Search

CN-122024862-A - State identification method of adaptive gene editing laboratory mice

CN122024862ACN 122024862 ACN122024862 ACN 122024862ACN-122024862-A

Abstract

The invention discloses a state identification method of an adaptive gene editing experiment mouse, and aims to solve the problem of insufficient specificity of state identification of the gene editing experiment mouse in the prior art. The method comprises the steps of obtaining genotype, disease model type and experimental environment parameters of an experimental mouse, optimizing detection and optical flow analysis parameters, generating a time sequence behavior track map, dividing a key region, extracting general and pathological specificity composite characteristics, standardizing, calling a genotype pathological characteristic association database calibration weight and a threshold value, inputting a random forest classifier identification state optimized by transfer learning, calculating scores through a weighted scoring model, judging the state, and calculating credibility through a multidimensional confidence assessment model. The invention adapts to various gene editing laboratory mice and experimental environments, improves the accuracy and adaptability of state identification, and provides powerful support for gene therapy research and development and disease mechanism research.

Inventors

  • WU LIMIN

Assignees

  • 上海谛科生物科技有限公司

Dates

Publication Date
20260512
Application Date
20260212

Claims (10)

  1. 1. A state identification method for an adaptive gene editing experiment mouse is characterized by comprising the steps of obtaining genotype information, disease model types and experiment environment parameters of the experiment mouse, optimizing an improved target detection algorithm and dense optical flow field analysis parameters based on the experiment environment parameters, capturing space coordinates and motion vectors of the experiment mouse in real time to generate a time sequence behavior track map, carrying out region segmentation on the time sequence behavior track map, identifying coordinate boundaries of a drinking region, a nest region and a pathology sensitive region, extracting a composite feature set in the time sequence behavior track map, carrying out standardization processing on feature data, wherein the composite feature set comprises general behavior features and pathology specific features, calling a preset genotype pathology feature association database, calibrating feature weights and dynamic judgment thresholds through an adaptive learning algorithm, inputting the standardized composite features into a random forest classifier optimized through transfer learning to carry out state identification, calculating a comprehensive score of the experiment mouse through an improved weighting model, judging as a pathology related anxiety state when the comprehensive score is higher than the calibrated dynamic judgment threshold, otherwise judging as a physiological state, and adopting a multidimensional state reliability evaluation model to evaluate the reliability.
  2. 2. The method for identifying the state of the adaptive gene editing laboratory mice according to claim 1, wherein the genotype information comprises mutant genes and mutation sites, the disease model type comprises a single-gene genetic disease model and a spontaneous inflammation model, the experimental environment parameters comprise feeding level, cage size, illumination intensity and background noise related information, the general behavior characteristics comprise activity density, regional stay time ratio, movement acceleration variance, drinking interval period and resting event frequency, and the pathology specific characteristics comprise affected limb activity ratio, feeding drinking time difference, inflammation related contracture frequency and metabolic abnormal drinking fluctuation coefficient.
  3. 3. The method for identifying the state of the adaptive gene editing laboratory mice according to claim 2 is characterized in that the pathological specificity is defined as follows, the affected limb activity ratio is the ratio of the pathological limb activity duration to the total activity duration of the gene editing inflammatory model mice, the feeding drinking time difference is the mean value and standard deviation of the time interval from the end of single feeding to the beginning of the next drinking, the inflammation-related crimping frequency is the number of times that the crimping amplitude of the laboratory mice exceeds a preset threshold value in unit time, the abnormal metabolism drinking fluctuation coefficient is the variation coefficient of the drinking times in a continuous period, and the metabolic disease model is adapted.
  4. 4. The method for identifying the state of the adaptive gene editing laboratory mouse according to claim 1, wherein the technical process of optimizing the improved target detection algorithm and the dense optical flow field analysis parameters and generating the time sequence behavior track map comprises the steps of collecting body type data and movement characteristic data of the laboratory mouse, wherein the body type data comprises body type dimension parameters related to body length, body width and body weight, and the movement characteristic data comprises movement speed, movement direction change frequency and movement track continuity parameters; performing cluster analysis on the body type data to obtain a plurality of anchor frame size cluster centers, generating initial anchor frames based on the cluster centers, calculating the intersection ratio of the initial anchor frames and a target area of an experimental mouse, judging whether the intersection ratio mean value meets a preset standard, if not, adjusting the length and width parameters of the anchor frames according to the body type data corresponding to the anchor frames with lower intersection ratio, recalculating the intersection ratio of the adjusted anchor frames and the target area, repeating the adjusting and calculating steps until the intersection ratio mean value meets the preset standard, completing anchor frame size optimization, counting the motion speed distribution and track change amplitude of the experimental mouse based on the motion characteristic data, setting an initial detection threshold value, reducing the detection threshold value when the motion speed of the experimental mouse is higher than the preset speed threshold value, increasing the detection threshold value when the motion track change amplitude is lower than the preset amplitude threshold value, completing detection threshold value optimization, extracting illumination intensity data and background noise related data from the experimental environment parameters, wherein the background noise related data comprises environment noise intensity and noise frequency distribution, establishing an illumination intensity, background noise sensitivity relation, and mapping when the motion speed of the experimental mouse is higher than the preset threshold value, the method comprises the steps of reducing brightness sensitivity parameters calculated by an optical flow field, improving contrast sensitivity parameters calculated by the optical flow field when illumination intensity is lower than a preset weak light threshold value, adjusting time window parameters of the optical flow field according to noise frequency distribution when background noise intensity is higher than the preset noise threshold value, capturing space coordinates and motion vectors of a laboratory mouse in real time by adopting an optimized improved target detection algorithm and adjusted dense optical flow field analysis parameters, and generating a time sequence behavior track map based on the space coordinates and the motion vectors.
  5. 5. The method for identifying the status of an adaptive gene editing laboratory mouse according to claim 1, wherein the calibration formula of the dynamic determination threshold is: ; Wherein, the As a basic threshold for the pathological state of the corresponding genotype, As a result of the baseline weight coefficient, For feature sequences within a sliding window, median is a Median function, And dynamically updating the characteristic distribution of the lead time node for self-adaptive learning factors.
  6. 6. The method for identifying the state of an adaptive gene editing laboratory mouse according to claim 1, wherein the feature weights are obtained through a calculation model based on mutual information entropy, and the formula is: ; Wherein, the For the mutual information entropy of feature f and state class S, In order to combine the feature sets of the feature set, For other feature sets than feature f, For the average correlation coefficient of feature f with other features, Is a redundancy inhibitor.
  7. 7. The method for identifying the status of an adaptive gene editing laboratory mouse according to claim 1, wherein the calculation formula of the improved weighted scoring model is: ; Wherein, the As the characteristic weight of the object to be processed, As a result of the normalized feature values, For the enhancement coefficient of the pathological relevance, Is characterized by And disease model Is a degree of association of (a) with each other.
  8. 8. The method for identifying the status of an adapted gene editing laboratory mouse according to claim 1, wherein the formula of the multidimensional confidence assessment model is: ; Wherein, the For the current state composite score, For a set of scores for all possible states, Is the minimum in the score set.
  9. 9. The method for identifying the state of the adaptive gene editing laboratory mice according to claim 1 is characterized in that the training process of the random forest classifier optimized through transfer learning comprises the steps of constructing a gene editing laboratory mice behavior sample library, wherein the sample library comprises time sequence behavior track maps of laboratory mice with different genotypes and different growth stages, each sample is marked with genotype, disease state and pathological index data, adopting a pre-training model to conduct initial training on the behavior data of a common mouse to obtain a basic classification model, extracting sample characteristics of the gene editing laboratory mice in the sample library, updating decision tree node splitting threshold and characteristic weight of the basic classification model through a fine-tuning strategy, introducing a cross verification mechanism to evaluate model performance, completing model training when the model performance reaches a preset standard, storing the trained model, and supporting rapid loading and self-adaptive adjustment when experimental environment parameters change.
  10. 10. The method for identifying the state of the adaptive gene editing laboratory mice is characterized in that an adaptive flow of the sizes of cages in experimental environment parameters comprises the steps of presetting coordinate transformation rules corresponding to various standard cage sizes, storing the coordinate transformation rules in a rule base, obtaining cage size information in the experimental environment parameters, judging whether the cage sizes are standard cage sizes, directly calling the corresponding coordinate transformation rules in the rule base if the cage sizes are standard cage sizes, generating the adaptive transformation rules through a coordinate scaling algorithm based on the standard coordinate transformation rules if the cage sizes are non-standard cage sizes, applying the generated adaptive transformation rules to time sequence behavior track patterns, correcting regional proportion parameters of the time sequence behavior track patterns, and guaranteeing consistency of feature extraction under different cage specifications.

Description

State identification method of adaptive gene editing laboratory mice Technical Field The invention relates to the technical field of image processing, in particular to a state identification method of an adaptive gene editing laboratory mouse. Background In the fields of life science research and drug development, the accurate identification of the state of a laboratory mouse is a key link for evaluating the physiological characteristics, disease progression and intervention effect of animals. In the prior art, a method for realizing anxiety and calm binary state identification by analyzing an experimental mouse behavior path diagram through machine learning is provided, and the method is characterized in that the optical flow method and a target detection algorithm are utilized to extract mouse position information, state judgment is completed through a random forest classifier and dynamic threshold calculation based on the general behavior characteristics such as the edge activity proportion, the drinking frequency and the like, and the error and the labor cost of traditional manual observation are effectively reduced. With the rapid development of gene editing technology, CRISPR/Cas9, base editing and other technologies are widely applied to disease animal model construction, and the gene editing disease model has definite genotype characteristics and specific pathological phenotypes. However, the existing laboratory mouse state identification method has the obvious limitations that firstly, the behavior characteristics of the prior art are designed only for common mice, the pathological association behaviors of a gene editing disease model (such as asymmetrical limb activities, pain-induced contracture behaviors, and the like of an arthritic mouse) are not covered, so that the characteristic pertinence is insufficient, secondly, the existing dynamic threshold and characteristic weight calibration mechanism do not consider genotype differences, the behavior baselines (such as resting time proportion and activity area range) of different gene editing models are essentially different from those of the common mice, the identification precision is greatly reduced due to the fact that the existing threshold is directly applied, thirdly, the existing training sample library depends on the human record data of the common mice, the standardized phenotype data support of the gene editing mice is lacked, the model generalization capability is limited, fourthly, the existing method is not adapted to the SPF-level feeding environment and standardized cage specification which are commonly used in the gene editing experiments, and the environmental factors are easy to cause interference on position extraction and characteristic analysis. The defects cause that the prior art cannot meet the condition identification requirement of a gene editing disease model, is difficult to accurately reflect the pathological condition of a model mouse, limits the application of the model mouse in related research of gene editing, and needs an experimental mouse condition identification technology capable of adapting to the specificity of the gene editing disease model. Disclosure of Invention The invention provides a state identification method of an adaptive gene editing experiment mouse, which comprises the steps of obtaining genotype information, disease model types and experiment environment parameters of the experiment mouse, optimizing an improved target detection algorithm and dense optical flow field analysis parameters based on the experiment environment parameters, capturing space coordinates and motion vectors of the experiment mouse in real time to generate a time sequence behavior track map, carrying out region segmentation on the time sequence behavior track map, identifying coordinate boundaries of a drinking region, a nest region and a pathology sensitive region, extracting a composite feature set in the time sequence behavior track map, carrying out standardization processing on feature data, wherein the composite feature set comprises general behavior features and pathology specific features, calling a preset genotype pathology feature association database, carrying out state identification on the standardized composite features by a random forest classifier optimized by transfer learning through an adaptive learning algorithm calibration feature weight and a dynamic judgment threshold, calculating the comprehensive score of the experiment mouse state by an improved weighting model, judging that the experiment mouse is in a related state is not the state when the comprehensive score is higher than the calibrated dynamic judgment threshold, and adopting a multidimensional degree of reliability evaluation model to evaluate the pathological state. Further, the genotype information comprises mutant genes and mutation sites, the disease model type comprises a single-gene genetic disease model and a spontaneous inflammati