Method for predicting the proportion of different cycle cells based on chromatin conformational profiling

Abstract

The application provides a method for predicting proportions of different cell cycles based on chromatin conformation spectrum, which can be used in the field of biological detection technology. The method comprises the following steps: obtaining to-be-detected Bulk Hi-C data; inputting the to-be-detected Bulk Hi-C data into a cell cycle proportion prediction model to obtain a cell cycle proportion prediction result output by the cell cycle proportion prediction model; wherein the cell cycle proportion prediction model is obtained based on training of a deep neural network model. The method of the application can quickly and accurately estimate the composition proportions of cells in different cell cycles in a Bulk Hi-C data sample based on the Bulk Hi-C data through the cell cycle proportion prediction model, which helps to ensure the reliability of the detection result of the Bulk Hi-C test.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biological detection, and in particular to a method for predicting proportions of different cycle cells based on chromatin conformation spectrum. BACKGROUND

[0002] Hi-C technology is a genomics technology combining high-throughput sequencing technology and chromatin conformation capture, which is used to study the spatial relationship of whole chromatin DNA in the whole genome range and obtain high-resolution three-dimensional structure information of chromatin. Bulk Hi-C is a Hi-C technology based on a million cells to complete the experiment at the tissue sample level, so this technology can capture the average three-dimensional conformation of cells in different cycles.

[0003] In the study of Bulk Hi-C, the cells are treated with cell cycle synchronization, so that the cells are all in the G0 / G1 phase, avoiding the interference of cells in different cycles, especially G2 / M phase cells, on the final results. This is because G2 / M phase cells have highly condensed chromatin, which is quite different from the loose chromatin state of G0 / G1 phase cells. The transcriptional activity level of cells in this cycle is low. Allowing a large number of cells in the G2 / M phase to exist in the tissue will introduce noise into the data analysis results and may lead to incorrect conclusions.

[0004] Therefore, estimating the composition proportion of cells in different cell cycles in the sample from which the Bulk Hi-C data is derived is a key step for quality control of tissue Bulk Hi-C data, and is crucial to ensure the reliability of the results of Bulk Hi-C experiments. However, there is currently a lack of methods for estimating the composition proportion of cells in different cell cycles based on Bulk Hi-C data itself. SUMMARY

[0005] The present application provides a method for predicting proportions of different cycle cells based on chromatin conformation spectrum, to solve the technical problem that there is a lack of methods for estimating the composition proportion of cells in different cell cycles in the sample from which the Bulk Hi-C data is derived in the prior art Bulk Hi-C technology, affecting the reliability of the detection results.

[0006] According to a first aspect of the present application, the present application provides a method for predicting proportions of different cycle cells based on chromatin conformation spectrum, comprising:

[0007] obtaining Bulk Hi-C data to be detected;

[0008] input the to-be-tested Bulk Hi-C data into a cell cycle proportion prediction model to obtain a cell cycle proportion prediction result output by the cell cycle proportion prediction model; wherein the cell cycle proportion prediction model is obtained based on training of a deep neural network model.

[0009] In an implementable embodiment, the training of the deep neural network model comprises:

[0010] a plurality of training samples are obtained to construct a training set; wherein the training samples are labeled with true values, and the true values comprise real proportions of cells in different cell cycles in the training samples;

[0011] the training samples included in the training set are input into the deep neural network model to obtain predicted values; wherein the predicted values comprise predicted proportions of cells in different cell cycles in the training samples;

[0012] based on a preset loss function, a loss amount between the predicted values and the true values in the training set is obtained;

[0013] network parameters of the deep neural network model are optimized based on the loss amount until the loss function converges or a preset training iteration number is reached, to obtain a cell cycle proportion prediction model.

[0014] In an implementable embodiment, the deep neural network model comprises a first DNN model, a second DNN model and a third DNN model, and inputting the training samples included in the training set into the deep neural network model to obtain predicted values comprises:

[0015] the training samples are input into the first DNN model to obtain first predicted proportions;

[0016] the training samples are input into the second DNN model to obtain second predicted proportions;

[0017] the training samples are input into the third DNN model to obtain third predicted proportions;

[0018] the predicted proportions of cells in different cell cycles in the first predicted proportions, the second predicted proportions and the third predicted proportions are averaged to obtain the predicted values.

[0019] In an implementable embodiment, the first DNN model comprises a first input layer, a first one-round linear layer, a first two-round linear layer, a first three-round linear layer, a first four-round linear layer and a first five-round linear layer, and inputting the training samples into the first DNN model to obtain the first predicted proportions comprises:

[0020] extracting features in the training sample through the first input layer to obtain first input extracted features with a dimension of 1000;

[0021] reducing the feature dimension of the first input extracted features to 256 through the first one-round linear layer, and then activating through a ReLu function to obtain first one-round extracted features;

[0022] reducing the feature dimension of the first one-round extracted features to 128 through the first two-round linear layer, and then activating through a ReLu function to obtain first two-round extracted features;

[0023] reducing the feature dimension of the first two-round extracted features to 64 through the first three-round linear layer, and then activating through a ReLu function to obtain first three-round extracted features;

[0024] reducing the feature dimension of the first three-round extracted features to 32 through the first four-round linear layer, and then activating through a ReLu function to obtain first four-round extracted features;

[0025] reducing the feature dimension of the first four-round extracted features to 4 through the first five-round linear layer, and then normalizing the features through a Softmax function to obtain a first prediction ratio.

[0026] In a feasible implementation, the second DNN model comprises a second input layer, a second one-round linear layer, a second two-round linear layer, a second three-round linear layer, a second four-round linear layer, and a second five-round linear layer, the training sample is input into the second DNN model to obtain a second prediction ratio, comprising:

[0027] extracting features in the training sample through the second input layer to obtain second input extracted features with a dimension of 1000;

[0028] reducing the feature dimension of the second input extracted features to 512 through the second one-round linear layer, and then activating through a ReLu function to obtain second one-round extracted features;

[0029] reducing the feature dimension of the second one-round extracted features to 256 through the second two-round linear layer, and then activating through a ReLu function to obtain second two-round extracted features by randomly discarding 30% of parameters through Dropout;

[0030] reducing the feature dimension of the second two-round extracted features to 128 through the second three-round linear layer, and then activating through a ReLu function to obtain second three-round extracted features by randomly discarding 20% of parameters through Dropout;

[0031] The second four-wheel linear layer reduces the feature dimension of the second three-wheel extracted feature to 64, and then the ReLu function is activated, and the Dropout is applied to randomly discard 10% of the parameters, to obtain a second four-wheel extracted feature;

[0032] The second five-wheel linear layer reduces the feature dimension of the second four-wheel extracted feature to 4, and then the Softmax function is applied to normalize the feature, to obtain a second predicted proportion.

[0033] In a feasible implementation, the third DNN model includes a third input layer, a third one-wheel linear layer, a third two-wheel linear layer, a third three-wheel linear layer, a third four-wheel linear layer, and a third five-wheel linear layer, the training sample is input into the third DNN model, and a third predicted proportion is obtained, including:

[0034] The third input layer extracts features in the training sample, to obtain a third input extracted feature with a dimension of 1000;

[0035] The third one-wheel linear layer increases the feature dimension of the third input extracted feature to 1024, and then the ReLu function is activated, to obtain a third one-wheel extracted feature;

[0036] The third two-wheel linear layer reduces the feature dimension of the third one-wheel extracted feature to 512, and then the ReLu function is activated, and the Dropout is applied to randomly discard 60% of the parameters, to obtain a third two-wheel extracted feature;

[0037] The third three-wheel linear layer reduces the feature dimension of the third two-wheel extracted feature to 256, and then the ReLu function is activated, and the Dropout is applied to randomly discard 30% of the parameters, to obtain a third three-wheel extracted feature;

[0038] The third four-wheel linear layer reduces the feature dimension of the third three-wheel extracted feature to 128, and then the ReLu function is activated, and the Dropout is applied to randomly discard 10% of the parameters, to obtain a third four-wheel extracted feature;

[0039] The third five-wheel linear layer reduces the feature dimension of the third four-wheel extracted feature to 4, and then the Softmax function is applied to normalize the feature, to obtain a third predicted proportion.

[0040] In a feasible implementation, the method includes:

[0041] Obtaining a plurality of single-cell Hi-C data; wherein the single-cell Hi-C data is an interaction matrix;

[0042] The single-cell Hi-C data is pre-processed to obtain pre-processed Hi-C data; wherein the pre-processing comprises depth filtering and depth normalization.

[0043] A plurality of pre-processed Hi-C data is randomly selected for fusion to obtain pseudo-Bulk Hi-C data as a training sample.

[0044] According to a second aspect of the present application, the present application provides a device for predicting the proportion of different cycle cells based on chromatin conformational spectrum, comprising:

[0045] A data acquisition module is configured to acquire to-be-tested Bulk Hi-C data.

[0046] A proportion prediction module is configured to input the to-be-tested Bulk Hi-C data into a cell cycle proportion prediction model to obtain a cell cycle proportion prediction result output by the cell cycle proportion prediction model; wherein the cell cycle proportion prediction model is obtained based on training of a deep neural network model.

[0047] According to a third aspect of the present application, the present application provides an electronic device, comprising a processor and a memory connected with the processor in communication;

[0048] The memory stores computer execution instructions.

[0049] The processor executes the computer execution instructions stored in the memory to implement the method of any one of the first aspect.

[0050] According to a fourth aspect of the present application, the present application provides a computer readable storage medium, wherein the computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed to implement the method of any one of the first aspect.

[0051] According to a fifth aspect of the present application, the present application provides a computer program product, comprising a computer program, wherein the computer program is executed to implement the method of any one of the first aspect.

[0052] Compared with the prior art, the present application has the following beneficial effects:

[0053] The application provides a method for predicting the proportion of different cycle cells based on chromatin conformation spectrum, which trains a deep neural network model, and then inputs the Bulk Hi-C data into the trained cell cycle proportion prediction model to obtain the cell cycle proportion prediction result of the Bulk Hi-C data. The cell cycle proportion prediction model can quickly and accurately estimate the proportion of cells in different cell cycles in the sample data based on the Bulk Hi-C data, which provides an effective means for quality control of the tissue Bulk Hi-C data, helps to ensure the reliability of the Bulk Hi-C test detection result, avoids the average three-dimensional conformation interference caused by the mixing of different cycle cells and the noise of the data analysis result, and further prevents the drawing of wrong conclusions, thereby providing key technical support for ensuring the reliability of the Bulk Hi-C test result. BRIEF DESCRIPTION OF DRAWINGS

[0054] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and serve to explain the principles of the application.

[0055] Figure 1 A flowchart of a method for predicting the proportion of different cycle cells based on chromatin conformation spectrum provided by the embodiments of the application;

[0056] Figure 2 A flowchart of training a deep neural network model provided by the embodiments of the application;

[0057] Figure 3 A flowchart of obtaining a prediction value by a deep neural network model provided by the embodiments of the application;

[0058] Figure 4 A structural diagram of a deep neural network model provided by the embodiments of the application;

[0059] Figure 5 An interaction matrix diagram of single-cell Hi-C data conversion provided by the embodiments of the application;

[0060] Figure 6 A comparative analysis diagram of predicted values and true values of pseudo Bulk Hi-C data provided by the embodiments of the application;

[0061] Figure 7 A data analysis diagram of predicted values and true values of cross-tissue type Bulk Hi-C data provided by the embodiments of the application;

[0062] Figure 8 A structural diagram of a device for predicting the proportion of different cycle cells based on chromatin conformation spectrum provided by the embodiments of the application;

[0063] Figure 9 Figure 1 shows a schematic diagram of an electronic device according to an embodiment of the present application.

[0064] The above-described embodiments of the present application have been shown by way of illustration, and thus should not be construed to limit the scope of the present application. The following detailed description will be made with reference to the accompanying drawings, and the detailed description is not intended to limit the scope of the present application by any means. DETAILED DESCRIPTION

[0065] Reference will now be made to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. Unless otherwise indicated, the same numbers on different drawings represent the same or similar elements. The following detailed description does not limit the application thereto, as the claimed application can be practiced in other ways. Rather, it is intended to cover all modifications, variations, and equivalents that fall within the scope of the application, which is limited only by the appended claims.

[0066] In the nucleus of mammalian cells, chromatin is organized in micron-sized territories, each occupied by a specific chromatin. Different nuclear territories occupied by chromatin are highly correlated with gene density and transcriptional activity. Dynamic changes in chromatin structure are associated with many biological processes, especially with the cell cycle. Chromatin in animal cells undergoes dramatic structural changes during each different cell cycle, typically alternating between a highly condensed state that facilitates chromosome segregation and interphase structures that accommodate transcription, gene silencing, and DNA replication.

[0067] Hi-C technology is a combination of high-throughput sequencing technology and chromatin conformation capture, which is used to study the spatial relationship of whole chromatin DNA in the whole genome range, and obtain high-resolution three-dimensional structure information of chromatin. Hi-C technology can analyze the high-level structural characteristics of chromatin, such as A / B compartment, topological associated domain (TAD) and chromatin loop, and provides an important tool for understanding genome function, gene expression regulation and disease mechanism.

[0068] Bulk Hi-C is a common form of application of Hi-C technology, which refers to obtaining population-level chromatin three-dimensional structure information by performing Hi-C experiments on a large number of cell populations. It is a Hi-C technology based on millions of cells to complete the experiment at the tissue sample level, so this technology may capture the average three-dimensional conformation of mixed cells in different cycles. In some Bulk Hi-C studies, scientists will perform cell cycle synchronization treatment on cells, so that cells are in G0 / G1 phase, avoiding the interference of cells in different cycles, especially G2 / M phase cells, with the final results.

[0069] This is because the cells in G2 / M phase have highly condensed chromatin, which is quite different from the loose chromatin state of cells in G0 / G1 phase. Cells in this cycle have low transcriptional activity, and scientists studying transcriptional regulatory mechanisms are generally not interested in this state, so they try to keep cells in G0 / G1 phase. Allowing a large number of cells in G2 / M phase in the tissue will introduce noise into the data analysis results and may lead researchers to draw incorrect conclusions.

[0070] For some normal adult tissue samples, the vast majority of cells are naturally in G0 / G1 phase, so the simultaneous conversion can be skipped and the Bulk Hi-C experiment can be directly performed. However, not all samples have similar characteristics. Therefore, estimating the proportion of cells in different cell cycles in the sample from which the tissue Bulk Hi-C data is derived is a key step in the quality control of tissue Bulk Hi-C data and is crucial to ensure the reliability of the Bulk Hi-C test results. However, there is currently a lack of methods for estimating the proportion of cells in different cell cycles based on Bulk Hi-C data itself.

[0071] In view of the above technical problems, the present application provides a method for predicting the proportion of cells in different cycles based on chromatin conformation spectrum. The cell cycle proportion prediction model can quickly and accurately estimate the proportion of cells in different cell cycles in the Bulk Hi-C data sample based on the Bulk Hi-C data, which helps to ensure the reliability of the Bulk Hi-C test detection results.

[0072] The technical solutions of the method for predicting the proportion of cells in different cycles based on chromatin conformation spectrum provided by the present application will be described in detail below through specific embodiments. It should be noted that the following embodiments can exist independently or in combination. For the same or similar content, it may not be repeated in different embodiments.

[0073] Figure 1 A flowchart of a method for predicting the proportion of cells in different cycles based on chromatin conformation spectrum provided by the present application is shown in Figure 1 In some embodiments, the method for predicting the proportion of cells in different cycles based on chromatin conformation spectrum includes the following steps:

[0074] S101, obtaining the Bulk Hi-C data to be tested.

[0075] The obtained Bulk Hi-C data to be tested is used for subsequent cell cycle proportion prediction.

[0076] S102, input the to-be-tested Bulk Hi-C data into the cell cycle proportion prediction model to obtain a cell cycle proportion prediction result output by the cell cycle proportion prediction model; wherein the cell cycle proportion prediction model is obtained based on training of a deep neural network model.

[0077] The Bulk Hi-C data is input into the cell cycle proportion prediction model, and the cell cycle proportion prediction model trained based on the deep neural network model can output the cell cycle proportion prediction result.

[0078] Specifically, the cell cycle proportion prediction result includes proportions of cells in G0, G1, G2 and M cycles. The G0 cycle represents cells that temporarily leave the cell cycle, stop cell division, but can re-enter the cell cycle under certain stimulation; the G1 cycle represents a period from the completion of mitosis to before DNA replication, which is the main stage of cell growth; the G2 cycle represents a period from the completion of DNA replication to before the start of mitosis; and the M cycle represents a period in which cells undergo mitosis, and through mitosis, the replicated genetic material is evenly distributed to two daughter cells.

[0079] In this embodiment, by training the deep neural network model, subsequent input of the Bulk Hi-C data into the trained cell cycle proportion prediction model can obtain the cell cycle proportion prediction result of the Bulk Hi-C data. The cell cycle proportion prediction model can quickly and accurately estimate the proportion of cells in different cell cycles in the sample data based on the Bulk Hi-C data, which provides an effective means for quality control of the organization Bulk Hi-C data, helps to ensure the reliability of the Bulk Hi-C test detection result, avoids the average three-dimensional conformational interference caused by the mixing of different cycle cells and the noise of the data analysis result, and further prevents the drawing of incorrect conclusions, thereby providing key technical support for ensuring the reliability of the Bulk Hi-C test result.

[0080] In Figure 1 The method for predicting the proportion of different cycle cells based on the chromatin conformation spectrum shown in the figure needs to train the deep neural network model. The following will be combined with Figure 2 The technical scheme of the method for predicting the proportion of different cycle cells based on the chromatin conformation spectrum is further introduced.

[0081] Figure 2 A flowchart for training the deep neural network model is provided for the embodiments of the present application, which is shown in Figure 2 In some embodiments, the flow of training the deep neural network model includes the following steps:

[0082] S201, obtain a plurality of training samples to construct a training set; wherein the training samples are labeled with true values, and the true values include real proportions of cells in different cell cycles in the training samples.

[0083] Optionally, the plurality of training samples are obtained to construct the training set, specifically including:

[0084] Step 1, obtain a plurality of single-cell Hi-C data; wherein the single-cell Hi-C data is an interaction matrix.

[0085] Wherein, after the Hi-C raw data (generally paired 150bp long sequences) is subjected to sequence alignment and filtering with the genome, it is represented as a pair of reads (Read Pair) aligned to two different one-dimensional regions of the genome; in order to better display and analyze the results of these paired reads, the Hi-C data is usually converted into an interaction matrix; refer to Figure 5 Fig. 1 shows an interaction matrix of the first 10kb region of chromosome 1 with a precision of 1kb;

[0086] Wherein, 10 represents that the number of paired reads connecting the 1-2kb region and the 6-7kb region of chromosome 1 is 10, i.e. the non-normalized interaction intensity is 10; 8 represents that the number of paired reads connecting the 3-4kb region and the 8-9kb region of chromosome 1 is 8, i.e. the non-normalized interaction intensity is 8; because the interaction intensity is disturbed by the genomic sequence bias such as GC content (the proportion of guanine G and cytosine C bases in the genomic sequence), Mappability (the ability of sequencing reads to uniquely align to the genomic reference sequence), etc., therefore, the above interference factors are usually corrected by the method of ICE (iterative correction and eigenvector decomposition), which is called the intra-sample normalization process of the original interaction matrix.

[0087] Step 2, pre-process the single-cell Hi-C data to obtain pre-processed Hi-C data; wherein the pre-processing includes depth filtering and depth normalization.

[0088] Wherein, the single-cell Hi-C data is pre-processed, and the matrices with depth too deep greater than 500 k and the matrices with depth too shallow less than 100 k are filtered out; the depth refers to the total number of paired reads aligned to the genome; in general, the sample with high total depth will show an overall increase in interaction matrix interaction strength than the sample with low total depth, and this effect is considered as noise caused by sequencing process bias and needs to be corrected; since samples with comparable depth are needed to construct training and test sets, and the depth of these samples needs to be normalized so that the final interaction matrix data is not affected by the depth factor, data that is too deep or shallow will introduce a large bias that cannot be corrected by conventional depth normalization, so it needs to be filtered out and the data depth is normalized to the shallowest data level; there are two strategies for normalization, downsampling and upsampling, and here the more robust downsampling method is used to randomly sample the effective paired reads of single-cell Hi-C data, so that the effective paired read number, i.e. data depth, is reduced from greater than 100000 to 100000.

[0089] Step 3, randomly select multiple pre-processed Hi-C data for fusion to obtain pseudo Bulk Hi-C data as training samples.

[0090] Wherein, 500 kb matrices of multiple pre-processed Hi-C data are randomly selected and fused together, i.e. the values of each corresponding cell of the matrices are summed to form a new interaction matrix as the matrix of pseudo Bulk Hi-C data, and intrasample ICE normalization processing is performed; since each single-cell Hi-C data sample has been normalized to the same depth, the depth of pseudo Bulk Hi-C data should be basically consistent in theory, and no depth normalization processing is needed, so the purpose of intrasample ICE normalization processing in this step is to remove the interference of genomic sequence bias such as GC content and Mappability. Randomly mixing single-cell Hi-C data to generate pseudo Bulk Hi-C data as training samples.

[0091] Since the cell cycle of single-cell Hi-C data is known, and each pseudo Bulk Hi-C data comes from the mixing of multiple single-cell Hi-C data with known cell cycles, the proportion of cells in different cell cycles in the pseudo Bulk Hi-C data formed by mixing single-cell Hi-C data is also known.

[0092] Optionally, this step is repeated multiple times, and finally multiple pseudo Bulk Hi-C data mixed with different cycle cells can be generated, and a training set is constructed therefrom.

[0093] S202, inputting the training samples contained in the training set into the deep neural network model to obtain a predicted value; wherein the predicted value includes the predicted proportion of cells in different cell cycles in the training samples.

[0094] Among them, after the training set is input into the deep neural network model, the deep neural network model will output a predicted value, which includes the predicted proportion of cells in different cell cycles in the training sample.

[0095] S203: Based on a preset loss function, obtain the loss between the predicted value and the true value in the training set.

[0096] Among them, the loss function can be used to obtain the loss between the predicted value and the true value, and then the loss is used to update the network parameters of the model, thereby gradually improving the model's cell composition ratio prediction performance.

[0097] Optionally, the loss function satisfies the following formula:

[0098]

[0099] Among them, L1 represents the loss function, Indicates the predicted values, Indicates the A true value.

[0100] The absolute error loss, L1, is simple and intuitive. It simply calculates the absolute value of the difference between the predicted and true values ​​and then averages them. This makes it easy to understand and implement, has low computational complexity, is insensitive to outliers, and more robustly reflects the model's predictive performance on mostly normal data, ensuring the stability of the model's overall predictions. Furthermore, its loss value aligns with the actual dimension of the error, making it easy to directly interpret the error magnitude. This provides an intuitive reference for evaluating and improving model performance, helping users quickly understand the model's predictions and make appropriate adjustments.

[0101] S204, optimizing the network parameters of the deep neural network model based on the loss amount until the loss function converges or reaches a preset number of training iterations, thereby obtaining a cell cycle ratio prediction model.

[0102] Among them, the deep neural network model is iteratively trained based on the collected training samples. The network parameters of the deep neural network model are continuously optimized according to the loss amount during the training process until the loss function converges or the preset number of training iterations is reached. The training of the model is terminated. At this time, a cell cycle ratio prediction model for predicting the proportion of cells in different cycles is obtained.

[0103] Specifically, the number of iterations refers to the number of times the model fully traverses the training dataset and updates the model parameters during training. A higher number of iterations can improve model accuracy because the model has more opportunities to learn features from the data. However, too many iterations can also lead to overfitting, meaning the model performs well on the training set but poorly on unseen data. Too few iterations can lead to underfitting, meaning the model does not fully learn the features of the data. Therefore, the number of training iterations should be set based on a combination of factors, such as the size of the training set and the complexity of the model.

[0104] Preferably, the training samples can be divided into a training set, a validation set and a test set.

[0105] Among them, the validation set is used to adjust the network parameters of the model and evaluate the performance of the model during training to avoid overfitting; the test set is used to finally evaluate the generalization ability of the model to ensure that the model can perform well on new data; these three together constitute the complete process of model training, evaluation and optimization.

[0106] In this embodiment, the cell cycle ratio prediction model obtained by training the deep neural network model through the above-mentioned training method can effectively detect the cell cycle ratio, and has the advantages of high detection efficiency and high accuracy.

[0107] exist Figure 2 In the method for training a deep neural network model shown in FIG, it is necessary to train the deep neural network model. Figure 3 , in the technical solution of the above-mentioned method for training a deep neural network model, the content of obtaining the prediction value through the deep neural network model is further introduced.

[0108] Figure 3 A schematic diagram of a process for obtaining a prediction value through a deep neural network model provided in an embodiment of the present application, see Figure 3 In some embodiments, combined with Figure 4 , the deep neural network model includes a first DNN model, a second DNN model and a third DNN model, and the process of obtaining a prediction value through the deep neural network model includes the following steps:

[0109] S301: Input the training sample into the first DNN model to obtain a first prediction ratio.

[0110] Optionally, the first DNN model includes a first input layer, a first first-round linear layer, a first second-round linear layer, a first third-round linear layer, a first fourth-round linear layer, and a first fifth-round linear layer. Inputting the training sample into the first DNN model to obtain a first prediction ratio includes:

[0111] Step 1, extract features in the training sample through the first input layer, obtain the first input extraction features with a dimension of 1000.

[0112] Among them, the input layer receives the training sample and represents the training sample in the form of a 1000-dimensional feature vector as the initial input of the model, providing a basis for subsequent feature extraction and processing.

[0113] Step 2, reduce the feature dimension of the first input extraction features to 256 through the first linear layer, and then activate the ReLu function to obtain the first round of extraction features.

[0114] Among them, the first linear layer weights and sums the input 1000-dimensional first extraction features to reduce the feature dimension to 256, and the ReLu function nonlinearly transforms the output of the linear layer to introduce nonlinear factors, enabling the model to learn more complex feature representations.

[0115] Specifically, the ReLu function sets negative values to 0 and keeps positive values unchanged, which helps to alleviate the gradient vanishing problem and accelerate model convergence.

[0116] Step 3, reduce the feature dimension of the first round of extraction features to 128 through the second linear layer, and then activate the ReLu function to obtain the second round of extraction features.

[0117] Among them, the second linear layer weights and sums the 256-dimensional features to reduce the feature dimension to 128. The ReLu function continues to nonlinearly process the features, gradually refining more representative features.

[0118] Step 4, reduce the feature dimension of the second round of extraction features to 64 through the third linear layer, and then activate the ReLu function to obtain the third round of extraction features.

[0119] Among them, the third linear layer weights and sums the 128-dimensional features to reduce the feature dimension to 64. The ReLu function continues to nonlinearly process the features, gradually refining more representative features.

[0120] Step 5, reduce the feature dimension of the third round of extraction features to 32 through the fourth linear layer, and then activate the ReLu function to obtain the fourth round of extraction features.

[0121] Among them, the fourth linear layer weights and sums the 64-dimensional features to reduce the feature dimension to 32. The ReLu function further compresses the feature dimension while enhancing the discriminability of the features through nonlinear transformation, enabling the model to better capture key information in the data.

[0122] Step 6, the feature dimension of the first four-wheel extracted features is reduced to 4 by the first five-wheel linear layer, and then the features are standardized by the Softmax function to obtain the first prediction ratio.

[0123] wherein the first five-wheel linear layer weights and sums the 32-dimensional features to reduce the feature dimension to 4. The Softmax function standardizes the output of the linear layer and converts the output into a probability distribution, so that the probability of each to-be-tested Bulk Hi-C data sample being in four different cell cycles is 1, thereby obtaining the ratio of the sample being in each cell cycle.

[0124] wherein the first DNN model starts from an input layer feature dimension of 1000, sequentially passes through linear layers to reduce the dimension to 256, 128, 64, and 32, and finally outputs a dimension of 4, thereby gradually extracting key features in the data, reducing the amount of calculation and memory consumption, preventing overfitting, and finally mapping the features to a 4-dimensional space suitable for classification to obtain the ratio of the sample being in four different cell cycles.

[0125] S302, input the training sample into the second DNN model to obtain the second prediction ratio.

[0126] Optionally, the second DNN model includes a second input layer, a second one-wheel linear layer, a second two-wheel linear layer, a second three-wheel linear layer, a second four-wheel linear layer, and a second five-wheel linear layer, and the training sample is input into the second DNN model to obtain the second prediction ratio, comprising:

[0127] Step 1, extract the features in the training sample through the second input layer to obtain the second input extracted features with a dimension of 1000.

[0128] wherein the training sample is received as the initial input of the model.

[0129] Step 2, reduce the feature dimension of the second input extracted features to 512 by the second one-wheel linear layer, and then activate it by the ReLu function to obtain the second one-wheel extracted features.

[0130] wherein the 1000-dimensional features input by the second one-wheel linear layer are weighted and summed to reduce the feature dimension to 512, and the ReLu function performs a nonlinear transformation on the output of the linear layer to introduce a nonlinear factor and enhance the expression ability of the model.

[0131] Step 3, reduce the feature dimension of the second one-wheel extracted features to 256 by the second two-wheel linear layer, and then activate it by the ReLu function and apply Dropout to randomly discard 30% of the parameters to obtain the second two-wheel extracted features.

[0132] The second two-wheel linear layer reduces the feature dimension to 256 by weighted summation of 512-dimensional features. The ReLu function further nonlinearly transforms the features to extract higher-level features. The Dropout randomly discards 30% of the parameters, which is to prevent model overfitting. By randomly discarding part of the neurons, the model will not rely too much on certain specific neurons during training, thereby improving the generalization ability of the model.

[0133] In step 4, the feature dimension of the second two-wheel extracted features is reduced to 128 by the second three-wheel linear layer, and then activated by the ReLu function. The Dropout randomly discards 20% of the parameters to obtain the second three-wheel extracted features.

[0134] The second three-wheel linear layer reduces the feature dimension to 128 by weighted summation of 256-dimensional features. The ReLu function continues to nonlinearly process the features to extract more representative features. The Dropout randomly discards 20% of the parameters to further prevent model overfitting and enhance the robustness of the model.

[0135] In step 5, the feature dimension of the second three-wheel extracted features is reduced to 64 by the second four-wheel linear layer, and then activated by the ReLu function. The Dropout randomly discards 10% of the parameters to obtain the second four-wheel extracted features.

[0136] The second four-wheel linear layer reduces the feature dimension to 64 by weighted summation of 128-dimensional features. The ReLu function further compresses the feature dimension to enhance the discriminability of the features. The Dropout randomly discards 10% of the parameters to appropriately reduce the proportion of Dropout when the model approaches the output layer, so as to retain more effective information while still playing a certain role in preventing overfitting.

[0137] In step 6, the feature dimension of the second four-wheel extracted features is reduced to 4 by the second five-wheel linear layer, and then normalized by the Softmax function to obtain the second prediction ratio.

[0138] The second five-wheel linear layer reduces the feature dimension to 4 by weighted summation of 64-dimensional features. The Softmax function normalizes the output of the linear layer to obtain the proportion of the sample in each cell cycle.

[0139] The input layer of the second DNN model has a feature dimension of 1000, which is first reduced to 512, then to 256, 128, and finally to 4. By gradually reducing the dimension to extract important features, and combining the Dropout technology to prevent overfitting, the model maintains good generalization ability while reducing the dimension. Finally, the prediction ratio of the sample in the four cell cycles is obtained.

[0140] S303, input the training sample into the third DNN model to obtain a third prediction ratio.

[0141] Optionally, the third DNN model comprises a third input layer, a third one-round linear layer, a third two-round linear layer, a third three-round linear layer, a third four-round linear layer, and a third five-round linear layer, inputting the training sample into the third DNN model to obtain the third prediction ratio, comprising:

[0142] Step 1, extracting features in the training sample through the third input layer to obtain third input extracted features with a dimension of 1000.

[0143] Among them, the training sample is received as the initial input of the model.

[0144] Step 2, raising the feature dimension of the third input extracted features to 1024 through the third one-round linear layer, and then activating through the ReLu function to obtain the third one-round extracted features.

[0145] Among them, the third one-round linear layer weights and sums the input 1000-dimensional features to raise the feature dimension to 1024. The non-linear transformation is performed on the output of the linear layer, and raising the feature dimension can increase the capacity of the model, so that the model can learn more complex feature representations. The ReLu activation function introduces a non-linear factor, which helps the model to capture the non-linear relationship in the data.

[0146] Step 3, reducing the feature dimension of the third one-round extracted features to 512 through the third two-round linear layer, and then activating through the ReLu function, and applying Dropout to randomly discard 60% of the parameters to obtain the third two-round extracted features.

[0147] Among them, the third two-round linear layer weights and sums the 1024-dimensional features to reduce the feature dimension to 512. The ReLu function further performs non-linear transformation on the features to extract higher-level features. Dropout randomly discards 60% of the parameters, and since the feature dimension is high, a larger Dropout ratio can more effectively prevent model overfitting and avoid the model being too complex to lose generalization ability.

[0148] Step 4, reducing the feature dimension of the third two-round extracted features to 256 through the third three-round linear layer, and then activating through the ReLu function, and applying Dropout to randomly discard 30% of the parameters to obtain the third three-round extracted features.

[0149] The third third linear layer weights and sums the 512-dimensional features to reduce the feature dimension to 256. The ReLu function continues to perform nonlinear processing on the features to extract more representative features. The Dropout randomly discards 30% of the parameters, and as the feature dimension decreases, the proportion of Dropout is appropriately reduced to retain more effective information while still preventing overfitting.

[0150] In step 5, the feature dimension of the third third extracted feature is reduced to 128 by the third fourth linear layer, and then activated by the ReLu function. The Dropout randomly discards 10% of the parameters to obtain the third fourth extracted feature.

[0151] The third fourth linear layer weights and sums the 256-dimensional features to reduce the feature dimension to 128. The ReLu function further compresses the feature dimension to enhance the discriminability of the features. The Dropout randomly discards 10% of the parameters, and as the model approaches the output layer, the proportion of Dropout is further reduced to ensure that the model can fully utilize the learned features for accurate classification.

[0152] In step 6, the feature dimension of the third fourth extracted feature is reduced to 4 by the third fifth linear layer, and then normalized by the Softmax function to obtain the third prediction proportion.

[0153] The third fifth linear layer weights and sums the 128-dimensional features to reduce the feature dimension to 4. The Softmax function normalizes the output of the linear layer to obtain the proportion of the sample in each cell cycle.

[0154] The input layer of the third DNN model has a feature dimension of 1000, which is first increased to 1024, then decreased to 512, 256, and 128, and finally output with a dimension of 4. The purpose of increasing the dimension of the third DNN model is to increase the capacity of the model, so that it can learn more complex feature representations. Subsequently, through step-by-step dimension reduction, noise and redundant information are removed, and the most important features are retained. Combined with the Dropout technique to prevent overfitting, the final prediction result of the sample in the four cell cycles is obtained.

[0155] In S304, the prediction proportions of cells in different cell cycles in the first prediction proportion, the second prediction proportion, and the third prediction proportion are averaged to obtain the prediction value.

[0156] By integrating the prediction results of multiple DNN models, the advantages of each DNN model can be combined, the errors and biases of individual models can be reduced, and the stability and generalization ability of the cell cycle proportion prediction model can be improved, thereby obtaining more accurate and reliable prediction results of the proportion of samples in four different cell cycles.

[0157] In this embodiment, the cell cycle proportion prediction model is composed of three deep neural network models with different structures and using Dropout regularization technology, each model has four layers of hidden layers with different sizes. The score is calculated using the Softmax function to distinguish the cell cycle proportion, and supervised training is performed on the simulation data.

[0158] Among them, the second DNN model and the third DNN model use Dropout with gradually reduced dropout ratio to adapt to the needs of different training stages. Because the higher dropout rate in the early stage of training can enhance the regularization effect, force the model to learn more robust feature combinations, and avoid premature convergence or overfitting; Especially in the scene of less data or more noise, strong regularization helps to suppress excessive sensitivity to noise. And gradually reducing the dropout rate in the later stage of training can reduce the artificial interference to neurons, allowing the model to fully utilize all neurons for parameter fine-tuning, improving the convergence stability and finality of the model.

[0159] The model uses ReLu as the activation function, which can speed up the training, alleviate the gradient vanishing problem, introduce non-linear characteristics, and enable sparse activation of neurons, improving the efficiency and accuracy of the model.

[0160] The model uses linear layers with gradually reduced size, which can gradually filter redundant information and extract high-order features by reducing the dimension layer by layer. The low-dimensional layer has fewer parameters, which can suppress the risk of overfitting. This structure is equivalent to implicit L2 regularization, which is complementary to explicit regularization methods. Reducing the number of parameters in each layer can directly reduce memory usage and computational complexity, and hierarchical dimension reduction can also alleviate the gradient explosion / vanishing problem.

[0161] By combining the first DNN model, the second DNN model and the third DNN model, the first DNN model uses simple stepwise dimension reduction to realize basic feature extraction and classification, the second DNN model introduces Dropout to enhance generalization ability and reduce overfitting, and the third DNN model increases model capacity by increasing dimension to capture complex features. The three DNN models process data from different angles, can more comprehensively mine data features, and adapt to different complexity of data. Finally, the results of the three DNN models are integrated for ensemble learning, which can reduce overfitting, improve the accuracy and stability of the model, and enhance the generalization ability of the model. The three models use hidden layers of different sizes and dropout ratios to capture and extract data features from multiple angles. The cell cycle proportion prediction model trained in this way can effectively reduce the error and bias of a single model, and the stability, accuracy and robustness of the prediction result, more accurately predicting the proportion of BulkHi-C samples in four different cell cycles.

[0162] Specifically, the training parameters include an Adam optimizer, a learning rate of 0.0001, and L1 loss as the optimization objective. Each model is independently trained for 5000 steps, and the "early stopping" method is used to avoid overfitting. Due to the large number of 2D bins in the two-dimensional matrix, only the 2D bins with the highest variance in the pseudo Bulk Hi-C matrix are selected for modeling. The interaction values of the selected 2D bins are pulled into a vector and input into the DNN model.

[0163] Assuming mouse embryonic stem cell single-cell Hi-C data, prepare the data set for training and testing. Using the deep neural network model (DNN) and ensemble learning (Ensemble Learning) method in deep learning, randomly mix single-cell Hi-C data to generate pseudo Bulk Hi-C data, and construct and test a model that can identify the characteristics of chromatin conformation changes in different cell cycles. A set of bulk Hi-C data (a total of 9 samples, including long-term hematopoietic stem cells, short-term hematopoietic stem cells, multipotent progenitor cells, common myeloid progenitor cells, granulocyte-macrophage progenitor cells, megakaryocyte-erythrocyte progenitor cells, common lymphoid progenitor cells, megakaryocyte lineage progenitor cells, and granulocytes) containing different cycle cells and the corresponding known cell cycle proportion are used to verify the model.

[0164] Referring to Figure 6 In the test set, the MSE (Mean Square Error) value between the predicted value and the true value of the model is 3.43x10-4, and the Pearson correlation coefficient is 0.99, indicating that the prediction effect is better when the modeled cell type (mouse ESC stem cells) and the tested cell type (mouse ESC stem cells) are consistent.

[0165] Referring to Figure 7 In the validation set, the Pearson correlation coefficient between the predicted value and the true value is 0.40, and the MSE value is 0.028, which has a certain accuracy, but the prediction effect is not as good as the test set result, indicating that the prediction effect is not good when the modeled cell type (mouse ESC stem cells) and the tested cell type (mouse adult blood cells) are inconsistent. However, with the gradual popularization of single-cell Hi-C technology, if single-cell Hi-C data of mouse adult blood cells can be obtained, a model with high prediction accuracy can be established. That is, when the modeled cell type and the tested cell type are consistent, the model accuracy will be very high.

[0166] Figure 8 is a structural schematic diagram of a device for predicting different cycle cell proportions based on chromatin conformation spectrum provided by the embodiments of the present application, referring to Figure 8The device for predicting proportions of different cycle cells based on the chromatin conformation spectrum comprises various functional modules for implementing the foregoing method for predicting proportions of different cycle cells based on the chromatin conformation spectrum, and any functional module can be implemented in the form of software and / or hardware.

[0167] In some embodiments, the device 800 for predicting proportions of different cycle cells based on the chromatin conformation spectrum comprises a data acquisition module 801 and a proportion prediction module 802. In this regard:

[0168] The data acquisition module 801 is configured to acquire Bulk Hi-C data to be tested.

[0169] The proportion prediction module 802 is configured to input the Bulk Hi-C data to be tested into a cell cycle proportion prediction model to obtain a cell cycle proportion prediction result output by the cell cycle proportion prediction model. The cell cycle proportion prediction model is obtained based on training of a deep neural network model.

[0170] In some embodiments, the device 800 further comprises a model training module 803, which is specifically configured to:

[0171] acquire a plurality of training samples to construct a training set. The training samples are labeled with true values, and the true values include real proportions of cells in different cell cycles in the training samples.

[0172] input the training samples included in the training set into a deep neural network model to obtain predicted values. The predicted values include predicted proportions of cells in different cell cycles in the training samples.

[0173] acquire a loss amount between the predicted values and the true values in the training set based on a preset loss function;

[0174] optimize network parameters of the deep neural network model based on the loss amount until the loss function converges or a preset training iteration number is reached, to obtain the cell cycle proportion prediction model.

[0175] In some embodiments, the deep neural network model comprises a first DNN model, a second DNN model, and a third DNN model. The model training module 803 is specifically further configured to:

[0176] input the training samples into the first DNN model to obtain a first predicted proportion;

[0177] input the training samples into the second DNN model to obtain a second predicted proportion;

[0178] input the training samples into the third DNN model to obtain a third predicted proportion;

[0179] The predicted proportions in different cell cycles are averaged to obtain a prediction value.

[0180] In some embodiments, the first DNN model comprises a first input layer, a first one-round linear layer, a first two-round linear layer, a first three-round linear layer, a first four-round linear layer, and a first five-round linear layer, and the model training module 803 is specifically further configured to:

[0181] The features in the training sample are extracted through the first input layer to obtain first input extracted features with a dimension of 1000;

[0182] The feature dimension of the first input extracted features is reduced to 256 through the first one-round linear layer, and then activated by a ReLu function to obtain first one-round extracted features;

[0183] The feature dimension of the first one-round extracted features is reduced to 128 through the first two-round linear layer, and then activated by a ReLu function to obtain first two-round extracted features;

[0184] The feature dimension of the first two-round extracted features is reduced to 64 through the first three-round linear layer, and then activated by a ReLu function to obtain first three-round extracted features;

[0185] The feature dimension of the first three-round extracted features is reduced to 32 through the first four-round linear layer, and then activated by a ReLu function to obtain first four-round extracted features;

[0186] The feature dimension of the first four-round extracted features is reduced to 4 through the first five-round linear layer, and then normalized by a Softmax function to obtain the first predicted proportion.

[0187] In some embodiments, the second DNN model comprises a second input layer, a second one-round linear layer, a second two-round linear layer, a second three-round linear layer, a second four-round linear layer, and a second five-round linear layer, and the model training module 803 is specifically further configured to:

[0188] The features in the training sample are extracted through the second input layer to obtain second input extracted features with a dimension of 1000;

[0189] The feature dimension of the second input extracted features is reduced to 512 through the second one-round linear layer, and then activated by a ReLu function to obtain second one-round extracted features;

[0190] The feature dimension of the second one-round extracted features is reduced to 256 through the second two-round linear layer, and then activated by a ReLu function to obtain second two-round extracted features, and 30% of the parameters are randomly discarded by applying Dropout;

[0191] The feature dimension of the second three-round extraction feature is reduced to 128 through a second four-round linear layer, and then activated through a ReLu function, and 10% of the parameters are randomly discarded through Dropout, to obtain a second four-round extraction feature;

[0192] The feature dimension of the second three-round extraction feature is reduced to 64 through a second four-round linear layer, and then activated through a ReLu function, and 10% of the parameters are randomly discarded through Dropout, to obtain a second four-round extraction feature;

[0193] The feature dimension of the second four-round extraction feature is reduced to 4 through a second five-round linear layer, and then the feature is standardized through a Softmax function, to obtain a second prediction ratio.

[0194] In some embodiments, the third DNN model includes a third input layer, a third one-round linear layer, a third two-round linear layer, a third three-round linear layer, a third four-round linear layer, and a third five-round linear layer, the model training module 803 is specifically further configured to:

[0195] Features in the training sample are extracted through the third input layer, to obtain third input extraction features with a dimension of 1000;

[0196] The feature dimension of the third input extraction feature is increased to 1024 through the third one-round linear layer, and then activated through a ReLu function, to obtain a third one-round extraction feature;

[0197] The feature dimension of the third one-round extraction feature is reduced to 512 through the third two-round linear layer, and then activated through a ReLu function, and 60% of the parameters are randomly discarded through Dropout, to obtain a third two-round extraction feature;

[0198] The feature dimension of the third two-round extraction feature is reduced to 256 through the third three-round linear layer, and then activated through a ReLu function, and 30% of the parameters are randomly discarded through Dropout, to obtain a third three-round extraction feature;

[0199] The feature dimension of the third three-round extraction feature is reduced to 128 through the third four-round linear layer, and then activated through a ReLu function, and 10% of the parameters are randomly discarded through Dropout, to obtain a third four-round extraction feature;

[0200] The feature dimension of the third four-round extraction feature is reduced to 4 through the third five-round linear layer, and then the feature is standardized through a Softmax function, to obtain a third prediction ratio.

[0201] In some embodiments, the model training module 803 is specifically further configured to:

[0202] Obtain a plurality of single-cell Hi-C data; wherein the single-cell Hi-C data is an interaction matrix;

[0203] The single-cell Hi-C data is pre-processed to obtain pre-processed Hi-C data; wherein the pre-processing includes depth filtering and depth normalization;

[0204] Randomly select a plurality of pre-processed Hi-C data for fusion to obtain pseudo-Bulk Hi-C data as a training sample;

[0205] The device 800 for predicting different cycle cell proportions based on chromatin conformation spectrum provided by the embodiment of the application is used to execute the technical solutions of the method for predicting different cycle cell proportions based on chromatin conformation spectrum provided by the foregoing embodiment, and has similar implementation principles and technical effects to those of the foregoing embodiment, which will not be described here in detail.

[0206] It should be noted that the division of each module of the above device is only a logical functional division, and all or part of it can be integrated into one physical entity, or can be physically separated. And these modules can all be implemented in the form of software called by a processing element, or all be implemented in the form of hardware, or part of the modules are implemented in the form of software called by a processing element, and part of the modules are implemented in the form of hardware. For example, the model training module 803 can be a separate processing element, or can be integrated into a chip of the above device, in addition, it can also be stored in the form of program code in the memory of the above device, and the functions of the above model training module 803 are called and executed by a processing element of the above device. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or can be independently implemented. The processing element here can be an integrated circuit with signal processing capability. In the implementation process, each step of the above method or each module can be completed by the integrated logic circuit of hardware or the instruction of software in the processing element.

[0207] Figure 9 For the structure schematic diagram of an electronic device provided by the embodiment of the application, refer to Figure 9 The electronic device 900 includes a processor 901 and a memory 902 in communication connection with the processor 901;

[0208] The memory 902 stores computer execution instructions;

[0209] The processor 901 executes the computer execution instructions stored in the memory 902 to realize the technical solutions of the method for predicting different cycle cell proportions based on chromatin conformation spectrum.

[0210] In the electronic device 900 described above, the memory 902 and the processor 901 are directly or indirectly electrically connected to realize the transmission or interaction of data. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines, such as through a bus connection. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc., but it does not mean that there is only one bus or only one type of bus. The memory 902 stores computer execution instructions for implementing the aforementioned method for predicting the proportion of different cycle cells based on chromatin conformation spectrum, including at least one software function module stored in the memory 902 in the form of software or firmware, and the processor 901 executes various function applications and data processing by running the software program and module stored in the memory 902.

[0211] The memory 902 includes at least one type of readable storage medium, not limited to a Random Access Memory (RAM), a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electric Erasable Programmable Read-Only Memory (EEPROM), etc. Among them, the memory 902 is used to store programs, and the processor 901 executes the programs after receiving execution instructions. Further, the software program and module in the memory 902 can also include an operating system, which can include various software components and / or drivers for managing system tasks (such as memory management, storage device control, power management, etc.), and can communicate with various hardware or software components to provide an operating environment for other software components.

[0212] The processor 901 can be an integrated circuit chip having a processing capability of signals. The processor 901 described above can be a general processor, including a central processing unit (CPU), a network processor (NP), a digital signal processor (DSP), an application specific integrated circuit (ASIC), etc. The processor 901 can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application. The general processor can be a microprocessor, or the processor 901 can also be any conventional processor or the like.

[0213] The electronic device 900 is configured to implement the technical solutions provided by the method embodiments described above for predicting the proportion of different cycle cells based on chromatin conformation spectrum, and the implementation principle and technical effects are similar to those of the method embodiments described above, which will not be repeated here.

[0214] The embodiments of the present application further provide a computer readable storage medium, and the computer readable storage medium stores computer execution instructions. When the computer execution instructions are executed, the technical solutions of the method for predicting the proportion of different cycle cells based on chromatin conformation spectrum are implemented.

[0215] The computer readable storage medium described above can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk. The computer readable storage medium can be any available medium that can be accessed by a general or special purpose computer.

[0216] An exemplary readable storage medium is coupled to the processor, so that the processor can read information from the readable storage medium, and can write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the control device of the device for predicting the proportion of different cycle cells based on chromatin conformation spectrum.

[0217] The embodiment of the present application further provides a computer program product comprising a computer program which, when executed, implements the technical solutions of the method for predicting the proportion of different cycle cells based on the chromatin conformation spectrum.

[0218] In the above embodiments, those skilled in the art can understand that the implementation of the above-mentioned method embodiments can be realized by software, hardware, firmware or any combination thereof, wholly or partially. When implemented by using software, the computer program product can be realized wholly or partially in the form of a computer program product. The computer program product comprises one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions according to the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired (for example, coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (for example, infrared, wireless network, microwave, etc.) manner. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. containing one or more available media. The available medium can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD) or a semiconductor medium (for example, solid state disk (SSD)) and the like.

[0219] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0220] Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. The application is intended to cover any variations, uses or adaptations of the application following, in general, the principles of the application and including such departures from the present disclosure as come within known or customary practice in the art to which the application pertains. The specification and examples are to be regarded as exemplary only, and the true scope and spirit of the application are indicated by the appended claims.

[0221] It is to be understood that the application is not limited to the precise construction already described above and shown in the drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the application. The scope of the application should only be limited by the claims appended hereto.

Claims

1. A method for predicting the proportion of cells in different cell cycles based on chromatin conformation profiles, characterized in that: include: Obtain bulk Hi-C data to be tested; Inputting the bulk Hi-C data to be tested into a cell cycle ratio prediction model to obtain a cell cycle ratio prediction result output by the cell cycle ratio prediction model; wherein the cell cycle ratio prediction model is obtained based on training a deep neural network model; The obtaining of bulk Hi-C data to be tested includes: Acquire a plurality of single-cell Hi-C data; wherein the single-cell Hi-C data is an interaction matrix; Preprocessing the single-cell Hi-C data to obtain preprocessed Hi-C data; wherein the preprocessing includes depth filtering and depth normalization; Randomly select multiple preprocessed Hi-C data for fusion to obtain pseudo Bulk Hi-C data as training samples; The training of the deep neural network model includes: Acquire multiple training samples and construct a training set; wherein the training samples are annotated with true values, and the true values ​​include the true proportions of cells in different cell cycles in the training samples; Inputting the training samples contained in the training set into the deep neural network model to obtain a predicted value; wherein the predicted value includes the predicted proportion of cells in different cell cycles in the training samples; Based on a preset loss function, obtaining the loss between the predicted value and the true value in the training set; Optimizing the network parameters of the deep neural network model based on the loss amount until the loss function converges or a preset number of training iterations is reached, thereby obtaining a cell cycle ratio prediction model; The deep neural network model includes a first DNN model, a second DNN model, and a third DNN model. The training samples included in the training set are input into the deep neural network model to obtain a prediction value, including: Inputting the training sample into a first DNN model to obtain a first prediction ratio; Inputting the training sample into a second DNN model to obtain a second prediction ratio; Inputting the training sample into a third DNN model to obtain a third prediction ratio; The predicted proportions of cells in different cell cycles in the first predicted proportion, the second predicted proportion and the third predicted proportion are averaged to obtain a predicted value.

2. The method according to claim 1, characterized in that The first DNN model includes a first input layer, a first first-round linear layer, a first second-round linear layer, a first third-round linear layer, a first fourth-round linear layer, and a first fifth-round linear layer. The training sample is input into the first DNN model to obtain a first prediction ratio, including: Extracting features from the training sample through the first input layer to obtain first input extracted features of dimension 1000; The feature dimension of the first input extracted features is reduced to 256 by the first round of linear layers, and then activated by the ReLu function to obtain the first round of extracted features; The feature dimensions of the first round of extracted features are reduced to 128 by the first and second rounds of linear layers, and then activated by the ReLu function to obtain the first and second rounds of extracted features; The feature dimensions of the features extracted from the first and second rounds are reduced to 64 by the first three rounds of linear layers, and then activated by the ReLu function to obtain the features extracted from the first and third rounds; The feature dimensions of the features extracted from the first three rounds are reduced to 32 by the first four rounds of linear layers, and then activated by the ReLu function to obtain the first four rounds of extracted features; The feature dimensions of the features extracted in the first four rounds are reduced to 4 through the first five rounds of linear layers, and then the features are normalized by the Softmax function to obtain the first prediction ratio.

3. The method according to claim 1, characterized in that The second DNN model includes a second input layer, a second first-round linear layer, a second second-round linear layer, a second third-round linear layer, a second fourth-round linear layer, and a second fifth-round linear layer. The training sample is input into the second DNN model to obtain a second prediction ratio, including: Extract features from the training sample through the second input layer to obtain a second input extracted feature with a dimension of 1000; The feature dimension of the second input extracted features is reduced to 512 by the second round of linear layers, and then activated by the ReLu function to obtain the second round of extracted features; The feature dimension of the second round of extracted features is reduced to 256 by the second round of linear layers, and then activated by the ReLu function. Dropout is applied to randomly discard 30% of the parameters to obtain the second round of extracted features; The feature dimensions of the second-second round extracted features are reduced to 128 through the second-third round linear layer, and then activated by the ReLu function. Dropout is applied to randomly discard 20% of the parameters to obtain the second-third round extracted features; The feature dimensions of the features extracted in the second and third rounds are reduced to 64 by the second four rounds of linear layers, and then activated by the ReLu function. Dropout is applied to randomly discard 10% of the parameters to obtain the second four rounds of extracted features; The feature dimension of the features extracted in the second four rounds is reduced to 4 through the second five rounds of linear layers, and then the features are normalized by the Softmax function to obtain the second prediction ratio.

4. The method according to claim 1, wherein The third DNN model includes a third input layer, a third first-round linear layer, a third second-round linear layer, a third third-round linear layer, a third fourth-round linear layer, and a third fifth-round linear layer. The training sample is input into the third DNN model to obtain a third prediction ratio, including: Extracting features from the training sample through the third input layer to obtain a third input extracted feature of dimension 1000; The feature dimension of the third input extracted feature is increased to 1024 by the third round of linear layer, and then activated by the ReLu function to obtain the third round of extracted features; The feature dimension of the third round of extracted features is reduced to 512 by the third round of linear layers, and then activated by the ReLu function, and 60% of the parameters are randomly discarded by Dropout to obtain the third round of extracted features; The feature dimension of the third round of extracted features is reduced to 256 by the third round of linear layer, and then activated by ReLu function, and 30% of the parameters are randomly discarded by Dropout to obtain the third round of extracted features; The feature dimension of the third-fourth round of extracted features is reduced to 128 by the third-fourth round of linear layers, and then activated by the ReLu function, and 10% of the parameters are randomly discarded by Dropout to obtain the third-fourth round of extracted features; The feature dimension of the features extracted in the third and fourth rounds is reduced to 4 by the third and fifth rounds of linear layers, and then the features are normalized by the Softmax function to obtain the third prediction ratio.

5. A device for predicting the proportion of cells in different life cycles based on chromatin conformation profiles, characterized in that: include: Data acquisition module, used to obtain bulk Hi-C data to be tested; The obtaining of bulk Hi-C data to be tested includes: Acquire a plurality of single-cell Hi-C data; wherein the single-cell Hi-C data is an interaction matrix; Preprocessing the single-cell Hi-C data to obtain preprocessed Hi-C data; wherein the preprocessing includes depth filtering and depth normalization; Randomly select multiple preprocessed Hi-C data for fusion to obtain pseudo Bulk Hi-C data as training samples; a ratio prediction module, configured to input the bulk Hi-C data to be tested into a cell cycle ratio prediction model to obtain a cell cycle ratio prediction result output by the cell cycle ratio prediction model; wherein the cell cycle ratio prediction model is obtained based on training a deep neural network model; The training of the deep neural network model includes: Acquire multiple training samples and construct a training set; wherein the training samples are annotated with true values, and the true values ​​include the true proportions of cells in different cell cycles in the training samples; Inputting the training samples contained in the training set into the deep neural network model to obtain a predicted value; wherein the predicted value includes the predicted proportion of cells in different cell cycles in the training samples; Based on a preset loss function, obtaining the loss between the predicted value and the true value in the training set; Optimizing the network parameters of the deep neural network model based on the loss amount until the loss function converges or a preset number of training iterations is reached, thereby obtaining a cell cycle ratio prediction model; The deep neural network model includes a first DNN model, a second DNN model, and a third DNN model. The training samples included in the training set are input into the deep neural network model to obtain a prediction value, including: Inputting the training sample into a first DNN model to obtain a first prediction ratio; Inputting the training sample into a second DNN model to obtain a second prediction ratio; Inputting the training sample into a third DNN model to obtain a third prediction ratio; The predicted proportions of cells in different cell cycles in the first predicted proportion, the second predicted proportion and the third predicted proportion are averaged to obtain a predicted value.

6. An electronic device, characterized in that: comprising a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method according to any one of claims 1 to 4.

Images