CN112086199A - Liver cancer data processing system based on multiple groups of mathematical data - Google Patents

Abstract

A liver cancer data processing system based on multi-omics data provided by the present invention includes a preprocessing module, a data dimensionality reduction processing module, a classification processing module and a classifier module; the preprocessing module is used for multi-omics data analysis of liver cancer. Perform screening, and output the screened target data to the data dimensionality reduction processing module; the data dimensionality reduction processing module is used to receive the target data output by the preprocessing module, perform dimensionality reduction processing on the target data, and reduce the The dimensionally processed target data is output to the data dimensionality reduction processing module; the classification processing module is used to receive the dimensionally reduced target data output by the data dimensionality reduction processing module, and perform classification processing according to the dimensionally reduced target data, and output the classification label; the classifier module is used to receive the classification label, use the classification label to train the classifier module, and then the classifier module receives the real-time multi-omics liver cancer data and predicts the survival time of liver cancer; The multi-omics data is well fused, and the complementarity of the data is effectively used to fuse the multi-omics data of liver cancer, thereby effectively avoiding the loss of characteristic information in the process of data processing, effectively ensuring the accuracy of data processing, and ensuring the survival of subsequent liver cancer. guarantee the accuracy of the forecast.

Description

Liver cancer data processing system based on multi-omics data

technical field

The invention relates to a data processing system, in particular to a liver cancer data processing system based on multi-omics data.

Background technique

Surgical resection is the main method for early stage liver cancer, but clinical data show that the recurrence rate of liver cancer after surgery is about 70%, which seriously hinders the long-term survival of patients. If we establish classification standards for HCC, carry out more detailed stratified management of high-risk recurrence patients, and first screen out the potentially beneficial population from the source before performing surgery, it may be more important to improve patient survival and achieve precise treatment of HCC. meaning. To establish a classification standard for liver cancer based on multi-omics data, and to carry out more accurate prognostic treatment and management for different patients, will improve the survival rate of patients. Therefore, it is of great significance to fuse multi-omics data to classify patients at the molecular level and predict the prognosis of patients, which also has clinical significance for the treatment of patients.

In recent years, there are also methods that fuse RNA sequencing data, miRNA data, methylation data and clinical survival data of liver cancer patients to classify liver cancer and predict prognosis. However, in the prior art, few researchers consider the patient's survival status when studying molecular subtypes. Survival rates have important clinical implications for the study of molecular subtypes, and large differences in survival rates often have a large impact on molecular subtypes. Using the fusion of multi-omics data to perform molecular typing and predict prognosis has the following two characteristics: (1) The fusion period of multi-omics data is generally divided into early fusion, mid-term fusion and late fusion. There is a big impact. (2) The fusion method also has a great influence. The fusion method or system of the prior art has the following defects: on the one hand, an automatic encoder is used to integrate the input data, but it is easy to cause the loss of characteristic data; The fusion of different data is poor, the data cannot complement each other, and accurate information cannot be extracted.

Therefore, in order to solve the above technical problems, it is urgent to propose a new technical means.

SUMMARY OF THE INVENTION

In view of this, the purpose of the present invention is to provide a liver cancer data processing system based on multi-omics data, which can well fuse the liver cancer multi-omics data, and effectively utilize the complementarity of the data to fuse the liver cancer multi-omics data together. , so as to effectively avoid the loss of characteristic information in the process of data processing, effectively ensure the accuracy of data processing, and provide a guarantee for the accuracy of subsequent liver cancer survival prediction.

The invention provides a liver cancer data processing system based on multi-omics data, comprising a preprocessing module, a data dimensionality reduction processing module, a classification processing module and a classifier module;

The preprocessing module is used for screening liver cancer multi-omics data, and outputting the screened target data to the data dimension reduction processing module;

The data dimensionality reduction processing module is used to receive the target data output by the preprocessing module, perform dimensionality reduction processing on the target data, and output the dimensionality reduction processed target data to the data dimensionality reduction processing module;

The classification processing module is configured to receive the dimensionally reduced target data output by the data dimensionality reduction processing module, perform classification processing according to the dimensionally reduced target data, and output a classification label;

The classifier module is used for receiving classification labels, and using the classification labels to train the classifier module, and then the classifier module receives real-time multi-omics liver cancer data and predicts the survival time of liver cancer.

Further, the preprocessing module for screening liver cancer multi-omics data includes:

The preprocessing module scores each feature of the multi-omics data of liver cancer based on the univariate Cox-PH model, then compares the score Per1 with the set threshold P _y to screen out the features with Per1 < P _y The filtered data are fused to form target data.

Further, the dimensionality reduction processing performed by the data dimensionality reduction processing module on the target data specifically includes:

SA1. Build a K-layer autoencoder in the data dimensionality reduction processing module, where the output function of the K-layer autoencoder is:

x'=Relu(W _i · _Relu (W _i x+ _bi )); wherein, Wi is the weight matrix between adjacent self-encoders, _bi is the offset of the weight matrix Wi _, and x is m Dimensional target data X = eigenvalues in (x ₁ , x ₂ ,...,x _m );

SA2. The data dimensionality reduction processing module constructs a loss function, where the loss function is:

其中，L(x,x')为损失函数，β_w为正则化惩罚系数，

Among them, L(x,x') is the loss function, _βw is the regularization penalty coefficient,

SA3. Perform an iterative operation through the loss function, update the weight matrix _Wi and the offset _bi of the weight matrix _Wi , until the number of iterations is reached, the data dimensionality reduction processing module outputs the target data after dimensionality reduction.

Further, the survival prediction of the classification processing module specifically includes:

SB1. The classification processing module uses the univariate Cox-PH model to score the features in the target data after dimensionality reduction processing again, and then compares the score value Per2 of the feature with the set threshold P _y , and filters out Per2 < P _y . features, and fuse the filtered data;

SB2. The classification processing module constructs a normalization processing model, and normalizes the data processed in step SB1, wherein the normalization processing model is:

p为步骤SB1输出的特征数据，P为归一化处理后的特征数据，Var(p)为特征数据p的方差，E(p)为特征数据p的经验平均值；

p is the characteristic data output in step SB1, P is the normalized characteristic data, Var(p) is the variance of the characteristic data p, and E(p) is the empirical average value of the characteristic data p;

SB3. The classification processing module builds the similarity function:

其中，W(i,j)为第i个样本z_i与第j个样本z_j的相似性，θ_ij为归一化因子；其中：

Among them, W(i,j) is the similarity between the ith sample _zi and the jth sample z _j , and θ _ij is the normalization factor; where:

λ_i为第i个样本z_i的k个近邻，λ_j为第j个样本z_j的k个近邻；z_r表示λ_i里的第r个样本。

λ _i is the k nearest neighbors of the ith sample zi _i , λ _j is the k nearest neighbors of the j th sample z _j ; z _r represents the r th sample in λ _i .

SB4. The classification processing module determines the classification label according to the similarity function, and outputs it to the classifier module.

Beneficial effects of the present invention: through the present invention, the multi-omics data of liver cancer can be well fused, and the complementarity of the data can be effectively used to fuse the multi-omics data of liver cancer, thereby effectively avoiding the loss of characteristic information in the process of data processing , which can effectively ensure the accuracy of data processing and provide a guarantee for the accuracy of subsequent liver cancer survival prediction.

Description of drawings

Below in conjunction with accompanying drawing and embodiment, the present invention is further described:

FIG. 1 is a schematic structural diagram of the present invention.

FIG. 2 is a schematic diagram of a classification label of the present invention.

3 is a comparison diagram of a specific example of the present invention.

Detailed ways

The present invention is further described in detail below in conjunction with the accompanying drawings:

The invention provides a liver cancer data processing system based on multi-omics data, comprising a preprocessing module, a data dimensionality reduction processing module, a classification processing module and a classifier module;

The preprocessing module is used for screening liver cancer multi-omics data, and outputting the screened target data to the data dimension reduction processing module;

The data dimensionality reduction processing module is used to receive the target data output by the preprocessing module, perform dimensionality reduction processing on the target data, and output the dimensionality reduction processed target data to the data dimensionality reduction processing module;

The classification processing module is configured to receive the dimensionally reduced target data output by the data dimensionality reduction processing module, perform classification processing according to the dimensionally reduced target data, and output a classification label;

The classifier module is used to receive the classification label, and use the classification label to train the classifier module, and then the classifier module receives the real-time multi-omics liver cancer data and predicts the survival period of liver cancer; The omics data is well fused, and the complementarity of the data is effectively used to fuse the multi-omics data of liver cancer, thereby effectively avoiding the loss of characteristic information in the process of data processing, effectively ensuring the accuracy of data processing, and improving the survival of subsequent liver cancer. The accuracy of the forecast is guaranteed.

In this embodiment, the preprocessing module for screening liver cancer multi-omics data includes:

The preprocessing module scores each feature of the multi-omics data of liver cancer based on the univariate Cox-PH model, then compares the score Per1 with the set threshold P _y to screen out the features with Per1 < P _y The filtered data are fused to form target data, wherein the set threshold P _y is generally set to 0.5. Through the above, the loss of information during processing can be effectively prevented, thereby ensuring the accuracy of the final result.

In this embodiment, the dimensionality reduction processing performed on the target data by the data dimensionality reduction processing module specifically includes:

SA1. Build a K-layer autoencoder in the data dimensionality reduction processing module, where the output function of the K-layer autoencoder is:

x'=Relu(W _i · _Relu (W _i x+ _bi )); wherein, Wi is the weight matrix between adjacent self-encoders, _bi is the offset of the weight matrix Wi _, and x is m Dimensional target data X = eigenvalues in (x ₁ , x ₂ ,...,x _m );

SA2. The data dimensionality reduction processing module constructs a loss function, where the loss function is:

其中，L(x,x')为损失函数，β_w为正则化惩罚系数，

Among them, L(x,x') is the loss function, _βw is the regularization penalty coefficient,

SA3. Perform an iterative operation through the loss function, update the weight matrix _Wi and the offset _bi of the weight matrix _Wi , until the number of iterations is reached, the data dimensionality reduction processing module outputs the target data after dimensionality reduction.

In this embodiment, the survival prediction of the classification processing module specifically includes:

SB1. The classification processing module uses the univariate Cox-PH model to score the features in the target data after dimensionality reduction processing again, and then compares the score value Per2 of the feature with the set threshold P _y , and filters out Per2 < P _y . features, and fuse the filtered data, wherein, in the data fusion process, a feature matrix is formed by combining multiple features;

SB2. The classification processing module constructs a normalization processing model, and normalizes the data processed in step SB1, wherein the normalization processing model is:

p为步骤SB1输出的特征数据，P为归一化处理后的特征数据，Var(p)为特征数据p的方差，E(p)为特征数据p的经验平均值；

p is the characteristic data output in step SB1, P is the normalized characteristic data, Var(p) is the variance of the characteristic data p, and E(p) is the empirical average value of the characteristic data p;

SB3. The classification processing module builds the similarity function:

其中，W(i,j)为第i个样本z_i与第j个样本z_j的相似性，θ_ij为归一化因子；其中：

Among them, W(i,j) is the similarity between the ith sample _zi and the jth sample z _j , and θ _ij is the normalization factor; where:

λ_i为第i个样本z_i的k个近邻，λ_j为第j个样本z_j的k个近邻；z_r表示λ_i里的第r个样本。

λ _i is the k nearest neighbors of the ith sample zi _i , λ _j is the k nearest neighbors of the j th sample z _j ; z _r represents the r th sample in λ _i .

SB4. The classification processing module determines the classification label according to the similarity function, and outputs it to the classifier module. Among them, the classifier module uses the XGBoost classifier, and the multi-omics liver cancer data includes RNA sequencing data, miRNA data, and DNA methylation data; taking RNA sequencing data as an example: when the preprocessing module performs screening, the RNA sequencing data is screened. Feature data that meet the screening criteria are obtained, and then the screening data of each RNA sequencing data are recombined to form a new RNA sequencing data.

In step SB1, the features selected from the three kinds of multi-omics data are fused to form a data matrix, and the data matrix is of order n×n, and each column of the matrix is used as a sample. With n samples {z ₁ , z ₂ ,..., z _n }, the classifier module obtains the final classification label by performing cluster analysis on each sample as described above. Generally speaking, the classification label is set to 2.

Datasets GSE14520 and GSE31384 were mined from the GEO database as validation cohorts for RNA-seq and miRNA training classifiers, respectively. For both confirmation cohorts, we first select common features in the training set samples, and then normalize the data using the same method as for multicomponent data normalization. In our study, we need to select M features based on cluster labels for the training set and two cohorts. In this way, the two cohorts will be used as validation datasets to test the model and finally get the classification result. Here, we set the value of M (50-100) and found that when the value of M is set to 50, the resulting trained model can achieve the best prediction results.

Using TCGA as the training data set, the RNA-seq, miRNA-seq and DNA methylation data of liver cancer were obtained, and the prediction processing module constructed a univariate Cox-PH model to obtain the feature of Per1<0.05, and then processed the multi-omics data. After input to the dimensionality reduction processing module for processing, input to the classification processing module to construct a univariate Cox-PH model again for screening to obtain features with Per1 < 0.05. Finally, the classifier module uses spectral clustering to obtain two subtypes with significant survival differences. , based on the obtained cluster labels, the classifier module also uses the XGBoost classifier to train with the cluster labels, and then input real-time multi-omics liver cancer data for survival prediction. To verify the effectiveness of this classifier in predicting survival, we used two sets of data from GEO, namely GSE1452 and GES31384, to validate the model as shown in Figure 2. For the survival curves of the two survival subtypes, our results outperformed those of the other models, showing a significant improvement in the predictive power of our model compared to other published models.

Finally, we also compare our results with those of other models. Both the log-rank P value and the C index, our experimental results are significantly better than other experimental results, as shown in Figure 3.

In differential gene expression analysis, we could identify 1465 up-regulated genes and 930 down-regulated genes, including tumor marker gene BIRC5 (P=2.07e-41) and stem cell marker genes CD24 (P=2.83e-11), KRT19 (P=2.83e-11) = 2.82e-26) and EPCAM (P = 1.01e-6). In addition, we also found 28 genes (SLC2A2, AQP9, RGN, SULT2A1, CRYL1, SERPINC1, PAH, CDO1, PLG, APOC3, CYP27A1, PFKFB3, TM4SF1, ACSL5, RGS2, HN1, SERPINA10, CYB5A, EPHX2, SPHX2, RGS1, ADH1B, LECT2, TBX3, RNASE4, ALDOA, ADH6, SLC38A1) were different between the two survival risk groups we identified and were strongly associated with HCC survival.

For differentially expressed genes obtained by differential analysis, we also performed Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis on both subgroups. The PI3K-Akt signaling pathway, cell cycle signaling pathway, and p53 signaling pathway are enriched in tumor-related pathways in the aggressive subtype (C2), among which the P13K-Akt signaling pathway is also associated with CD8+ T cell infiltration. The low-risk survival subtype (C1) has related pathways such as drug metabolism, cytochrome P450, metabolic pathways and fatty acid degradation. These pathways have important implications for studying the prognosis of liver cancer.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be Modifications or equivalent replacements without departing from the spirit and scope of the technical solutions of the present invention should be included in the scope of the claims of the present invention.

Claims (4)

1. a liver cancer data processing system based on multi-omics data, is characterized in that: comprise preprocessing module, data dimensionality reduction processing module, classification processing module and classifier module; The preprocessing module is used for screening liver cancer multi-omics data, and outputting the screened target data to the data dimension reduction processing module; The data dimensionality reduction processing module is used to receive the target data output by the preprocessing module, perform dimensionality reduction processing on the target data, and output the dimensionality reduction processed target data to the data dimensionality reduction processing module; The classification processing module is configured to receive the dimensionally reduced target data output by the data dimensionality reduction processing module, perform classification processing according to the dimensionally reduced target data, and output a classification label; The classifier module is used for receiving classification labels, and using the classification labels to train the classifier module, and then the classifier module receives real-time multi-omics liver cancer data and predicts the survival time of liver cancer. 2. The liver cancer data processing system based on multi-omics data according to claim 1, characterized in that: the screening of liver cancer multi-omics data by the preprocessing module comprises: The preprocessing module scores each feature of the multi-omics data of liver cancer based on the univariate Cox-PH model, and then compares the score Per1 with the set threshold P _y to screen out the features with Per1 < P _y The filtered data are fused to form target data. 3. The liver cancer data processing system based on multi-omics data according to claim 2, wherein the data dimensionality reduction processing module performs dimensionality reduction processing on the target data specifically comprising: SA1. Build a K-layer autoencoder in the data dimensionality reduction processing module, where the output function of the K-layer autoencoder is: x'=Relu(W _i · _Relu (W _i x+ _bi )); wherein, Wi is the weight matrix between adjacent self-encoders, _bi is the offset of the weight matrix Wi _, and x is m Dimensional target data X = eigenvalues in (x ₁ , x ₂ ,...,x _m ); SA2. The data dimensionality reduction processing module constructs a loss function, where the loss function is:

Among them, L(x,x') is the loss function, _βw is the regularization penalty coefficient,

SA3. Perform an iterative operation through the loss function, update the weight matrix _Wi and the offset _bi of the weight matrix _Wi , until the number of iterations is reached, the data dimensionality reduction processing module outputs the target data after dimensionality reduction. 4. The liver cancer data processing system based on multi-omics data according to claim 3, wherein the survival prediction of the classification processing module specifically includes: SB1. The classification processing module uses the univariate Cox-PH model to score the features in the target data after dimensionality reduction processing again, and then compares the score value Per2 of the feature with the set threshold P _y , and filters out Per2 < P _y . features, and fuse the filtered data; SB2. The classification processing module constructs a normalization processing model, and normalizes the data processed in step SB1, wherein the normalization processing model is:

p is the characteristic data output in step SB1, P is the normalized characteristic data, Var(p) is the variance of the characteristic data p, and E(p) is the empirical average value of the characteristic data p; SB3. The classification processing module builds the similarity function:

Among them, W(i,j) is the similarity between the ith sample _zi and the jth sample z _j , and θ _ij is the normalization factor; where:

λ _i is the k nearest neighbors of the ith sample zi _i , λ _j is the k nearest neighbors of the j th sample z _j ; z _r represents the r th sample in λ _i . SB4. The classification processing module determines the classification label according to the similarity function, and outputs it to the classifier module.

Images