A feature construction method of synthetic lethal gene combination based on miRNA-mRNA regulation relationship

Abstract

The application belongs to the field of biological information, and discloses a feature construction method of synthetic lethal gene combination based on miRNA-mRNA regulation relationship, comprising the following steps: acquiring isomiR sequences, quantifying the uncertainty of the isomiR sequences corresponding to the miRNA by using information entropy as the feature of the miRNA; acquiring the miRNA-mRNA regulation relationship, finding the miRNA having the regulation relationship with the synthetic lethal related genes, and calculating the weighted average value of the miRNA features by using the expression correlation coefficient as the weight; and the feature of the synthetic lethal gene combination is the average value of the calculation results of the two related genes. The application starts from the isomiR, integrates the regulation effect of the miRNA to describe the multi-molecule correlation, constructs the synthetic lethal gene combination feature with the system biology significance, and improves the biological explainability of the model when constructing the synthetic lethal gene combination prediction model.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of biological information, and particularly relates to a feature construction method of synthetic lethal gene combination based on miRNA-mRNA regulation relationship. BACKGROUND

[0002] Synthetic lethality effect, also known as synergistic lethal effect, is a new idea for the research and development of anticancer drugs in recent years, and is concerned due to small toxicity and good effect. However, how to obtain a batch of high-quality synthetic lethal gene combinations is the main challenge currently faced.

[0003] Machine learning prediction is an efficient method for obtaining a large number of synthetic lethal gene combinations. Most of the methods using machine learning prediction use limited feature values, such as gene expression, protein network topology and gene function. However, the research on mRNA-miRNA regulation network feature values related to synthetic lethality is still lacking, especially the research method based on big data. SUMMARY

[0004] Objective: In order to overcome the deficiencies in the prior art, the application provides a feature construction method of synthetic lethal gene combination based on miRNA-mRNA regulation relationship, and the feature values constructed based on the method can effectively contribute to machine learning prediction and improve the prediction effect.

[0005] To solve the above technical problems, the technical scheme adopted by the application is as follows:

[0006] A feature construction method of synthetic lethal gene combination based on miRNA-mRNA regulation relationship, comprising the following steps:

[0007] Obtain isomiR sequence, use information entropy to quantify the uncertainty of miRNA corresponding isomiR sequence as the feature of miRNA;

[0008] Obtain miRNA-mRNA regulation relationship, find miRNA having regulation relationship with synthetic lethal related genes, and use expression correlation coefficient as weight to calculate weighted average value of miRNA feature;

[0009] The feature of synthetic lethal gene combination is the average value of the above calculation results of the two related genes, which is used for inputting a machine learning model.

[0010] Further, according to the isomiR expression data, the chromosome coordinates of all isomiRs and the names of their corresponding precursor and mature miRNAs are obtained; according to the name of the precursor, the chromosome coordinates of the precursor are obtained from hsa.gff3, and the sequence of the precursor is obtained from hairpin.fa.

[0011] Further, the chromosome coordinates of all isomiRs are obtained, and the sequence at the corresponding position is obtained.

[0012] Further, according to the relative position of the chromosome coordinates of the isomiR and the precursor coordinates, the isomiR sequence is extracted from the precursor sequence, wherein the coordinate starting point of the positive strand is at the head of the sequence, and the coordinate starting point of the negative strand is at the tail of the sequence.

[0013] Further, from the precursor corresponding mature miRNA in hsa.gff3, the chromosome coordinates of the isomiR corresponding mature miRNA are found according to the mature miRNA name.

[0014] Further, from the precursor corresponding mature miRNA in hsa.gff3, the chromosome coordinates of the isomiR corresponding mature miRNA are found according to the mature miRNA name, and the isomiR whose coordinate offset relative to the mature miRNA coordinate exceeds 4 nt is removed.

[0015] Further, all miRNA-mRNA regulatory relationship results are downloaded from the starBase database, and each set of regulatory relationship downloaded should be predicted by at least three kinds of prediction software.

[0016] Further, the uncertainty of the miRNA to the 5' end and 3' end sequence of the isomiR is quantified by using information entropy as the feature of the miRNA, denoted as MIH:

[0017]

[0018] MIH represents the feature of the miRNA, L represents the length variation range of the classical miRNA, which ranges from ±4 nt, i is a certain position, represents the frequency of the occurrence of a certain position base, represents the frequency of the occurrence of a certain position base.

[0019] Further, a pair of miRNA-mRNA regulatory relationship is analyzed by Spearman correlation analysis using TCGA expression data, and the results of significant negative correlation are retained.

[0020] Further, for all miRNAs regulating a certain mRNA, the weighted average value of their MIH is calculated by using the expression correlation coefficient as the weight, as the feature of the mRNA, denoted as m:

[0021]

[0022] m represents the characteristics of mRNA, N is the number of miRNAs regulating a certain mRNA, i represents one of the miRNAs, r is the expression correlation coefficient, and MIH represents the characteristics of miRNA;

[0023] The characteristics of the synthetic lethal gene combination are the average of the above calculation results of the two related genes.

[0024] A feature construction method of a synthetic lethal gene combination based on miRNA-mRNA regulatory relationship, characterized in that it comprises the following steps

[0025] Step 1: Obtain the chromosome coordinates of all isomiRs from the isomiR expression data of the TCGA database, and collect the isomiR sequences of each miRNA. Remove the isomiRs whose coordinate offset relative to the mature miRNA is more than 4 nt.

[0026] The information entropy is used to quantify the uncertainty of the 5' end and 3' end sequences of the miRNA, which is used as the characteristics of the miRNA, denoted as MIH:

[0027]

[0028] MIH represents the characteristics of the miRNA, L represents the length variation range (±4 nt) of the classical miRNA, and i represents a certain position, represents the frequency of the occurrence of a certain position base, represents the frequency of the non-occurrence of a certain position base.

[0029] Step 2: Download all miRNA-mRNA regulatory relationship results from the starBase database, and each set of regulatory relationship downloaded should be predicted by at least three prediction software. Find the miRNAs that have regulatory relationship with the synthetic lethal related genes. A pair of miRNA-mRNA regulatory relationship, use TCGA expression data for Spearman correlation analysis, use the expression correlation coefficient as the weight to calculate the weighted average value of the miRNA characteristics MIH, as the characteristics of the mRNA.

[0030] Step 3: The characteristics of the synthetic lethal gene combination are the average of the above calculation results of the two related genes, which are used as input for the machine learning model.

[0031] Beneficial effects

[0032] The application provides a feature construction method of synthetic lethal gene combination based on miRNA-mRNA regulation relationship, which utilizes miRNA regulation and short and flexible isomiR to further explore the features of synthetic lethal gene combination, effectively contributes to machine learning prediction, and improves the prediction effect. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 is a method flowchart of an embodiment of the application;

[0034] Figure 2 is a model prediction effect comparison when the features constructed by the application are added or not in the embodiment. DETAILED DESCRIPTION

[0035] The application will be further described below in combination with the drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the application, and cannot be used to limit the protection scope of the application.

[0036] Embodiment 1

[0037] The chromosome coordinate file hsa.gff3 of human miRNA is downloaded from the miRBase database, and the Fasta format sequence file hairpin.fa of all precursors is downloaded.

[0038] The chromosome coordinates of all isomiRs and the names of the corresponding precursors and mature miRNAs are obtained from the isomiR expression data of the TCGA database. According to the name of the precursor, the chromosome coordinates of the precursor are obtained from hsa.gff3, and the sequence of the precursor is obtained from hairpin.fa. According to the relative position of the chromosome coordinates of the isomiR and the precursor coordinates, the isomiR sequence is extracted from the precursor sequence, wherein the coordinate starting point of the positive strand is at the head of the sequence, and the coordinate starting point of the negative strand is at the tail of the sequence. According to the name of the mature miRNA, the chromosome coordinates of the isomiR corresponding to the mature miRNA are found in the mature miRNA corresponding to the precursor in hsa.gff3, and the isomiR with a coordinate offset of more than 4 nt relative to the mature miRNA coordinates is removed.

[0039] All miRNA-mRNA regulation relationship results are downloaded from the starBase database, and each set of regulation relationship downloaded should be predicted by at least three prediction software. For the mRNA involved in the synthetic lethal gene combination, the miRNA set having regulation effect on the mRNA is extracted from the regulation relationship result, and the uncertainty of the 5' end and 3' end sequences of the miRNA corresponding isomiR is quantified by using information entropy as the feature of the miRNA, which is denoted as MIH:

[0040]

[0041] MIH represents the feature of miRNA, L represents the length variation range of classical miRNA (±4 nt), i represents a position, represents the frequency of a base appearing at a position, represents the frequency of a base not appearing at a position.

[0042] A pair of miRNA-mRNA regulatory relationship, Spearman correlation analysis of the same sample set using TCGA expression data, and the results of significant negative correlation are retained. For all miRNAs regulating a certain mRNA, the weighted average value is calculated using the expression correlation coefficient as the weight as the feature of the mRNA, denoted as m:

[0043]

[0044] m represents the feature of mRNA, N is the number of miRNAs regulating a certain mRNA, i represents one of the miRNAs, r is the expression correlation coefficient, and MIH represents the feature of miRNA.

[0045] The feature of the synthetic lethal gene combination is the average of the above calculation results of the two genes.

[0046] According to the feature of the synthetic lethal gene combination based on the miRNA-mRNA regulatory relationship obtained by the above steps, a feature matrix of the synthetic lethal gene combination verified by multi-omics data for experimental verification is constructed. Using a support vector machine model (SVM), 90% of the samples are used for training, and 10% of the samples are used for testing. Repeat ten times, and take the average of the test results as the final evaluation result of the model. The evaluation index is AUC (Area Under Curve), defined as the area under the ROC curve. Among them, the ROC (receiver operating characteristic curve) is a curve drawn with the true positive rate as the vertical coordinate and the false positive rate as the horizontal coordinate. The prediction effect of the synthetic lethal gene combination of 30 cancer types was evaluated, and the prediction effect of the prediction model using only the multi-omics data features was also evaluated (Table 1, model1 represents the model using the features of the present application, model2 represents the model using only the multi-omics data features, and the other steps are the same):

[0047] Table 1

[0048] Cancer Type model1 model2 ACC 0.853 0.681 BLCA 0.858 0.818 BRCA 0.850 0.807 CESC 0.844 0.784 CHOL 0.930 0.811 COAD 0.830 0.771 DLBC 0.929 0.716 ESCA 0.853 0.794 GBM 0.865 0.812 HNSC 0.908 0.664 KICH 0.837 0.777 KIRC 0.861 0.767 KIRP 0.933 0.854 LAML 0.892 0.832 LGG 0.832 0.770 LIHC 0.868 0.831 LUAD 0.835 0.781 LUSC 0.848 0.748 OV 0.915 0.832 PAAD 0.869 0.661 PCPG 0.893 0.744 PRAD 0.858 0.721 READ 0.868 0.784 SARC 0.865 0.821 SKCM 0.891 0.847 STAD 0.882 0.753 TGCT 0.869 0.738 THCA 0.924 0.806 THYM 0.904 0.870 UCEC 0.878 0.731

[0049] The AUC values of the model using the features of the present application in 30 cancers are higher than those of the model using only the multi-omics data features. Overall, the AUC values of the model using the features of the present application are significantly higher than those of the model using only the multi-omics data features ( Figure 2).

[0050] In another machine learning model random forest, the "MeanDecreaseAccuracy" index of the feature of the synthetic lethal gene combination constructed based on the miRNA-mRNA regulation relationship reaches 40%, indicating that the prediction accuracy will be significantly negatively affected after random disturbance of the feature constructed by the application, proving the effectiveness and importance of the feature construction method.

[0051] The above only describes the preferred embodiments of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A method for characterizing synthetic lethal gene combinations based on miRNA-mRNA regulatory relationships, characterized in that, Includes the following steps: Obtain the isomiR sequence. Based on the isomiR expression data, obtain the chromosomal coordinates of all isomiRs and the names of their corresponding precursor and mature miRNAs. Based on the precursor name, obtain the precursor's chromosomal coordinates from hsa.gff3 and the precursor sequence from hairpin.fa. Based on the relative position of the isomiR's chromosomal coordinates and the precursor's coordinates, extract the isomiR sequence from the precursor sequence, where the positive strand's coordinates start at the beginning of the sequence and the negative strand's coordinates start at the end of the sequence. From the mature miRNAs corresponding to the precursors in hsa.gff3, find the chromosomal coordinates of the mature miRNAs corresponding to the isomiRs based on the mature miRNA names, and remove isomiRs whose offset relative to the mature miRNA coordinates exceeds 4 nt. The uncertainty of the 5' and 3' end sequences of the corresponding isomiR of miRNA is quantified using information entropy, and this uncertainty is used as a characteristic of the miRNA, denoted as MIH: ； MIH represents the characteristics of the miRNA, L represents the range of classical miRNA length variation (±4 nt), and i represents a specific position within this range. This indicates the frequency of a base at a certain position. Indicates the frequency at which a base does not appear at a certain position; To obtain miRNA-mRNA regulatory relationships, identify miRNAs that regulate synthetic lethal genes. For all miRNAs regulating a particular mRNA, calculate a weighted average of their MIHs using expression correlation coefficients as weights. This average is then used as a characteristic of the mRNA, denoted as m. ； m represents the characteristics of the mRNA, N is the number of miRNAs that regulate a certain mRNA, i represents one of the miRNAs, r is the expression correlation coefficient, and MIH represents the characteristics of the miRNA. The characteristic of the synthetic lethal gene combination is the average of the above calculation results of the two related genes, which is used as input into the machine learning model.

2. The method for characterizing synthetic lethal gene combinations based on miRNA-mRNA regulatory relationships according to claim 1, characterized in that, Obtain the chromosome coordinates of all isomiRs and retrieve the sequence at the corresponding positions.

3. The method for characterizing synthetic lethal gene combinations based on miRNA-mRNA regulatory relationships according to claim 1, characterized in that, From the precursor corresponding to the mature miRNA in hsa.gff3, find the chromosomal coordinates of the mature miRNA corresponding to isomiR based on the mature miRNA name.

4. The method for characterizing synthetic lethal gene combinations based on miRNA-mRNA regulatory relationships according to claim 1, characterized in that, All miRNA-mRNA regulatory relationships were downloaded from the starBase database. Each downloaded regulatory relationship was predicted by at least three prediction software programs.

5. The method for characterizing synthetic lethal gene combinations based on miRNA-mRNA regulatory relationships according to claim 1, characterized in that, The regulatory relationship between a miRNA and mRNA pair was analyzed using Spearman correlation analysis of TCGA expression data, retaining results showing significant negative correlations.

Images