Colorectal cancer onset risk prediction model based on multi-omics data as well as construction method and application of colorectal cancer onset risk prediction model

Abstract

The invention discloses a colorectal cancer onset risk prediction model based on multi-omics data, and a construction method and application thereof. The construction method comprises the following steps: firstly, constructing a protein score ProS, a non-genetic risk factor score QCancer-S and a multi-gene risk score PRS; and then constructing a combined model according to the ProS, the QCancer-S and the PRS, wherein the combined model is a colorectal cancer onset risk prediction model. The colorectal cancer onset risk prediction model is non-invasive prediction, has good prediction efficiency, has obvious layering ability for colorectal cancer onset risks of people, and deduces colorectal cancer screening recommendation initial ages suitable for people with different risks.

Description

Technical Field

[0001] The present invention relates to the field of biomedical technology, and in particular to a colorectal cancer risk prediction model based on multi-omics data, a construction method thereof, and a use of the colorectal cancer risk prediction model obtained by the construction method of the colorectal cancer risk prediction model based on multi-omics data in predicting the risk of colorectal cancer. Background Art

[0002] Colorectal cancer (CRC) is the most common digestive tract malignancy and the second leading cause of cancer death worldwide. Currently, early diagnosis and treatment of colorectal cancer through population screening remains the most effective way to reduce the burden of colorectal cancer. Although screening methods vary internationally, current screening programs are mostly based on age and family history without fully considering other risk determinants.

[0003] However, increasing evidence suggests that differences in CRC risk among the population are not only attributed to age and family history, but also to genetic and environmental risk factors. In addition, given that colonoscopy (the gold standard) is invasive, time-consuming, and expensive, there is an urgent need to further identify non-invasive early screening and diagnostic biomarkers and customize risk-based personalized screening recommendations to improve screening efficiency.

[0004] As the main bearer of life activities, circulating proteins reflect the overall status of human health and disease, and have become the main source of biomarkers and therapeutic targets. They have the most intrinsic disease prediction potential and are non-invasive and easy to obtain. Some studies have shown a close association between circulating proteins and CRC risk, and suggest that protein markers have the potential to distinguish CRC cases from controls. In addition, genetic factors and established traditional risk factors (such as QCancer-[15yr, colorectal cancer]) also play an important role in the development of CRC and contribute to the risk stratification of CRC.

[0005] However, most of the existing colorectal cancer prediction models are based on case-control designs and use single-omics information (only genetics, traditional risk factors or biomarkers), providing moderate classification and discrimination capabilities, but have not yet solved the problem of how to accurately predict and provide early warning for the future risk of colorectal cancer in the natural population.

[0006] Therefore, in view of the shortcomings of the existing technology, it is necessary to provide a colorectal cancer risk prediction model based on multi-omics data, a construction method thereof, and a use of a colorectal cancer risk prediction model obtained by the construction method of a colorectal cancer risk prediction model based on multi-omics data in predicting the risk of colorectal cancer to solve the shortcomings of the existing technology. Summary of the invention

[0007] One of the purposes of the present invention is to avoid the shortcomings of the prior art and provide a method for constructing a colorectal cancer risk prediction model based on multi-omics data. The colorectal cancer risk prediction model obtained by the method for constructing a colorectal cancer risk prediction model based on multi-omics data has good predictive efficiency and has obvious stratification ability for the risk of colorectal cancer in the population.

[0008] The above-mentioned purpose of the present invention is achieved by the following technical measures:

[0009] A method for constructing a colorectal cancer risk prediction model based on multi-omics data is provided, which is constructed by the following steps:

[0010] S1. Screening out core protein markers associated with colorectal cancer risk through a two-stage strategy consisting of a discovery stage and a validation stage, and then constructing a protein score ProS based on the core protein markers;

[0011] The non-genetic risk factor score QCancer-S was constructed using the traditional risk factor QCancer-15;

[0012] A polygenic risk score (PRS) for colorectal cancer was constructed using a modeling database and a cross-ethnic colorectal cancer genome-wide association summary dataset.

[0013] S2. A joint model is constructed based on the protein score ProS, non-genetic risk factor score QCancer-S and polygenic risk score PRS obtained in S1. The joint model is a colorectal cancer risk prediction model.

[0014] In S1, the protein score ProS is constructed by the following steps:

[0015] A1. Discovery stage: Compared with the blood samples of the control group without colorectal cancer, the proteins whose protein expression is upregulated exceeding the first threshold and the proteins whose protein expression is downregulated exceeding the second threshold are screened in the blood of the colorectal cancer cases to obtain the first screening protein set, wherein the first threshold is Log2FC>0 and FDR<0.05, and the second threshold is Log2FC<0 and FDR<0.05;

[0016] A2, verification stage: Screen out proteins that coexist with the first screening protein set obtained in A1 in the verification database to obtain the second screening protein set;

[0017] A3, screening the proteins in the second screening protein set belonging to the verification database, screening out the successfully verified proteins, wherein the successfully verified proteins are proteins that are consistently associated with colorectal cancer in the discovery stage of A1 and the verification stage of A2 and are statistically significant, as the third screening protein set;

[0018] A4, directly inputting the third screening protein set obtained in A3 into LASSO-Cox regression to obtain the core protein marker;

[0019] A5. The core protein marker is directly input into the Cox model, and then the output of the Cox model is directly input into the "predict" function to obtain the protein score ProS.

[0020] In S1, the construction of the non-genetic risk factor score QCancer-S is specifically to input the traditional risk factor QCancer-15 into the Cox model, and then directly input the output of the Cox model into the "predict" function to obtain the non-genetic risk factor score QCancer-S.

[0021] In S2, the protein score ProS, the non-genetic risk factor score QCancer-S and the polygenic risk score PRS obtained in S1 are directly input into the Cox model to obtain the joint model.

[0022] In S1, the polygenic risk score PRS is constructed by the following steps:

[0023] B1, using the genome-wide SNPs covered in the cross-ethnic colorectal cancer genome-wide association summary dataset as the first dataset;

[0024] B2, screening the first data set obtained in B1 again, screening out SNPs in the first data set that coexist with SNPs in the HapMap3+ data set and SNPs in the modeling database, and obtaining a corresponding second data set;

[0025] B3, based on the second data set obtained in B2, the LDpred2-auto model and the LDpred2-grid model are used to construct PRSs to obtain two PRSs;

[0026] B4. Input the two PRSs into the Cox model respectively, and obtain two corresponding PRS prediction models, which are defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively;

[0027] B5. Calculate the HR value and C-statistic value of the two PRS prediction models obtained in B4 respectively, take the PRS with the largest C-statistic value as the optimal PRS, and use the optimal PRS as the polygenic risk score PRS in S2.

[0028] Preferably, the above non-genetic risk factor score QCancer-S is composed of gender factors, age factors, Townsend deprivation index factors, smoking status factors, drinking status factors, family history of colorectal cancer factors, race factors and previous medical history factors.

[0029] Preferably, the above-mentioned past medical history factors are diabetes factors, ulcerative colitis factors, colorectal polyp factors, breast cancer factors, uterine cancer factors, ovarian cancer factors, cervical cancer factors, lung cancer factors, blood system tumor factors and oral cancer factors.

[0030] Preferably, the core protein markers are IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1 and CR2.

[0031] Preferably, the above-mentioned validation database and modeling database are both prospective natural population cohort UK Biobank UKBB.

[0032] Preferably, the above-mentioned colorectal cancer cases are colorectal cancer cases in people of Chinese descent.

[0033] Preferably, the above-mentioned cross-ethnic colorectal cancer genome-wide association summary dataset is a cross-ethnic CRC meta-GWASs dataset comprising Asian and European ancestry.

[0034] Preferably, the above-mentioned traditional risk factor QCancer-15 is QCancer-15 of the British population.

[0035] The second purpose of the present invention is to avoid the shortcomings of the prior art and provide a colorectal cancer risk prediction model based on multi-omics data. The colorectal cancer risk prediction model based on multi-omics data has good predictive efficiency and has obvious stratification ability for the risk of colorectal cancer in the population.

[0036] The above-mentioned purpose of the present invention is achieved by the following technical measures:

[0037] A colorectal cancer risk prediction model based on multi-omics data is provided, which is represented by formula (1):

[0038] h(t,X)=h0(t)×exp(θ1×QCancer-S+θ2×PRS+θ3×ProS)…Equation (1);

[0039] Among them, h(t,X) is the risk score, h0(t) is the baseline risk of the Cox model for colorectal cancer risk, QCancer-S is the non-genetic risk factor score of the traditional risk factor QCancer-15, ProS is the protein score of the core protein marker, PRS is the polygenic risk score, θ1 is the non-genetic risk factor score coefficient and 0.100<θ1≤0.500, θ2 is the polygenic risk score coefficient and 1.400<θ1≤1.700, θ3 is the protein score coefficient and 0.100<θ3≤0.400.

[0040] Preferably, the above core protein markers are IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1 and CR2.

[0041] Preferably, the above protein score ProS is represented by formula (2):

[0042] ProS = h1 × exp(β ADGRG2 ×X ADGRG2 +β APLP1 ×X APLP1 +β CCN3 ×X CCN3 +β CDON ×X CDON +β COL18A1 ×X COL18A1 +β CR2 ×X CR2 +β IGF2R ×X IGF2R +β IGFBP4 ×X IGFBP4 +β LTBP2 ×X LTBP2 +β PON3 ×X PON3 +β PZP ×X PZP +β SELL ×X SELL +β TFF1 ×X TFF1 +β TNXB ×X TNXB +β WFDC2 ×X WFDC2 ) ……Formula (2);

[0043] Among them, h1 is the baseline risk of the Cox model for colorectal cancer-related protein score, and h1 is 0.800-1.200, β ADGRG2 is the ADGRG2 coefficient, β ADGRG2 0.050~0.250, X ADGRG2is the protein level of ADGRG2, β APLP1 is the APLP1 coefficient, β APLP1 -0.150~-0.001, X APLP1 is the APLP1 protein level, β CCN3 is the CCN3 coefficient, β CCN3 0.001~0.150, X CCN3 is the CCN3 protein level, β CDON is the CDON coefficient, β CDON 0.300~0.500, X CDON is the CDON protein level, β COL18A1 is the COL18A1 coefficient, β COL18A1 0.001~0.150, X COL18A1 is the protein level of COL18A1, β CR2 is the CR2 coefficient, β CR2 0.180~0.380, X CR2 is the CR2 protein level, β IGF2R is the IGF2R coefficient, β IGF2R -0.550~-0.350, X IGF2R is the protein level of IGF2R, β IGFBP4 is the IGFBP4 coefficient, β IGFBP4 -0.300~-0.100, X IGFBP4 is the protein level of IGFBP4, β LTBP2 is the LTBP2 coefficient, β LTBP2 -0.200~-0.001, X LTBP2 is the LTBP2 protein level, β PON3 is the PON3 coefficient, β PON3 0.300~0.500, X PON3 is the protein level of PON3, β PZP is the PZP coefficient, β PZP -0.350~-0.150, X PZP is the PZP protein level, β SELL is the SELL coefficient, β SELL -0.250~-0.050, X SELL is the SELL protein level, β TFF1 is the TFF1 coefficient, β TFF1 -0.250~-0.050, X TFF1 is the protein level of TFF1, β TNXB is the TNXB coefficient, β TNXB -0.600~-0.400, X TNXBis the protein level of TNXB, β WFDC2 is the WFDC2 coefficient, β WFDC2 -0.400~-0.200, X WFDC2 is the WFDC2 protein level.

[0044] Preferably, the non-genetic risk factor score QCancer-S is calculated by formula (3):

[0045] QCancer-S=h2×exp(β1+β2X2+β3X3+β4X4+β5+β6+β7+β8+β9+β 10 +β 11 +β 12 +β 13 +β 14 +β 15 +β 16 +β 17 +β c ) ……Formula (3);

[0046] Among them, h2 is the Cox model baseline risk of non-genetic risk factors related to colorectal cancer and h2 is 0.010-0.030; β1 is the sex coefficient and β1 is 0-0.4500; X2 is age and the unit is year, β2 is the age coefficient and β2 is 0.055-0.100; X3 is the Townsend deprivation index, β3 is the Townsend deprivation index coefficient, and β3 is 0.001-0.100; X4 is the body mass index and the unit is kg / m 2 , β4 is the body mass index coefficient and β4 is 0.001~0.010; β5 is the smoking status coefficient and β5 is -0.300~0.200; β6 is the drinking status coefficient and β6 is -0.090~0.550; β7 is the family history coefficient of colorectal cancer and β7 is 0~0.5000; β c is the racial coefficient, and β c is -14.000~1.200; β8 is the diabetes coefficient and β8 is 0~0.100; β9 is the ulcerative colitis coefficient and β9 is 0~1.000; β 10 is the colorectal polyp coefficient and β 10 -14.500~0; β 11 is the breast cancer coefficient and β 11 0~0.200; β 12 is the uterine cancer coefficient and β 12 0~0.300; β 13 is the ovarian cancer coefficient and β 13 0~0.100; β 14 is the cervical cancer coefficient and β 14 -13.800~0; β 15 is the lung cancer coefficient and β15 0~0.010; β 16 is the hematological tumor coefficient and β 16 0~0.450; β 17 is the oral cancer coefficient and β 17 It is 0~0.450.

[0047] The polygenic risk score PRS in formula (1) is obtained by the following steps:

[0048] B1, using the genome-wide SNPs covered by the cross-ethnic colorectal cancer genome-wide association summary dataset as the first dataset;

[0049] B2, screening the first data set obtained in B1 again, screening out SNPs in the first data set that coexist with SNPs in the HapMap3+ data set and SNPs in the modeling database, and obtaining a corresponding second data set;

[0050] B3, based on the second data set obtained in B2, the LDpred2-auto model and the LDpred2-grid model are used to construct PRSs to obtain two PRSs;

[0051] B4. Input the two PRSs into the Cox model respectively, and obtain two corresponding PRS prediction models, which are defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively;

[0052] B5. Calculate the HR value and C-statistic value of the two PRS prediction models obtained in B4 respectively, take the PRS with the largest C-statistic value as the optimal PRS, and use the optimal PRS as the polygenic risk score PRS in formula (1).

[0053] Preferably, the above colorectal cancer risk prediction model is represented by formula (4):

[0054] h(t,

[0055] Preferably, the protein score ProS is represented by formula (5):

[0056] ProS = 1.001 × exp (0.15158 × X ADGRG2 +(-0.06612)×X APLP1 +0.05061×X CCN3 +0.40746×X CDON+0.08616×X COL18A1 +0.28257×X CR2 +(-0.47348)×X IGF2R +(-0.19442)×X IGFBP4 +(-0.1128)×X LTBP2 +0.39655×X PON3 +(-0.24582)×X PZP +(-0.15248)×X SELL +(-0.15048)×X TFF1 +(-0.4726)×X TNXB +(-0.28659)×X WFDC2 ) ……Formula (5).

[0057] Preferably, the non-genetic risk factor score QCancer-S is calculated by formula (6):

[0058] QCancer-S=0.018×exp(β1+β2X2+β3X3+β4X4+β5+β6+β7+β8+β9+β 10 +β 11 +β 12 +β 13 +β 14 +β 15 +β 16 +β 17 +β c )

[0059] ...Formula (6);

[0060] In β1, when the person is female, β1 is 0, and when the person is male, β1 is 0.3752. β2 is 0.0685. β3 is 0.0172. β4 is 0.0056. In β5, when the person is not smoking, β5 is 0; when the person smoked before and is not smoking currently, β5 is 0.1231; when the person is currently smoking and the amount of cigarettes smoked is ≤10 cigarettes / d, β5 is -0.1505; when the person is currently smoking and the amount of cigarettes smoked is 10 cigarettes / d<the amount of cigarettes smoked is ≤19 cigarettes / d, β5 is -0.2253; when the person is currently smoking and the amount of cigarettes smoked is>19 cigarettes / d, β5 is 0.0803; when the smoking status is missing, β5 is 0.0248. In β6, when the person does not drink alcohol, β6 is 0; when the amount of alcohol consumed is ≤8g alcohol / d, β6 is -0.0273; when 8g alcohol / d<drinking amount ≤24g alcohol / d, β6 is -0.0443; when 24g alcohol / d<drinking amount ≤56g alcohol / d, β6 is 0.0658; when 56g alcohol / d<drinking amount ≤80g alcohol / d, β6 is 0.1696; when the amount of alcohol consumed is >80g alcohol / d, β6 is 0.4874; when the drinking situation is missing, β6 is -0.0649. In β7, when there is no family history of colorectal cancer, β7 is 0; when there is a family history of colorectal cancer, β7 is 0.4132; when the family history of colorectal cancer is missing, β7 is 0.1511. In β8, β8 is 0 when there is no diabetes, and β8 is 0.0654 when there is diabetes. In β9, β9 is 0 when there is no ulcerative colitis, and β9 is 0.7213 when there is ulcerative colitis. 10 In the absence of colorectal polyps, 10 0, when there are colon polyps 10 is -14.2300. 11 In the absence of breast cancer, 11 is 0, when there is breast cancer, 11 is 0.1482. 12 In the absence of uterine cancer, 12 is 0, when there is uterine cancer 12 is 0.2712. 13 In the absence of ovarian cancer, 13 is 0, when there is ovarian cancer 13 is 0.04049. 14 In the absence of cervical cancer, 14 is 0, when there is cervical cancer 14 is -13.7900. 15 In the absence of lung cancer, 15 is 0, when there is lung cancer, 15 is 0.0018. 16 In the absence of hematologic tumors, 16 0, when there is a blood system tumor16 is 0.3576. 17 In the absence of oral cancer, 17 is 0, when there is oral cancer 17 is 0.3760. c When race is white, β c is 0; when the race is Indian, β c is -1.3000~-1.0000; when the ethnicity is Pakistani, β c 0.3000~0.5000, when the ethnicity is Bangladeshi c -14.0000~-13.5000; when the race is other Asians, β c 0.2000~0.4500; When the race is Caribbean, β c is -0.6500~-0.4500; when the race is African American, β c 0.1000~0.3000; when the race is Chinese, β c 0.8000~1.3000; when the race is other, β c It is -0.5500 to -0.3500.

[0061] The C-statistic of each PRS prediction model of the present invention is obtained by the following steps:

[0062] C1. Randomly divide the modeling database into training data set and validation data set in a ratio of 7:3;

[0063] C2, randomly divide the training data set obtained in C1 into 5 subsets, arrange the 5 subsets from front to back, select one subset as the test set in turn, and combine the remaining subsets as the training set for training, and perform 5 trainings in total;

[0064] C3. Use the roc function to calculate the C-statistic of the current test set during each training, and get a total of 5 C-statistics;

[0065] C4. Select the largest C-statistic among the five C-statistics as the C-statistic value of the PRS prediction model.

[0066] Preferably, the HR value of each PRS is estimated by a CPH regression equation.

[0067] The third purpose of the present invention is to avoid the shortcomings of the prior art and provide a method for constructing a colorectal cancer risk prediction model based on multi-omics data to predict the risk of colorectal cancer. This method has good predictive efficiency and has obvious stratification ability for the risk of colorectal cancer in the population.

[0068] The present invention is a colorectal cancer risk prediction model based on multi-omics data, a construction method and use thereof, wherein the construction method is constructed by the following steps: S1, core protein markers related to colorectal cancer risk are screened out through a two-stage strategy consisting of a discovery stage and a verification stage, and then a protein score ProS is constructed based on the core protein marker; a non-genetic risk factor score QCancer-S is constructed through the traditional risk factor QCancer-15; a polygenic risk score PRS for colorectal cancer is constructed through a modeling database and a cross-ethnic colorectal cancer genome-wide association summary data set; S2, a joint model is constructed based on the protein score ProS, non-genetic risk factor score QCancer-S and polygenic risk score PRS obtained in S1, and the joint model is a colorectal cancer risk prediction model. The colorectal cancer risk prediction model of the present invention is a non-invasive prediction with good predictive efficiency, has obvious stratification ability for the risk of colorectal cancer in the population, and derives the recommended starting age for colorectal cancer screening suitable for different risk groups. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The present invention is further described with reference to the accompanying drawings, but the contents in the accompanying drawings do not constitute any limitation to the present invention.

[0070] Figure 1 Results of differential expression analysis (discovery phase) and Cox proportional hazards regression analysis (validation phase) of 15 plasma proteins used to construct the protein risk score for colorectal cancer (Pro-S).

[0071] Figure 2 Receiver operating characteristic (ROC) curves of QCancer-S alone, PRS alone, ProS alone and the combined model of the present invention.

[0072] Figure 3 Calibration curves for QCancer-S alone, PRS alone, ProS alone and the combined model of the present invention predict the probability of CRC.

[0073] Figure 4 Decision curves of QCancer-S alone and the combined model of the present invention.

[0074] Figure 5 This is a nomogram for predicting the probability of being free of colorectal cancer within 5, 10 and 15 years using the combined model of the present invention.

[0075] Figure 6 Figure 2 shows the Kaplan-Meier cumulative incidence curves for different risk groups of QCancer-15 alone.

[0076] Figure 7 Kaplan-Meier cumulative incidence curves for different risk groups for individual polygenic risk scores (PRS).

[0077] Figure 8 Kaplan-Meier cumulative incidence curves among different risk groups for individual protein scores ProS.

[0078] Fig. 9 It is the Kaplan-Meier cumulative incidence curve between different risk groups of the combined risk score (ComS) of the joint model of the present invention.

[0079] Fig.10 Age-specific 10-year cumulative risk of colorectal cancer in different risk groups using a single polygenic risk score (PRS).

[0080] Fig.11 Age-specific 10-year cumulative risk of colorectal cancer in different risk groups for individual protein scores ProS.

[0081] Fig.12 The combined risk score (ComS) of the combined model of the present invention is the age-specific 10-year cumulative risk of colorectal cancer in different risk groups. DETAILED DESCRIPTION

[0082] The technical solution of the present invention is further described in conjunction with the following embodiments.

[0083] Example 1

[0084] A method for constructing a colorectal cancer risk prediction model based on multi-omics data is constructed by the following steps:

[0085] S1. Through a two-stage strategy consisting of a discovery phase and a validation phase, core protein markers associated with colorectal cancer risk were screened out, and then a protein score ProS was constructed based on the core protein markers; a non-genetic risk factor score QCancer-S was constructed through traditional risk factors QCancer-15; a polygenic risk score PRS for colorectal cancer was constructed through a modeling database and a cross-ethnic colorectal cancer genome-wide association summary data set;

[0086] S2. A joint model is constructed based on the protein score ProS, non-genetic risk factor score QCancer-S and polygenic risk score PRS obtained in S1. The joint model is a colorectal cancer risk prediction model.

[0087] In S1, the protein score ProS is constructed by the following steps:

[0088] A1. Discovery stage: Compared with the blood samples of the control group without colorectal cancer, the proteins whose protein expression is upregulated exceeding the first threshold and the proteins whose protein expression is downregulated exceeding the second threshold are screened in the blood of the colorectal cancer cases to obtain the first screening protein set. The first threshold is Log2FC>0 and FDR<0.05, and the second threshold is Log2FC<0 and FDR<0.05;

[0089] A2, verification stage: Screen out proteins that coexist with the first screening protein set obtained in A1 in the verification database to obtain the second screening protein set;

[0090] A3, screening the proteins in the second screening protein set belonging to the verification database, screening out the successfully verified proteins, wherein the successfully verified proteins are proteins that are consistently associated with colorectal cancer in the discovery stage of A1 and the verification stage of A2 and are statistically significant, as the third screening protein set;

[0091] A4, directly input the third screening protein set obtained in A3 into LASSO-Cox regression to obtain core protein markers;

[0092] A5. Input the core protein markers directly into the Cox model, and then input the output of the Cox model directly into the "predict" function to obtain the protein score ProS.

[0093] In S1, the construction of the non-genetic risk factor score QCancer-S is specifically to input the traditional risk factor QCancer-15 into the Cox model, and then directly input the output of the Cox model into the “predict” function to obtain the non-genetic risk factor score QCancer-S;

[0094] In S2, the protein score ProS, non-genetic risk factor score QCancer-S and polygenic risk score PRS obtained in S1 were directly input into the Cox model to obtain a joint model.

[0095] In S1, the polygenic risk score PRS is constructed by the following steps:

[0096] B1, using the genome-wide SNPs covered in the Trans-ethnic Colorectal Cancer Genome-wide Association Summary Dataset as the first dataset;

[0097] B2, screening the first data set obtained in B1 again, screening out SNPs in the first data set that coexist with SNPs in the HapMap3+ data set and SNPs in the modeling database, and obtaining a corresponding second data set;

[0098] B3, based on the second data set obtained in B2, the LDpred2-auto model and the LDpred2-grid model are used to construct PRSs to obtain two PRSs;

[0099] B4. Input the two PRSs into the Cox model respectively, and obtain two corresponding PRS prediction models, which are defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively;

[0100] B5. Calculate the HR value and C-statistic value of the two PRS prediction models obtained in B4 respectively, take the PRS with the largest C-statistic value as the optimal PRS, and use the optimal PRS as the polygenic risk score PRS in S2.

[0101] It should be noted that the Cox model of the present invention is a Cox proportional hazard (CPH) model. The Cox model, LASSO-Cox, and "predict" function of the present invention are all common models, regression methods, and functions in the art. You only need to directly input the input into these tools to get the corresponding output. The input and output of these models, regression methods, and functions are also conventional operations in the art.

[0102] The non-genetic risk factor score QCancer-S is composed of gender, age, Townsend deprivation index, smoking status, drinking status, family history of colorectal cancer, race and previous medical history. The previous medical history factors include diabetes, ulcerative colitis, colorectal polyps, breast cancer, uterine cancer, ovarian cancer, cervical cancer, lung cancer, blood system tumors and oral cancer; the core protein markers are IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1 and CR2.

[0103] The validation database and subsequent modeling database for the protein screening process of the present invention are both the prospective natural population cohort UK Biobank UKBB; the colorectal cancer cases and the control samples without colorectal cancer in the discovery phase are colorectal cancer cases and the control samples without colorectal cancer in the Chinese descent population; the colorectal cancer whole genome association summary data are cross-ethnic CRC meta-GWASs data including Asian and European descents; the traditional risk factor QCancer-15 is QCancer-15 of the British population.

[0104] The validation database and modeling database of the present invention can both be the prospective natural population cohort UK Biobank UKBB; or they can be databases from other regions. The colorectal cancer cases of the present invention are colorectal cancer cases of the Chinese descent population, and colorectal cancer cases from other regions can also be selected as needed. The cross-ethnic colorectal cancer genome-wide association summary database of the present invention is a cross-ethnic CRC meta-GWASs database, and other ancestry colorectal cancer genome-wide association summary databases can also be selected. The traditional risk factor QCancer-15 of the present invention is QCancer-15 of the British population, and traditional risk factors from other regions can also be selected.

[0105] The gene sequence of the core protein marker of the present invention is as follows, wherein the underlined region is the coding region:

[0106] 1. IGFBP:

[0107]

[0108]

[0109] 2. WFDC2:

[0110]

[0111] 3. TFF1:

[0112]

[0113] 4. LTBP2:

[0114]

[0115]

[0116] 5. COL18A1:

[0117]

[0118]

[0119] 6. CCN3:

[0120]

[0121] 7、SELL:

[0122]

[0123] 8、PON3:

[0124]

[0125] 9、ADGRG2:

[0126]

[0127]

[0128] 10、IGF2R:

[0129]

[0130]

[0131]

[0132] 11、PZP:

[0133]

[0134] 12、CDON:

[0135]

[0136]

[0137] 13、TNXB:

[0138]

[0139]

[0140]

[0141] 14、APLP1:

[0142]

[0143] 15、CR2:

[0144]

[0145]

[0146] The colorectal cancer risk prediction model obtained by the method of constructing a colorectal cancer risk prediction model based on multi-omics data is a non-invasive prediction with good predictive efficiency. The area under the ROC curve (AUC) is 0.79 (0.75-0.84), and it has obvious stratification ability for the colorectal cancer risk of the population (Log-rank P<0.001), and the recommended starting age for colorectal cancer screening suitable for different risk groups is derived.

[0147] Example 2

[0148] A colorectal cancer risk prediction model based on multi-omics data, the colorectal cancer risk prediction model constructed by the method for constructing a colorectal cancer risk prediction model based on multi-omics data in Example 1 is represented by formula (1):

[0149] h(t,X)=h0(t)×exp(θ1×QCancer-S+θ2×PRS+θ3×ProS)…Equation (1);

[0150] Wherein, h(t,X) is the risk score (also called risk function), h0(t) is the baseline risk of the Cox model for colorectal cancer risk and h0(t) is 0.400-0.800, QCancer-S is the non-genetic risk factor score of the traditional risk factor QCancer-15, ProS is the protein score of the core protein marker, PRS is the polygenic risk score, θ1 is the non-genetic risk factor score coefficient and 0.100<θ1≤0.500, θ2 is the polygenic risk score coefficient and 1.400<θ1≤1.700, θ3 is the protein score coefficient and 0.100<θ3≤0.400.

[0151] It should be noted that since the COX model is a semi-parametric model, only the exp(βX) part is parameterized, while h0(t) is the baseline risk function, which represents the risk function without the influence of covariates. This function is related to time t, and this part is non-parametric.

[0152] The core protein markers are IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1 and CR2. The non-genetic risk factor score QCancer-S is composed of gender, age, Townsend deprivation index, smoking status, drinking status, family history of colorectal cancer, race and previous medical history. The previous medical history factors are diabetes, ulcerative colitis, colorectal polyps, breast cancer, uterine cancer, ovarian cancer, cervical cancer, lung cancer, blood system tumors and oral cancer.

[0153] The protein score ProS of the present invention is calculated by formula (2):

[0154] ProS = h1 × exp(β ADGRG2 ×X ADGRG2 +β APLP1 ×X APLP1 +β CCN3 ×X CCN3 +β CDON ×X CDON +β COL18A1 ×X COL18A1 +β CR2 ×X CR2 +β IGF2R ×X IGF2R +β IGFBP4 ×X IGFBP4 +β LTBP2 ×X LTBP2 +β PON3 ×X PON3 +β PZP ×X PZP +β SELL ×X SELL +β TFF1 ×X TFF1 +β TNXB ×X TNXB +β WFDC2 ×X WFDC2 ) ……Formula (2);

[0155] Among them, h1 is the baseline risk of the Cox model for colorectal cancer-related protein score, and h1 is 0.800-1.200, β ADGRG2 is the ADGRG2 coefficient, β ADGRG2 0.050~0.250, X ADGRG2 is the protein level of ADGRG2, β APLP1 is the APLP1 coefficient, β APLP1 -0.150~-0.001, X APLP1 is the APLP1 protein level, β CCN3 is the CCN3 coefficient, β CCN3 0.001~0.150, X CCN3 is the CCN3 protein level, β CDON is the CDON coefficient, β CDON 0.300~0.500, X CDON is the CDON protein level, β COL18A1 is the COL18A1 coefficient, β COL18A1 0.001~0.150, X COL18A1 is the protein level of COL18A1, β CR2is the CR2 coefficient, β CR2 0.180~0.380, X CR2 is the CR2 protein level, β IGF2R is the IGF2R coefficient, β IGF2R -0.550~-0.350, X IGF2R is the protein level of IGF2R, β IGFBP4 is the IGFBP4 coefficient, β IGFBP4 -0.300~-0.100, X IGFBP4 is the protein level of IGFBP4, β LTBP2 is the LTBP2 coefficient, β LTBP2 -0.200~-0.001, X LTBP2 is the LTBP2 protein level, β PON3 is the PON3 coefficient, β PON3 0.300~0.500, X PON3 is the protein level of PON3, β PZP is the PZP coefficient, β PZP -0.350~-0.150, X PZP is the PZP protein level, β SELL is the SELL coefficient, β SELL -0.250~-0.050, X SELL is the SELL protein level, β TFF1 is the TFF1 coefficient, β TFF1 -0.250~-0.050, X TFF1 is the protein level of TFF1, β TNXB is the TNXB coefficient, β TNXB -0.600~-0.400, X TNXB is the protein level of TNXB, β WFDC2 is the WFDC2 coefficient, β WFDC2 -0.400~-0.200, X WFDC2 is the WFDC2 protein level.

[0156] It should be noted that UKBBolink is a relative quantification, and the protein level is specifically the protein value, which is NPX (Normalized protein expression).

[0157] The non-genetic risk factor score QCancer-S is calculated by formula (3):

[0158] QCancer-S=h2×exp(β1+β2X2+β3X3+β4X4+β5+β6+β7+β8+β9+β 10+β 11 +β 12 +β 13 +β 14 +β 15 +β 16 +β 17 +β c ) ……Formula (3);

[0159] h2 is the baseline risk of the Cox model for non-genetic risk factors related to colorectal cancer and h2 is 0.010-0.030; β1 is the sex coefficient and β1 is 0-0.4500; X2 is age and the unit is year, β2 is the age coefficient and β2 is 0.055-0.100; X3 is the Townsend deprivation index, β3 is the Townsend deprivation index coefficient, and β3 is 0.001-0.100; X4 is the body mass index and the unit is kg / m 2 , β4 is the body mass index coefficient and β4 is 0.001~0.010; β5 is the smoking status coefficient and β5 is -0.300~0.200; β6 is the drinking status coefficient and β6 is -0.090~0.550; β7 is the family history coefficient of colorectal cancer and β7 is 0~0.5000; β c is the racial coefficient, and β c is -14.000~1.200; β8 is the diabetes coefficient and β8 is 0~0.100; β9 is the ulcerative colitis coefficient and β9 is 0~1.000; β 10 is the colorectal polyp coefficient and β 10 -14.500~0; β 11 is the breast cancer coefficient and β 11 0~0.200; β 12 is the uterine cancer coefficient and β 12 0~0.300; β 13 is the ovarian cancer coefficient and β 13 0~0.100; β 14 is the cervical cancer coefficient and β 14 -13.800~0; β 15 is the lung cancer coefficient and β 15 0~0.010; β 16 is the hematological tumor coefficient and β 16 0~0.450; β 17 is the oral cancer coefficient and β 17 It is 0~0.450.

[0160] The polygenic risk score (PRS) for colorectal cancer is obtained by the following steps:

[0161] B1. Use the genome-wide SNPs covered by the Trans-Ethnic Colorectal Cancer Genome-Wide Association Summary Dataset as the first dataset.

[0162] B2, screening the first data set obtained in B1 again, screening out the intersection of the first data set with the SNPs in the HapMap3+ data set and the SNPs in the modeling database, and obtaining the corresponding second data set;

[0163] B3, based on the second data set obtained in B2, the LDpred2-auto model and LDpred2-grid model were used to construct the colorectal cancer polygenic risk score, and two PRSs were obtained;

[0164] B4. Input the two PRSs into the Cox model respectively, and obtain two corresponding PRS prediction models, which are defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively;

[0165] B5. Calculate the HR value and C-statistic value of the two PRS prediction models obtained in B4 respectively, take the PRS with the largest C-statistic value as the optimal PRS, and use the optimal PRS as the polygenic risk score PRS in formula (1).

[0166] The C-statistic of each PRS prediction model in B5 is obtained by the following steps:

[0167] C1. Randomly divide the modeling database into training data set and validation data set in a ratio of 7:3;

[0168] C2, randomly divide the training data set obtained in C1 into 5 subsets, arrange the 5 subsets from front to back, select one subset as the test set in turn, and combine the remaining subsets as the training set for training, and perform 5 trainings in total;

[0169] C3. Use the roc function to calculate the C-statistic of the current test set during each training, and get a total of 5 C-statistics;

[0170] C4. Select the largest C-statistic among the five C-statistics as the C-statistic value of the PRS prediction model.

[0171] The HR value of each PRS was estimated by the CPH regression equation.

[0172] This colorectal cancer risk prediction model based on multi-omics data is a non-invasive prediction with good predictive efficiency. The area under the ROC curve (AUC) is 0.78 (0.72-0.84). It has obvious stratification ability for the risk of colorectal cancer in the population (Log-rank P<0.001), and derives the recommended starting age for colorectal cancer screening suitable for different risk groups.

[0173] Example 3

[0174] A colorectal cancer risk prediction model based on multi-omics data, the other features of which are the same as those of Example 2, except that: the colorectal cancer risk prediction model is represented by formula (1):

[0175] h(t,

[0176] It should be noted that since the COX model is a semi-parametric model, only the exp(βX) part is parameterized, and h0(t) is the baseline risk function, which represents the risk function without the influence of covariates. This function is related to time t, and this part is non-parametric. When the maximum follow-up time of the current data population is used for estimation, h0(t) is 0.609.

[0177] The protein score ProS is calculated by formula (5):

[0178] ProS = 1.001 × exp (0.15158 × X ADGRG2 +(-0.06612)×X APLP1 +0.05061×X CCN3 +0.40746×X CDON +0.08616×X COL18A1 +0.28257×X CR2 +(-0.47348)×X IGF2R +(-0.19442)×X IGFBP4 +(-0.1128)×X LTBP2 +0.39655×X PON3 +(-0.24582)×X PZP +(-0.15248)×X SELL +(-0.15048)×X TFF1 +(-0.4726)×X TNXB +(-0.28659)×X WFDC2 ) ……Formula (5).

[0179] The non-genetic risk factor score QCancer-S is calculated by formula (3):

[0180] QCancer-S=0.018×exp(β1+β2X2+β3X3+β4X4+β5+β6+β7+β8+β9+β 10 +β 11 +β 12 +β 13 +β 14 +β 15 +β 16 +β 17 +β c )……Formula (6).

[0181] In β1, when the person is female, β1 is 0, and when the person is male, β1 is 0.3752. β2 is 0.0685. β3 is 0.0172. β4 is 0.0056. In β5, when the person is not smoking, β5 is 0; when the person smoked before and is not smoking currently, β5 is 0.1231; when the person is currently smoking and the amount of cigarettes smoked is ≤10 cigarettes / d, β5 is -0.1505; when the person is currently smoking and the amount of cigarettes smoked is 10 cigarettes / d<the amount of cigarettes smoked is ≤19 cigarettes / d, β5 is -0.2253; when the person is currently smoking and the amount of cigarettes smoked is>19 cigarettes / d, β5 is 0.0803; when the smoking status is missing, β5 is 0.0248. In β6, when the person does not drink alcohol, β6 is 0; when the amount of alcohol consumed is ≤8g alcohol / d, β6 is -0.0273; when 8g alcohol / d<drinking amount ≤24g alcohol / d, β6 is -0.0443; when 24g alcohol / d<drinking amount ≤56g alcohol / d, β6 is 0.0658; when 56g alcohol / d<drinking amount ≤80g alcohol / d, β6 is 0.1696; when the amount of alcohol consumed is >80g alcohol / d, β6 is 0.4874; when the drinking situation is missing, β6 is -0.0649. In β7, when there is no family history of colorectal cancer, β7 is 0; when there is a family history of colorectal cancer, β7 is 0.4132; when the family history of colorectal cancer is missing, β7 is 0.1511. In β8, β8 is 0 when there is no diabetes, and β8 is 0.0654 when there is diabetes. In β9, β9 is 0 when there is no ulcerative colitis, and β9 is 0.7213 when there is ulcerative colitis. 10 In the absence of colorectal polyps, 10 0, when there are colon polyps 10 is -14.2300. 11 In the absence of breast cancer, 11 is 0, when there is breast cancer, 11 is 0.1482. 12 In the absence of uterine cancer, 12 is 0, when there is uterine cancer12 is 0.2712. 13 In the absence of ovarian cancer, 13 is 0, when there is ovarian cancer 13 is 0.04049. 14 In the absence of cervical cancer, 14 is 0, when there is cervical cancer 14 is -13.7900. 15 In the absence of lung cancer, 15 is 0, when there is lung cancer, 15 is 0.0018. 16 In the absence of hematologic tumors, 16 0, when there is a blood system tumor 16 is 0.3576. 17 In the absence of oral cancer, 17 is 0, when there is oral cancer 17 is 0.3760. c When race is white, β c is 0; when the race is Indian, β c is -1.3000~-1.0000; when the ethnicity is Pakistani, β c 0.3000~0.5000, when the ethnicity is Bangladeshi c -14.0000~-13.5000; when the race is other Asians, β c 0.2000~0.4500; When the race is Caribbean, β c is -0.6500~-0.4500; when the race is African American, β c 0.1000~0.3000; when the race is Chinese, β c 0.8000~1.3000; when the race is other, β c It is -0.5500 to -0.3500.

[0182] The different coefficients in the non-genetic risk factor score QCancer-S are summarized as follows, as shown in Table 1:

[0183] Table 1. Coefficients of non-genetic risk factors (QCancer-15) and QCancer-S scores

[0184]

[0185]

[0186] Verification Example

[0187] This invention is based on the research of Zhejiang Natural Science Foundation Outstanding Youth Project (LR22H260001), Zhejiang Key R&D Program (2020C03002) and Healthy Zhejiang Million People Cohort (K-20230085). The verification process of this verification example is as follows:

[0188] 1. Research subjects

[0189] The present invention adopts a two-stage strategy, namely a discovery stage and a verification stage.

[0190] In the discovery phase, 150 newly diagnosed colorectal cancer cases and 50 age- and sex-matched controls (age: 44-89 years) from the Second Affiliated Hospital of Zhejiang University School of Medicine were enrolled in the study. Ethical approval was obtained from the Ethics Committee of Zhejiang University School of Medicine (2023-1190).

[0191] In the validation phase, new CRC cases and controls from the prospective natural population cohort UK Biobank (UKBB) were included. UKBB completed baseline recruitment from 2006 to 2010, enrolled approximately 500,000 people, and has been followed up to date. Among them, 50,000 people underwent proteomic measurements. After excluding individuals with baseline colorectal cancer and missing genotype data, a total of 52,231 people (age: 39-70 years) (731 new CRC cases and 51,500 controls) were included in the validation phase. UKBB has been approved by the North West Multicenter Research Ethics Committee (11 / NW / 0382). All participants provided informed consent.

[0192] 2. Definition of colorectal cancer cases

[0193] The primary outcome was the diagnosis of CRC, defined as a malignancy of the colon and rectum, diagnosed by clinicians and pathologists in the discovery cohort and defined using the International Classification of Disease (ICD, ICD-9:153, 154.0, 154.1; ICD-10:C18-C20) in the validation database (UKBB) through linkage with hospital data and cancer and death registries. Follow-up began on the date the participant enrolled in the UKBB and ended on the date of a CRC event, death, loss to follow-up, or the end of available registry follow-up (March 31, 2021), whichever came first.

[0194] 3. Plasma proteomics assay

[0195] Blood samples were collected from subjects at enrollment, and plasma samples were prepared and stored at -80°C. Protein measurements were determined using TMT6plex quantitative proteomics (i.e., colorectal cancer cases and controls, also known as the discovery cohort) and Olink (UKBB, validation cohort) proteomics technology.

[0196] In colorectal cancer cases and controls (discovery cohort), protein measurements were relatively quantified and normalized based on the median ratio of the quantified peptides corresponding to the proteins and the median normalization method. After quality control, a total of 606 proteins were identified and included in subsequent analysis. In UKBB, the relative abundance of 2923 proteins was quantified, and the measurements were normalized by Log2 transformation. Protein values ​​below the limit of detection (LOD) were replaced by the LOD divided by the square root of 2. Missing protein measurements were interpolated with the mean. For the differentially expressed proteins between CRC cases and controls identified in the discovery phase, their relationship with the risk of CRC was further verified based on the UKBB prospective cohort design.

[0197] 4. Statistical analysis

[0198] 4.1 Identification of colorectal cancer-related protein biomarkers and construction of protein scores

[0199] A two-stage strategy was used to identify CRC-associated proteomic signatures. In the discovery stage based on a case-control design, proteome-wide differential expression analysis was performed to identify proteins that were upregulated (Log2-fold change, Log2FC>0) or downregulated (Log2FC<0) between CRC cases and controls.

[0200] The false discovery rate (FDR) was used for multiple testing correction, and FDR<0.05 was used as the significance level. For the differentially expressed proteins identified in the discovery phase (FDR<0.05), the Cox proportional hazards (CPH) model was further used to evaluate their relationship with the risk of CRC in the validation database (UKBB). Proteins with the same effect direction as the discovery set and significant P value (P<0.05) were considered to be successfully validated.

[0201] In the validation population, LASSO-Cox regression was used to further screen the core protein features associated with CRC for the proteins that passed the two-stage test, and a total of 15 proteins were retained.

[0202] The present invention uses the predict function to construct a weighted protein score for 15 selected proteins as follows:

[0203] ProS = 1.001 × exp (0.15158 × X ADGRG2 +(-0.06612)×X APLP1 +0.05061×X CCN3 +0.40746×XCDON +0.08616×X COL18A1 +0.28257×X CR2 +(-0.47348)×X IGF2R +(-0.19442)×X IGFBP4 +(-0.1128)×X LTBP2 +0.39655×X PON3 +(-0.24582)×X PZP +(-0.15248)×X SELL +(-0.15048)×X TFF1 +(-0.4726)×X TNXB +(-0.28659)×X WFDC2 ).

[0204] 4.2 Definition of QCancer-15 and Construction of QCancer-15 Score

[0205] In the validation database (UKBB), QCancer-15 (non-genetic risk factors) predictors were coded based on baseline information. These predictors included: age (at enrollment), sex, Townsend deprivation index, race, body mass index (BMI), smoking status, alcohol intake, family history of CRC, and past medical history (diabetes, ulcerative colitis, colorectal polyps, breast cancer, uterine cancer, ovarian cancer, cervical cancer, lung cancer, hematological tumors, oral cancer).

[0206] These predictors were used to construct a weighted QCancer-15 score, such as QCancer-S of Example 3, using the "predict" function.

[0207] 4.3 Gene sequencing and construction of polygenic risk score

[0208] Genotyping was performed using the UK BiLEVE Axiom array and the UK Biobank Axiom array. Effect sizes for associations of each single nucleotide polymorphism (SNP) with CRC risk were obtained based on pooled data from the largest transethnic CRC meta-GWASs to date (100,204 cases and 154,587 controls) of Asian and European ancestry.

[0209] The R package “bigsnpr” was used to construct the PRS using LDpred2, which is based on a Bayesian approach and calculates the average posterior effect of all variants (SNPs) based on the prior (i.e., the effect size and significance of SNPs in published GWAS) and takes into account linkage disequilibrium (LD) between SNPs.

[0210] A total of 22,203,197 SNPs were included in the cross-ethnic CRC meta-GWASs dataset. These SNPs were then intersected with the HapMap3+ dataset and the SNPs in the modeling population (UKBB). The HapMap3+ dataset covers common genetic polymorphisms in the human genome (mainly single nucleotide polymorphisms, SNPs) and contains more than one million SNPs. Finally, 1,325,278 SNPs that passed quality control after the intersection were included and PRS was constructed in the UKBB population using two different LDpred2 models (i.e., LDpred2-auto and LDpred2-grid).

[0211] Specifically, this verification example uses the above-mentioned CRC meta-GWASs summary data as input data to provide the effect value Beta and P value of SNPs on CRC risk. The number of SNPs (first data set) is shown in Table 2.

[0212] Table 2. Number of SNPs in the first dataset

[0213] Number of SNPs in the first dataset 22203197

[0214] Then the above SNPs were further limited to those included in the HapMap3+ dataset and the current UKBB population and passed quality control (the second dataset). The number of SNPs in the second dataset is shown in Table 3. Finally, a total of 1,325,278 SNPs met the conditions and were included in the construction of PRS.

[0215] Table 3. Number of SNPs in the second dataset

[0216] Number of SNPs in the second dataset 1325278

[0217] PRS were constructed using the LDpred2-auto and LDpred2-grid models, respectively, for a total of 2 PRSs.

[0218] Then the two PRSs were input into the Cox model respectively to obtain two PRS prediction models, which were defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively.

[0219] Table 4. Two PRS prediction models

[0220]

[0221]

[0222] The discrimination of the prediction model was evaluated using the area under the ROC curve (AUC, also known as C-statistic). The C-statistic (95% confidence interval, CI) was estimated using 5-fold cross validation. Specifically, the C-statistic (95% confidence interval, CI) for each PRS prediction model was obtained by the following steps:

[0223] C1. Randomly divide the modeling database into training data set and validation data set in a ratio of 7:3;

[0224] C2, randomly divide the training data set obtained in C1 into 5 subsets, arrange the 5 subsets from front to back, select one subset as the test set in turn, and combine the remaining subsets as the training set for training, and perform 5 trainings in total;

[0225] C3. Use the roc function to calculate the C-statistic of the current test set during each training, and get a total of 5 C-statistics;

[0226] C4. Select the largest C-statistic among the five C-statistics as the C-statistic value of the PRS prediction model.

[0227] The C-statistic (95% CI) of the two PRS prediction models of the present invention are shown in Table 5:

[0228] Table 5. C-statistics of the two PRS prediction models

[0229] PRS C-statistic (95% CI) LDpred2-auto-PRS 0.64(0.58,0.70) LDpred2-grid-PRS 0.74(0.69,0.80)

[0230] 4.4 Development and evaluation of colorectal cancer risk prediction models

[0231] The modeling database was randomly divided into a training data set and a validation data set at a ratio of 7:3 using the R package caret. In the training data set, the Cox model (CPH model) was used to establish a comparative model for CRC risk prediction based on QCancer-S, two PRSs or ProS, and four comparative models were obtained, of which two PRSs corresponded to two PRS prediction models. Then, the PRS with the best performance was selected from the two PRS prediction models, and it was combined with QCancer-S and ProS to construct a joint model to predict the risk of colorectal cancer, wherein the joint model of the present invention was expressed as h(t,X)=0.609×exp(0.32865QCancer-S+1.58435PRS+0.21920ProS), and the prediction efficiency of the joint model in the validation data set was further evaluated.

[0232] The present invention adopts the Bootstrap method, repeats the sampling 500 times, and compares the performance differences of different models. The "riskRegression" software package is used to draw a calibration curve to describe the consistency between the observed event rate and the risk predicted based on the CPH model. Using the R package "rms", a parametric survival model based on the joint model is first constructed, and then a nomogram is drawn, setting the prediction time points to 5 years, 10 years and 15 years, respectively predicting the probability of being free of colorectal cancer in 5 years, 10 years and 15 years.

[0233] To evaluate the potential clinical utility of the prediction model for recommended CRC interventions (such as screening), decision curve analysis was performed using the "ggDCA" package to calculate the net benefit (NB) obtained by selecting subjects for colonoscopy screening based only on the non-genetic risk factor score QCancer-S or based on the combined model of the present invention.

[0234] NB = ((true positive) / n)-((false positive) / n)*(Pt / (1-Pt)), where n is the total number of participants and Pt is the risk (or probability) threshold. The combined model of the present invention was compared with a model based on the non-genetic risk factor score QCancer-S alone, and models assuming no intervention and full intervention.

[0235] 4.5 Hazard ratio estimation and risk stratification analysis

[0236] The CPH regression equation was used to estimate the hazard ratio (HR) and 95% CI for CRC incidence per standard deviation (SD) increase in QCancer-S, PRS and ProS. Then, the single non-genetic risk factor score QCancer-S, the single polygenic risk score PRS, the single protein score ProS and the comprehensive risk score of the combined model of the present invention (hereinafter referred to as ComS) were divided into quartiles (Q1-lowest, Q2, Q3, Q4-highest) according to the population distribution. Further, referring to the 10-year cumulative risk of colorectal cancer in the general population, the participants in the Q1 and Q2 groups were classified as low-scoring groups because their risks were lower than those of the general population, the participants in the Q2 group were classified as medium-scoring groups because their risks were similar to those of the general population, and the participants in the Q4 group were classified as high-scoring groups because their risks were higher than those of the general population. The population was divided into three groups (low, medium and high). The HR (95% CI) of the three categorical variables was estimated with the medium quantile group as a reference. The Kaplan-Meier curve and Log-rank test were used to evaluate the risk stratification value of the risk score for CRC incidence.

[0237] 4.6 Risk Advance Period (RAP) Analysis

[0238] In order to convert the HRs contributed by PRS, ProS and ComS into the number of years later or earlier that individuals in the low or high PRS, ProS and ComS groups reach a risk equivalent to that of the reference group (i.e., individuals with medium PRS, ProS or ComS levels), RAP analysis was further performed. Since age information itself is contained in QCancer-S and dominates, RAP cannot be estimated for QCancer-S. Using a multivariate CPH model, the model included age, PRS (tertiles) and ProS (tertiles), and included sex and the first 10 genetic principal components (PCs) as covariates. The point estimate of RAP was calculated by dividing the regression coefficient of PRS or ProS by the regression coefficient of age. When estimating the RAP of ComS, the model included age, ComS (tertiles), sex and the first 10 PCs. The 95% CI of RAP was estimated by the Bootstrap method, and the sampling was repeated 500 times. 4.7 Estimation of risk-adapted screening start age

[0239] In order to further estimate the risk-adapted screening start age for different risk groups, the age-specific 10-year cumulative risk was further calculated. 10-year cumulative risk = 1-exp (-10-year cumulative incidence), where the 10-year cumulative incidence is equal to the sum of the age-specific annual incidence rates of each age group in the following 10 years. The age-specific annual incidence rate is calculated by dividing the number of cases at each age by the person-years of that age. Given that the current guidelines recommend a starting age of 50 for screening of general-risk adults, the risk-adapted screening start age is defined as the age at which individuals with a specific CRC risk reach a 10-year cumulative risk level similar to that of the 50-year-old general population. All statistical analyses were performed using R4.2.2 software.

[0240] 5. Results

[0241] 5.1 Proteomic signature identification of colorectal cancer

[0242] In the discovery phase, there were significant differences in the levels of 421 plasma proteins in CRC patients compared with the control group, of which 243 proteins were upregulated (Log2FC>0, FDR<0.05) and 178 proteins were downregulated (Log2FC<0, FDR<0.05).

[0243] In the validation database (UKBB), 731 participants were diagnosed with new CRC during a median follow-up of 13.05 years (IQR, 12.38-13.75). 253 of the 421 proteins were also measured in UKBB, of which 88 were successfully validated, including 75 proteins positively associated with the risk of CRC (HR>1, P<0.05) and 13 proteins negatively associated with the risk of CRC (HR<1, P<0.05).

[0244] 5.2. Construction of risk score

[0245] Among the 88 proteins identified in the two stages, LASSO-Cox regression further screened 15 core protein markers associated with CRC risk (IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1, CR2) and constructed ProS, such as Figure 1 .in Figure 1 In the data, FDR is the false discovery rate and HR is the hazard ratio.

[0246] from Figure 1 It can be seen that the 15 core plasma protein markers used to construct the protein risk score showed significant statistical associations with the risk of colorectal cancer and incidence in the discovery phase (case-control design) and validation phase (independent prospective cohort population), with the effect direction consistent.

[0247] The ProS constructed by the present invention (HR: 1.24 / SD increment, 95% CI: 1.21-1.28, P = 1.85 × 10 -48 ) and QCancer-S (HR: 1.55 / SD increment, 95% CI: 1.48-1.62, P = 1.51 × 10 -72 ) was significantly positively correlated with the risk of CRC.

[0248] Two PRS were constructed using the LDpred2-auto model and the LDpred2-grid model, and all PRS were positively correlated with the incidence of CRC.

[0249] The HR values ​​of the two PRS were estimated using the CPH regression equation. Specifically, the CPH regression equation was used to estimate the hazard ratio (HR) and 95% CI for CRC incidence for each standard deviation (SD) increase in the two PRS prediction models.

[0250] The HR (95% CI) values ​​of the two PRSs are shown in Table 6.

[0251] Table 6. HR values ​​of the two PRS prediction models

[0252] PRS HR (95% CI) LDpred2-auto-PRS 1.46(1.36,1.57) LDpred2-grid-PRS 2.31(2.15,2.48)

[0253] The CPH regression equation was also used to estimate the hazard ratio (HR) and 95% CI for CRC incidence per standard deviation (SD) increase in QCancer-S and ProS, respectively.

[0254] It can be seen from Tables 5 and 6 that LDpred2-grid-PRS has the largest HR value and the largest C-statistic value, so LDpred2-grid-PRS is the optimal PRS.

[0255] 5.3 Development and evaluation of prediction models

[0256] In the training dataset of a prospective cohort population (UKBB), CRC incidence risk prediction models based on a single non-genetic risk factor score QCancer-S, a single polygenic risk score PRS, and a single protein score ProS were constructed as control models.

[0257] Among them, the C-statistic (95% CI) of the model based on the single non-genetic risk factor score QCancer-S and the single protein score ProS were 0.71 (0.66, 0.76) and 0.66 (0.61, 0.71), respectively. The LDpred2-grid-PRS was integrated with QCancer-S and ProS to construct the joint model of the present invention. The C-statistic (95% CI) of the joint model reached 0.79 (0.75, 0.84), which was significantly improved (P<0.05) compared with the single QCancer-S model. Figure 2 In the validation data set, the prediction performance of the combined model was still significantly better than that of the single QCancer-S model (P<0.05), with the C-statistic (95% CI) of the combined model being 0.75 (0.72, 0.79) and the C-statistic (95% CI) of the QCancer-S model being 0.69 (0.66, 0.72). The calibration curve showed good consistency between the observed rate and the model prediction value, as shown in Figure 2. Figure 3 Decision curve analysis showed that the combined model had a greater net benefit than the non-genetic risk factor score QCancer-S alone, as Figure 4 .exist Figure 4 The median follow-up time (13.05 years) was used for calculation, where the gray line in the net benefit curve indicates no intervention, and the black line indicates that all people have been intervened. The joint model of the present invention was then used to evaluate the overall CRC risk of the participants, and a nomogram was used to display it, such as Figure 5 .

[0258] from Figures 2 to 5It can be seen that among the four colorectal cancer (CRC) risk prediction models, namely, the model based on a single non-genetic risk factor score (QCancer-S), a single polygenic risk score (PRS), a single protein score (ProS) and the combined model of the present invention, the combined model of the present invention has the highest effectiveness in predicting the risk of colorectal cancer, and at the same time has better calibration and higher net benefit.

[0259] 5.4 Colorectal cancer risk stratification

[0260] The population was divided into quartiles (Q1-lowest, Q2, Q3, Q4-highest) according to the individual non-genetic risk factor score, the individual polygenic risk score PRS, the individual protein score ProS, and the combined score ComS of the joint model. Further, referring to the 10-year cumulative risk of colorectal cancer in the general population, participants in groups Q1 and Q2 had lower risks than the general population and were classified as the low score group, participants in groups Q2 had risks similar to the general population and were classified as the medium score group, and participants in groups Q4 had higher risks than the general population and were classified as the high score group. The population was divided into three groups (low, medium, and high), as shown in Figures 6 to 9 .

[0261] Compared with the medium score group, the risk of CRC in the high QCancer-S group increased by 1.91 times (HR: 1.91, 95% CI: 1.60-2.29), the risk of CRC in the high PRS group increased by 2.71 times (HR: 2.71, 95% CI: 2.24-3.27), the risk of CRC in the high ProS group increased by 1.67 times (HR: 1.67, 95% CI: 1.40-1.99), and the risk of CRC in the high ComS group increased by 3.27 times (HR: 3.27, 95% CI: 2.71-3.96).

[0262] Figures 6 to 9 The Kaplan-Meier curves based on the four risk score groups of low, medium and high were respectively displayed, indicating that the four scores can significantly stratify CRC risk (Log-rank P<0.0001), among which the combined score ComS of the joint model of the present invention has the strongest risk stratification ability.

[0263] 5.5 Risk Advance Period (RAP) for Colorectal Cancer in Different Risk Groups

[0264] Compared with participants with a medium PRS level, participants in the high PRS group reached equivalent CRC risk approximately 17.9 years earlier (RAP: 17.87, 95% CI: 12.91, 22.83), while participants in the low PRS group reached equivalent CRC risk approximately 8.9 years later (RAP: -8.93, 95% CI: -13.34, -4.52).

[0265] Compared with the medium ProS group, participants in the high ProS group reached equivalent CRC risk approximately 4.7 years earlier (RAP: 4.67, 95% CI: 1.08, 8.27), while participants in the low ProS group reached equivalent CRC risk approximately 4.4 years later (RAP: -4.42, 95% CI: -8.17, -0.66).

[0266] Compared with the medium ComS group, participants in the high ComS group even reached the same CRC risk about 27.5 years earlier (RAP: 27.52, 95% CI: 16.66, 38.37), while participants in the low ComS group reached the same CRC risk about 15.2 years later (RAP: -15.15, 95% CI: -23.78, -6.52).

[0267] 5.6 Recommended starting age for risk-adapted colorectal cancer screening

[0268] The age-specific 10-year cumulative risk was then calculated to estimate the recommended starting age for individualized screening for different risk groups.

[0269] Taking the average risk level of the general population at age 50 as a reference, for a single polygenic risk score PRS, the age at which the 10-year cumulative CRC risk of participants in the low, medium, and high PRS groups reached the reference level was >60, 53, and 41 years old, respectively. Fig.10 For the individual protein scores, the age at which the low, medium, and high ProS groups reached the reference level was 57, 47, and 46 years, respectively. Fig.11 After integrating ProS, PRS and QCancer-S, the combined model of the present invention showed that the age of participants in the low, medium and high groups reaching the reference level was >60 years old, 48 years old and <40 years old, respectively. Fig.12 .

[0270] From Figures 6 to 12 As can be seen in the age-specific 10-year cumulative risk curve, the dashed line represents the risk at age 50 (the age at which current guidelines recommend screening for average-risk adults). Figures 6 to 12 The results showed that the four risk scores, namely, the single non-genetic risk factor score QCancer-S, the single polygenic risk score PRS, the single protein score ProS and the combined risk score ComS of the combined model of the present invention, all showed a certain stratification ability for the risk of colorectal cancer (CRC). Among them, the ComS of the combined model of the present invention had the strongest risk stratification ability, and derived the recommended starting age for risk-adapted colorectal cancer screening.

[0271] In summary, the combined model of the present invention, which is composed of a protein score constructed from 15 core protein markers, a PRS constructed from genetic risk factors, and a QCancer-S constructed from traditional risk factors QCancer-15, can effectively predict the risk of colorectal cancer, improve the individualized risk stratification of colorectal cancer and the recommendation for the starting age for screening, and contribute to risk warning, screening decisions, and early diagnosis and treatment of colorectal cancer.

[0272] Example 4

[0273] A colorectal cancer risk prediction model obtained by the construction method of Example 1, or the use of the colorectal cancer risk prediction model of Example 2 or Example 3 in predicting the risk of colorectal cancer.

[0274] The colorectal cancer risk prediction model constructed by the present invention is used to predict the risk of colorectal cancer. The colorectal cancer risk prediction model is a non-invasive prediction and has good predictive efficiency. The area under the ROC curve (AUC) is 0.79 (0.75, 0.84) (training data set) and 0.75 (0.72, 0.79) (validation data set). It has obvious stratification ability for the risk of colorectal cancer in the population (Log-rank P<0.001), and the recommended starting age for colorectal cancer screening suitable for different risk groups is derived.

[0275] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for constructing a colorectal cancer risk prediction model based on multi-omics data, characterized in that: Constructed by the following steps: S1. Screening out core protein markers associated with colorectal cancer risk through a two-stage strategy consisting of a discovery stage and a validation stage, and then constructing a protein score ProS based on the core protein markers; The non-genetic risk factor score QCancer-S was constructed using the traditional risk factor QCancer-15; A polygenic risk score (PRS) for colorectal cancer was constructed using a modeling database and a cross-ethnic colorectal cancer genome-wide association summary dataset. S2. A joint model is constructed based on the protein score ProS, non-genetic risk factor score QCancer-S and polygenic risk score PRS obtained in S1. The joint model is a colorectal cancer risk prediction model.

2. The method for constructing a colorectal cancer risk prediction model based on multi-omics data according to claim 1, characterized in that: In S1, the protein score ProS is constructed by the following steps: A1. Discovery stage: Compared with the blood samples of the control group without colorectal cancer, the proteins whose protein expression is upregulated exceeding the first threshold and the proteins whose protein expression is downregulated exceeding the second threshold are screened in the blood of the colorectal cancer cases to obtain the first screening protein set, wherein the first threshold is Log2FC>0 and FDR<0.05, and the second threshold is Log2FC<0 and FDR<0.05; A2, verification stage: Screen out proteins that coexist with the first screening protein set obtained in A1 in the verification database to obtain the second screening protein set; A3, screening the proteins in the second screening protein set belonging to the verification database, screening out the successfully verified proteins, wherein the successfully verified proteins are proteins that are consistently associated with colorectal cancer in the discovery stage of A1 and the verification stage of A2 and are statistically significant, as the third screening protein set; A4, directly inputting the third screening protein set obtained in A3 into LASSO-Cox regression to obtain the core protein marker; A5. The core protein marker is directly input into the Cox model, and then the output of the Cox model is directly input into the "predict" function to obtain the protein score ProS.

3. The method for constructing a colorectal cancer risk prediction model based on multi-omics data according to claim 2, characterized in that: In S1, the construction of the non-genetic risk factor score QCancer-S is specifically to input the traditional risk factor QCancer-15 into the Cox model, and then directly input the output of the Cox model into the "predict" function to obtain the non-genetic risk factor score QCancer-S.

4. The method for constructing a colorectal cancer risk prediction model based on multi-omics data according to claim 3, characterized in that: In S2, the protein score ProS, the non-genetic risk factor score QCancer-S and the polygenic risk score PRS obtained in S1 are directly input into the Cox model to obtain the joint model.

5. The method for constructing a colorectal cancer risk prediction model based on multi-omics data according to claim 4, characterized in that: In S1, the polygenic risk score PRS is constructed by the following steps: B1, using the genome-wide SNPs covered in the cross-ethnic colorectal cancer genome-wide association summary dataset as the first dataset; B2, screening the first data set obtained in B1 again, screening out SNPs in the first data set that coexist with SNPs in the HapMap3+ data set and SNPs in the modeling database, and obtaining a corresponding second data set; B3, based on the second data set obtained in B2, the LDpred2-auto model and the LDpred2-grid model are used to construct PRS, and two PRSs are obtained; B4. Input the two PRSs into the Cox model respectively, and obtain two corresponding PRS prediction models, which are defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively; B5. Calculate the HR value and C-statistic value of the two PRS prediction models obtained in B4 respectively, take the PRS with the largest C-statistic value as the optimal PRS, and use the optimal PRS as the polygenic risk score PRS in S2.

6. The method for constructing a colorectal cancer risk prediction model based on multi-omics data according to claim 5, characterized in that: The non-genetic risk factor score QCancer-S is composed of gender factors, age factors, Townsend deprivation index factors, smoking status factors, drinking status factors, family history of colorectal cancer factors, ethnic factors and previous medical history factors; The past medical history factors are diabetes, ulcerative colitis, colorectal polyps, breast cancer, uterine cancer, ovarian cancer, cervical cancer, lung cancer, blood system tumors and oral cancer; The core protein markers are IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1 and CR2.

7. The method for constructing a colorectal cancer risk prediction model based on multi-omics data according to claim 6, characterized in that: The validation database and modeling database are both prospective natural population cohort UK Biobank UKBB; The colorectal cancer cases are colorectal cancer cases of people of Chinese descent; The cross-ethnic colorectal cancer genome-wide association summary data set is a cross-ethnic CRCmeta-GWASs data containing Asian and European ancestry; The traditional risk factor QCancer-15 is QCancer-15 of the British population.

8. A colorectal cancer risk prediction model based on multi-omics data, characterized in that: The colorectal cancer risk prediction model is expressed by formula (1): h(t,X)=h0(t)×exp(θ1×QCancer-S+θ2×PRS+θ3×ProS)…Equation (1); Wherein, h(t,X) is the risk score, h0(t) is the baseline risk of the Cox model for colorectal cancer, QCancer-S is the non-genetic risk factor score of the traditional risk factor QCancer-15, ProS is the protein score of the core protein marker, PRS is the polygenic risk score, θ1 is the non-genetic risk factor score coefficient and 0.100<θ1≤0.500, θ2 is the polygenic risk score coefficient and 1.400<θ1≤1.700, θ3 is the protein score coefficient and 0.100<θ3≤0.400; The core protein markers are IGFBP4, WFDC2, TFF1, LTBP2, COL18A1, CCN3, SELL, PON3, ADGRG2, IGF2R, PZP, CDON, TNXB, APLP1 and CR2; The protein score ProS is represented by formula (2): ProS=h1×exp(β ADGRG2 ×X ADGRG2 +b APLP1 ×X APLP1 +b CCN3 ×X CCN3 +b CDON ×X CDON +b COL18A1 ×X COL18A1 +b CR2 ×X CR2 +b IGF2R ×X IGF2R +b IGFBP4 × X IGFBP4 +β LTBP2 ×X LTBP2 +β PON3 ×X PON3 +β PZP ×X PZP +β SELL ×X SELL +β TFF1 ×X TFF1 +β TNXB ×X TNXB +β WFDC2 ×X WFDC2 )... Equation (2); Among them, h1 is the baseline risk of the Cox model for colorectal cancer-related protein score, and h1 is 0.800-1.200, β ADGRG2 is the ADGRG2 coefficient, β ADGRG2 0.050~0.250, X ADGRG2 is the protein level of ADGRG2, β APLP1 is the APLP1 coefficient, β APLP1 -0.150~-0.001, X APLP1 is the APLP1 protein level, β CCN3 is the CCN3 coefficient, β CCN3 0.001~0.150, X CCN3 is the CCN3 protein level, β CDON is the CDON coefficient, β CDON 0.300~0.500, X CDON is the CDON protein level, β COL18A1 is the COL18A1 coefficient, β COL18A1 0.001~0.150, X COL18A1 is the protein level of COL18A1, β CR2 is the CR2 coefficient, β CR2 0.180~0.380, X CR2 is the CR2 protein level, β IGF2R is the IGF2R coefficient, β IGF2R -0.550~-0.350, X IGF2R is the protein level of IGF2R, β IGFBP4 is the IGFBP4 coefficient, β IGFBP4 -0.300~-0.100, X IGFBP4 is the protein level of IGFBP4, β LTBP2 is the LTBP2 coefficient, β LTBP2 -0.200~-0.001, X LTBP2 is the LTBP2 protein level, β PON3 is the PON3 coefficient, β PON3 0.300~0.500, X PON3 is the protein level of PON3, β PZP is the PZP coefficient, β PZP -0.350~-0.150, X PZP is the PZP protein level, β SELL is the SELL coefficient, β SELL -0.250~-0.050, X SELL is the SELL protein level, β TFF1 is the TFF1 coefficient, β TFF1 -0.250~-0.050, X TFF1 is the protein level of TFF1, β TNXB is the TNXB coefficient, β TNXB -0.600~-0.400, X TNXB is the protein level of TNXB, β WFDC2 is the WFDC2 coefficient, β WFDC2 -0.400~-0.200, X WFDC2 is the protein level of WFDC2; The non-genetic risk factor score QCancer-S is calculated by formula (3): QCancer-S = h2×exp(β1 + β2X2 + β3X3 + β4X4 + β5 + β6 + β7 + β8 + β9 + β 10 +β 11 +β 12 +β 13 +β 14 +β 15 +β 16 +β 17 +β c )……Equation (3); Among them, h2 is the Cox model baseline risk of non-genetic risk factors related to colorectal cancer and h2 is 0.010-0.030; β1 is the sex coefficient and β1 is 0-0.4500; X2 is age and the unit is year, β2 is the age coefficient and β2 is 0.055-0.100; X3 is the Townsend deprivation index, β3 is the Townsend deprivation index coefficient, and β3 is 0.001-0.100; X4 is the body mass index and the unit is kg / m 2 , β4 is the body mass index coefficient and β4 is 0.001~0.010; β5 is the smoking status coefficient and β5 is -0.300~0.200; β6 is the drinking status coefficient and β6 is -0.090~0.550; β7 is the family history coefficient of colorectal cancer and β7 is 0~0.5000; β c is the racial coefficient, and β c is -14.000~1.200; β8 is the diabetes coefficient and β8 is 0~0.100; β9 is the ulcerative colitis coefficient and β9 is 0~1.000; β 10 is the colorectal polyp coefficient and β 10 -14.500~0; β 11 is the breast cancer coefficient and β 11 0~0.200; β 12 is the uterine cancer coefficient and β 12 0~0.300; β 13 is the ovarian cancer coefficient and β 13 0~0.100; β 14 is the cervical cancer coefficient and β 14 -13.800~0; β 15 is the lung cancer coefficient and β 15 0~0.010; β 16 is the hematological tumor coefficient and β 16 0~0.450; β 17 is the oral cancer coefficient and β 17 0~0.450; The polygenic risk score PRS in formula (1) is obtained by the following steps: B1, using the genome-wide SNPs covered by the cross-ethnic colorectal cancer genome-wide association summary dataset as the first dataset; B2, screening the first data set obtained in B1 again, screening out SNPs in the first data set that coexist with SNPs in the HapMap3+ data set and SNPs in the modeling database, and obtaining a corresponding second data set; B3, based on the second data set obtained in B2, the LDpred2-auto model and the LDpred2-grid model are used to construct PRS, and two PRSs are obtained; B4. Input the two PRSs into the Cox model respectively, and obtain two corresponding PRS prediction models, which are defined as LDpred2-auto-PRS and LDpred2-grid-PRS respectively; B5. Calculate the HR value and C-statistic value of the two PRS prediction models obtained in B4 respectively, take the PRS with the largest C-statistic value as the optimal PRS, and use the optimal PRS as the polygenic risk score PRS in formula (1).

9. The colorectal cancer risk prediction model based on multi-omics data according to claim 8, characterized in that: The colorectal cancer risk prediction model is expressed by formula (4): h(t, The protein score ProS is expressed by formula (5): ProS=1.001×exp(0.15158×X ADGRG2 +(-0.06612)×X APLP1 +0.05061×X CCN3 +0.40746×X CDON +0.08616×X COL18A1 +0.28257×X CR2 +(- 0.47348)×X IGF2R +(-0.19442)×X IGFBP4 +(-0.1128)×X LTBP2 +0.39655×X PON3 +(-0.24582)×X PZP +(-0.15248)×X SELL +(-0.15048)×X TFF1 +(-0.4726)×X TNXB +(-0.28659)×X WFDC2 )……Formula (5); The non-genetic risk factor score QCancer-S is calculated by formula (6): QCancer-S = 0.018×exp(β1 + β2X2 + β3X3 + β4X4 + β5 + β6 + β7 + β8 + β9 + β 10 +β 11 +β 12 +β 13 +β 14 +β 15 +β 16 +β 17 +β c )……Equation (6); In β1, when the person is female, β1 is 0, and when the person is male, β1 is 0.3752. β2 is 0.0685. β3 is 0.0172. β4 is 0.0056. In β5, when the person is not smoking, β5 is 0; when the person smoked before and is not smoking currently, β5 is 0.1231; when the person is currently smoking and the amount of cigarettes smoked is ≤10 cigarettes / d, β5 is -0.1505; when the person is currently smoking and the amount of cigarettes smoked is 10 cigarettes / d<the amount of cigarettes smoked is ≤19 cigarettes / d, β5 is -0.2253; when the person is currently smoking and the amount of cigarettes smoked is>19 cigarettes / d, β5 is 0.0803; when the smoking status is missing, β5 is 0.0248. In β6, when the person does not drink alcohol, β6 is 0; when the amount of alcohol consumed is ≤8g alcohol / d, β6 is -0.0273; when 8g alcohol / d<drinking amount ≤24g alcohol / d, β6 is -0.0443; when 24g alcohol / d<drinking amount ≤56g alcohol / d, β6 is 0.0658; when 56g alcohol / d<drinking amount ≤80g alcohol / d, β6 is 0.1696; when the amount of alcohol consumed is >80g alcohol / d, β6 is 0.4874; when the drinking situation is missing, β6 is -0.0649. In β7, when there is no family history of colorectal cancer, β7 is 0; when there is a family history of colorectal cancer, β7 is 0.4132; when the family history of colorectal cancer is missing, β7 is 0.1511. In β8, β8 is 0 when there is no diabetes, and β8 is 0.0654 when there is diabetes. In β9, β9 is 0 when there is no ulcerative colitis, and β9 is 0.7213 when there is ulcerative colitis. 10 In the absence of colorectal polyps, 10 0, when there are colon polyps 10 is -14.2300. 11 In the absence of breast cancer, 11 is 0, when there is breast cancer, 11 is 0.1482. 12 In the absence of uterine cancer, 12 is 0, when there is uterine cancer 12 is 0.2712. 13 In the absence of ovarian cancer, 13 is 0, when there is ovarian cancer 13 is 0.04049. 14 In the absence of cervical cancer, 14 is 0, when there is cervical cancer 14 is -13.7900. 15 In the absence of lung cancer, 15 is 0, when there is lung cancer, 15 is 0.0018. 16 In the absence of hematologic tumors, 16 0, when there is a blood system tumor 16 is 0.3576. 17 In the absence of oral cancer, 17 is 0, when there is oral cancer 17 is 0.3760. c When race is white, β c is 0; when the race is Indian, β c is -1.3000~-1.0000; when the ethnicity is Pakistani, β c 0.3000~0.5000, when the ethnicity is Bangladeshi c -14.0000~-13.5000; when the race is other Asians, β c 0.2000~0.4500; When the race is Caribbean, β c is -0.6500~-0.4500; when the race is African American, β c 0.1000~0.3000; when the race is Chinese, β c 0.8000~1.3000; when the race is other, β c It is -0.5500 to -0.3500.

10. Use of a colorectal cancer risk prediction model obtained by the method for constructing a colorectal cancer risk prediction model based on multi-omics data as described in any one of claims 1 to 7 in predicting the risk of colorectal cancer.

Images