Intelligent liquid nitrogen preservation effect evaluation system

Abstract

The invention discloses an intelligent liquid nitrogen preservation effect evaluation system. The system comprises a data acquisition module, an alignment preprocessing module, a membership evaluation module, a reliability weight calculation module, a liquid nitrogen preservation evaluation model establishment module and a liquid nitrogen preservation effect evaluation module. The invention belongs to the field of liquid nitrogen preservation evaluation, and particularly relates to an intelligent liquid nitrogen preservation effect evaluation system. According to the scheme, false anomalies introduced by accidental measurement errors in a neighborhood are inhibited through double-center membership depiction and refined critical distinguishing and local disharmony; projection residual measurement is introduced, and global projection distortion samples caused by instrument deviation are eliminated; performing automatic exception elimination based on a reliability weight; a reliability weight is introduced to slack variable penalty of each sample, low-credibility detection result interference is suppressed, and a decision surface is prevented from being distorted by noise samples; on the basis of upper and lower boundary separation loss and hyperbolic soft boundary bandwidth, samples which are most difficult to distinguish are focused, and the distinguishing sensitivity of critical survival rate samples is improved.

Description

Technical Field

[0001] The present invention relates to the field of liquid nitrogen preservation evaluation, and specifically refers to an intelligent liquid nitrogen preservation effect evaluation system. Background Art

[0002] A liquid nitrogen preservation effect evaluation system is a comprehensive system for evaluating the preservation effect when samples are preserved in a liquid nitrogen low-temperature environment. However, the general liquid nitrogen preservation effect evaluation system has problems such as being unable to give a delicate judgment on sublethal damage and critical state samples, resulting in a high misjudgment rate for critical samples, and poor accuracy in identifying false anomalies easily introduced by measurement noise, batch drift, or local instrument errors; the general liquid nitrogen preservation effect evaluation system has problems such as being unable to impose additional constraints on some of the most difficult-to-distinguish liquid nitrogen preservation samples, with insufficient discrimination ability, and noise samples dominating the decision surface, resulting in unstable evaluation. Summary of the Invention

[0003] In view of the above situation, to overcome the defects of the prior art, the present invention provides an intelligent liquid nitrogen preservation effect evaluation system. Aiming at the problems that the general liquid nitrogen preservation effect evaluation system is unable to give a delicate judgment on sublethal damage and critical state samples, resulting in a high misjudgment rate for critical samples, and poor accuracy in identifying false anomalies easily introduced by measurement noise, batch drift, or local instrument errors, this solution uses double-center membership degree to depict, refine the critical distinction, and can distinguish both mildly damaged samples and severely inactivated samples; through local disharmony degree, it suppresses the false anomalies introduced by accidental measurement errors in the neighborhood; introduces a projection residual metric to eliminate the globally projected distorted samples caused by batch drift or instrument deviation; automatically eliminates anomalies based on reliability weights; thereby improving the accuracy of liquid nitrogen preservation effect evaluation; aiming at the problems that the general liquid nitrogen preservation effect evaluation system is unable to impose additional constraints on some of the most difficult-to-distinguish liquid nitrogen preservation samples, with insufficient discrimination ability, and noise samples dominating the decision surface, resulting in unstable evaluation, this solution introduces reliability weights to the penalty of the slack variable of each sample, suppresses the interference of low-confidence detection results, and prevents noise samples from distorting the decision surface; based on the upper and lower bound separation loss and the hyperbolic soft boundary bandwidth, it focuses on the most difficult-to-distinguish samples and improves the discrimination sensitivity to critical survival rate samples; thereby improving the evaluation effect.

[0004] The technical solution adopted by the present invention is as follows: An intelligent liquid nitrogen preservation effect evaluation system provided by the present invention includes a data acquisition module, an alignment preprocessing module, a membership degree evaluation module, a reliability weight calculation module, a liquid nitrogen preservation evaluation model establishment module, and a liquid nitrogen preservation effect evaluation module;

[0005] The data acquisition module constructs a historical liquid nitrogen preservation sample set and a sample set to be evaluated;

[0006] The alignment preprocessing module maps samples to a common subspace by solving a projection matrix;

[0007] The membership evaluation module calculates the distances from the samples to be evaluated to the positive and negative class centers respectively in the projection subspace to characterize the sample membership;

[0008] The reliability weight calculation module introduces projection residuals and generates sample reliability weights based on the aggregation index;

[0009] The liquid nitrogen preservation evaluation model establishment module uses a weighted SVM with a hyperbolic soft margin bandwidth to train a classifier in the objective function with weighted slack variable penalties;

[0010] The liquid nitrogen preservation effect evaluation module evaluates the liquid nitrogen preservation effect of the samples to be evaluated based on the liquid nitrogen preservation evaluation model.

[0011] Furthermore, the data acquisition module collects a historical liquid nitrogen preservation sample set and a sample set to be evaluated; uses excellent preservation and failed preservation as supervision labels; the liquid nitrogen preservation samples include biological indicators, physical parameters, and environmental parameters.

[0012] Furthermore, the alignment preprocessing module first projects the source domain of the historical liquid nitrogen preservation samples and the target domain of the samples to be evaluated. Let the source domain data matrix be X S , and the target domain matrix be X T , and find the projection matrix P, expressed as: ; obtain the projection representation: ; ; where, and are the mean vectors of the source domain and target domain samples respectively; is a scalar hyperparameter used to balance the importance of the source domain and target domain scatter matrices; T is the transpose operation; and are the positions of the source domain and target domain samples after projection respectively; tr(·) is the matrix trace.

[0013] Furthermore, in the projection subspace, the membership evaluation module first defines two class centers: the positive class center ; the negative class center ; where, is the positive class center; is the negative class center; and are the sets of samples with excellent preservation and failed preservation after projection respectively; and are the total numbers of samples with excellent preservation and failed preservation respectively; x is the sample index; for any sample to be evaluated , calculate: the distance to the center, and the formula used is: ; ; where, and are the distances from the sample to be evaluated to the positive class center and the negative class center respectively; membership degree, the formula used is: ; ; non-membership degree, the formula used is: ; ; ; where, is the positive class limit distance; is the positive class membership degree; is the smoothing term; is the minimum distance difference baseline; is the distance between centers; is the non-membership degree; is the local disharmony degree, taking the k nearest neighbor sets N of i , ; is the sample in the nearest neighbor set N i ; and are the sample labels.

[0014] Furthermore, the reliability weight calculation module fuses the membership degree and the non-membership degree, generates the reliability weight of each sample, and introduces the projection residual metric, the formula used is: ; where, is the reliability weight; is the weight balance factor; is the aggregation index.

[0015] Furthermore, the liquid nitrogen preservation evaluation model establishment module is based on SVM, weights the relaxation variable penalty with the reliability weight of each sample, and punishes the upper and lower bound violations respectively; introduces the hyperbolic soft boundary bandwidth , and the objective function is expressed as: ;

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] where, is the weight vector in the kernel space; b is the bias term; is the relaxation variable; is the relaxation variable of the i-th sample; and is the loss coefficient; and are the boundary violation losses for the two types of boundary samples of over - conservatism and over - confidence respectively; is the upper - bound scaling factor; is the true label; is the predicted label; C is the penalty coefficient; m is the total number of samples; is the kernel mapping function; is the initial bandwidth, is the boundary coefficient.

[0021] Furthermore, the liquid nitrogen preservation effect evaluation module classifies the historical liquid nitrogen preservation sample set and the to - be - evaluated sample set jointly based on the liquid nitrogen preservation evaluation model; divides the total sample data into a test set and a training set, initializes the SVM parameters, updates the parameters iteratively based on the gradient of the objective function, and completes the training when the objective function converges or reaches the maximum number of iterations; sets a test threshold. If the prediction accuracy of the trained liquid nitrogen preservation evaluation model for the test set is higher than the test threshold, the liquid nitrogen preservation evaluation model is established, and the classification result of the to - be - evaluated sample at this time is used as the liquid nitrogen preservation evaluation result; otherwise, retrain.

[0022] The beneficial effects achieved by the present invention using the above - mentioned solution are as follows:

[0023] (1) Aiming at the problems existing in the general liquid nitrogen preservation effect evaluation system, such as the inability to give a delicate judgment on sub - lethal damage and critical - state samples, resulting in a high misjudgment rate for critical samples and poor recognition accuracy for false anomalies easily introduced by measurement noise, batch drift or local instrument errors. This solution uses double - center membership to depict, refine the critical distinction, and can distinguish both mildly damaged samples and severely inactivated samples; uses local disharmony to suppress false anomalies introduced by occasional measurement errors within the neighborhood; introduces a projection residual metric to eliminate globally projected distorted samples caused by batch drift or instrument deviation; automatically eliminates anomalies based on reliability weights; thereby improving the accuracy of liquid nitrogen preservation effect evaluation.

[0024] (2) Aiming at the problems existing in the general liquid nitrogen preservation effect evaluation system, such as the inability to impose additional constraints on some of the most difficult - to - distinguish liquid nitrogen preservation samples, insufficient discriminative ability, and noise samples dominating the decision surface, resulting in unstable evaluation. This solution introduces reliability weights for the penalty of the slack variable of each sample to suppress the interference of low - credibility detection results and prevent noise samples from distorting the decision surface; based on the upper - and - lower - bound separation loss and the hyperbolic soft - boundary bandwidth, focuses on the most difficult - to - distinguish samples and improves the discriminative sensitivity to critical survival rate samples; thereby improving the evaluation effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1Schematic flow diagram of an intelligent liquid nitrogen preservation effect evaluation system provided by the present invention.

[0026] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. Detailed implementation manners

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0028] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0029] Example 1, refer to Figure 1 An intelligent liquid nitrogen preservation effect evaluation system provided by the present invention includes a data acquisition module, an alignment preprocessing module, a membership degree evaluation module, a reliability weight calculation module, a liquid nitrogen preservation evaluation model establishment module, and a liquid nitrogen preservation effect evaluation module;

[0030] The data acquisition module constructs a historical liquid nitrogen preservation sample set and a sample set to be evaluated; and sends the data to the alignment preprocessing module;

[0031] The alignment preprocessing module maps the samples to a common subspace by solving a projection matrix; and sends the data to the membership degree evaluation module;

[0032] The membership degree evaluation module calculates the distances from the samples to be evaluated to the positive and negative class centers respectively in the projection subspace to characterize the sample attribution; and sends the data to the reliability weight calculation module;

[0033] The reliability weight calculation module introduces a projection residual and generates sample reliability weights based on the aggregation index; and sends the data to the liquid nitrogen preservation evaluation model establishment module;

[0034] The liquid nitrogen preservation evaluation model establishment module uses a weighted SVM with a hyperbolic soft boundary bandwidth to train a classifier in the objective function with weighted slack variable penalties; and sends the data to the liquid nitrogen preservation effect evaluation module;

[0035] The liquid nitrogen preservation effect evaluation module evaluates the liquid nitrogen preservation effect of the sample set to be evaluated based on the liquid nitrogen preservation evaluation model.

[0036] Example 2, refer to Figure 1 , this example is based on the above example. The data acquisition module collects the historical liquid nitrogen preservation sample set and the sample set to be evaluated, including excellent preservation samples with high post-thaw cell survival rate, intact membrane, low ROS level, and small DNA fragmentation index, and failed preservation samples with significantly decreased survival rate / function or cell inactivation after thawing; using excellent preservation and failed preservation as supervision labels; the liquid nitrogen preservation samples include biological indicators, physical parameters, and environmental parameters; the biological indicators include post-thaw survival rate, membrane integrity, ROS level, and DNA fragmentation index; the physical parameters include freezing rate, thawing rate, freezing program curve, and CPA concentration; the environmental parameters include liquid nitrogen temperature, temperature change rate, humidity, and storage container pressure.

[0037] Example 3, refer to Figure 1 , this example is based on the above example. Due to the systematic drift of the cooling rate curve and CPA batch differences between different laboratories and batches, the alignment preprocessing module first projects the source domain of the historical liquid nitrogen preservation samples and the target domain of the samples to be evaluated. Let the source domain data matrix be X S , and the target domain matrix be X T , and find the projection matrix P, expressed as: ; obtain the projection representation: ; ; where, and are the mean vectors of the source domain and target domain samples respectively; is a scalar hyperparameter used to balance the importance of the source domain and target domain divergence matrices; T is the transpose operation; and are the positions of the projected source domain and target domain samples respectively; tr(·) is the matrix trace; eliminating different experimental conditions and instrument drift, and improving the subsequent cross-batch generalization ability.

[0038] Example 4, refer to Figure 1 , this example is based on the above example. In the projection subspace, the membership degree evaluation module measures the membership degree and non-membership degree of each sample to be evaluated using the centers of the two types of samples of excellent preservation and failed preservation respectively; first define two class centers: the positive class center ; the negative class center ; where, is the positive class center; is the negative class center; and They are the sets of excellent preservation and failed preservation samples after projection, respectively; and They are the total numbers of excellent preservation and failed preservation samples, respectively; x is the sample index; for any sample to be evaluated , calculate: the distance to the center, and the formula used is: ; ; where and are the distances from the sample to be evaluated to the positive class center and the negative class center, respectively; the membership degree, and the formula used is: ; ; the non-membership degree, and the formula used is: ; ; ; where is the positive class limit distance; is the positive class membership degree; is the smoothing term; is the minimum distance difference baseline; is the distance between centers; is the non-membership degree; is the local disharmony degree, taking the k nearest neighbor set N of i , ; is the sample in the nearest neighbor set N i ; and are the sample labels, and the target domain samples are not considered; the dual centers can simultaneously characterize the good and bad poles, greatly improving the ability to distinguish the critical state of sub-lethal damage samples; the non-membership degree term can quickly approach 1 when the sample has deviated to the failed side, automatically eliminating those obvious thawing failure outliers; the local disharmony degree suppresses the false outliers caused by accidental measurement noise; also, at the same distance to the excellent center, different samples can be reasonably distinguished according to the distance difference from the failed center.

[0039] Example 5, refer to Figure 1 , this example is based on the above example. The reliability weight calculation module fuses the membership degree and the non-membership degree, generates the reliability weight of each sample, and introduces the projection residual metric to balance the local and global credibility, better eliminating outliers. The formula used is: ; where is the reliability weight; is the weight balance factor; is the aggregation index; samples with high membership + low non-membership are given greater weights to weaken the suspicious points introduced by measurement noise or batch drift.

[0040] By performing the above operations, for the general liquid nitrogen preservation effect evaluation system, there are problems such as the inability to give a delicate judgment on sub-lethal damage and critical state samples, resulting in a high misjudgment rate for critical samples and poor accuracy in identifying false anomalies easily introduced by measurement noise, batch drift, or local instrument errors. This solution uses double-center membership to characterize and refine the critical distinction, and can distinguish both mildly damaged samples and severely inactivated samples; through local discordance, it suppresses false anomalies introduced by occasional measurement errors in the neighborhood; introduces a projection residual metric to eliminate globally projected distorted samples caused by batch drift or instrument deviation; automatically eliminates anomalies based on reliability weights; thereby improving the accuracy of liquid nitrogen preservation effect evaluation.

[0041] Example 6, refer to Figure 1 , based on the above example, the liquid nitrogen preservation evaluation model establishment module is based on SVM, with the reliability weights of each sample weighted relaxation variable penalty, and separate penalties for upper and lower bound violations; introduce a hyperbolic soft boundary bandwidth , automatically amplify the discrimination penalty for near-boundary samples, and the objective function is expressed as: ;

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] Among them, is the weight vector in the kernel space; b is the bias term; is the relaxation variable; is the relaxation variable of the i-th sample; and are the loss coefficients; and are the boundary violation losses for two types of boundary samples, namely over-conservative and over-confident, respectively; is the upper bound scaling factor; is the true label; is the predicted label; C is the penalty coefficient; m is the total number of samples; is the kernel mapping function; is the initial bandwidth, is the boundary coefficient.

[0047] By performing the above operations, in view of the problems existing in the general liquid nitrogen preservation effect evaluation system, such as the inability to impose additional constraints on some of the most difficult-to-distinguish liquid nitrogen preservation samples, insufficient discrimination ability, and noise samples dominating the decision surface, resulting in unstable evaluation, this solution introduces a reliability weight to the penalty of the slack variable for each sample to suppress the interference of low-confidence detection results and prevent noise samples from distorting the decision surface; based on the upper and lower bound separation loss and the hyperbolic soft margin bandwidth, it focuses on the most difficult-to-distinguish samples and improves the discrimination sensitivity for samples with critical survival rates; thereby improving the evaluation effect.

[0048] Embodiment Seven. Refer to Figure 1 , this embodiment is based on the above embodiment. The liquid nitrogen preservation effect evaluation module classifies the historical liquid nitrogen preservation sample set and the to-be-evaluated sample set together based on the liquid nitrogen preservation evaluation model; divides the total sample data into a test set and a training set, initializes the SVM parameters, and updates the parameters iteratively based on the gradient of the objective function for training. When the objective function converges or reaches the maximum number of iterations, the training is completed; sets a test threshold. If the prediction accuracy of the trained liquid nitrogen preservation evaluation model for the test set is higher than the test threshold, the liquid nitrogen preservation evaluation model is established, and the classification result of the to-be-evaluated sample at this time is used as the liquid nitrogen preservation evaluation result; otherwise, retrain.

[0049] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0050] The above describes the present invention and its implementation manners. This description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural manners and embodiments without creative efforts without departing from the purpose of the present invention's creation, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent liquid nitrogen preservation effect evaluation system, characterized in that: The system includes a data acquisition module, an alignment preprocessing module, a membership degree evaluation module, a reliability weight calculation module, a liquid nitrogen preservation evaluation model establishment module, and a liquid nitrogen preservation effect evaluation module; The data acquisition module constructs a historical liquid nitrogen preservation sample set and a sample set to be evaluated; The alignment preprocessing module maps the samples to a common subspace by solving a projection matrix; The membership degree evaluation module calculates the distances from the samples to be evaluated to the positive and negative class centers respectively in the projection subspace to characterize the sample attribution; The reliability weight calculation module introduces projection residuals and generates sample reliability weights based on the aggregation index; The liquid nitrogen preservation evaluation model establishment module uses a weighted SVM with a hyperbolic soft boundary bandwidth to train a classifier in the objective function with weighted slack variable penalties; The liquid nitrogen preservation effect evaluation module evaluates the liquid nitrogen preservation effect of the sample set to be evaluated based on the liquid nitrogen preservation evaluation model.

2. The intelligent liquid nitrogen preservation effect evaluation system according to claim 1, characterized in that: The alignment preprocessing module first projects the source domain of historical liquid nitrogen-preserved samples and the target domain of samples to be evaluated, and let the source domain data matrix be X S , and the target domain matrix be X T . The projection matrix P is obtained and expressed as: ; The projected representation is obtained: ; ; where and are the mean vectors of the source domain and target domain samples respectively; is a scalar hyperparameter used to balance the importance of the source domain and target domain divergence matrices; T is the transpose operation; and are the positions of the projected source domain and target domain samples respectively; tr(·) is the matrix trace.

3. The intelligent liquid nitrogen preservation effect evaluation system according to claim 2, characterized in that: In the projection subspace, the membership degree evaluation module first defines two class centers: the positive class center ; the negative class center ; where is the positive class center; is the negative class center; and are the excellent preservation and failed preservation sample sets after projection respectively; and are the total numbers of excellent preservation and failed preservation samples respectively; x is the sample index; for any sample to be evaluated , calculate: the distance to the center, and the formula used is: ; ; where and are the distances from the sample to be evaluated to the positive class center and the negative class center respectively; the membership degree, and the formula used is: ; ; the non - membership degree, and the formula used is: ; ; ; where is the positive class limit distance; is the positive class membership degree; is the smoothing term; is the minimum distance difference baseline; is the distance between centers; is the non - membership degree; is the local disharmony degree, take the k - nearest neighbor set N of i , ; is the sample in the nearest neighbor set N i ; and are the sample labels.

4. An intelligent liquid nitrogen preservation effect evaluation system according to claim 3, characterized in that: The reliability weight calculation module fuses membership and non-membership degrees to generate the reliability weight of each sample, and introduces a projection residual metric. The formula used is: ; where is the reliability weight; is the weight balance factor; is the aggregation index.

5. An intelligent liquid nitrogen preservation effect evaluation system according to claim 4, characterized in that: The liquid nitrogen storage evaluation model establishment module is based on SVM, with the weighted relaxation variable penalty of the reliability weight of each sample, and separately penalizes the upper and lower bound violations; a hyperbolic soft margin bandwidth is introduced , and the objective function is expressed as: ; ； ； ； ； Among them, is the weight vector in the kernel space; b is the bias term; is the slack variable; is the slack variable of the i-th sample; and are the loss coefficients; and are the boundary violation losses for the two types of boundary samples of over-conservatism and over-confidence respectively; is the upper bound scaling factor; is the true label; is the predicted label; C is the penalty coefficient; m is the total number of samples; is the kernel mapping function; is the initial bandwidth, is the boundary coefficient.

6. The intelligent liquid nitrogen preservation effect evaluation system according to claim 5, wherein: The data acquisition module acquires a historical liquid nitrogen preservation sample set and a sample set to be evaluated; uses excellent preservation and failed preservation as supervision labels; the liquid nitrogen preservation samples include biological indicators, physical parameters, and environmental parameters.

7. An intelligent liquid nitrogen preservation effect evaluation system according to claim 6, characterized in that: The liquid nitrogen preservation effect evaluation module jointly classifies the historical liquid nitrogen preservation sample set and the sample set to be evaluated based on the established liquid nitrogen preservation evaluation model; divides the total sample data into a test set and a training set, initializes the SVM parameters, and updates the parameters iteratively based on the gradient of the objective function for training. When the objective function converges or reaches the maximum number of iterations, the training is completed; Set a test threshold. If the prediction accuracy of the trained liquid nitrogen preservation evaluation model for the test set is higher than the test threshold, the establishment of the liquid nitrogen preservation evaluation model is completed, and the classification result of the sample to be evaluated at this time is used as the liquid nitrogen preservation evaluation result; otherwise, retrain.

Images