A method for estimating rat death time based on nuclear magnetic resonance hydrogen spectrum technology

Abstract

The present invention relates to the field of forensic medicine, and specifically to a method for inferring the time of death of rats based on nuclear magnetic resonance hydrogen spectroscopy technology. The method includes the following steps: collecting skeletal muscle samples of rats with different times of death and extracting small molecule compounds, and performing metabolomics detection and analysis on the small molecule compounds in the skeletal muscle samples by <supgt;1< / supgt;H-NMR technology; using pattern recognition analysis, t-test, random forest feature selection and database comparison to screen for differential compounds with characteristic changes with different times of death, and establishing a time-of-death inference model using an ensemble learning algorithm; substituting the differential compound data of an unknown sample into the time-of-death inference model to predict the time of death of the unknown sample. The present invention uses an ensemble learning algorithm to predict the time of death and establishes a time-of-death inference model. The method of the present invention has high accuracy and high sensitivity.

Description

Technical Field

[0001] The invention relates to the field of forensic medicine, and in particular to a method for estimating rat death time based on nuclear magnetic resonance hydrogen spectrum technology. Background Art

[0002] In forensic medicine, time of death refers to the time experienced by the body after death or the postmortem interval (PMI), that is, the time interval between the examination of the body and the occurrence of death. Time of death inference refers to the inference of the postmortem experience or interval, which is not only an important part of forensic identification work, but also a scientific issue that has received much attention in the forensic profession. Accurate inference of time of death is of great significance for determining the time of crime, reconstructing the crime scene, and limiting the scope of criminal suspects. At present, there are many research methods for inferring time of death at home and abroad, such as immunohistochemistry, entomology, spectroscopy, and multi-omics. However, most of them are still in the theoretical research stage and cannot meet the needs of forensic practice. Therefore, it is particularly important to find more effective technical methods or objective biological indicators to infer time of death and improve the accuracy of time of death prediction.

[0003] In recent years, metabolomics has developed into a hot field in the study of life sciences. It mainly quantitatively studies the multivariate dynamic response of the level of metabolites in the body produced by living organisms to external stimuli. It can provide very comprehensive information on small molecule metabolites through high-throughput analysis of human muscles, body fluids and other samples. At present, forensic scientists have applied metabolomics technology to conduct research in different directions: changes in compounds in tissues or body fluids after death of organisms to infer PMI; metabolic changes in tissues or body fluids after body damage; changes in microorganisms after death of organisms, etc. Compared with liquid chromatography-mass spectrometry (LC-MS) and gas chromatography-mass spectrometry (GC-MS) technology, NMR technology has the advantages of wide application, stability and high reproducibility. Applying NMR technology to the study of time of death inference may be able to obtain more characteristic differential compounds, and through the combined use of multiple differential compounds, the accuracy and universality of time of death inference can be improved to a certain extent.

[0004] With the rapid development of artificial intelligence, machine learning algorithms have been widely used in various fields of scientific research. Machine learning algorithms mainly refer to the process of solving optimization problems through mathematical and statistical methods. According to different data and different model requirements, appropriate algorithms are selected to solve practical problems more efficiently. Stacking ensemble learning methods combine multiple machine learning algorithms to generate stronger models. The application of ensemble learning algorithms can more comprehensively explore the biological significance and relationships hidden behind the metabolomics data, and improve the accuracy of death time prediction and the universality of application. Summary of the invention

[0005] In order to improve the accuracy of predicting the time of death of rats, the present invention provides a method for estimating the time of death of rats based on nuclear magnetic resonance hydrogen spectrum technology.

[0006] The present invention is achieved through the following technical scheme: a method for estimating the time of death of rats based on nuclear magnetic resonance hydrogen spectrum technology, comprising the following steps:

[0007] Skeletal muscle samples of rats at different times of death were collected and small molecule compounds were extracted. 1 H-NMR technology was used to perform metabolomics analysis on small molecule compounds in skeletal muscle samples;

[0008] Pattern recognition analysis, t-test, random forest feature selection and database comparison were used to screen for differential compounds with characteristic changes at different death times, and an ensemble learning algorithm was used to establish a death time inference model.

[0009] The differential compound data of the unknown samples are substituted into the death time inference model to predict the death time of the unknown samples.

[0010] As a further improvement of the technical solution of the present invention, the pattern recognition includes principal component analysis and orthogonal partial least squares method-discriminant analysis.

[0011] As a further improvement of the technical solution of the present invention, the death time inference model adopts the Stacking integrated model.

[0012] As a further improvement of the technical solution of the present invention, the base learner of the Stacking integrated model includes a multilayer perceptron neural network, a linear discriminant and a random forest, and the meta learner includes logistic regression.

[0013] As a further improvement of the technical solution of the present invention, the differential compounds include choline, glycerol, succinate, phenylalanine, acetoacetate, leucine, glutamate, fumarate, valine, pyruvate, lactate dehydrogenase, betaine, hypoxanthine, propylene glycol, allopurindiol, tyrosine, inosine, acetone, hydroxybutyric acid, and creatinine.

[0014] As a further improvement of the technical solution of the present invention, the differential compound data of the unknown sample is obtained by 1 The results were obtained by H-NMR technology.

[0015] As a further improvement of the technical solution of the present invention, the evaluation methods that can be used for the death time inference model include calculating accuracy, precision, recall rate, F1 score and area under the receiver operating curve.

[0016] The present invention discloses a method for estimating the time of death of rats based on nuclear magnetic resonance hydrogen spectrum technology, which is used to explore the relationship between the metabolic changes of skeletal muscle and the time of death after death of rats, develop potential biomarkers related to the time of death, and mine differential compounds suitable for estimating the time of death. The time of death is predicted by an integrated learning algorithm, and a model for estimating the time of death is established. The unknown sample is substituted into the model to predict the time of death of the unknown sample. The method of the present invention has high accuracy and high sensitivity. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 Different time of death 1 H-NMR spectrum characteristics. In the figure, control group: skeletal muscle of rats immediately after death.

[0019] Figure 2 Orthogonal partial least squares-discriminant analysis was used to screen differential compounds at different death times.

[0020] Figure 3 This is a graph showing the internal validation results of the death time prediction model.

[0021] Figure 4 This is a diagram showing the external validation results of the death time prediction model.

[0022] Figure 5 The figure is a flow chart of the inference method of the present invention. DETAILED DESCRIPTION

[0023] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0024] The embodiment of the present invention provides a method for estimating the time of death of rats based on nuclear magnetic resonance hydrogen spectrum technology, comprising the following steps:

[0025] S101: Collect skeletal muscle samples of rats at different times of death and extract small molecule compounds. 1 H-NMR technology was used to perform metabolomics detection and analysis of small molecule compounds in skeletal muscle samples.

[0026] It should be noted that the rat skeletal muscle samples collected by the present invention can not only include samples with a death time of 3 days, but also samples with a longer death time. Moreover, even if the collected rat skeletal muscle samples are samples with a death time of 3 days, samples at other time points other than immediate death, postmortem 6h, 12h, 18h, 24h, 48h, 72h, etc. can also be used. The larger the death time span of the sample, the longer the predictable sample death time. The larger the number of samples, the higher the accuracy of the method of the embodiment of the present invention.

[0027] S102: Pattern recognition analysis, t-test, random forest feature selection and database comparison were used to screen differential compounds with characteristic changes at different death times, and an integrated learning algorithm was used to establish a death time inference model.

[0028] In this embodiment, the pattern recognition includes principal component analysis and orthogonal partial least squares-discriminant analysis. The difference compounds that can be used in this embodiment include: the difference compounds include choline, glycerol, succinate, phenylalanine, acetoacetate, leucine, glutamate, fumarate, valine, pyruvate, lactate dehydrogenase, betaine, hypoxanthine, propylene glycol, allopurinol, tyrosine, inosine, acetone, hydroxybutyric acid, and creatinine.

[0029] It is clear that the death time inference model established by the above-mentioned difference compounds can effectively estimate the death time of rats. Of course, the difference compounds that can be used in the present invention include but are not limited to the above-mentioned compounds.

[0030] Specifically, the death time inference model adopts the Stacking ensemble model. The base learners of the Stacking ensemble model include Multilayer Perception (MLP), Linear Discriminant Analysis (LDA) and Random Forest (RF), and the meta-learner includes Logistic Regression. However, the death time inference model that can be adopted by the present invention is by no means limited to the above-mentioned models.

[0031] When establishing a death time inference model, the content of differential compounds can be used as a variable, and the samples can be randomly divided into a training set, an external validation set, and an internal validation set according to a set ratio.

[0032] S103: Substitute the differential compound data of the unknown sample into the death time inference model to predict the death time of the unknown sample.

[0033] Similarly, the differential compound data of the unknown sample is obtained using 1 The H-NMR technique is used to detect the obtained results. Specifically, the same detection method as S101 can be used.

[0034] The present invention provides a 1 The method of inferring the time of death of rats by combining H-NMR technology with multivariate analysis includes the following steps in specific implementation:

[0035] 1. Sample collection:

[0036] (1) The animal experiments involved in this example strictly complied with the animal experiment regulations and systems of Shanxi Medical University. 70 Sprague-Dawley (SD) rats were randomly divided into 7 groups, 10 rats in each group, and placed in a climate chamber (humidity 50%, 16°C, and lighting conditions of 12 hours alternating day and night). The rats were killed immediately after death (control group), 6h, 12h, 18h, 24h, 48h, and 72h after death (10 rats in each group), and the corresponding rat skeletal muscle samples (>200 mg) were collected, quick-frozen in liquid nitrogen, and stored in a -80°C refrigerator for later use.

[0037] Among them, Sprague-Dawley (SD) rats were adult male rats, SPF grade (provided by Sprague-Dawley (Beijing) Biotechnology Co., Ltd., animal license number SCXK (Beijing) 2019-0010), with a body weight of 280-300 g, 10-12 weeks old, and the experiment was conducted after being raised at room temperature (24±2) ℃ for 1 week.

[0038] 2. Sample preprocessing:

[0039] Weigh 200 mg of rat skeletal muscle and put it into a 2 ml EP tube. Add 800 µl of pre-cooled 50% methanol-water (v:v). Add two zirconium oxide beads to the tube for tissue homogenization (30 times / second × 30 seconds × 4 times). After placing at 4°C for 10 minutes, centrifuge at 4°C and 12000r for 20 minutes, take 600 µl of the supernatant, and freeze-dry for 5 hours. Finally, dissolve the freeze-dried powder in 600 µl of phosphate buffer (0.2M Na2HPO4, 0.2M NaH2PO4, 0.01% TSP, pH 7.4) prepared with heavy water, centrifuge at 4°C and 12000r for 20 min, and transfer the supernatant into a 5 mm NMR tube for detection.

[0040] 3. 1 H-NMR detection:

[0041] The samples were analyzed using a presaturated water peak suppression (1D NOESY) pulse sequence with the following parameters: spin relaxation delay of 320 ms, free induction decay of 64 kb data points, spectrum width of 8000 Hz, and 64 scans.

[0042] 4. Data pre-processing:

[0043] MestReNova 9.0 (Mestrelab Research, Spain) was used to adjust the phase and baseline of the one-dimensional NMR spectra, and the spectra were calibrated using the chemical shift of TSP (δ=0). To eliminate the influence of residual water peaks, water peaks in the δ4.7~5.1 region were removed. The δ0.5~9.0 region was segmented integrated with a δ of 0.004, and the integral values ​​were normalized according to the total integral area of ​​each spectrum. The spectra were identified according to the Chenomx NMR Suite 8.2 database (Chenomx, Canada).

[0044] 5. Pattern recognition:

[0045] In SIMCA-P 14.0, the data were imported into SIMCA-P 14.0 software (Umetrics, Sweden) for principal component analysis (PCA) and orthogonal partial least squares-discriminant analysis (OPLS-DA). The model was verified by 200 permutation tests, and the variable importance in projection (VIP) was greater than 1 as the standard for screening differential variables. Student t test was performed in SPSS 16.0 software, with the test level α = 0.05.

[0046] 6. Screening of differential markers:

[0047] In SPSS 16.0 software, all time points in OPLS-DA were grouped and compared with the control group (immediate death) for t-test (α=0.05) to obtain the corresponding p value. Log in to the Chenomx NMR Suite 8.2 database (Chenomx, Canada) to identify the spectrum, and combine the VIP value obtained in OPLS-DA to screen potential differential markers that meet p<0.05 and VIP>1. Further feature selection by random forest model, 20 characteristic differential compounds were finally screened. The specific differential compounds are shown in the following table:

[0048] Table 1 Different compounds

[0049]

[0050] The metabolic pathways in which the above 20 differential compounds are mainly involved include methyl butyrate metabolism, alanine, aspartic acid and glutamate metabolism, phenylalanine, tyrosine and tryptophan, and the synthesis and degradation metabolism of ketone bodies. The biological functions involved are mainly amino acid metabolism and energy metabolism.

[0051] 7. Establish an integrated learning model for death time inference:

[0052] The contents of 20 different compounds were imported into Python software to establish an integrated learning model for death time inference. The Stacking integrated model was used, and its base learners included multilayer perceptron neural network (MultilayerPerception, MLP), linear discriminant analysis (LinearDiscriminantAnalysis, LDA) and random forest (RandomForest, RF), and logistic regression (LogisticRegression) was the meta-learner of the integrated learning model. One sample data was randomly selected from each death time point for external validation (a total of 7 sample data), and each of the remaining death time points contained nine biological replicates, of which 49 sample data were uniformly randomly selected as training sets, and another 14 sample data were used as internal validation sets.

[0053] 8. Evaluation of the performance of the death time inference model:

[0054] The internal validation accuracy, recall, F1 score, and precision of the Stacking ensemble model are all 92.86%, and the ROC area is 0.92 ( Figure 3 A in the internal validation data set includes 14 samples, and only one sample has an error in predicting the time of death ( Figure 3 B in the figure). The external validation accuracy, recall, F1 score, and precision are all 85.71%, and the ROC area is 0.85 ( Figure 4 A in the external validation data set, there are 7 samples in total, and the death time prediction of 1 sample is wrong ( Figure 4 B in the figure).

[0055] 9. Preparation, testing and result inference of unknown samples

[0056] The rat skeletal muscle sample to be tested is tested according to the above steps 1-5 to obtain the nuclear magnetic resonance hydrogen spectrum of the sample, and the data is pre-processed to obtain the content data of 20 differential compounds. The contents of the differential compounds are imported into the Stacking integrated model for prediction to infer the time of death of the rat.

[0057] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for estimating the time of death of rats based on nuclear magnetic resonance hydrogen spectrum technology, characterized in that: The following steps are involved: Skeletal muscle samples of rats at different times of death were collected and small molecule compounds were extracted. 1 H-NMR technology was used to perform metabolomics detection and analysis of small molecule compounds in skeletal muscle samples; when collecting rat skeletal muscle samples at different times of death, the samples were pretreated, and the specific steps were as follows: 50% methanol-water was added to the rat skeletal muscle, and zirconium oxide beads were added for tissue homogenization. After being placed at 4°C for 10 minutes, the samples were centrifuged at 4°C and 12000r for 20 minutes, the supernatant was taken, and freeze-dried for 5 hours; finally, the freeze-dried powder was dissolved in a phosphate buffer prepared with heavy water, and the supernatant was transferred to a nuclear magnetic tube for detection after centrifugation at 4°C and 12000r for 20 min. Pattern recognition analysis, t-test, random forest feature selection and database comparison were used to screen differential compounds with characteristic changes at different death times, and an integrated learning algorithm was used to establish a death time inference model; the pattern recognition included principal component analysis and orthogonal partial least squares-discriminant analysis; Substituting the differential compound data of the unknown sample into the death time inference model to predict the death time of the unknown sample; The death time inference model adopts a Stacking ensemble model; the base learners of the Stacking ensemble model include a multi-layer perceptron neural network, a linear discriminant and a random forest, and the meta-learner includes a logistic regression; The differential compounds include choline, glycerol, succinate, phenylalanine, acetoacetate, leucine, glutamate, fumarate, valine, pyruvate, lactate dehydrogenase, betaine, hypoxanthine, propylene glycol, allopurindiol, tyrosine, inosine, acetone, hydroxybutyric acid, and creatinine; the differential compound data of the unknown sample is obtained by 1 The results were obtained by H-NMR technology.

2. The method for estimating rat death time based on nuclear magnetic resonance hydrogen spectrum technology according to claim 1, characterized in that: The evaluation methods that can be used for the death time inference model include calculating accuracy, precision, recall, F1 score and area under the receiver operating curve.

Images