018] Furthermore, in step S103, the method for extracting single molecule data is:
[0019] In the acquired image sequence, the emission spot was fitted to a two-dimensional Gaussian function to determine its center position;
[0020] Drift correction is performed on the positioning of each frame;
[0021] Taking the center position as the center, the spatial distribution of each located light spot is extracted within the set pixel range;
[0022] These localization coordinates are clustered to obtain the localization sequence of each single molecule at different times;
[0023] An imaging dataset for single-molecule localization microscopy is established, including image sequences of single molecules, each of which contains localization coordinates, frame number, and photon number information.
[0024] Furthermore, the random forest classification model training method is:
[0025] Using the random forest classification algorithm, a corresponding number of spots are randomly selected from the single-molecule library and their center points are placed at random positions within the simulated diffraction limit region to simulate ultra-bright spots of a set type. The spatial distribution of overlapping spots is submitted as an input vector, the number of spots is assigned as a label, and a random tree predictor is generated from each sample score.
[0026] Then, a portion of the data from each type of ultrabright spot and single molecule type is selected for training, and the other portion of the data is used for verification. The parameters of the random forest algorithm are optimized, and finally the trained random forest classification model is obtained.
[0027] The random forest classification algorithm can be trained using a random forest function library of a language platform, including a random forest function library on a Matlab, Python or C language platform, and the like.
[0028] Furthermore, the method for identifying the positioning number and calibrating the positioning information of the super bright spot is:
[0029] Uncalibrated emission spots were extracted from image sequences of single-molecule localization super-resolution microscopy, and a trained random forest classification model was applied to determine the number of potential localizations per spot, thereby deriving the number of calibrated localizations per ultrastructure.
[0030] Compared with the prior art, the beneficial effects of the present invention are: by constructing a random forest classification model, the classification accuracy of different types of superbright spots is improved, the positioning number of superbright spots is identified, the accuracy of superbright spot positioning is effectively improved, the quantitative and comparative analysis capabilities of single-molecule localization super-resolution microscopy for biological specimens are improved, and the imaging capability of single-molecule localization super-resolution microscopy is significantly improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the solutions in the present application or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0032] Figure 1 is a flow chart of the method of the present invention;
[0033] Figure 2 Schematic diagram of the classification effect of the random forest classification model of the present invention, wherein a is a schematic diagram of the classification precision and accuracy of the random forest classification model of the present invention, b is a schematic diagram of the total number of localization effects of different labeling densities of simulated microtubule segments, and c is a schematic diagram of the percentage of correct localization at label densities of 10% to 90%;
[0034] Figure 3 This is a schematic diagram of calibration and positioning in a non-uniform biological superstructure simulating STORM imaging of the present invention, wherein a is a schematic diagram of simulated STORM imaging of a non-uniform superstructure, and b is a random forest classification model for the spot type. DETAILED DESCRIPTION
[0035] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of this application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit this application; the terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.
[0036] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0037] The present invention solves the technical problem of how to determine the number of different types of ultra-bright light spots in the process of reconstructing images by single-molecule localization super-resolution microscopy and how to improve the quantitative process of single-molecule super-resolution microscopy. Ultra-bright light spots are light spots formed by the simultaneous emission of multiple molecules within the diffraction limit. Although the light emission of densely distributed molecules can be separated in time with the help of light scintillation technology, due to the random characteristics of light scintillation, most biological samples may still have light spots formed by the simultaneous emission of multiple molecules within the diffraction limit during single-molecule localization super-resolution microscopy. This type of light spot is called an ultra-bright light spot. According to the number of molecules within the diffraction limit (N), the corresponding light spot is called Ultra-N. In order to enable those skilled in the art to better understand the present application, the technical solution in the embodiment of the present application will be clearly and completely described below in conjunction with the accompanying drawings.
[0038] The single-molecule localization super-resolution microscopy imaging includes single-molecule localization super-resolution microscopy imaging technologies such as STORM, PALM, and DNA-PAINT. The ultra-bright spot localization recognition method based on the random forest classification algorithm is applicable to the above single-molecule localization microscopy imaging technologies, and STORM is mainly described below.
[0039] like Figure 1 As shown, the ultra-bright spot positioning and identification method based on the random forest classification algorithm of the present invention includes the following steps:
[0040] Step 1: Single molecule library establishment step: used to establish a single molecule library simulating STORM super-resolution imaging;
[0041] Step 2: Simulation of ultra-bright spot data: randomly generate multiple position coordinates in the diffraction limit area, randomly extract a corresponding number of single-molecule spots from the single-molecule library, and overlap and place them in the simulation area so that the positioning coordinates of the single-molecule spots coincide with the generated position coordinates respectively;
[0042] Step 3: random forest classification model training step: using a random forest algorithm to train the single molecules in the single molecule library to obtain the random forest classification model;
[0043] Step 4: Identification and calibration step: using the random forest classification model, through biological sub-diffraction limit structure STORM imaging analysis, the ultra-bright spot is identified and the positioning information is calibrated.
[0044] The ultra-bright spot positioning and identification method based on the random forest classification algorithm of the present invention can be applied to the quantification, comparison, analysis and other technologies of biological specimens, thereby improving the accuracy of quantitative analysis. It can also be applied to the reconstruction of super-resolution image technology, which can effectively improve the imaging capability of super-resolution images.
[0045] As a preferred embodiment of the present invention, this example is applied to the technology of reconstructing super-resolution images. The present invention also includes step five - super-resolution image reconstruction step: based on the identified and calibrated super-bright spot information, reconstructing a calibrated STORM super-resolution image.
[0046] Specifically, as a detailed embodiment of the present invention, the step of establishing a single molecule library of the present invention includes the following sub-steps:
[0047] S101: Preparation of single molecule samples, obtaining single molecule samples after fluorescence labeling.
[0048] The specific preparation method of this example is:
[0049] First, the cover glass was cleaned intensively by ultrasonic treatment in Milli-Q water (ultrapure water);
[0050] Then, an oligonucleotide conjugated with a single labeling functional group molecule at the 5' end is fixed on the glass surface, wherein the concentration of the oligonucleotide is 0.1-1.0 nM and then prepared in 4% paraformaldehyde;
[0051] Thirdly, fluorescent microspheres with a diameter of 200 nm were fixed on the glass surface for a certain period of time at room temperature to serve as reference markers for drift correction of molecular samples in the XY plane during image acquisition;
[0052] Finally, the samples were washed with PBS (phosphate buffered saline) to remove unbound molecules.
[0053] The method for preparing the single molecule sample of the present invention can be prepared by covalent coupling or non-covalent coupling, and the above preparation method is only one embodiment thereof.
[0054] This example uses AlexaFluor647 fluorescent dye, but any fluorescent dye including AlexaFluor647 and CF568 can be used.
[0055] S102: STORM super-resolution imaging: Use imaging buffer to perform STORM super-resolution imaging of single-molecule samples to obtain image sequences of each single-molecule sample.
[0056] The imaging buffer has a variety of optimized versions. In this example, an imaging buffer containing 50 mM Tris (trishydroxymethylaminomethane, pH 8.0), 10 mM NaCl, 1% β-mercaptoethanol by volume, 10% glucose by mass concentration, 0.5 mg / mL glucose oxidase (G2133, Sigma), and 40 μg / mL catalase (C30, Sigma) was used. STORM imaging was performed on each single molecule sample, and an image sequence of 8000 frames was collected at a frequency of 33 Hz for each field of view.
[0057] This example can also use an imaging buffer containing 50mM Tris (pH 8.0), 10mM NaCl, 0.7% β-mercaptoethanol (v / v), 10% glucose (w / v), 0.55mg / mL glucose oxidase (G2133, Sigma), and 40μg / mL catalase (C30, Sigma), with NaOH used to adjust the pH to 8.10±0.05.
[0058] S103: Extraction of single molecule data: Based on the image sequence, a single molecule STORM imaging data set is acquired, wherein the single molecule STORM imaging data set includes an image sequence of a single molecule, wherein the image sequence includes an emission point and corresponding positioning coordinates, a frame number, and a photon number.
[0059] Specifically, in this example, the emission spot in the acquired image sequence is fitted to a two-dimensional Gaussian function to determine its center position (i.e., positioning). Then, the image sequence is drift corrected, a reference marker such as fixed fluorescent beads is added to the sample or software for analyzing the correlation of the image sequence is used, and hardware correction is performed by fine-tuning the stage. The photon number distribution of each spot is extracted in 7×7 pixels^2 around the center position of the emission spot. The center coordinates of these spots are then clustered to obtain the positioning sequence of each single molecule.
[0060] In this example, a single-molecule library is established, including more than 12,000 single-molecule image sequences. Each sequence contains different emission points and their corresponding positioning coordinates, frame numbers, and photon numbers, etc., which are used to further simulate four types of ultrabright spots (including Ultra1-4) and simulate STORM imaging of biological sub-diffraction limit structures.
[0061] Of course, the classification types in this example are not limited to 4 and can be changed according to needs.
[0062] This example provides a training process for a random forest classification model. The training process is as follows:
[0063] The TreeBagger function in the modeling tool MATLAB was used for training. A corresponding number of spots were randomly selected from the single-molecule STORM imaging dataset and their centers were placed at random positions within the simulated diffraction limit region to simulate the set type of ultrabright spots (Ultra2-4). The spatial distribution of overlapping spots was submitted as an input vector, the number of spots was assigned as a label, and a random tree predictor was generated from each sample score.
[0064] In this example, treebagger is only an embodiment. The random forest algorithm in this example can also use random forest function libraries of major language platforms, such as Matlab, Python, C language, etc.
[0065] Then, each type of the three ultrabright spots and the single molecule type contained 150,000 pairs of input data and labels, of which 120,000 pairs were used for training and 30,000 pairs were used for validation, and finally a trained random forest classification model was obtained.
[0066] In the identification and calibration step, this example uses single-molecule imaging data of various biological structures and simulated data to obtain benchmark facts, and corrects the experimental data to evaluate the results of identification and calibration. This example simulates the STORM imaging sequence of biological sub-diffraction limit structures for analysis.
[0067] This example simulates a STORM image of a 1-micron microtubule segment, with labeling densities ranging from 10% to 90%. In this example, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, and 90% density distributions are selected. In this simulation, fluorophore molecules are randomly distributed on the microtubule structure according to the required density, and then randomly selected flashing sequences from the single-molecule STORM imaging dataset are placed at specified locations on the simulated microtubule. The simulation is repeated five times for each simulated density to improve the simulation accuracy.
[0068] In this example, the STORM imaging of biological sub-diffraction limit structures also includes the simulation of STORM imaging of diffraction-limited cell ultrastructures. The simulation method is as follows:
[0069] A regular or irregular circular model was used with a diameter of about 240 nanometers, which was determined by the Abbe limit and contained up to 550 molecules. The molecules were assumed to be randomly distributed in the regular circular superstructure. In order to mimic the irregular circular structure of receptor clusters on the cell membrane, the density in the center of the structure was higher than the density at the periphery. Each fluorophore was randomly sampled as the ultrastructure of the single-molecule STORM imaging dataset, which included the corresponding localization, spot, number of photons, and blinking information. In the simulated image sequence of the ultrastructure, the emission spots in the same frame from different emitting molecules were overlapped according to their corresponding center positions.
[0070] As an embodiment of the present invention, the method for identifying the positioning number and calibrating the positioning information of the super bright spot is as follows:
[0071] From the image sequences of single-molecule localization super-resolution microscopy, uncalibrated emission spots were extracted and the trained random forest classification model was applied to determine the potential type of each localization, thereby deriving a calibrated localization number for each ultrastructure, which corresponds to the localization position, photon number, and blinking information of the target spot.
[0072] Model Validation
[0073] like Figure 2 As shown in Figure a, the random forest classification model in this example has verified the classification of four types of spots. The horizontal axis is the type, and the four types are single-molecule spots, Ultra2, Ultra3, and Ultra4 simulated in this example. The vertical axis is the classification accuracy (Accuracy), recall rate (Recallrate) or accuracy (Precision). The verification results show that the accuracy (gray bar) and recall rate (white bar) of single-molecule spots reach more than 96%, the accuracy of Ultra-2 and Ultra-4 reaches about 80%, and the classification accuracy of the verification set reaches 81.75% (a straight line parallel to the horizontal axis in the figure).
[0074] like Figure 2 As shown in Figure b, this example simulates STORM images of a 1-micron microtubule segment with labeling densities ranging from 10% to 90%. In this simulation, fluorophore molecules are randomly distributed on the microtubule structure according to the desired density. Then, a randomly selected blinking sequence from the single-molecule STORM imaging dataset is placed at a specified position on the simulated microtubule. The simulation is repeated five times for each simulated density. The total number of ground-truth localizations (the long bar on the far left of each simulated density) is calculated by summing the number of localizations of all sampled molecules, while the number of uncalibrated localizations (the long bar in the middle of each simulated density) is calculated by fitting the spots in the simulated sequence to a two-dimensional Gaussian function. It can be found that the number of uncalibrated localizations deviates from the number of ground-truth localizations as the labeling density increases due to the increase in the number of superbright localizations.
[0075] like Figure 2As shown in Figure c in , the random forest classification model classifier is then applied to identify potential spot types in the simulated microtubules to determine how many basic molecules contribute to each spot. For all simulated densities, the calibrated localization numbers from the random forest classification model are closer to the benchmark facts than the uncalibrated localization numbers. It is worth noting that when considering the efficiency of classifying the localization into the correct type, it can be found that for those uncalibrated localizations, up to 50% of the localizations are not correctly assigned. In contrast, after using TreeBagger, there is a great improvement in classifying localization into the correct type, and the percentage of correct localizations (Percentage of correct localizations) has reached more than 82%. It can be seen that the use of the random forest classification model in the present invention can effectively improve the accuracy of classification and the accuracy of positioning.
[0076] like Figure 3 As shown in Figure a, each ultrastructure exhibits a diffraction-limited size, containing 50-550 molecules respectively. This example simulates a series of non-uniform ultrastructures with 50, 150, 250, 350, 450, and 550 molecules. The center has a higher number of photons than the periphery, which simulates different biological ultrastructures, such as receptor clusters on cell membranes. From the simulated image sequence, uncalibrated emission spots were extracted, and a trained random forest classification model TreeBagger was applied to determine the potential type of each localization, thereby deriving the calibrated number of localizations for each ultrastructure. Compared with the uncalibrated localizations, the number of localizations from the random forest classification model TreeBagger shows a better correlation with the molecular density and has a higher consistency with the ground-truth, as shown in Figure 2. Figure 3 It is worth noting that TreeBagger also shows high accuracy in classifying the light spots into the correct type, as shown in the curve b in Figure 2. Figure 3 As shown in Figure c.
[0077] In summary, although the ubiquitous existence of ultrabright localization destroys the quantitative and comparative capabilities of single-molecule super-resolution microscopy, through the establishment of a single-molecule library provided in the embodiments of the present application and the ultrabright spot localization identification method based on the random forest classification algorithm, a sufficient number of biological super-resolution images are analyzed and these ultrabright localizations are calibrated, which greatly improves the quantitative and comparative analysis capabilities of single-molecule super-resolution microscopy for biological specimens, thereby significantly improving the capabilities of single-molecule localization super-resolution microscopy.
[0078] The specific implementation modes described above are preferred implementation modes of the present invention, and are not intended to limit the specific implementation scope of the present invention. The scope of the present invention includes but is not limited to the specific implementation modes, and all equivalent changes made according to the present invention are within the protection scope of the present invention.
Claims
1. A super bright spot positioning and identification method based on random forest classification algorithm, characterized in that: The steps include: Steps for establishing a single molecule library: used to establish a single molecule library for simulating single molecule localization super-resolution microscopy imaging; The simulation steps of the ultra-bright spot data are as follows: randomly generating a plurality of position coordinates in the diffraction limit region, randomly extracting a corresponding number of single-molecule spots from the single-molecule library, and overlapping and placing them in the simulation region, so that the positioning coordinates of the single-molecule spots coincide with the generated position coordinates respectively; Random forest classification model training step: using a random forest algorithm to train the single molecule light spot and super bright light spot data to obtain the random forest classification model, where the super bright light spot is a light spot formed by multiple molecules emitting light simultaneously within the diffraction limit; Identification and calibration step: using the random forest classification model, the ultra-bright spot of biological sub-diffraction limit structure single molecule localization super-resolution microscopy imaging is identified and the positioning information is calibrated. The method for identifying the positioning number and calibrating the positioning information of the super bright spot is: Uncalibrated emission spots were extracted from image sequences of single-molecule localization super-resolution microscopy, and a trained random forest classification model was applied to determine the number of potential localizations per spot, thereby deriving the number of calibrated localizations per ultrastructure.
2. The ultra-bright spot positioning and identification method based on the random forest classification algorithm according to claim 1 is characterized in that: The method also includes a super-resolution image reconstruction step: based on the identified and calibrated super-bright spot information, a calibrated single-molecule localization microscopy super-resolution image is reconstructed.
3. The ultra-bright spot positioning and identification method based on the random forest classification algorithm according to claim 1 or 2, characterized in that: The step of establishing the single molecule library includes the following sub-steps: S101: Preparation of single molecule samples, obtaining single molecule samples after fluorescence labeling; S102: Single-molecule localization super-resolution microscopy imaging: Use imaging buffer to perform single-molecule localization super-resolution microscopy imaging of single-molecule samples to obtain image sequences of each single-molecule sample; S103: Extraction of single-molecule data: Based on the image sequence, the imaging data of each single-molecule localization microscopy imaging is obtained to form a single-molecule library. The single-molecule library is an imaging data set of each single-molecule imaging data. The imaging data set includes an image sequence of a single molecule. The image sequence includes an emission point and a corresponding positioning coordinate, a frame number, and a photon number.
4. The ultra-bright spot positioning and identification method based on the random forest classification algorithm according to claim 3 is characterized in that: In step S101, the single-molecule localization super-resolution microscopy imaging method includes but is not limited to STORM, PALM, and DNA-PAINT single-molecule localization super-resolution microscopy imaging techniques.
5. The ultra-bright spot positioning and identification method based on the random forest classification algorithm according to claim 3 is characterized in that: In step S103, the method for extracting single molecule data is: In the acquired image sequence, the emission spot was fitted to a two-dimensional Gaussian function to determine its center position; Drift correction is performed on the positioning of each frame; Taking the center position as the center, the spatial distribution of each located light spot is extracted within the set pixel range; These localization coordinates are clustered to obtain the localization sequence of each single molecule at different times; An imaging dataset for single-molecule localization microscopy is established, including image sequences of single molecules, each of which contains localization coordinates, frame number, and photon number information.
6. The ultra-bright spot positioning and identification method based on the random forest classification algorithm according to claim 1 or 2, characterized in that: The random forest classification model training method is: Using the random forest classification algorithm, a corresponding number of spots are randomly selected from the single-molecule library and their center points are placed at random positions within the simulated diffraction-limited region to simulate the set type of ultra-bright spots. The spatial distribution of overlapping spots is submitted as an input vector, the number of spots is assigned as a label, and a random tree predictor is generated from each sample score. Then, a portion of the data from each type of ultrabright spot and single molecule type is selected for training, and the other portion of the data is used for verification. The parameters of the random forest algorithm are optimized, and finally the trained random forest classification model is obtained.
Citation Information
Patent Citations
System and method for virtual reality simulation of monomolecular dynamics in nano-environment
CN102063580A
Super-resolution fluorescence fluctuation microscopy imaging method and device and storage medium
CN108318464A
Super-resolution stimulated Raman microimaging method and device for realizing near resonance enhancement
CN112240880A
Sub-ten-nanometer positioning direction finding method and device based on one-dimensional dark spot time-sharing illumination
CN112485232A
Methods and systems for cytometry
CN112638529A
