Biological radiation dose evaluation method and device and terminal equipment

Abstract

The invention discloses a biological radiation dose evaluation method and device and terminal equipment. The method comprises the following steps: extracting expression information of a first target gene in a biological sample to be evaluated; inputting the expression information of the first target gene into a preset dose evaluation model, so that the model determines the radiation dose of the biological sample to be evaluated according to the expression information of the first target gene; wherein the model finds out differential genes by comparing transcriptome sequencing information of a radiated training sample with transcriptome sequencing information of a corresponding non-radiated training sample, and performs characteristic correlation screening and quantitative gene amplification fluorescence detection on the differential genes to obtain expression information of a second target gene; and taking the actual radiation dose of the radiated training sample and the expression information of the corresponding second target gene as input, taking the predicted radiation dose as output, training a dose evaluation model to be trained, and obtaining a trained dose evaluation model. According to the invention, the radiation dose of the to-be-evaluated sample can be rapidly predicted.

Description

Technical Field

[0001] The present invention relates to the field of biological dose assessment, and in particular to a biological radiation dose assessment method, device and terminal equipment. Background Art

[0002] At present, there are relatively few methods available for calculating the radiation dose to organisms. Conventional calculation methods include the use of genetic methods, such as chromosomal aberrations, cytokinesis-block micronucleus, and fluorescence in situ hybridization, as well as various omics analyses, such as genomics, metabolomics, and proteomics, and the search for biomarkers for dose reconstruction.

[0003] Although conventional radiation dose assessment methods provide important tools for scientific research, their efficiency and economic limitations make these methods challenging to apply in practice. Conventional genetic methods and omics analysis methods require complex experimental operations and data analysis, which are time-consuming and require a lot of manpower and time investment. In emergency radiation accidents, it is necessary to quickly and accurately assess the dose, but conventional methods often cannot meet the requirements of real-time and rapidity. Summary of the invention

[0004] The embodiments of the present invention provide a biological radiation dose assessment method, apparatus and terminal device, which can quickly predict the radiation dose of a sample to be assessed.

[0005] An embodiment of the present invention provides a biological radiation dose assessment method, comprising:

[0006] Extracting expression information of a first target gene in the biological sample to be evaluated;

[0007] Inputting the expression information of the first target gene into a preset dose assessment model, so that the dose assessment model determines the radiation dose of the biological sample to be assessed according to the expression information of the first target gene;

[0008] The construction of the dose assessment model includes:

[0009] Obtaining several groups of training samples of training subjects that have been exposed to known radiation doses and training samples that have not been exposed to radiation;

[0010] Performing transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples;

[0011] Compare and analyze the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differentially expressed genes;

[0012] Perform feature correlation screening on the differentially expressed genes to obtain the second target gene;

[0013] Performing quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene;

[0014] Taking the actual radiation dose of the irradiated training sample and the expression information of the corresponding second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model.

[0015] Furthermore, the differentially expressed genes are screened for correlation of features to obtain the second target genes, including:

[0016] For each differentially expressed gene, the log-fold change, significance level, and correlation coefficient of the differentially expressed gene were calculated;

[0017] The differentially expressed genes that simultaneously meet the second preset condition, the third preset condition and the fourth preset condition are screened out from the differentially expressed genes as the second target genes for preliminary screening; the second preset condition is that the absolute value of the logarithmic fold change is greater than the first preset threshold; the third preset condition is that the significance level is less than the second preset threshold; the fourth preset condition is that the absolute value of the correlation coefficient is greater than or equal to the third preset threshold;

[0018] The Boruta algorithm was used to perform feature screening on the preliminarily screened second target genes to obtain the final second target genes.

[0019] Furthermore, the dose assessment model to be trained is trained with the actual radiation dose of the irradiated training sample and the corresponding expression information of the second target gene as input and the predicted radiation dose as output, so as to obtain a trained dose assessment model, including:

[0020] Dividing the irradiated training samples into a first training set and a first test set according to a preset ratio;

[0021] The actual radiation dose of each training sample in the first training set and the corresponding expression information of the second target gene are input into the dose assessment model to be trained for training until the number of training times exceeds the preset number, thereby obtaining a trained dose assessment model.

[0022] Furthermore, the dose assessment model includes: a basic sub-model and an optimization sub-model;

[0023] The actual radiation dose of each training sample in the first training set and the corresponding expression information of the second target gene are input into the dose assessment model to be trained for training until the first training times exceed the first preset times, thereby obtaining a trained dose assessment model, including:

[0024] Divide the first training set into several training subsets;

[0025] In each training, one training subset is selected from all training subsets in order as the second test set, and the remaining training subsets are used as the second training set; wherein the training subsets can be selected cyclically;

[0026] The basic sub-model is trained according to each data in the second training set until the second training times exceed the second preset times, thereby obtaining a trained basic sub-model and each first predicted radiation dose obtained from each training of the basic sub-model;

[0027] According to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set, the optimization sub-model is trained until a third training number exceeds a third preset number, thereby obtaining a trained optimization sub-model;

[0028] Inputting the second test set into the trained basic sub-model and the trained optimization sub-model to obtain the predicted radiation dose of each data in the second test set;

[0029] The values ​​of the coefficient of determination and the root mean square error are calculated according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set;

[0030] Determine whether the first training times is less than the first preset times. If so, continue with the next training. If not, compare the values ​​of the coefficient of determination and the root mean square error obtained in each training, and use the dose assessment model corresponding to the maximum coefficient of determination and the minimum root mean square error as the trained dose assessment model.

[0031] Furthermore, the basic sub-model includes a combination of the following multiple regression sub-models: random forest regression sub-model, linear regression sub-model, XGBoost regression sub-model, CatBoost regression sub-model, elastic network regression sub-model and support vector regression sub-model.

[0032] Further, the basic sub-model is trained according to each data in the second training set until the second training times exceed the second preset times, and the trained basic sub-model and each first predicted radiation dose obtained by each training of the basic sub-model are obtained, including:

[0033] The basic sub-model is trained according to each data in the second training set; wherein, in each training, for each regression sub-model, the expression information of the second target gene of each data in the second training set is used as input, and the first predicted radiation dose of each data in the second training set is used as output for training;

[0034] For each regression sub-model, according to the first predicted radiation dose of each data in each second training set and the corresponding actual radiation dose, the root mean square error of each regression sub-model is calculated, and the root mean square error of each regression sub-model is weighted to obtain the first loss function of the basic sub-model;

[0035] Determine whether the second training times is less than the second preset times. If so, repeat the next training; if not, compare the first loss function values ​​obtained from each training, and take the basic sub-model corresponding to the minimum first loss function value as the basic sub-model that has completed the training.

[0036] Further, according to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set, the optimization sub-model is trained until the third training times exceeds the third preset times, and a trained optimization sub-model is obtained, including:

[0037] Divide each first predicted radiation dose obtained by each training of the basic sub-model into several groups of predicted data; wherein the predicted data includes: each first predicted radiation dose obtained by the same training of the basic sub-model;

[0038] The optimization sub-model is trained according to each first predicted radiation dose obtained from each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set; wherein, during each training, a group of prediction data is selected in sequence from a plurality of groups of prediction data, and for each first predicted radiation dose in the group of prediction data and the data corresponding to the first predicted radiation dose in the second training set, the first predicted radiation dose and the data corresponding to the first predicted radiation dose in the second training set are used as input, and the second predicted radiation dose of the second training set data is used as output for training;

[0039] A second loss function is calculated based on the second predicted radiation dose of each second training set data and the corresponding actual radiation dose;

[0040] Determine whether the third training times is less than the third preset times. If so, continue with the next training; if not, compare the second loss function values ​​obtained from each training, and take the optimized sub-model corresponding to the minimum second loss function value as the optimized sub-model that has completed the training.

[0041] Furthermore, according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the values ​​of the coefficient of determination and the root mean square error are calculated, including:

[0042] According to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the determination coefficient is calculated by the following formula:

[0043]

[0044] According to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the root mean square error is calculated by the following formula:

[0045]

[0046] Among them, R 2 represents the coefficient of determination, RMSE represents the root mean square error, and Y i represents the actual radiation dose of the i-th data in the second test set, represents the predicted radiation dose of the i-th data in the second test set, represents the average value of the actual radiation dose in the second test set, and n represents the total amount of data in the test set.

[0047] Based on the above method embodiment, the present invention provides a corresponding device embodiment, including: a first target gene expression information extraction module, and a dosage assessment module;

[0048] A first target gene expression information extraction module, used to extract the expression information of the first target gene in the biological sample to be evaluated;

[0049] A dose assessment module, used to input the expression information of the first target gene into a preset dose assessment model, so that the dose assessment model determines the radiation dose of the biological sample to be assessed according to the expression information of the first target gene;

[0050] The construction of the dose assessment model includes:

[0051] Obtaining several groups of training samples of training objects that have been irradiated and training samples that have not been irradiated;

[0052] Performing transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples;

[0053] Compare and analyze the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differentially expressed genes;

[0054] Perform feature correlation screening on the differentially expressed genes to obtain the second target gene;

[0055] Performing quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene;

[0056] Taking the actual radiation dose of the irradiated training sample and the expression information of the corresponding second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model.

[0057] Based on the above-mentioned method embodiment, the present invention provides a corresponding terminal device embodiment, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps of the biological radiation dose assessment method described in the present invention are implemented.

[0058] Compared with the prior art, the beneficial effects of the embodiment of this solution are:

[0059] The present invention obtains the expression information of specific target genes of the biological sample to be evaluated. Compared with the traditional gene detection method, the target genes targeted by the present invention have been pre-determined as genes associated with radiation doses, eliminating the step of identifying differential genes caused by radiation by comparison with normal biological individual samples, thereby improving the efficiency of dose evaluation; then, the expression information of the specific target genes is input into a preset dose evaluation model, so that the dose evaluation model determines the radiation dose of the sample to be evaluated according to the specific target gene expression information. Through a large amount of sample data, the model can learn the relationship between different gene expressions and their corresponding radiation doses, so that the radiation dose can be evaluated in a shorter time. The present invention can quickly predict the radiation dose of the sample to be evaluated according to the relationship between differential gene expression and radiation dose that has been learned by the dose evaluation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of a biological radiation dose assessment method provided by an embodiment of the present invention;

[0061] Figure 2 is a flow chart of a dose assessment model training process provided by an embodiment of the present invention;

[0062] Figure 3 is a dose reconstruction diagram of a dose assessment model of a first test set provided by an embodiment of the present invention;

[0063] Figure 4 It is a schematic diagram of the structure of a biological radiation dose assessment device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features.

[0066] like Figure 1 As shown, an embodiment of the present invention provides a biological radiation dose assessment method, the method at least comprising the following steps:

[0067] Step S1: extracting the expression information of the first target gene in the biological sample to be evaluated;

[0068] For step S1, firstly, it is necessary to obtain a biological sample to be evaluated of the evaluation object, which can be blood, urine, hair, oral swab, etc. of the evaluation object that can provide genetic information of the evaluation object. In this embodiment, mouse blood samples are taken as an example to obtain blood of mice exposed to unknown radiation.

[0069] Then, the expression information of the first target gene in the blood of mice subjected to unknown radiation is measured using real-time quantitative gene amplification fluorescence detection (qPCR) technology. qPCR technology is a highly sensitive and quantitative method that can accurately measure the expression level of a gene, and the measured results have high accuracy and repeatability. In this embodiment, for the sequence of the first target gene, corresponding primers and probes are designed, RNA is extracted and reverse transcribed to synthesize cDNA, and then amplification reaction is performed using a qPCR instrument, and the progress and results of the PCR product are tracked by monitoring the fluorescent signal, and the expression amount of the first target gene is calculated accordingly.

[0070] It should be noted that the first target gene involved in the present invention is a gene that has been predetermined to be associated with the radiation dose. Therefore, when the biological radiation dose is actually evaluated, these genes associated with the radiation dose can be directly extracted, and the expression of these target genes can be calculated by qPCR technology. Compared with the traditional gene detection method, the present invention does not need to identify the differential genes affected by radiation by comparing with normal biological individual samples, thereby eliminating the step of comparing normal biological individual samples and effectively improving the efficiency of evaluating the dose. The first target genes involved in this embodiment include: Net_sig, Ccnd1, Grn, Ifitm3, Phlda3, Ccng1, Sgta, Tcn2, Lrg1, Tmbim6, Xdh, Bax, Cd5l, Cd19, Glul, Spib, Swap70, Ccr7, Rhoc, Ly6d.

[0071] Step S2: inputting the expression information of the first target gene into a preset dose assessment model, so that the dose assessment model determines the radiation dose of the biological sample to be assessed according to the expression information of the first target gene;

[0072] The construction of the dose assessment model includes:

[0073] Obtaining several groups of training samples of training subjects that have been irradiated with known doses and training samples that have not been irradiated;

[0074] Performing transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples;

[0075] Compare and analyze the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differentially expressed genes;

[0076] Perform feature correlation screening on the differentially expressed genes to obtain the second target gene;

[0077] Performing quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene;

[0078] Taking the actual radiation dose of the irradiated training sample and the expression information of the corresponding second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model.

[0079] For step S2, the first target gene expression information obtained in step S1 is input into a preset dose assessment model, and the model can determine the radiation dose of the biological sample to be evaluated based on this information. The dose assessment model takes the actual radiation dose of the irradiated training sample and the corresponding expression information of the second target gene as input, and is trained with the predicted radiation dose of the training sample as output. Through training, the dose assessment model can learn the relationship between gene expression information and radiation dose, so that when the gene expression information of the biological sample to be evaluated is given, the radiation dose of the sample can be accurately predicted. This model can play an important role in the field of radiation dose assessment, helping scientists and doctors to more quickly assess the radiation dose of biological samples for further analysis and processing.

[0080] The following is a detailed description of the training process of the dose assessment model:

[0081] like Figure 2 As shown in Figure 1, the training process of the dose assessment model includes the following steps:

[0082] Step S201: obtaining several groups of training samples of training subjects that have been irradiated with known doses and training samples that have not been irradiated;

[0083] For step S201, multiple groups of irradiated training samples with known radiation doses and normal training samples of the same training subject without radiation are obtained. It should be noted that the normal training samples without radiation used as controls and the training samples with known radiation doses in the same group need to come from the same training subject, which can reduce the impact of individual gene differences on the results, thereby better obtaining the relationship between gene expression information and radiation dose.

[0084] Step S202: performing transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples;

[0085] For step S202, transcriptome sequencing is performed on the irradiated training samples and the corresponding non-irradiated training samples in each group of training samples. In transcriptome sequencing, RNA in the sample needs to be extracted first and converted into cDNA; then, the cDNA is sequenced using a sequencer to obtain a large number of sequencing reads; next, the expression level of each gene can be determined by comparing the sequencing reads with normal biological samples, and the expression levels of all genes in each training sample are measured to obtain the corresponding transcriptome sequencing information. Among them, this information includes numerical data of gene expression levels, which can be used to compare the differences in gene expression between irradiated training samples and normal training samples. By analyzing these transcriptome sequencing data, we can obtain detailed information about the impact of radiation on gene expression.

[0086] Step S203: Compare and analyze the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differential genes;

[0087] For step S203, the transcriptome sequencing information of the irradiated training samples obtained in step S202 is compared and analyzed with the transcriptome sequencing information of the corresponding non-irradiated training samples. By comparing the gene expression levels of the two groups of samples, genes whose expression is significantly changed in the irradiated group compared with the non-irradiated group can be found, thereby obtaining differential genes.

[0088] Step S204: screening the correlation of features of the differentially expressed genes to obtain a second target gene;

[0089] In a preferred embodiment, the differentially expressed genes are screened for correlation of features to obtain a second target gene, including:

[0090] For each differentially expressed gene, the log-fold change, significance level, and correlation coefficient of the differentially expressed gene were calculated:

[0091] The differentially expressed genes that simultaneously meet the second preset condition, the third preset condition and the fourth preset condition are screened out from the differentially expressed genes as the second target genes for preliminary screening; the second preset condition is that the absolute value of the logarithmic fold change is greater than the first preset threshold; the third preset condition is that the significance level is less than the second preset threshold; the fourth preset condition is that the absolute value of the correlation coefficient is greater than or equal to the third preset threshold;

[0092] The Boruta algorithm was used to perform feature screening on the preliminarily screened second target genes to obtain the final second target genes.

[0093] For step S204, by comparing and analyzing the irradiated samples with the non-irradiated samples in step S203, the genes with different transcriptome sequencing information due to radiation changes are identified. However, in order to ensure the accuracy and reliability of model training, it is necessary to further screen out the differential genes with strong correlation with radiation dose among the genes with different transcriptome sequencing information.

[0094] First, the differentially expressed genes that were correlated with the dosage were preliminarily screened out using statistical methods. Specifically, the log fold change (log2 Fold Change, log2FC), significance level (p-value), and correlation coefficient of each gene in the genes with different transcriptome sequencing information were calculated. Next, differentially expressed genes that met the following preset thresholds were screened out from the genes with different transcriptome sequencing information: |log2FC|>1, indicating that the gene expression level had changed at least twice; p-value <0.05 to ensure significant differences in differential gene expression; at the same time, |Pearson correlation coefficient|≤0.3, indicating that there was a strong correlation between the differentially expressed genes and the dosage.

[0095] After preliminary screening by statistical methods, the Boruta algorithm was further used for feature screening. The Boruta algorithm is a feature selection algorithm that comprehensively considers the correlation between genes and dosages and their importance in random forests. It iteratively evaluates each differentially expressed gene, compares the importance index of the original gene with the importance index of the randomly generated features in the random forest model, and finally selects the second target gene that has a strong correlation with dosage.

[0096] It should be noted that the first target gene used in the actual radiation dose assessment is consistent with the second target gene obtained through the dose assessment model training, and both represent genes that are strongly correlated with the dose.

[0097] Step S205: performing quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene;

[0098] For step S205, quantitative gene amplification fluorescence detection (qPCR) technology is used to quantitatively measure the expression level of a specific second target gene, and suitable primers and probes are designed for the second target gene to amplify and detect these genes. During the amplification process, the primers or probes are labeled to emit fluorescent signals, and then the intensity of the fluorescent signals is measured to obtain the expression information of the second target gene. By performing quantitative gene amplification fluorescence detection on the second target gene, quantitative data on the expression level of the second target gene can be obtained.

[0099] Step S206: taking the actual radiation dose of the irradiated training sample and the corresponding expression information of the second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model.

[0100] In a preferred embodiment, the dose assessment model to be trained is trained with the actual radiation dose of the irradiated training sample and the corresponding expression information of the second target gene as input and the predicted radiation dose as output to obtain a trained dose assessment model, including:

[0101] Dividing the irradiated training samples into a first training set and a first test set according to a preset ratio;

[0102] The actual radiation dose of each training sample in the first training set and the corresponding expression information of the second target gene are input into the dose assessment model to be trained for training until the number of training times exceeds the preset number, thereby obtaining a trained dose assessment model.

[0103] For step S206, the irradiated training samples are first divided into two parts according to a preset ratio (7:3): a first training set N_train and a first test set N_test. The training set N_train is used to train the dose assessment model, while the test set N_test is used to verify and evaluate the trained model.

[0104] Then, the first training set N_train is input into the dose assessment model to be trained, with the expression information of each second target gene in the first training set N_train and the corresponding actual radiation dose as input, and the predicted radiation dose as output for training until the number of training times reaches a preset number of times. In this implementation, the preset number of times is 50 times.

[0105] The following is a detailed description of each training process of the dose assessment model:

[0106] In a preferred embodiment, the dose assessment model includes: a basic sub-model and an optimization sub-model;

[0107] The actual radiation dose of each training sample in the first training set and the corresponding expression information of the second target gene are input into the dose assessment model to be trained for training until the first training times exceed the first preset times, thereby obtaining a trained dose assessment model, including:

[0108] Divide the first training set into several training subsets;

[0109] In each training, one training subset is selected from all training subsets in order as the second test set, and the remaining training subsets are used as the second training set; wherein the training subsets can be selected cyclically;

[0110] The basic sub-model is trained according to each data in the second training set until the second training times exceed the second preset times, thereby obtaining a trained basic sub-model and each first predicted radiation dose obtained from each training of the basic sub-model;

[0111] According to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set, the optimization sub-model is trained until a third training number exceeds a third preset number, thereby obtaining a trained optimization sub-model;

[0112] Inputting the second test set into the trained basic sub-model and the trained optimization sub-model to obtain the predicted radiation dose of each data in the second test set;

[0113] The values ​​of the coefficient of determination and the root mean square error are calculated according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set;

[0114] Determine whether the first training times is less than the first preset times. If so, continue with the next training. If not, compare the values ​​of the coefficient of determination and the root mean square error obtained in each training, and use the dose assessment model corresponding to the maximum coefficient of determination and the minimum root mean square error as the trained dose assessment model.

[0115] In a preferred embodiment, the values ​​of the coefficient of determination and the root mean square error are calculated based on the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, including:

[0116] According to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the determination coefficient is calculated by the following formula:

[0117]

[0118] According to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the root mean square error is calculated by the following formula:

[0119]

[0120] Among them, R 2 represents the coefficient of determination, RMSE represents the root mean square error, and Y i represents the actual radiation dose of the i-th data in the second test set, represents the predicted radiation dose of the i-th data in the second test set, represents the average value of the actual radiation dose in the second test set, and n represents the total amount of data in the test set.

[0121] In one embodiment of the present invention, the dose assessment model includes two parts: a basic sub-model and an optimization sub-model. The basic sub-model is used to establish an initial dose assessment model, while the optimization sub-model further integrates and corrects the prediction results based on the prediction results of the basic sub-model.

[0122] During the training of the dose assessment model, the first training set is first divided into five training subsets, i.e., five-fold cross validation. Each time training, one training subset is selected in sequence as the second test set N_train_test, and the remaining training subsets are selected as the second training set N_train_train. In this way, multiple training and testing combinations can be obtained, thereby increasing the number of samples for the training model and making more effective use of available data. These five training subsets can be selected cyclically to ensure that the model is fully trained;

[0123] In each training, firstly, the expression information of each second target gene in the second training set N_train_train is input into the basic sub-model for training until the number of training times of the basic sub-model meets the preset number of times, and the optimal basic sub-model is selected according to the root mean square error obtained from each training of the basic sub-model, so that the trained basic sub-model and the first predicted radiation dose of the basic sub-model for the data in the second training set N_train_train can be obtained; next, according to the first predicted radiation dose and the corresponding data in the second training set, the optimized sub-model is trained until the number of training times of the optimized sub-model meets the preset number of times, and the optimal optimized sub-model is selected according to the root mean square error obtained from each training of the optimized sub-model, so that the trained basic sub-model and the optimized sub-model can be obtained; then, by inputting the second test set into the trained basic sub-model and the optimized sub-model, the prediction results of the basic sub-model and the optimized sub-model on the second test set are calculated, and according to the comparison between the predicted results and the actual radiation dose, the determination coefficient R of the basic sub-model and the optimized sub-model is calculated by the formula. 2 The value of the root mean square error (RMSE) is used to determine whether the number of training times meets the requirements according to the preset determination coefficient requirements, root mean square error threshold requirements, and the preset number of iterations threshold. If the requirements are met, the determination coefficient R obtained from each training is 2 Compared with the value of the root mean square error RMSE, in this embodiment, 50 groups of determination coefficients R 2 and the value of the root mean square error RMSE, and select the determination coefficient R 2 The dose assessment model corresponding to the minimum maximum simultaneous root mean square error RMSE is taken as the trained dose assessment model. If the number of training times does not meet the requirement, the next training is carried out, and the next training subset is selected as the second test set for the next training to continue training.

[0124] It should be noted that if there is no determination coefficient R 2 When the maximum root mean square error RMSE is the smallest, the determination coefficient R is selected according to a certain weight. 2 The dose assessment model corresponding to a larger value and a smaller RMSE.

[0125] In a preferred embodiment, the basic sub-model includes a combination of the following multiple regression sub-models: a random forest regression sub-model, a linear regression sub-model, an XGBoost regression sub-model, a CatBoost regression sub-model, an elastic network regression sub-model and a support vector regression sub-model.

[0126] In a preferred embodiment, the basic sub-model is trained according to each data in the second training set until the second training times exceed the second preset times, and the trained basic sub-model and each first predicted radiation dose obtained by each training of the basic sub-model are obtained, including:

[0127] The basic sub-model is trained according to each data in the second training set; wherein, in each training, for each regression sub-model, the expression information of the second target gene of each data in the second training set is used as input, and the first predicted radiation dose of each data in the second training set is used as output for training;

[0128] For each regression sub-model, according to the first predicted radiation dose of each data in each second training set and the corresponding actual radiation dose, the root mean square error of each regression sub-model is calculated, and the root mean square error of each regression sub-model is weighted to obtain the first loss function of the basic sub-model;

[0129] Determine whether the second training times is less than the second preset times. If so, repeat the next training; if not, compare the first loss function values ​​obtained from each training, and take the basic sub-model corresponding to the minimum first loss function value as the basic sub-model that has completed the training.

[0130] In one embodiment of the present invention, the basic sub-model is composed of a stacked combination of a random forest regression sub-model, a linear regression sub-model and an XGBoost regression sub-model. Therefore, in the training process of the basic sub-model, the expression information of each second target gene in the second training set N_train_train is input into the basic sub-model, that is, the expression information of each second target gene in the second training set is respectively input into the three regression sub-models. For each second target gene expression information, it is used as input, and the random forest regression sub-model, the linear regression sub-model and the XGBoost regression sub-model are used for training respectively, and the predicted radiation dose results of the three regression sub-models are obtained. n represents the expression information of the nth second target gene in the second training set, and i represents the i-th training of the basic sub-model.

[0131] Each regression sub-model has its own learning algorithm and feature representation capabilities. By inputting the expression information of the second target gene into these models, their diversity can be used to explore different patterns and associations in the data and obtain different results. The random forest regression sub-model has good robustness and generalization capabilities, and can effectively handle the nonlinear relationship between multiple input features; the linear regression sub-model is suitable for processing data with strong linear correlation and can provide better interpretability; and the XGBoost regression sub-model is a powerful model based on gradient boosting, which has high prediction performance when dealing with complex data and nonlinear relationships.

[0132] For each regression sub-model, the calculated predicted radiation dose is compared with the corresponding actual radiation dose, the root mean square error of each regression sub-model is calculated, and the weighted value of the root mean square error of each regression sub-model is used as the loss function; when the number of training times of the basic sub-model meets the preset number of training times, the first loss function obtained from each training is compared, and the basic sub-model corresponding to the minimum value of the first loss function is used as the basic sub-model that has completed the training, and each first predicted radiation dose obtained from each training of the basic sub-model is output If the preset number of training times is not reached, the training will be repeated on the data of the second training set.

[0133] In a preferred embodiment, the optimization sub-model is trained according to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set, until the third training times exceeds the third preset times, and the optimized sub-model after training is obtained, including:

[0134] Divide each first predicted radiation dose obtained by each training of the basic sub-model into several groups of predicted data; wherein the predicted data includes: each first predicted radiation dose obtained by the same training of the basic sub-model;

[0135] The optimization sub-model is trained according to each first predicted radiation dose obtained from each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set; wherein, during each training, a group of prediction data is selected in sequence from a plurality of groups of prediction data, and for each first predicted radiation dose in the group of prediction data and the data corresponding to the first predicted radiation dose in the second training set, the first predicted radiation dose and the data corresponding to the first predicted radiation dose in the second training set are used as input, and the second predicted radiation dose of the second training set data is used as output for training;

[0136] A second loss function is calculated based on the second predicted radiation dose of each second training set data and the corresponding actual radiation dose;

[0137] Determine whether the third training times is less than the third preset times. If so, continue with the next training; if not, compare the second loss function values ​​obtained from each training, and take the optimized sub-model corresponding to the minimum second loss function value as the optimized sub-model that has completed the training.

[0138] In one embodiment of the present invention, the result obtained by the basic sub-model is based on the predicted radiation dose results obtained by three independent regression sub-models, but we hope to further optimize these results and integrate them to obtain more accurate prediction results. To this end, each first predicted radiation dose output by the basic sub-model And the expression information of the second target gene corresponding to the first predicted radiation dose in the second training set is used as input, that is, Each time j is taken as 1, 2, 3, ..., i, the second predicted radiation dose of the second training set data is Train for output; the second predicted radiation dose obtained by this training By comparing with the actual radiation dose, the root mean square error of this training can be calculated and used as the second loss function to determine whether the number of training times of the optimization sub-model has reached the preset number of times. If the optimization sub-model has reached the preset number of training times, the second loss function of each training is obtained. The second loss function values ​​obtained from each training are compared, and the optimization sub-model corresponding to the minimum second loss function value is selected as the optimization sub-model that has completed training. The minimum second loss function value means that the optimization sub-model has reached a good training state.

[0139] It should be noted that the optimization sub-model aims to integrate the advantages of different models and obtain a final prediction result by integrating the results of each regression sub-model obtained by the basic sub-model.

[0140] Finally, by comparing the determination coefficient and root mean square error values ​​obtained from each training, the optimal dose assessment model is selected as the trained dose assessment model. The trained dose assessment model is used to evaluate the radiation dose of the data of the first test set N_test. The results are as follows: Figure 3 As shown, the coefficient of determination R 2 The coefficient of determination is 0.852, which is greater than 85%. The coefficient of determination is an indicator for evaluating the degree of fit of the model to the data. Its value range is between 0 and 1. The closer it is to 1, the stronger the model's ability to explain the data. In this case, the coefficient of determination R 2The root mean square error (RMSE) is 1.292 Gy, which is less than 1.5 Gy. The root mean square error is an indicator of the difference between the predicted value and the true value. The smaller the value, the higher the prediction accuracy of the model. In this case, the root mean square error is 1.292 Gy, which is less than the threshold of 1.5 Gy, indicating that the prediction accuracy of the model is high and the difference between the predicted result and the true value is relatively small. In summary, the trained dose assessment model performs well and has high prediction ability and accuracy.

[0141] like Figure 4 As shown, based on the above method embodiment, a corresponding device embodiment is provided;

[0142] An embodiment of the present invention provides a biological radiation dose assessment device, comprising: a first target gene expression information extraction module, and a dose assessment module;

[0143] A first target gene expression information extraction module, used to extract the expression information of the first target gene in the biological sample to be evaluated;

[0144] A dose assessment module, used to input the expression information of the first target gene into a preset dose assessment model, so that the dose assessment model determines the radiation dose of the biological sample to be assessed according to the expression information of the first target gene;

[0145] The dosage assessment module includes: a training sample acquisition submodule, a transcriptome sequencing submodule, a differential gene analysis submodule, a differential gene screening submodule, a gene expression information detection submodule, and a dosage assessment model training submodule;

[0146] The training sample acquisition submodule is used to acquire several groups of training samples of training subjects with known radiation doses and training samples without radiation;

[0147] The transcriptome sequencing submodule is used to perform transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples;

[0148] The differential gene analysis submodule is used to compare and analyze the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differential genes;

[0149] The differential gene screening submodule is used to screen the correlation of the characteristics of the differential genes to obtain the second target gene;

[0150] A gene expression information detection submodule is used to perform quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene;

[0151] The dose assessment model training submodule is used to take the actual radiation dose of the irradiated training sample and the expression information of the corresponding second target gene as input and the predicted radiation dose as output to train the dose assessment model to be trained, thereby obtaining a trained dose assessment model.

[0152] In a preferred embodiment, the differential gene screening submodule includes: a differential expression calculation unit, a first screening unit, and a second screening unit;

[0153] A differential expression calculation unit, used to calculate the logarithmic fold change, significance level and correlation coefficient of each differential gene;

[0154] The first screening unit is used to screen out the differential genes that meet the second preset condition, the third preset condition and the fourth preset condition from the differential genes as the second target genes for preliminary screening; the second preset condition is that the absolute value of the logarithmic fold change is greater than the first preset threshold; the third preset condition is that the significance level is less than the second preset threshold; the fourth preset condition is that the absolute value of the correlation coefficient is greater than or equal to the third preset threshold;

[0155] The second screening unit is used to perform feature screening on the preliminarily screened second target gene using the Boruta algorithm to obtain the second target gene.

[0156] In a preferred embodiment, the dose assessment model training submodule includes: a first data division unit, and a training unit;

[0157] A first data division unit, used for dividing the irradiated training samples into a first training set and a first test set according to a preset ratio;

[0158] The training unit is used to input the actual radiation dose of each training sample in the first training set and the expression information of the corresponding second target gene into the dose assessment model to be trained for training until the number of training times exceeds a preset number, thereby obtaining a trained dose assessment model.

[0159] In a preferred embodiment, the training unit includes: a second data partitioning subunit, a second training set selection subunit, a basic submodel training subunit, an optimization submodel training subunit, a second test set prediction subunit, a model performance calculation subunit, and a first training times determination subunit;

[0160] A second data division subunit, used for dividing the first training set into a plurality of training subsets;

[0161] The second training set selection subunit is used to select one training subset from all training subsets in order as the second test set during each training, and the remaining training subsets as the second training set; wherein the training subsets can be selected cyclically;

[0162] A basic sub-model training sub-unit is used to train the basic sub-model according to each data in the second training set until the second training times exceed the second preset times, thereby obtaining a trained basic sub-model and each first predicted radiation dose obtained from each training of the basic sub-model;

[0163] The optimization sub-model training sub-unit is used to train the optimization sub-model according to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set, until a third training number exceeds a third preset number, thereby obtaining a trained optimization sub-model;

[0164] A second test set prediction sub-element is used to input the second test set into the trained basic sub-model and the trained optimization sub-model to obtain the predicted radiation dose of each data in the second test set;

[0165] a model performance calculation subunit, for calculating the values ​​of the coefficient of determination and the root mean square error according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set;

[0166] The first training times discrimination subunit is used to judge whether the first training times is less than the first preset times. If so, continue with the next training; if not, compare the values ​​of the determination coefficient and the root mean square error obtained in each training, and use the dose assessment model corresponding to the maximum determination coefficient and the minimum root mean square error as the trained dose assessment model.

[0167] In a preferred embodiment, the basic sub-model training sub-unit includes: a basic sub-model iterative training component, a first loss function calculation component, and a second training times discrimination component;

[0168] A basic sub-model iterative training component is used to train the basic sub-model according to each data in the second training set; wherein, during each training, for each regression sub-model, the expression information of the second target gene of each data in the second training set is used as input, and the first predicted radiation dose of each data in the second training set is used as output for training;

[0169] A first loss function calculation component is used to calculate, for each regression sub-model, a root mean square error of each regression sub-model according to a first predicted radiation dose of each data in each second training set and a corresponding actual radiation dose, and perform weighted calculation on the root mean square error of each regression sub-model to obtain a first loss function of the basic sub-model;

[0170] The second training times discrimination component is used to determine whether the second training times is less than the second preset times. If so, repeat the next training; if not, compare the first loss function values ​​obtained from each training, and use the basic sub-model corresponding to the minimum first loss function value as the basic sub-model that has completed the training.

[0171] In a preferred embodiment, the optimization sub-model training sub-unit includes: a prediction data grouping component, an optimization sub-model iterative training component, a second loss function calculation component, and a third training times discrimination component;

[0172] The prediction data grouping component is used to divide each first predicted radiation dose obtained by each training of the basic sub-model into a plurality of groups of prediction data; wherein the prediction data includes: each first predicted radiation dose obtained by the same training of the basic sub-model;

[0173] an optimization sub-model iterative training component, for training the optimization sub-model according to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set; wherein, during each training, a group of prediction data is selected in sequence from a plurality of groups of prediction data, and for each first predicted radiation dose in the group of prediction data and the data corresponding to the first predicted radiation dose in the second training set, the first predicted radiation dose and the data corresponding to the first predicted radiation dose in the second training set are used as input, and the second predicted radiation dose of the second training set data is used as output for training;

[0174] A second loss function calculation component is used to calculate a second loss function according to a second predicted radiation dose of each second training set data and a corresponding actual radiation dose;

[0175] The third training times discrimination component is used to determine whether the third training times is less than the third preset times. If so, continue with the next training; if not, compare the second loss function values ​​obtained from each training, and use the optimized sub-model corresponding to the minimum second loss function value as the optimized sub-model that has completed the training.

[0176] In a preferred embodiment, the model performance calculation subunit includes: a determination coefficient calculation component and a root mean square error calculation component;

[0177] The determination coefficient calculation component is used to calculate the determination coefficient according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set by the following formula:

[0178]

[0179] The root mean square error calculation component is used to calculate the root mean square error according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set by the following formula:

[0180]

[0181] Among them, R 2 represents the coefficient of determination, RMSE represents the root mean square error, and Y i represents the actual radiation dose of the i-th data in the second test set, represents the predicted radiation dose of the i-th data in the second test set, represents the average value of the actual radiation dose in the second test set, and n represents the total amount of data in the test set.

[0182] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention, and can implement the biological radiation dose assessment method provided by any one of the above-mentioned method embodiments of the present invention.

[0183] It should be noted that the device embodiments described above are merely illustrative, and some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. Those of ordinary skill in the art may understand and implement the present invention without creative work.

[0184] Based on the above-mentioned embodiment of the biological radiation dose assessment method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the biological radiation dose assessment method of any embodiment of the present invention is implemented.

[0185] Exemplarily, in this embodiment, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more module elements may be a series of computer program instruction segments capable of implementing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0186] The terminal device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0187] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.

[0188] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A biological radiation dose assessment method, characterized in that: include: Extracting expression information of a first target gene in the biological sample to be evaluated; Inputting the expression information of the first target gene into a preset dose assessment model, so that the dose assessment model determines the radiation dose of the biological sample to be assessed according to the expression information of the first target gene; Wherein, the construction of the dose assessment model includes: Obtaining several groups of training samples of training subjects that have been exposed to known radiation doses and training samples that have not been exposed to radiation; Performing transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples; Comparing and analyzing the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differentially expressed genes; Perform feature correlation screening on the differentially expressed genes to obtain the second target gene; Performing quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene; Taking the actual radiation dose of the irradiated training sample and the expression information of the corresponding second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model.

2. The biological radiation dose assessment method according to claim 1, characterized in that: The differentially expressed genes were screened for correlation of features to obtain the second target genes, including: For each differentially expressed gene, the log-fold change, significance level, and correlation coefficient of the differentially expressed gene were calculated; The differentially expressed genes that simultaneously meet the second preset condition, the third preset condition and the fourth preset condition are screened out from the differentially expressed genes as the second target genes for preliminary screening; the second preset condition is that the absolute value of the logarithmic fold change is greater than the first preset threshold; the third preset condition is that the significance level is less than the second preset threshold; the fourth preset condition is that the absolute value of the correlation coefficient is greater than or equal to the third preset threshold; The Boruta algorithm is used to perform feature screening on the second target gene that is initially screened to obtain the final second target gene.

3. The biological radiation dose assessment method according to claim 2, characterized in that: Taking the actual radiation dose of the irradiated training sample and the corresponding expression information of the second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model, including: Dividing the irradiated training samples into a first training set and a first test set according to a preset ratio; The actual radiation dose of each training sample in the first training set and the expression information of the corresponding second target gene are input into the dose assessment model to be trained for training until the number of training times exceeds a preset number, thereby obtaining a trained dose assessment model.

4. The biological radiation dose assessment method according to claim 3, characterized in that: The dose assessment model includes: a basic sub-model and an optimization sub-model; The actual radiation dose of each training sample in the first training set and the corresponding expression information of the second target gene are input into the dose assessment model to be trained for training until the first training times exceed the first preset times, thereby obtaining a trained dose assessment model, including: Dividing the first training set into a plurality of training subsets; In each training, one training subset is selected from all training subsets in order as the second test set, and the remaining training subsets are used as the second training set; wherein the training subsets can be selected cyclically; The basic sub-model is trained according to each data in the second training set until the second training times exceed the second preset times, thereby obtaining a trained basic sub-model and each first predicted radiation dose obtained from each training of the basic sub-model; According to each first predicted radiation dose obtained by each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set, the optimization sub-model is trained until a third training number exceeds a third preset number, thereby obtaining a trained optimization sub-model; Inputting the second test set into the trained basic sub-model and the trained optimization sub-model to obtain the predicted radiation dose of each data in the second test set; Calculate the value of the coefficient of determination and the root mean square error according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set; Determine whether the first training times is less than the first preset times. If so, continue with the next training. If not, compare the values ​​of the coefficient of determination and the root mean square error obtained in each training, and use the dose assessment model corresponding to the maximum coefficient of determination and the minimum root mean square error as the trained dose assessment model.

5. The biological radiation dose assessment method according to claim 4, characterized in that: The basic sub-model includes a combination of the following multiple regression sub-models: a random forest regression sub-model, a linear regression sub-model, an XGBoost regression sub-model, a CatBoost regression sub-model, an elastic network regression sub-model and a support vector regression sub-model.

6. The biological radiation dose assessment method according to claim 5, characterized in that: The basic sub-model is trained according to each data in the second training set until the second training times exceed the second preset times, and the trained basic sub-model and each first predicted radiation dose obtained by each training of the basic sub-model are obtained, including: The basic sub-model is trained according to each data in the second training set; wherein, in each training, for each regression sub-model, the expression information of the second target gene of each data in the second training set is used as input, and the first predicted radiation dose of each data in the second training set is used as output for training; For each regression sub-model, according to the first predicted radiation dose of each data in each second training set and the corresponding actual radiation dose, the root mean square error of each regression sub-model is calculated, and the root mean square error of each regression sub-model is weighted to obtain the first loss function of the basic sub-model; Determine whether the second training times is less than the second preset times. If so, repeat the next training; if not, compare the first loss function values ​​obtained from each training, and take the basic sub-model corresponding to the minimum first loss function value as the basic sub-model that has completed the training.

7. The biological radiation dose assessment method according to claim 6, characterized in that: According to each first predicted radiation dose obtained from each training of the basic sub-model and data corresponding to each first predicted radiation dose in the second training set, the optimization sub-model is trained until a third training number exceeds a third preset number, thereby obtaining a trained optimization sub-model, including: Divide each first predicted radiation dose obtained by each training of the basic sub-model into a plurality of groups of predicted data; wherein the predicted data include: each first predicted radiation dose obtained by the same training of the basic sub-model; The optimization sub-model is trained according to each first predicted radiation dose obtained from each training of the basic sub-model and the data corresponding to each first predicted radiation dose in the second training set; wherein, during each training, a group of prediction data is selected in sequence from a plurality of groups of prediction data, and for each first predicted radiation dose in the group of prediction data and the data corresponding to the first predicted radiation dose in the second training set, the first predicted radiation dose and the data corresponding to the first predicted radiation dose in the second training set are used as input, and the second predicted radiation dose of the second training set data is used as output for training; A second loss function is calculated based on the second predicted radiation dose of each second training set data and the corresponding actual radiation dose; Determine whether the third training times is less than the third preset times. If so, continue with the next training; if not, compare the second loss function values ​​obtained from each training, and take the optimized sub-model corresponding to the minimum second loss function value as the optimized sub-model that has completed the training.

8. The biological radiation dose assessment method according to claim 7, characterized in that: The values ​​of the coefficient of determination and the root mean square error are calculated according to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, including: According to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the determination coefficient is calculated by the following formula: According to the predicted radiation dose of each data in the second test set and the actual radiation dose corresponding to each data in the second test set, the root mean square error is calculated by the following formula: Among them, R 2 represents the coefficient of determination, RMSE represents the root mean square error, and Y i represents the actual radiation dose of the i-th data in the second test set, represents the predicted radiation dose of the i-th data in the second test set, represents the average value of the actual radiation dose in the second test set, and n represents the total amount of data in the test set.

9. A biological radiation dose assessment device, characterized in that: include: A first target gene expression information extraction module, and a dosage assessment module; The first target gene expression information extraction module is used to extract the expression information of the first target gene in the biological sample to be evaluated; The dose assessment module is used to input the expression information of the first target gene into a preset dose assessment model, so that the dose assessment model determines the radiation dose of the biological sample to be assessed according to the expression information of the first target gene; Wherein, the construction of the dose assessment model includes: Obtaining several groups of training samples of training subjects that have been exposed to known radiation doses and training samples that have not been exposed to radiation; Performing transcriptome sequencing on each group of training samples to obtain transcriptome sequencing information of the irradiated training samples and transcriptome sequencing information of the corresponding non-irradiated training samples; Comparing and analyzing the transcriptome sequencing information of the irradiated training samples with the transcriptome sequencing information of the corresponding non-irradiated training samples to obtain differentially expressed genes; Perform feature correlation screening on the differentially expressed genes to obtain the second target gene; Performing quantitative gene amplification fluorescence detection on the second target gene to obtain expression information of the second target gene; Taking the actual radiation dose of the irradiated training sample and the expression information of the corresponding second target gene as input and the predicted radiation dose as output, the dose assessment model to be trained is trained to obtain a trained dose assessment model.

10. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the biological radiation dose assessment method according to any one of claims 1 to 8 is implemented.

Images