A method for quality evaluation of cryo-em electron tomography label-free point alignment

Abstract

The application discloses a kind of cryo-EM electron tomography label-free point alignment quality evaluation method, belong to cryo-EM electron tomography imaging technical field.The present application takes gold standard point residual as core evaluation index, realize indirect, objective and global judgment to label-free point alignment quality, gold standard point does not participate in the optimization process of label-free point registration, residual is independent external reference index, guarantee the objectivity, independence and globality of evaluation result, avoid the problem that existing evaluation method is interfered by optimization target, cannot be globally measured.The numerical value of residual, distribution characteristics and comparison of similar methods can realize accurate and scientific evaluation of label-free point alignment quality.The present application provides reliable quantitative reference for optimization, improvement and screening of registration alignment method in cryo-EM electron tomography imaging technology, and has wide practical application value.

Description

Technical Field

[0001] This invention relates to the field of cryo-electron microscopy electron tomography, and more specifically to a method for quality assessment of markerless point alignment in cryo-electron microscopy electron tomography. Background Technology

[0002] Cryo-electron microscopy (cryo-electron tomography) is a crucial technique for resolving the three-dimensional structures of biomolecules. Registration and alignment are key preprocessing steps in this technology, with the core objective of providing accurate geometric parameters for subsequent tomographic reconstruction. The quality of registration and alignment directly determines the topological realism and structural fidelity of the final reconstructed tomographic image. Registration and alignment methods are mainly divided into two paradigms: marker-based and markerless. Marker-based registration and alignment, due to the existence of a clear reference, can be quantitatively evaluated through reprojection residuals. However, markerless registration and alignment lacks an inherent reference basis, making objective and quantitative quality assessment difficult. Furthermore, current technologies lack a unified evaluation system that can simultaneously adapt to both registration paradigms, hindering effective comparison of the performance of different markerless registration and alignment methods and making it difficult to verify the geometric accuracy and reliability of markerless registration results.

[0003] Current quality assessments for markerless registration and alignment often suffer from problems such as a single assessment metric, lack of independence, and inability to measure globally. Some assessment methods directly use parameters from the optimization process as the assessment basis, leading to the assessment results being influenced by the optimization objective and lacking objectivity. At the same time, existing assessment methods do not integrate information from the entire tilted image sequence, making it impossible to globally quantify the geometric accuracy of the 3D-to-2D spatial mapping, and thus failing to fully reflect the overall quality of registration and alignment.

[0004] Therefore, there is an urgent need to build a quality evaluation framework that can achieve unified quantification of unmarked and marker-based registration and alignment, and possess independence, objectivity, and globality, to solve the technical problem of the difficulty in effectively evaluating the quality of unmarked registration and alignment. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies in evaluating the quality of markerless alignment in cryo-electron microscopy electron tomography, which lacks a unified quantitative system and suffers from insufficient objectivity and globality. It provides a method for evaluating the quality of markerless alignment in cryo-electron microscopy electron tomography to overcome the deficiencies of existing technologies.

[0006] This invention uses the residual of the gold standard benchmark point as the core evaluation index to build a unified quantitative evaluation system, so as to realize the indirect, objective and global evaluation of the alignment quality of unmarked points. At the same time, it can realize the performance comparison of different unmarked point methods, verify the robustness of the method and the effectiveness of the evaluation method.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0008] A quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography, comprising the following steps: S1: Obtain datasets containing markers aligned using different cryo-electron microscopy electron tomography marker-free point alignment methods, and identify the markers on the datasets; S2: Perform point chain matching on the marker points identified in S1, connect the two-dimensional coordinates of the marker points in different tilted images, and construct a two-dimensional point chain trajectory; S3: Obtain geometric parameters based on the dataset obtained from S1; S4: Using the geometric parameters obtained in S3, the two-dimensional point chain trajectory obtained in S2 is mapped to three-dimensional space one by one through affine transformation to obtain the unique three-dimensional coordinates of each marked point. S5: Reproject the three-dimensional coordinates of the mapped marker points back to the two-dimensional image plane using the reprojection formula; S6: Calculate the Euclidean distance between the reprojected coordinates of each marker point and the actual observed coordinates, and take the mean of the Euclidean distances of the marker points as the reference point residual; S7: Perform S2-S6 calculations on the dataset for each alignment method in S1 to obtain the corresponding reference point residuals. The alignment method with the smallest residual has the best alignment quality.

[0009] Furthermore, in S2, the specific method for point chain matching is as follows: based on the four-point affine invariance, a quadrilateral point group is randomly selected under the RANSAC framework, and the optimal correspondence is determined by matching its affine invariant proportions, thereby solving the affine transformation matrix.

[0010] Furthermore, the execution flow of the RANSAC framework is as follows:

[0011] (1) Based on the registration scenario of the tilted image of cryo-electron microscopy electron tomography, it is assumed that the relationship between the set of marker points between the tilted image to be registered and the reference image conforms to the law of affine transformation.

[0012] (2) In each iteration of RANSAC, four non-collinear markers are randomly selected from the moving point set M (the set of markers identified in the cryo-electron microscope tilted image to be registered) to form a quadrilateral.

[0013] (3) Using the two-dimensional coordinates of the four points of the selected quadrilateral, the two geometric metric ratios r1 and r2 that remain unchanged under the affine transformation are calculated. The combination of r1 and r2 constitutes the unique geometric "key" of the quadrilateral. The kd-tree algorithm is used to quickly search for candidate quadrilaterals with the same or similar "key" in the fixed point set S (the set of marker points identified in the cryo-electron microscopy reference image as the registration benchmark). The criteria for the same or similar judgment is that the Euclidean distance between the key of the candidate quadrilateral and the target quadrilateral is less than the preset feature matching threshold. Once a matching quadrilateral is found in S, a candidate affine transformation matrix T is calculated using the two-dimensional coordinates of the four points that correspond one-to-one.

[0014] (4) Apply the candidate transformation T to the entire moving point set M and count the number of points whose distance from the transformed points to points in S is less than a preset distance threshold, wherein the distance threshold is set in advance based on the pixel accuracy of the cryo-electron microscope image and the error of the marker point recognition.

[0015] (5) Repeat the sampling-verification process of steps (2)-(4) above. The number of iterations is preset according to the size of the point set and the registration accuracy requirements. Finally, the transformation model T that obtains the most consistent points (points in the moving point set M that, after being mapped by the candidate transformation T, are less than the preset distance threshold to a point in the fixed point set S, i.e., points that are successfully registered and matched) is recorded as the preliminary optimal model.

[0016] (6) To combat noise interference, a two-stage strategy was adopted: In the first stage, the transformation was calculated using the initial 4-point pairs, and a loose threshold (a pixel-level distance threshold set based on the coarse registration requirements, with a wider threshold range) was used to count consistent points. If the number of consistent points obtained met the preset effective threshold (the proportion of consistent points to the total number of points in the moving point set M reached a preset proportion, or the number of consistent points reached a preset number, which was set in advance according to the registration requirements), it was determined that there were enough consistent points, and the second stage was initiated. In the second stage, the affine transformation matrix T was recalculated accurately using the two-dimensional coordinates of all consistent point pairs obtained in the first stage by fitting with the least squares method, and then the number of consistent points was recounted using a strict threshold (a pixel-level distance threshold smaller than the loose threshold in the first stage, set based on the accuracy requirements of fine registration), and the registration effectiveness of the transformation model T was finally evaluated based on this number.

[0017] Furthermore, in S3, the specific definitions of the geometric parameters are: the tilt angle includes the pitch angle α=0 (fixed value), the yaw angle β (range -60°~+60°), and the roll angle γ, all in radians; the components t of the translation vector t... xThe unit tᵧ is pixels, and its value range matches the image size (≤10% of the image side length); the scaling factor s=1 (dimensionless, based on the statistical variance of the marker alignment results ≤10⁻). 6 The projection matrix P is a 3×2 orthogonal projection matrix (P=[[1,0,0],[0,1,0]]), determined according to the imaging principle of cryo-electron microscopy.

[0018] Furthermore, in S4, the specific formula for the affine transformation is: Where s=1 (based on experimental statistics, the mean of the scaling factor is close to 1, and the variance is ≤10⁻). 6 Rα is the rotation matrix around the X-axis (corresponding to pitch angle α=0, based on experimental statistics, the mean is close to 0°, and the variance is ≤0.134, so it is fixed at 0), Rβ is the rotation matrix around the Y-axis (corresponding to yaw angle β), Rγ is the rotation matrix around the Z-axis (corresponding to roll angle γ), P is the orthogonal projection matrix (P=[[1,0,0],[0,1,0]], which conforms to the principle of cryo-electron microscopy), and t is the translation vector (unit: pixels). , λ represents the translation component, and λ is the depth degree of freedom parameter.

[0019] Furthermore, in step S5, the reprojection formula used is: Where s=1 (based on experimental statistics, the mean of the scaling factor is close to 1, and the variance is ≤10⁻). 6 Rα is the rotation matrix around the X-axis (corresponding to pitch angle α=0, based on experimental statistics, the mean is close to 0°, the variance is ≤0.134, and it is fixed at 0), Rβ is the rotation matrix around the Y-axis (corresponding to yaw angle β), Rγ is the rotation matrix around the Z-axis (corresponding to roll angle γ), P is the orthogonal projection matrix (P=[[1,0,0],[0,1,0]], which conforms to the principle of cryo-electron microscopy imaging), and t is the translation vector (unit: pixels).

[0020] Furthermore, in S6, the formula for calculating the benchmark point residual is as follows: , where Res is the baseline residual and N is the number of two-dimensional marker points (N≥300 and uniformly distributed in the tilted image). , The two-dimensional coordinates of the marked point after reprojection. , These are the actual observed two-dimensional coordinates of the marked point.

[0021] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0022] This invention uses the residual of the gold standard reference point as the core evaluation index and establishes for the first time a unified quantitative evaluation system for registration and alignment based on unmarked points and marked points. It solves the technical problem of lacking effective evaluation criteria for the quality of unmarked point alignment, realizes the unification of evaluation standards for the two registration paradigms, and facilitates the performance comparison of different methods.

[0023] In this invention, the gold standard reference point does not participate in the optimization process of markerless point registration. The residual is an independent external reference index, ensuring the objectivity, independence, and globality of the evaluation results. This avoids the problems of existing evaluation methods being affected by optimization objectives and unable to measure globally. By comparing the magnitude and distribution characteristics of the residual with similar methods, a precise and scientific evaluation of the markerless point alignment quality can be achieved. Low and uniform residual values ​​indicate high geometric accuracy of registration and alignment, ensuring the topological realism and structural fidelity of the final tomographic reconstruction image.

[0024] This invention can not only evaluate the alignment quality of markerless points, but also verify the effectiveness of the method itself and the robustness of the markerless point method. It provides a reliable quantitative reference for the optimization, improvement and screening of registration and alignment methods in cryo-electron microscopy electron tomography, and has broad practical application value. Attached Figure Description

[0025] Figure 1 It is a quality assessment framework process based on markerless alignment of cryo-electron microscopy electron tomography.

[0026] Figure 2 This is a schematic diagram of the mapping of the two-dimensional coordinates of the gold standard reference point to three-dimensional space. Detailed Implementation

[0027] The technical solution of the present invention will be further described and illustrated below with reference to the embodiments.

[0028] Example 1

[0029] A quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography, such as... Figure 1 As shown, the method includes the following steps: S1: Obtain datasets containing markers aligned using different cryo-electron microscopy electron tomography marker-free point alignment methods, and identify the markers on the datasets; S2: Perform point chain matching on the marker points identified in S1, connect the two-dimensional coordinates of the marker points in different tilted images, and construct a two-dimensional point chain trajectory; The specific method of point chain matching is: based on the four-point affine invariance, randomly select quadrilateral point groups under the RANSAC framework, determine the optimal correspondence by matching their affine invariant proportions, and then solve the affine transformation matrix. S3: Obtain geometric parameters based on the dataset obtained in S1; the specific definitions of the geometric parameters are: tilt angle including pitch angle α=0 (fixed value), yaw angle β (range -60°~+60°), and roll angle γ, all in radians; components t of the translation vector t. x The unit tᵧ is pixels, and its value range matches the image size (≤10% of the image side length); the scaling factor s=1 (dimensionless, based on the statistical variance of the marker alignment results ≤10⁻). 6 The projection matrix P is a 3×2 orthogonal projection matrix (P=[[1,0,0],[0,1,0]]), determined according to the imaging principle of cryo-electron microscopy. S4: Using the geometric parameters obtained in S3, the two-dimensional point chain trajectory obtained in S2 is mapped one by one to three-dimensional space through affine transformation (e.g., Figure 2 As shown), the unique three-dimensional coordinates of each marker point are obtained; the specific formula for the affine transformation is: Where s=1 (based on experimental statistics, the mean of the scaling factor is close to 1, and the variance is ≤10⁻). 6 Rα is the rotation matrix around the X-axis (corresponding to pitch angle α=0, based on experimental statistics, the mean is close to 0°, and the variance is ≤0.134, so it is fixed at 0), Rβ is the rotation matrix around the Y-axis (corresponding to yaw angle β), Rγ is the rotation matrix around the Z-axis (corresponding to roll angle γ), P is the orthogonal projection matrix (P=[[1,0,0],[0,1,0]], which conforms to the principle of cryo-electron microscopy), and t is the translation vector (unit: pixels). , λ represents the translation component, and λ is the depth degree of freedom parameter. S5: Reproject the mapped 3D coordinates of the marker points back to the 2D image plane using the reprojection formula; the reprojection formula used is: Where s=1 (based on experimental statistics, the mean of the scaling factor is close to 1, and the variance is ≤10⁻). 6 Rα is the rotation matrix around the X-axis (corresponding to pitch angle α=0, based on experimental statistics, the mean is close to 0°, the variance is ≤0.134, and it is fixed at 0), Rβ is the rotation matrix around the Y-axis (corresponding to yaw angle β), Rγ is the rotation matrix around the Z-axis (corresponding to roll angle γ), P is the orthogonal projection matrix (P=[[1,0,0],[0,1,0]], which conforms to the principle of cryo-electron microscopy imaging), and t is the translation vector (unit: pixels). S6: Calculate the Euclidean distance between the reprojected coordinates and the actual observed coordinates of each marker point, and take the mean of the Euclidean distances of the marker points as the reference point residual; the formula for calculating the reference point residual is as follows: , where Res is the baseline residual and N is the number of two-dimensional marker points (N≥300 and uniformly distributed in the tilted image). , The two-dimensional coordinates of the marked point after reprojection. , The actual observed two-dimensional coordinates of the marked point; S7: Perform S2-S6 calculations on the dataset for each alignment method in S1 to obtain the corresponding reference point residuals. The alignment method with the smallest residual has the best alignment quality.

[0030] In one specific embodiment, the execution flow of the RANSAC framework is as follows:

[0031] (1) Based on the registration scenario of the tilted image of cryo-electron microscopy electron tomography, it is assumed that the relationship between the set of marker points between the tilted image to be registered and the reference image conforms to the law of affine transformation.

[0032] (2) In each iteration of RANSAC, four non-collinear markers are randomly selected from the moving point set M (the set of markers identified in the cryo-electron microscope tilted image to be registered) to form a quadrilateral.

[0033] (3) Using the two-dimensional coordinates of the four points of the selected quadrilateral, the two geometric metric ratios r1 and r2 that remain unchanged under the affine transformation are calculated. The combination of r1 and r2 constitutes the unique geometric "key" of the quadrilateral. The kd-tree algorithm is used to quickly search for candidate quadrilaterals with the same or similar "key" in the fixed point set S (the set of marker points identified in the cryo-electron microscopy reference image as the registration benchmark). The criteria for the same or similar judgment is that the Euclidean distance between the key of the candidate quadrilateral and the target quadrilateral is less than the preset feature matching threshold. Once a matching quadrilateral is found in S, a candidate affine transformation matrix T is calculated using the two-dimensional coordinates of the four points that correspond one-to-one.

[0034] (4) Apply the candidate transformation T to the entire moving point set M and count the number of points whose distance from the transformed points to points in S is less than a preset distance threshold, wherein the distance threshold is set in advance based on the pixel accuracy of the cryo-electron microscope image and the error of the marker point recognition.

[0035] (5) Repeat the sampling-verification process of steps (2)-(4) above. The number of iterations is preset according to the size of the point set and the registration accuracy requirements. Finally, the transformation model T that obtains the most consistent points (the points in the moving point set M that are mapped by the candidate transformation T and whose distance to a point in the fixed point set S is less than the preset distance threshold, i.e. the points that are successfully registered and matched) is recorded as the preliminary optimal model.

[0036] (6) To combat noise interference, a two-stage strategy was adopted: In the first stage, the transformation was calculated using the initial 4-point pairs, and a loose threshold (a pixel-level distance threshold set based on the coarse registration requirements, with a wider threshold range) was used to count consistent points. If the number of consistent points obtained met the preset effective threshold (the proportion of consistent points to the total number of points in the moving point set M reached a preset proportion, or the number of consistent points reached a preset number, which was set in advance according to the registration requirements), it was determined that there were enough consistent points, and the second stage was initiated. In the second stage, the affine transformation matrix T was recalculated accurately using the two-dimensional coordinates of all consistent point pairs obtained in the first stage by fitting with the least squares method, and then the number of consistent points was recounted using a strict threshold (a pixel-level distance threshold smaller than the loose threshold in the first stage, set based on the accuracy requirements of fine registration), and the registration effectiveness of the transformation model T was finally evaluated based on this number.

[0037] Example 2

[0038] To further verify the effectiveness of the evaluation framework of this invention and the robustness of different markerless registration and alignment methods, Example 1 was used as the application method. This example selected five typical biological structures—HIV-1 Gag, Vibrio, Adhesion belt, Nitrosop1, and Nitrosop2—as test objects. Three markerless registration and alignment methods—Markerfree, Areomo, and IMOD based on patch methods—were used for experiments. The residual of the gold standard reference point was used as the core evaluation index to quantitatively compare the alignment quality of the three methods. Each method was run independently multiple times, and the mean and standard deviation of the residuals were calculated to evaluate the stability and geometric accuracy of the methods.

[0039] The experimental results are shown in Table 1. Table 1 lists the mean and standard deviation of the residuals for the three methods on five different structures, where the mean reflects the alignment accuracy and the standard deviation reflects the stability and robustness of the method.

[0040] Table 1

[0041] As can be seen from the results in the table, the quality assessment method based on gold standard reference point residuals proposed in this invention can effectively quantify the geometric accuracy and stability of different markerless point registration and alignment methods, achieving an objective and global evaluation of markerless point alignment quality. Furthermore, through comparative verification using multiple structures and methods, it is demonstrated that this assessment framework has good universality and reliability, and can serve as a standard tool for evaluating the alignment quality of markerless points in cryo-electron microscopy electron tomography.

[0042] Based on the above embodiments, the present invention continues to describe in detail the technical features involved therein and the functions and roles of these technical features in the present invention, so as to help those skilled in the art to fully understand the technical solution of the present invention and reproduce it.

[0043] Finally, although this specification describes embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for quality assessment of markerless point alignment in cryo-electron microscopy electron tomography, characterized in that, The method includes the following steps: S1: Obtain labeled data aligned by different cryo-electron microscopy electron tomography marker-free point alignment algorithms, and identify the labeled points on the dataset; S2: Perform point chain matching on the marker points, connect the two-dimensional coordinates of the marker points in different tilted images, and construct a two-dimensional point chain trajectory; S3: Obtain geometric parameters, which are the core parameters output by the cryo-electron microscopy electron tomography markerless point alignment algorithm; S4: Using the geometric parameters obtained in S3, the two-dimensional point chain trajectory of the marked points obtained in S2 is mapped to the three-dimensional space one by one through affine transformation to obtain the unique three-dimensional coordinates of each marked point. S5: Reproject the three-dimensional coordinates of the mapped marker points back to the two-dimensional image plane using the reprojection formula; S6: Calculate the Euclidean distance between the reprojected coordinates of each marker point and the actual observed coordinates, and take the mean of the Euclidean distances of the marker points as the reference point residual; S7: Perform S2-S6 calculations on each type of alignment data in S1 to obtain the corresponding reference point residuals. The alignment quality is best when the residual is the smallest.

2. The quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography as described in claim 1, characterized in that, In S2, the specific method for point chain matching is as follows: based on the four-point affine invariance, a quadrilateral point group is randomly selected under the RANSAC framework, and the optimal correspondence is determined by matching its affine invariant proportions, thereby solving the affine transformation matrix.

3. The quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography as described in claim 1, characterized in that, In S3, the specific definitions of the geometric parameters are: the tilt angle includes pitch angle α, yaw angle β, and roll angle γ; the components t of the translation vector t are... x tᵧ is in pixels, and its value range matches the image size; scaling factor s; The projection matrix P is determined based on the imaging principle of cryo-electron microscopy.

4. The quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography as described in claim 1, characterized in that, In S4, the specific formula for the affine transformation is: Rα is the rotation matrix about the X-axis, Rβ is the rotation matrix about the Y-axis, Rγ is the rotation matrix about the Z-axis, P is the orthogonal projection matrix, and t is the translation vector. , λ represents the translation component, and λ is the depth degree of freedom parameter.

5. The quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography as described in claim 1, characterized in that, In S5, the reprojection formula used is: Rα is the rotation matrix around the X-axis, Rβ is the rotation matrix around the Y-axis, Rγ is the rotation matrix around the Z-axis, P is the orthogonal projection matrix, and t is the translation vector.

6. The quality assessment method for markerless point alignment in cryo-electron microscopy electron tomography as described in claim 1, characterized in that, In S6, the residual calculation formula used is as follows: Where Res is the baseline residual and N is the number of two-dimensional marker points. , The two-dimensional coordinates of the marked point after reprojection. , These are the actual observed two-dimensional coordinates of the marked point.

Images