Method for screening cross differential expression genes of lung cancer and novel coronavirus infection through machine learning and correspondingly screening prognosis genes of lung cancer

Abstract

The invention belongs to the technical field of bioinformatics, and discloses a method for screening cross differential expression genes of lung cancer and novel coronavirus infection through machine learning and correspondingly screening lung cancer prognosis genes. Comprising the following steps: acquiring cross differential expression genes of lung cancer and novel coronavirus, screening the cross differential expression genes of lung cancer and novel coronavirus infection, correspondingly screening lung cancer prognosis genes by using Cox regression and LASSO regression, and analyzing the screened lung cancer prognosis genes by using a K-M survival analysis method, a GO and KEGG enrichment analysis method and a protein interaction analysis method. According to the method, not only can data be processed, but also intricate modes in the data can be identified, high sensitivity and specificity are achieved, and the screening efficiency and accuracy are improved. The technology can be popularized to other disease research, and has universality.

Description

Technical Field

[0001] The present invention belongs to the technical field of bioinformatics, and specifically relates to a method for using machine learning to screen cross-differentially expressed genes of lung cancer and novel coronavirus infection and correspondingly screening lung cancer prognosis genes. Background Art

[0002] Cancer is one of the main factors threatening human life and health. In 2020, there were approximately 19.3 million new cancer cases and 9.96 million deaths worldwide. Among them, the incidence of lung cancer was approximately 2.21 million cases and the mortality rate was approximately 1.8 million cases, ranking first, far exceeding other cancers. In 2022, there were approximately 4.8247 million new cancer cases and 2.5742 million deaths in my country, among which the incidence (approximately 22%) and mortality (approximately 28%) of lung cancer ranked first. With the prevalence of novel coronavirus (COVID-19) infection, people with lung cancer may experience more severe symptoms, leading to an increase in mortality. Therefore, it is necessary to explore the relationship between the two and find cross-differentially expressed genes for lung tumors and novel coronavirus infection.

[0003] According to histopathological classification, lung cancer is mainly divided into non-small cell lung cancer (about 85%) and small cell lung cancer (about 15%). The current means of treating lung cancer are mainly surgical treatment, chemotherapy, radiotherapy and targeted therapy. However, since the early symptoms of lung cancer are not obvious and difficult to detect, and the recurrence and metastasis rate of late-stage surgery is high, the cure rate of lung cancer by traditional treatments is not high. Tumor markers can assist in cancer diagnosis, risk stratification and efficacy and prognosis evaluation. Exploring markers that are beneficial for early screening and prognostic effect evaluation of lung cancer is the development direction for improving the cure rate of lung cancer. However, due to the development of sequencing technology and the complexity of lung cancer types, a large amount of genomic data has emerged, which has brought great difficulties to the screening work. How to obtain key prognostic gene information of lung cancer has become a problem. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present invention provides a method for machine learning to screen cross-differentially expressed genes of lung cancer and novel coronavirus infection and correspondingly screen lung cancer prognosis genes.

[0005] The above-mentioned object of the present invention is achieved through the following technical scheme: a method for machine learning to screen cross-differentially expressed genes of lung cancer and novel coronavirus infection and correspondingly screen lung cancer prognosis genes, wherein the lung cancer is lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC), and the machine learning method is Cox regression and LASSO regression; the method for machine learning to screen cross-differentially expressed genes of lung cancer and novel coronavirus infection and correspondingly screen lung cancer prognosis genes is specifically:

[0006] 1. Obtaining cross-differentially expressed genes between lung cancer and novel coronavirus: Download gene expression data and clinical sample data of LUAD and LUSC, and download COVID-19 related gene information;

[0007] 2. Screening of cross-differentially expressed genes between lung cancer and novel coronavirus infection: Using the statistically significant difference value range and differential expression multiple range as screening criteria, screen differentially expressed genes and draw a volcano map. Then, take the intersection of the obtained LUAD and LUSC differentially expressed genes and COVID-19 related genes and draw a Venn diagram;

[0008] 3. Use Cox regression to screen lung cancer prognostic genes: Combine the clinical sample information of LUAD and LUSC patients, and use the univariate Cox regression method to screen the corresponding prognosis-related genes;

[0009] 4. Use LASSO regression to screen lung cancer prognostic genes: the number of samples in the training set and the validation set is allocated according to a certain ratio, and then LASSO regression is used for corresponding screening;

[0010] 5. Use KM survival analysis to analyze the selected lung cancer prognosis genes: estimate the survival function using the following formula

[0011]

[0012] Where S(t) refers to the survival probability of an individual at time t, n refers to the number of time points at which the event occurs, and d i refers to the number of individuals that experience an event at time t; n i It refers to the number of individuals still alive at time t, s(0) = 1;

[0013] The KM survival analysis method was used to analyze the 3-7 year survival rates of patients in the training set and validation set, and the KM survival curve (p<0.05) and the AUC value of the ROC curve were used to evaluate the performance of the model;

[0014] 6. Use GO and KEGG enrichment analysis methods to analyze the selected lung cancer prognosis genes: save the differentially expressed gene information in vector or data frame format, use the enrichGO function to perform GO enrichment analysis, and use the ggplot2 function to draw the top 10 results of each function into a bar graph; use the enrichKEGG function to perform KEGG enrichment analysis, and use the ggplot2 function to draw the results of the top 10 pathways into a bar graph;

[0015] 7. Use protein interaction analysis to analyze the selected lung cancer prognosis genes: Use the STRING database and Cytoscape software to explore the protein interaction (PPI) network in which lung cancer prognosis genes participate. Import the selected differentially expressed genes into the STRING database, and use the Cytoscape software to analyze the protein interaction network in which key genes participate.

[0016] Furthermore, the sources obtained in step 1 are: downloading gene expression data and clinical sample data of LUAD and LUSC from the Cancer Genome Atlas (TCGA) database; downloading COVID-19 related gene information from Gene Cards, KEGG, NCBI, and OMIM databases.

[0017] Furthermore, the specific screening criteria of step 2 are: using the statistically significant difference value p<0.05 and the differential expression fold (FC) greater than or equal to 4 times, that is, |log2FC|≥2 as the screening criteria, using the DESeq2 package of R language to screen differentially expressed genes, and drawing a volcano map; then the obtained differentially expressed genes of LUAD and LUSC are intersected with the COVID-19-related genes, and a Venn diagram is drawn.

[0018] Furthermore, the regression model risk score calculation formula of the Cox regression method in step 3 is as follows:

[0019] h(t,X)=h0(t)exp(β1X1+β2X2+…+β n X n ) (Formula 1)

[0020] Where β is the partial regression coefficient of Cox regression, X is the risk factor, h(t,X) represents the risk of an event occurring at time t under the condition that the risk factor X exists, and h0(t) refers to the basic risk when all X are 0. The hazard ratio (HR) refers to the risk ratio of the group exposed to the risk factor to the group not exposed to the risk factor:

[0021]

[0022] In the Cox regression analysis results, HR>1 is considered to increase the risk of death, that is, a risk factor; HR<1 is considered to reduce the risk of death, that is, a protective factor; HR=1 is considered to have no effect on prognosis; for the binary classification problem (survival or death) of univariate Cox regression, X has two values ​​of 1 and 0. The risk degree at this time is:

[0023]

[0024] Combined with clinical sample data, the genes associated with the prognosis of LUAD and LUSC were screened according to the univariate Cox regression method, and forest plots were drawn with 95% confidence intervals.

[0025] Furthermore, the LASSO regression objective function in step 4 is as follows:

[0026] minimize||Y-Xw|| 2 +λ||w||1 (Formula 4)

[0027] Where Y refers to the observed target variable, X is the feature matrix, w is the regression coefficient vector to be estimated, and λ is the parameter that controls the regularization strength. Formula 4 consists of two terms, the first of which is the residual sum of squares in the ordinary least squares method, and the second is the L1 regularization term, also known as the "penalty term"; LASSO regression achieves feature selection and model simplification while fitting data by adjusting λ; when λ is large, smaller coefficients will be compressed to zero, thereby achieving feature selection, and when λ is small, more features will be retained;

[0028] The risk score for each patient was calculated according to the following formula.

[0029]

[0030] where β j is the regression coefficient, x j is the expression level of the gene.

[0031] The beneficial effects of the present invention compared with the prior art are:

[0032] The bioinformatics method and model proposed in this invention explored the relationship between novel coronavirus (COVID-19) infection and lung cancer for the first time, and found biomarkers that are more effective for prognosis. This invention can not only process data, but also identify intricate patterns in the data, with high sensitivity and specificity, and improve the efficiency and accuracy of screening. This technology can be extended to other disease studies and has universal applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] Figure 1 Volcano plot of differentially expressed genes between LUAD (A) and LUSC (B);

[0035] Figure 2 The Venn diagram of cross-differentially expressed genes among LUAD, LUSC, and COVID-19;

[0036] Figure 3 This is the forest plot of LUSC prognostic genes;

[0037] Figure 4 This is the LUAD prognostic gene forest map;

[0038] Figure 5 The number of key genes screened by LASSO (A) and the relationship between L1 regularization and LASSO regression coefficient (B);

[0039] Figure 6 Gene expression analysis of ERO1A (A), AHNAK2 (B), and NTS (C);

[0040] Figure 7 Gene expression analysis of LGR4 (A), P3H4 (B), ABCC2 (C), TRIP13 (D), HMMR (E), KRT14 (F), UPK1B (G), IGF2BP1 (H), and GRIA1 (I);

[0041] Figure 8 Gene expression analysis of PTX3 (A), NEFL (B), SALL1 (C), EFNA2 (D), and CAMP (E);

[0042] Fig. 9 The risk score distribution (A) and patient survival status (B) are shown;

[0043] Fig.10 KM survival curves for the training set (A) and validation set (B);

[0044] Fig.11 ROC curves for the training set (A) and the validation set (B);

[0045] Fig.12 The bar graph is a GO enrichment analysis of 172 intersection genes;

[0046] Fig.13 The bar graph is a GO enrichment analysis of 17 prognostic genes;

[0047] Fig.14 The bar graph is a KEGG enrichment analysis of 172 intersection genes;

[0048] Fig.15 The bar graph is a KEGG enrichment analysis of 17 prognostic genes;

[0049] Fig.16 This is the protein interaction network diagram of 172 differentially expressed genes;

[0050] Fig.17 This is a diagram of the key protein interaction network involving EFNA2. DETAILED DESCRIPTION

[0051] The present invention is described in detail below by specific examples, but the protection scope of the present invention is not limited. Unless otherwise specified, the experimental methods used in the present invention are all conventional methods, and the experimental equipment, materials, reagents, etc. used can be obtained from commercial channels.

[0052] Example 1

[0053] The method of using machine learning to screen cross-differentially expressed genes between lung cancer and novel coronavirus infection and correspondingly screen lung cancer prognostic genes is as follows:

[0054] S1. Acquisition of cross-differentially expressed genes between lung cancer and novel coronavirus: Download gene expression data of LUAD and LUSC and clinical sample data from The Cancer Genome Atlas (TCGA) database; download COVID-19 related gene information from Gene Cards, KEGG, NCBI, and OMIM databases;

[0055] S2. Screening of cross-differentially expressed genes between lung cancer and novel coronavirus infection: With statistically significant difference value p<0.05 and differential expression fold (FC) greater than or equal to 4 times, that is, |log2FC|≥2 as the screening criteria, the DESeq2 package of R language was used to screen differentially expressed genes and draw a volcano map; then the obtained LUAD and LUSC differentially expressed genes were intersected with COVID-19 related genes, and a Venn diagram was drawn; 172 cross-differentially expressed genes of LUAD, LUSC, and COVID-19 were obtained.

[0056] S3. Use Cox regression to screen lung cancer prognostic genes: Combine the clinical sample information of LUAD and LUSC patients, and use the univariate Cox regression method to screen the corresponding prognosis-related genes. The risk score calculation formula of the Cox regression model is as follows:

[0057] h(t,X)=h0(t)exp(β1X1+β2X2+…+β n X n ) (Formula 1)

[0058] Where β is the partial regression coefficient of Cox regression, X is the risk factor, h(t,X) represents the risk of an event occurring at time t under the condition that the risk factor X exists, and h0(t) refers to the basic risk when all X are 0. The hazard ratio (HR) refers to the risk ratio of the group exposed to the risk factor to the group not exposed to the risk factor:

[0059]

[0060] In the Cox regression analysis results, HR>1 is considered to increase the risk of death, that is, a risk factor; HR<1 is considered to reduce the risk of death, that is, a protective factor; HR=1 is considered to have no effect on prognosis

[76] For the binary classification problem (survival or death) of univariate Cox regression, X has two values: 1 and 0. The risk at this time is:

[0061]

[0062] Combined with clinical sample data and using the univariate Cox regression method, genes associated with the prognosis of LUAD and LUSC were screened and forest plots were drawn, with 95% confidence intervals given;

[0063] S4. LASSO regression was used to screen lung cancer prognostic genes: the number of samples in the training set and the validation set (total sample size: 541) was allocated according to 7:3. The LASSO regression objective function is as follows:

[0064] minimize||Y-Xw|| 2 +λ||w|| 1 (Formula 4)

[0065] Where Y refers to the observed target variable, X is the feature matrix, w is the regression coefficient vector to be estimated, and λ is the parameter that controls the strength of regularization. Formula 4 consists of two terms, the first of which is the residual sum of squares in the ordinary least squares method, and the second is the L1 regularization term, also known as the "penalty term". LASSO regression can achieve feature selection and model simplification while fitting data by adjusting λ. When λ is large, smaller coefficients will be compressed to zero, thereby achieving feature selection, and when λ is small, more features will be retained. Therefore, LASSO regression can help us identify features that have an important impact on the target variable. Calculate the risk score for each patient according to the following formula.

[0066]

[0067] where β j is the regression coefficient, x j is the expression level of the gene. 17 lung cancer prognostic genes were obtained using Cox and LASSO regression methods.

[0068] S5. Use KM survival analysis to analyze the selected lung cancer prognosis genes: estimate the survival function using the following formula:

[0069]

[0070] Where S(t) refers to the survival probability of an individual at time t, n refers to the number of time points at which the event occurs, and d irefers to the number of individuals that experience an event at time t; n i It refers to the number of individuals still alive at time t, S(0)=1.

[0071] The KM survival analysis method was used to analyze the 3-7 year survival rates of patients in the training set and validation set. The KM survival curve (p<0.05) and the AUC value of the ROC curve were used to evaluate the performance of the model. The larger the AUC value, the higher the prediction accuracy (the AUC value greater than 0.7 means that the model is considered effective).

[0072] S6. Use GO and KEGG enrichment analysis methods to analyze the screened lung cancer prognosis genes: save the information of 172 differentially expressed genes in vector or data frame format, use the enrichGO function to perform GO enrichment analysis, and use the ggplot2 function to plot the top 10 results of each function into a bar graph; use the enrichKEGG function to perform KEGG enrichment analysis, and use the ggplot2 function to plot the results of the top 10 pathways into a bar graph.

[0073] S7. Protein interaction analysis was used to analyze the selected lung cancer prognosis genes: the protein interaction (PPI) network of key genes was explored using the STRING database and Cytoscape software. The 172 differentially expressed genes screened were imported into the STRING database, and the protein interaction network of key genes was analyzed using the Cytoscape software.

[0074] Figure 1 The distribution of differentially expressed genes is displayed in the form of a volcano plot. The horizontal axis is the logarithm (base 2) of the differential expression fold, where the positive or negative log2 Fold Change value represents the upregulation (red dots in the figure) or downregulation (green dots in the figure) of the gene in the tumor sample. The greater the difference, the more it is distributed toward both ends of the X-axis. The gray area in the figure represents genes with no significant changes. The vertical axis is the p-value after -log10 conversion. The smaller the p-value, the more significant it is, and the more it is distributed upward. Among them, 2597 differentially expressed genes ( Figure 1 A), 3954 differentially expressed genes were screened from the LUSC gene expression data (60660 genes) ( Figure 1 B). The intersection was taken with the genes of COVID-19 (4783 genes), and a total of 172 cross-differentially expressed genes of three lung diseases were obtained ( Figure 2 ).

[0075] Forest map ( Figure 3 , 4) to show the results of univariate Cox regression. The dotted line with HR=1 is used as the central axis. The green dots on the left side of the dotted line are factors that reduce the occurrence of death events, i.e., low-risk genes. The red dots on the right side of the dotted line are factors that promote the occurrence of death events, i.e., high-risk genes. The blue line segments where each point is located are 95% confidence intervals.

[0076] A total of 77 genes related to LUAD prognosis were screened through the calculation of univariate COX regression model, including 23 tumor suppressor genes and 54 tumor promoter genes ( Figure 3 ); 15 genes related to LUSC prognosis, 4 tumor suppressor genes, and 11 tumor promoter genes ( Figure 4 ), among which NTS, CDCA3, CHODL, and MEGF10 were highly expressed in lung squamous cell carcinoma tissues compared with normal tissues, while ABCA3, CD93, MARCO, CXCL2, LPL, ZBTB16, PTX3, C8B, PPBP, F11, and CLEC4M were lowly expressed in lung squamous cell carcinoma tissues.

[0077] A total of 17 key prognostic genes were screened through LASSO regression model calculation ( Figure 5 ), namely ERO1A, AHNAK2, NTS, LGR4, P3H4, ABCC2, TRIP13, HMMR, KRT14, UPK1B, IGF2BP1, GRIA1, PTX3, NEFL, SALL1, EFNA2, CAMP. The expression of 17 key prognostic genes in tumor samples and normal samples was further analyzed using R language ( Figure 6 , Figure 7 , Figure 8 ).

[0078] Among them, ERO1A, AHNAK2, LGR4, P3H4, ABCC2, TRIP13, HMMR, KRT14, IGF2BP1, NEFL, SALL1, and EFNA2 were highly expressed, while NTS, UPK1B, GRIA1, PTX3, and CAMP were lowly expressed in lung adenocarcinoma tissues. A prognostic model was established by combining the regression coefficients.

[0079] risk.score←0.030753725×ERO1A+0.009632997×AHNAK2+0.020734698×NT S+0.05924787×LGR4+0.025156105×P3H4+0.016861383×ABCC2-0.067831582×TRIP13+0.135319151×HMMR+0.025039238×KRT14+0.006908539×UPK1 B+0.0245833×IGF2BP1-0.074275033×GRIA1+0.058295983×PTX3+0.035083611×NEFL+0.039350275×SALL1+0.039424578×EFNA2-0.076919192×CAMP

[0080] As the risk score increases, the number of deaths also increases, that is, the survival rate of patients decreases ( Fig. 9 ).

[0081] In order to further verify the effectiveness of the model, survival analysis was performed on the samples of the training set and the validation set. This study used the KM survival curve to show the survival status of the high-risk and low-risk groups, and p<0.05 was used to determine whether there was a significant difference. The survival curves of the high-risk and low-risk groups in the training set and the validation set did not intersect, and the p values ​​were all less than 0.05 ( Fig.10 ), indicating that the prognostic model can significantly distinguish high-risk and low-risk groups. The 17 key genes screened by the model can be used as prognostic markers for this type of disease. In the training set, the AUC values ​​of the 3-year and 7-year overall survival of patients were 0.732 and 0.739, respectively, and in the validation set, the AUC values ​​of the 3-year and 7-year overall survival were 0.704 and 0.709, respectively. Fig.11 ). The results show that with the increase of time, the prognostic model has achieved good results in both the training set and the validation set samples, that is, the model has certain effectiveness.

[0082] In order to explore the cellular components, biological processes, molecular functions and signaling pathways involved in the 172 differentially expressed genes and 17 key prognostic genes, GO and KEGG enrichment analysis were performed on these genes. GO enrichment results ( Fig.12 , Fig.13) showed that the 172 cross-differentially expressed genes of the three lung diseases were mainly involved in biological processes such as vascular processes in the circulatory system, leukocyte migration, and humoral immune responses. The 17 key prognostic genes were mainly involved in biological processes such as bone remodeling, telencephalon development, synaptic protein complex assembly, and cell differentiation in glomerular development. When p<0.05, ephrin A2 (EFNA2) was involved in more biological processes and cellular components. The KEGG enrichment results showed that the cross-differentially expressed genes were related to coronavirus disease-COVID-19, IL-17 signaling pathways, etc. KEGG enrichment results ( Fig.14 , Fig.15 ) showed that the cross-differentially expressed genes were related to malaria, coronavirus disease-COVID-19, IL-17 signaling pathway, etc.

[0083] PPI Network Results( Fig.16 , 17) showed that EFNA2 protein and TEK protein have a direct interaction relationship. EFNA2 can bind to TEK protein and activate the TEK signaling pathway, thereby affecting the proliferation, migration and angiogenesis of vascular endothelial cells. This interaction plays an important role in regulating vascular development and homeostasis, and may also play an important role in the angiogenesis process of tumors.

[0084] The above-described embodiments are only preferred embodiments of the present invention, but not all feasible embodiments of the present invention. For those skilled in the art, any obvious changes made thereto without departing from the principles and spirit of the present invention should be considered to be included in the scope of protection of the claims of the present invention.

Claims

1. A method for machine learning to screen cross-differentially expressed genes between lung cancer and novel coronavirus infection and correspondingly screen lung cancer prognosis genes, characterized in that: Here are the steps: S1. Obtaining cross-differentially expressed genes between lung cancer and novel coronavirus: Downloading gene expression data and clinical sample data of LUAD and LUSC, and downloading COVID-19 related gene information; S2. Screening of cross-differentially expressed genes between lung cancer and novel coronavirus infection: Using the statistically significant difference value range and differential expression multiple range as screening criteria, screen differentially expressed genes and draw a volcano map. Then, intersect the obtained differentially expressed genes of LUAD and LUSC with the COVID-19-related genes and draw a Venn diagram; S3. Use Cox regression to screen lung cancer prognostic genes: Combine the clinical sample information of LUAD and LUSC patients, and use the univariate Cox regression method to screen the corresponding prognosis-related genes; S4. LASSO regression was used to screen lung cancer prognostic genes: the number of samples in the training set and the validation set was allocated according to a certain ratio, and then LASSO regression was used to screen; S5. Use KM survival analysis to analyze the selected lung cancer prognosis genes: estimate the survival function using the following formula Where S(t) refers to the survival probability of an individual at time t, n refers to the number of time points at which an event occurs, and di refers to the number of individuals at which the event occurs at time t; n i It refers to the number of individuals still alive at time t, S(0) = 1; The KM survival analysis method was used to analyze the 3-7 year survival rates of patients in the training set and validation set, and the KM survival curve was used to set p < 0.05 and the AUC value of the ROC curve to evaluate the performance of the model; S6. Use GO and KEGG enrichment analysis methods to analyze the screened lung cancer prognosis genes: save the differentially expressed gene information in vector or data frame format, use the enrichGO function to perform GO enrichment analysis, and use the ggplot2 function to plot the top 10 results of each function into a bar graph; use the enrichKEGG function to perform KEGG enrichment analysis, and use the ggplot2 function to plot the results of the top 10 pathways into a bar graph; S7. Use protein interaction analysis to analyze the selected lung cancer prognosis genes: Use the STRING database and Cytoscape software to explore the protein interaction network in which lung cancer prognosis genes participate. The selected differentially expressed genes were imported into the STRING database, and the protein interaction network in which key genes participated was analyzed using the Cytoscape software.

2. The method for machine learning screening of cross-differentially expressed genes between lung cancer and novel coronavirus infection and corresponding screening of lung cancer prognosis genes according to claim 1, characterized in that: The sources obtained in step S1 are: downloading the gene expression data and clinical sample data of LUAD and LUSC from the Cancer Genome Atlas database; downloading COVID-19 related gene information from Gene Cards, KEGG, NCBI, and OMIM databases.

3. The method for machine learning screening of cross-differentially expressed genes of lung cancer and novel coronavirus infection and corresponding screening of lung cancer prognosis genes according to claim 1, characterized in that: The specific screening criteria of step S2 are: using the statistically significant difference value p<0.05 and the differential expression multiple, i.e., FC, greater than or equal to 4 times, i.e., |log2FC|≥2 as the screening criteria, using the DESeq2 package of the R language to screen the differentially expressed genes, and drawing a volcano map; then, the obtained differentially expressed genes of LUAD and LUSC are intersected with the COVID-19-related genes, and a Venn diagram is drawn.

4. The method for machine learning screening of cross-differentially expressed genes between lung cancer and novel coronavirus infection and corresponding screening of lung cancer prognosis genes according to claim 1, characterized in that: The regression model risk score calculation formula of the Cox regression method in step S3 is as follows: h(t, X) = h0(t) exp(β1X1 + β2X2 + … + β n X n ) (Equation 1) Where β is the partial regression coefficient of Cox regression, X is the risk factor, h(t,X) represents the risk of an event occurring at time t under the condition that the risk factor X exists, and h0(t) refers to the basic risk when all X are 0. The hazard ratio (HR) refers to the risk ratio of the group exposed to the risk factor to the group not exposed to the risk factor: In the Cox regression analysis results, HR>1 is considered to increase the risk of death, that is, a risk factor; HR<1 is considered to reduce the risk of death, that is, a protective factor; HR=1 is considered to have no effect on prognosis; for the binary classification problem of univariate Cox regression, that is, survival or death, X has two values ​​of 1 and 0. The risk degree at this time is: Combined with clinical sample data, the genes associated with the prognosis of LUAD and LUSC were screened according to the univariate Cox regression method, and forest plots were drawn with 95% confidence intervals.

5. The method for machine learning screening of cross-differentially expressed genes between lung cancer and novel coronavirus infection and corresponding screening of lung cancer prognosis genes according to claim 1, characterized in that: The LASSO regression objective function in step S4 is as follows: minimize||Y-Xw|| 2 +λ||w||1 (Figure 4) Where Y is the observed target variable, X is the feature matrix, and w is the regression coefficient vector to be estimated. λ is a parameter that controls the regularization strength. Formula 4 consists of two terms, the first of which is the residual sum of squares in the ordinary least squares method, and the second is the L1 regularization term, also known as the "penalty term". LASSO regression is adjusted by λ It achieves feature selection and model simplification while fitting the data; λ When λ is large, smaller coefficients will be compressed to zero, thus achieving feature selection, while when λ is small, more features will be retained; The risk score for each patient was calculated according to the following formula. where β j is the regression coefficient, x j is the expression level of the gene.

Images