Multi-omics non-invasive intestinal cancer risk assessment method based on pks + intestinal bacteria, cancer driving genes and methylated molecular clocks

Abstract

The invention discloses a multi-omics non-invasive intestinal cancer risk assessment method based on pks + intestinal bacteria, cancer driver genes and methylated molecular clocks, which comprises the steps of excrement DNA sample extraction, primer design and screening, triple specificity verification, non-invasive detection and multi-omics quantitative scoring model, and detection of pks + escherichia coli in excrement. The pks + Escherichia coli can synthesize cytotoxin colicin, induce DNA double-chain breakage and chromosome instability in host cells, cause cell cycle arrest and finally cause cell death, key information can be provided for cancer prevention, diagnosis and precise treatment, and the application has the advantages that the application is wide in application range and high in practicability. In addition, deep understanding of the key cancer driver gene of the intestinal cancer is helpful for studying the occurrence and development mechanism of the intestinal cancer, a new therapeutic target can be provided for the intestinal cancer, new possibilities are provided for early discovery, risk assessment and monitoring of the intestinal cancer, the screening accuracy and sensitivity are further improved, and the screening method is suitable for popularization and application. And the method has the advantage of higher accuracy in risk assessment of intestinal cancer.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of non-invasive colorectal cancer risk assessment, in particular to a pks + A multi-omics non-invasive colorectal cancer risk assessment method based on gut bacteria, cancer driver genes and methylation molecular clock. BACKGROUND

[0002] The existing detection techniques for colorectal cancer mainly include the following methods, which are mainly divided into two types: invasive and non-invasive:

[0003] Colonoscopy: including anoscopy, sigmoidoscopy or fiberoptic colonoscopy, which can directly observe the inside of the intestinal tract, find lesions and perform tissue biopsy, has the advantages of high sensitivity, can directly view the inside of the colon, and has the ability to find early lesions, is the most simple, safe and effective method for finding intestinal tumors and precancerous lesions at present, and most early colorectal cancer patients can be found and diagnosed by endoscopy, but it is an invasive examination, which may cause certain discomfort and complication risk, requires bowel preparation before examination, requires anesthesia or sedation, involves the highest cost and resources;

[0004] Imaging examination: barium enema examination can observe the rigidity of the colon wall, disappearance of the folds, presence of filling defects and small niche, but has little diagnostic value for rectal cancer, ultrasonography and CT examination can help understand the depth of invasion and lymph node metastasis of colorectal cancer, and can also indicate whether there is abdominal implantation metastasis, whether it invades adjacent tissues or organs, or whether there are liver and lung metastases, magnetic resonance examination can assess the depth of tumor invasion in the intestinal wall, and has important value for the diagnosis and staging of middle and low rectal cancer, imaging examination has the advantages of non-invasiveness and can provide detailed images of the inside and outside of the intestinal tract, but may be affected by intestinal gas and feces, affecting imaging quality;

[0005] Fecal occult blood test (FOBT): this test detects blood components in feces, during the growth of cancer cells, a small amount of blood may seep into the feces, its working principle is to detect potential bleeding in fecal samples, which can be used as a potential indication of precancerous lesions or cancer, as a simple, non-invasive detection method, it has been widely used worldwide, it is simple, fast, non-invasive and painless, but has low specificity, relatively high false positive rate, and false negative results may occur, which may need to be combined with other screening methods, in order to ensure the accuracy of the test, it is usually necessary to repeat the test regularly;

[0006] Tumor marker testing: This method assesses the risk of colorectal cancer by measuring the levels of specific tumor markers in the blood (such as carcinoembryonic antigen and CA19-9). This method has the advantage of assisting in the diagnosis and postoperative monitoring of colorectal cancer. However, these tumor markers are not specific, meaning that an increase in their levels does not necessarily indicate colorectal cancer and may be associated with other diseases.

[0007] Non-invasive colorectal cancer gene testing (SOC2 gene methylation testing):

[0008] A non-invasive, painless, and non-invasive early colorectal cancer screening technology that assesses colorectal cancer risk by detecting SDC2 gene methylation levels in stool. It has a high detection rate and is accurate. It does not require bowel preparation and can be sampled at home and mailed at room temperature, making it convenient and enabling early diagnosis. However, it has certain inaccuracies and is relatively expensive.

[0009] Immunochemical fecal occult blood test: A chemical method more suitable for detecting lower gastrointestinal bleeding, it has higher specificity, sensitivity, and positive predictive value. It is inexpensive, convenient, and not affected by food or drugs, but the screening interval is still uncertain.

[0010] Fecal DNA testing: This method assesses colorectal cancer risk by detecting mutations in specific DNA sites and abnormal changes in epigenetic biomarkers in intestinal exfoliated cells in stool. Its advantages are that it is non-invasive, convenient, and accurate, but it is relatively expensive and the screening interval is uncertain.

[0011] In recent years, machine learning technology has been widely used in disease prediction research. Machine learning algorithms such as decision trees, random forests (RF), and support vector machines (SVM) can effectively process large-scale, nonlinear, and high-dimensional medical data. Compared with traditional statistical methods, machine learning models have shown significant advantages in predictive accuracy and generalization ability. Currently, the mainstream models include the following four:

[0012] Machine Learning Models:

[0013] Decision Tree: A classification and regression method originated in the 1980s. It is based on the recursive divide-and-conquer principle and simulates the human decision-making process. It recursively divides data through feature selection to form a tree structure (if-then rule set).

[0014] Random Forest (RF): proposed by Breiman in 2001, a Bagging method of ensemble learning, which improves the generalization ability through diversity, Random Forest is an ensemble learning method composed of multiple decision trees, which combines the prediction results of multiple decision trees through voting or averaging to improve the accuracy and stability of the model, and reduce the risk of overfitting;

[0015] Artificial Neural Network (ANN): inspired by biological neurons, the prototype was proposed by McCulloch-Pitts in 1943, and the development was promoted by backpropagation algorithm in the 1980s, composed of multiple layers of neurons, which can learn complex nonlinear relationships through training;

[0016] Support Vector Machine (SVM): proposed by Vapnik et al. in the 1990s, based on the structural risk minimization principle of statistical learning theory, Support Vector Machine is a supervised learning model that aims to find an optimal hyperplane to separate different classes of data while maximizing the classification boundary;

[0017] XGBoost (eXtreme Gradient Boosting): XGBoost is a high-performance implementation of Gradient Boosting Decision Tree (GBDT) algorithm, which has been optimized in many ways compared to traditional GBDT, and has become one of the most popular algorithms in modern machine learning competitions and practical applications, it works by sequentially constructing tree models, each new tree learns from the prediction errors (residuals) of all previous trees, and its core is to decompose the objective function into loss function and regularization term;

[0018] Detection and screening of intestinal cancer can greatly reduce the incidence and mortality of intestinal cancer. Although medical technology is constantly improving, the results of improving the disease-free survival rate of intestinal cancer patients are still limited. The existing technology is relatively ideal for accurate, economical, non-invasive and high-sensitivity detection of advanced intestinal cancer, but the screening rate for early intestinal cancer is still very low, so the incidence and mortality of intestinal cancer remain high. Therefore, developing an accurate and non-invasive screening method is particularly important for reducing the global health burden of intestinal cancer. SUMMARY

[0019] The present application provides a pks + The multi-omics non-invasive intestinal cancer risk assessment method based on intestinal flora, cancer driver genes and methylation molecular clock can effectively solve the problems raised in the above background technology.

[0020] To achieve the above purpose, the present application provides the following technical scheme: a pks +The method comprises the following steps: extracting a fecal DNA sample, designing and screening primers, triple specificity verification, non-invasive detection and multi-omics quantitative scoring model.

[0021] Based on the detection of pks+ E. coli in feces, pks+ E. coli can synthesize cytotoxin colibactin, induce DNA double-strand breaks and chromosome instability in host cells, cause cell cycle arrest, and ultimately lead to cell death.

[0022] According to the technical scheme, the operation of extracting the fecal DNA sample comprises collecting fecal samples of different sources, and extracting genomic DNA in the feces by using a fecal DNA extraction kit.

[0023] According to the technical scheme, the primer design and screening are designed for the primers related to the PKS gene island. Four existing primers of genes in the literature are selected, namely clbA, clbB, clbN and clbQ. PCR amplification is performed on the genomic DNA of pks positive bacteria, E. coli Nissle1917 and the synthesized primers. If a bright and single target band can be amplified, it indicates that the specificity of the primer is high, and the primer can be used for subsequent screening of PKS positive samples.

[0024] Five methylation molecular clock primers are designed, namely TFAP2B, GLRA3, CHAT, TCTEX1D1 and ZNF454. PCR amplification is performed on the DNA of A375 cell line and the designed primers. If a bright and single band can be run, it indicates that the specificity of the primer is high, and the primer can be used, otherwise the primer is optimized.

[0025] According to the technical scheme, the triple specificity verification is susceptible to diet and intestinal inflammation.

[0026] According to the technical scheme, the triple specificity verification is susceptible to diet and intestinal inflammation.

[0027] The positive rate of pks+ E. coli in the feces of healthy people, intestinal cancer patients and other intestinal disease patients is detected.

[0028] Marker1: PCR amplification is performed on the extracted fecal DNA and PKS gene primers. After the PCR amplification reaction, the product is subjected to agarose gel electrophoresis, and the band size is compared with the DNA marker of DS TM 2000, if the band size is consistent with the target product and has a specific nature, the sample is sent for sequencing to determine the gene sequence. After sequencing verification, the obtained amplification product is the target gene;

[0029] Screening of cancer driver gene mutation signatures, SBS88 / ID18 mutation enrichment analysis distinguishes oncogenic from non-oncogenic mutations:

[0030] Marker 2: Sequencing of 566 cancer driver genes in fecal DNA. In addition to counting characteristic mutations caused by PKS gene islands, the ratio of various mutations in cancer driver genes is also counted. The characteristic mutations caused by PKS+ E. coli are named SBS88 and ID18. If the mutation occurs in a cancer driver gene, the risk of cancer increases.

[0031] Methylation molecular clock age correction eliminates age-related interference based on the methylation baseline of healthy people, improving overall specificity:

[0032] Marker 3: The fecal DNA of the positive sample after bisulfite conversion was PCR amplified with the methylated molecular clock primers, and the transformed A375 cell line DNA was used as a positive control. TM 2000 DNA marker to compare the band size. If the band size is consistent with the target product and is specific, second-generation sequencing is performed to further determine the methylation status of the sample's methylation molecular clock and detect the methylation level of Marker3 methylation molecular clock. The number of cell divisions in a normal and healthy individual is positively correlated with age. If the methylation level is abnormally high, there is a risk of cancer.

[0033] According to the above technical solution, the non-invasive test requires the collection of stool samples, which can be completed by sampling at home and transporting at room temperature.

[0034] According to the above technical solution, the multi-omics quantitative scoring model integrates pks+ positive rate, driver gene mutation load, and methylation deviation index to output low, medium, and high risk stratification, providing a basis for individualized intervention.

[0035] Compared with the existing technology, the present invention has the following beneficial effects: the structure of the present invention is scientific and reasonable, and it is safe and convenient to use. The present invention is based on the detection of pks+ Escherichia coli in feces. pks+ Escherichia coli can synthesize the cytotoxin colibactin, which has genotoxicity and can induce DNA double-strand breaks and chromosomal instability in host cells, causing cell cycle arrest and ultimately leading to cell death. Therefore, pks+ Escherichia coli may profoundly affect the health of the host. Moreover, the gene island can cause characteristic mutations in the host DNA. Identifying "driver genes" that have an important impact on cancer is the current research focus on the mechanism of cancer occurrence. Genes containing driver mutations are "cancer driver genes (Cancer Driver Gene) The results of the study show that the multi-omics data analysis method can accurately predict the driver genes of the disease, provide a new method for the diagnosis and treatment of cancer, and find the correlation between the driver genes and the methylation molecular clock. The multi-omics combined screening of three indicators is used to conduct non-invasive colorectal cancer risk assessment. This multi-omics strategy provides new possibilities for the early detection, risk assessment and monitoring of colorectal cancer through the comprehensive application of different biological markers, further improves the accuracy and sensitivity of screening, and has the advantage of higher accuracy in colorectal cancer risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0037] Figure 1 It is a schematic diagram of the method flow structure of the present invention;

[0038] Figure 2 This is the map of 566 cancer driver genes of the present invention;

[0039] Figure 3 It is a schematic diagram of the comprehensive screening results of the present invention;

[0040] Figure 4 It is a schematic diagram of model establishment of the present invention;

[0041] Figure 5 This is a graph showing the electrophoresis results of detecting pks+ Escherichia coli samples in stool samples from different sources using the pks genomic island-specific primers of the present invention;

[0042] Figure 6 This is the first electrophoresis diagram of a positive sample detected by the clbA and clbN dual primers of the present invention;

[0043] Figure 7 This is the second electrophoresis diagram of the positive sample detected by the clbA and clbN dual primers of the present invention;

[0044] Figure 8 is a relationship diagram of the methylation degrees of the five methylated molecular clocks of the present invention;

[0045] Figure 9 is a first relationship diagram between the methylation levels of the five methylation molecular clocks of the present invention and age;

[0046] Figure 10 This is the second relationship diagram between the methylation level of the five methylation molecular clocks and age;

[0047] Figure 11 is a schematic diagram of the methylation levels of different intestinal tissues of different patients according to the present invention;

[0048] Figure 12 Schematic diagram of the methylation status of the ZNF454 gene in the DNA of stool samples from healthy individuals and colorectal cancer patients of the present invention;

[0049] Figure 13 It is a schematic diagram of the relationship between the methylation degree of the methylation molecular clock of the present invention and the number of mutations in cancer driver genes. DETAILED DESCRIPTION

[0050] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0051] Example 1:

[0052] The present invention provides a technical solution based on pks + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks, including extraction of fecal DNA samples, primer design and screening, triple specificity validation, non-invasive testing, and a multi-omics quantitative scoring model.

[0053] According to the above technical solution, the operation of extracting fecal DNA samples includes collecting fecal samples from different sources, such as 24 samples from healthy people and 28 samples from patients with intestinal cancer and other intestinal diseases, a total of 52 fecal samples, and using a fecal DNA extraction kit to extract genomic DNA from the feces.

[0054] According to the above technical solution, the primers were designed and screened. For the design of primers for the PKS gene island, primers for four genes already available in the literature were directly selected, namely clbA, clbB, clbN, and clbQ. PCR amplification was performed using genomic DNA of pks-positive bacteria (Escherichia coli Nissle 1917) and the synthesized primers. If a bright and single target band can be amplified, it indicates that the primers are highly specific and can be used for subsequent screening of PKS-positive samples.

[0055] Design five methylation molecular clock primers, namely TFAP2B, GLRA3, CHAT, TCTEX1D1, and ZNF454. Use A375 cell line DNA and the designed primers for PCR amplification. If a bright and single band is generated, it indicates that the primers are highly specific and can be used. Otherwise, the primers should be further optimized.

[0056] According to the above technical solution, the triple specificity verification is easily interfered by diet and intestinal inflammation through a single biomarker (such as occult blood test), and the false positive rate is as high as 20-30%.

[0057] According to the above technical solution, the present invention is verified by triple specificity:

[0058] (1) Detection of the positive rate of pks+ Escherichia coli in feces of healthy people, patients with intestinal cancer and other intestinal diseases

[0059] Marker 1: The extracted fecal DNA and PKS gene primers were used for PCR amplification. The PCR amplification products were run on agarose gel electrophoresis and compared with DS TM The band size is compared with a 2000 DNA marker. If the band size is consistent with the target product and is specific, the sample is sent for sequencing to determine the gene sequence of the sample. After sequencing verification, the amplified product obtained is the target gene.

[0060] (2) Screening of cancer driver gene mutation signatures (SBS88 / ID18 mutation enrichment analysis) to distinguish oncogenic from non-oncogenic mutations;

[0061] Marker 2: Sequencing of 566 cancer driver genes in fecal DNA. In addition to counting characteristic mutations caused by PKS gene islands, the ratio of various mutations in cancer driver genes is also counted. The characteristic mutations caused by PKS+ E. coli are named SBS88 and ID18. If the mutation occurs in a cancer driver gene, the risk of cancer is greatly increased.

[0062] (3) Age correction of the methylation molecular clock (based on the methylation baseline of healthy people) eliminates age-related interference and improves the overall specificity;

[0063] Marker3: PCR amplification reaction of fecal DNA of the positive sample after bisulfite conversion with methylation molecular clock primers, and the DNA of the A375 cell line after conversion as a positive control, compared with the DS TM 2000 DNA marker, if the band size is consistent with the desired product and has specific properties, then perform second-generation sequencing to further determine the methylation of the methylation molecular clock of the sample, and detect the methylation degree of Marker3 methylation molecular clock, because the number of cell divisions and age of normal healthy individuals are positively correlated, if the methylation degree is abnormally high, there is a risk of cancer.

[0064] According to the above technical solution, the non-invasive detection: invasive examination such as enteroscopy is difficult to popularize due to high cost, complex operation and strong discomfort of patients, the present application is completely non-invasive, only needs to collect fecal samples, and can complete detection through home sampling and normal temperature transportation, greatly improves patient acceptance, and is suitable for large-scale population screening.

[0065] According to the above technical solution, the multi-omics quantitative scoring model: the prior art is difficult to quantify the risk of cancer (such as the "positive / negative" result provided by the occult blood test), the present application constructs a multi-omics quantitative scoring model, integrates the pks+ positive rate, driver gene mutation load and methylation deviation index, and outputs low / medium / high risk stratification, to provide accurate basis for individualized intervention.

[0066] Among the related terms are explained as follows:

[0067] Intestinal cancer: colorectal cancer or colon cancer (Colorectal cancer, CRC), which originates from the colon or rectum, usually starts with abnormal hyperplasia of intestinal polyps, and is a heterogeneous disease of intestinal epithelium.

[0068] PKS: Polyketide Synthase (Polyketide Synthase), also known as clb gene, generally composed of 19 genes (from clbA to clbS) in series into a 54kb gene cluster, encoding colibactin with genetic gene toxicity.

[0069] PCR: Polymerase Chain Reaction (Polymerase Chain Reaction), which utilizes the principle of DNA double-strand replication to replicate specific DNA fragments in vitro, and is a nucleic acid synthesis technology that can amplify the target DNA fragment in a short time.

[0070] TP53: encodes a tumor suppressor protein containing transcriptional activation, DNA binding and oligomerization domains. This protein responds to various cellular stresses and regulates the expression of target genes, thereby inducing cell cycle arrest, apoptosis, senescence, DNA repair or metabolic changes. Mutations in this gene are associated with a variety of human cancers.

[0071] A375: Human malignant melanocytoma.

[0072] Nissle1917: A probiotic Escherichia coli containing a PKS gene island.

[0073] clbA, clbB, clbN, clbQ: genes on the PKS island, encoding different catalytic proteins.

[0074] Methylation molecular clock: An assessment method based on DNA methylation markers that are closely related to time or age. It can accurately quantify age-related phenotypes and outcomes and is currently the most promising biomarker of aging. This type of DNA gene usually shows low methylation. With increasing age, the degree of methylation continues to increase. When cells proliferate abnormally, the methylation of the gene will also increase abnormally.

[0075] TFAP2B: A member of the transcription factor AP-2 family, it plays an important regulatory role in biological processes such as embryonic development, cell proliferation and differentiation, and immune response.

[0076] GLRA3: glycine receptor α3 subunit, mainly mediates inhibitory signal transmission in the central nervous system, participates in pain regulation and movement coordination, and serves as a methylation molecular clock in the present invention.

[0077] ZNF454: encodes a zinc finger protein that can act as a transcription factor to regulate the transcription of target genes and serves as a methylation molecular clock in the present invention.

[0078] DS TM 2000: A DNA molecular weight marker composed of six double-stranded linear DNA fragments. Suitable for determining the size of linear double-stranded DNA fragments ranging from 100 bp to 2 kb. The indicator band is 750 bp for easy visualization after electrophoresis. Each band is rigorously physically quantified and can be used to determine the size and content of the target fragment.

[0079] Example 2:

[0080] The four clb gene primers of the pks island and the five methylation molecular clock primers were synthesized by Shanghai Biotech Co., Ltd. The main kits used are as follows: HiPure Stool DNA Mini Kit (fecal DNA extraction kit, purchased from Meiji Biotechnology, China), PCR kit Plus Mix (purchased from Dongsheng Biotechnology, China), EpiArt DNA Methylation Bisulfite Kit (EpiArt DNA methylation bisulfite kit, purchased from Novezan Biotechnology, China), KAPA2G Robust HotStart ReadyMix PCR Kit (PCR kit, purchased from Roche, China).

[0081] Extraction of fecal DNA:

[0082] Fecal DNA was extracted using the HiPure Stool DNA Mini Kit (Meiji Biotechnology, China). The specific steps are as follows:

[0083] Reagent preparation:

[0084] (1) Dissolve Proteinase K (20 mg / mL): Add an appropriate amount of Protease Dissolve Buffer to dissolve Proteinase K to a final concentration of 20 mg / mL. Gently invert the container to allow the proteinase K to fully dissolve. (Protease K powder can be stored at room temperature for one year, but the dissolved Proteinase K must be aliquoted and stored at -20°C).

[0085] (2) Dissolve RNase A (15 mg / mL): Add an appropriate amount of Protease Dissolve Buffer to dissolve RNase A to a final concentration of 15 mg / mL. Invert and mix to allow RNase A to fully dissolve. RNase A powder can be stored at room temperature, but dissolved RNase A must be stored at -20 to 8°C.

[0086] (3) Dilute Buffer GW1 by adding anhydrous ethanol as indicated on the bottle label;

[0087] (4) Dilute Buffer GW2 by adding anhydrous ethanol as indicated on the bottle label;

[0088] (5) Pour the PVP-10 powder into the Buffer ATL bottle, invert and mix thoroughly, and use after it is completely dissolved;

[0089] Experimental steps:

[0090] (1) In a 2 mL centrifuge tube, add 0.5 g Glass Beads (0.1-0.6 mm);

[0091] (2) Transfer 100-150 mg of fecal sample to the centrifuge tube containing Glass Beads, add 0.6 mL Buffer ATL / PVP-10 and 0.6 mL Buffer PCI to the sample, vortex at high speed for 10 minutes on a vortexer or bead-beat using a bead beater;

[0092] (3) Incubate at 65°C water bath for 15 minutes;

[0093] (4) Centrifuge at 14,000 x g for 10 minutes at room temperature;

[0094] (5) Transfer 500 μL of supernatant to a 2 mL centrifuge tube, add 10 μL RNase A to the supernatant, and incubate at room temperature for 10 minutes;

[0095] (6) Add 20 μL Proteinase K and 500 μL Buffer AL to the supernatant, mix by inverting, and incubate at 70°C for 10 minutes;

[0096] (7) Add 500 μL absolute ethanol to the mixture, vortex to mix for 15 seconds;

[0097] (8) Place a HiPure DNA Mini Column II in a 2 mL collection tube, transfer half of the volume of the mixture to the column, and centrifuge at 10,000 x g for 30-60 seconds;

[0098] (9) Discard the flow-through, place the column back in the collection tube, and transfer the remaining mixture to the column, centrifuge at 10,000 x g for 30-60 seconds;

[0099] (10) Discard the flow-through, place the column back in the collection tube, add 500 μL Buffer GW1 (diluted with absolute ethanol) to the column, and centrifuge at 10,000 x g for 30-60 seconds;

[0100] (Buffer GW1 must be diluted with absolute ethanol, as indicated on the bottle label or in the instructions);

[0101] (11) Discard the flow-through, place the column back in the collection tube, add 650 μL Buffer GW2 (diluted with ethanol) to the column, and centrifuge at 10,000 x g for 30-60 seconds;

[0102] (Buffer GW2 must be diluted with absolute ethanol, as indicated on the bottle label or in the instructions);

[0103] (12) Discard the filtrate, put the column back into the collection tube, add 650 μL Buffer GW2 (diluted with ethanol) into the column, centrifuge at 10,000 x g for 30-60 seconds;

[0104] (13) Discard the filtrate, put the column back into the collection tube, centrifuge at 13,000 x g for 2 minutes to dry the column;

[0105] (14) Put the column into a 1.5 mL centrifuge tube, add 30-100 μL Buffer AE or sterilized water preheated to 65°C into the center of the membrane of the column, let it stand at room temperature for 2 minutes, centrifuge at 13,000 x g for 1 minute;

[0106] (15) Discard the DNA binding column, store the DNA at 2-8°C, and store at -20°C for long-term storage;

[0107] Determine the concentration and purity of the extracted DNA by using a spectrophotometer Nanodrop, as a subsequent pks+ E. coli screening experiment and correlation experiment of methylation molecular clock;

[0108] Design and screening of primers:

[0109] Synthesize four PKS island gene primers (Table 6-1), use pks+ bacteria Nissle1917 and synthesized primers for PCR amplification, and the obtained target product is basically a single band;

[0110] Table 6-1 Sequence of pks island synthesized primers, size of target fragment and annealing temperature

[0111]

[0112]

[0113] Design methylation primers, use A375 cell line as a positive control and the designed primers for PCR amplification, if a bright and single band can be run out, it indicates that the specificity of this primer is high, which can be used for subsequent experimental operation, otherwise the primer should be optimized.

[0114] Table 6-2 Sequence of methylation site primers

[0115]

[0116]

[0117] pks gene amplification:

[0118] PCR reagent kit Plus Mix (Dongsheng Bio, China) was used to amplify the pks island specific primers with the extracted fecal DNA. The amplification primers were selected as clbA, clbB, clbN, clbQ, and the primer sequence length is shown in the table below:

[0119] (1) The PCR reaction system is as follows:

[0120] Components Volume (μL) 2×Plus Mix 25 Upstream primer (10 μM) 2 Downstream primer (10 μM) 2 DNA template 1-4 ddH20 Fixed volume to 50 Total volume 50

[0121] (2) The PCR reaction program is as follows:

[0122]

[0123]

[0124] Finally, the quality of the PCR product was detected by 1.2% agarose gel electrophoresis: 5 μL of PCR product was added to the agarose gel for electrophoresis, and the gel was imaged after electrophoresis. The suspected positive samples with the same size as the target gene band were selected, and all suspected positive samples were sequenced for analysis;

[0125] Agarose gel electrophoresis method:

[0126] (1) Preparation of agarose gel: weigh 0.36 g of agarose powder (pks experimental gel concentration is 1.2%) / 0.6 g (methylated experimental gel concentration is 2%), add to a conical flask, measure 30 mL of 1×TAE solution into the conical flask, place the flask in a microwave oven and heat to boiling, pay attention to avoid liquid overflow during boiling, when the liquid is cooled to about 60℃, add 3 μL of nucleic acid dye, shake gently, add the dissolved liquid to the gel slot and remove the bubbles generated during the gel pouring process, insert the comb, remove the comb slowly after about 30 min, and prepare for use.

[0127] (2) Electrophoresis: place the gel in an electrophoresis tank containing 1×TAE solution, TAE is appropriate to cover the gel, take 5 μL of PCR product, add 1 μL of 6×loading buffer, mix well and add to the gel hole, and the same for the following, cover the electrophoresis tank cover, connect the power supply, 120V, 35 minutes after gel imaging observation.

[0128] DNA modification and purification:

[0129] The extracted fecal DNA was subjected to bisulfite conversion using EpiArt DNA Methylation Bisulfite Kit (Novozyme Bio, China), and the specific operation steps are as follows:

[0130] Reagent preparation:

[0131] (1) Add 80 mL of anhydrous ethanol to EM101-01 and 160 mL of anhydrous ethanol to EM101-02.

[0132] (2) Prepare Conversion Mix: Add 1 mL ddH20, 200 μL CT Conversion Diluent, and 100 μL CT Conversion Buffer to a tube of CT Conversion Powder, and vortex at room temperature for about 1 minute to dissolve.

[0133] 6.5.2 Bisulfite Conversion

[0134] (1) Equilibrate the CT Conversion Mix to room temperature and prepare the following reaction in a 200 μL sterile PCR tube:

[0135] Components volume Input DNA XμL (100pg-2μg) CT Conversion Mix 130μL ddH20 To 150μL

[0136] (2) Mix by inverting the tube or using a pipette, and centrifuge briefly to collect the reaction solution at the bottom of the tube.

[0137] (3) Place the PCR tube in a PCR instrument and perform the following reaction:

[0138]

[0139]

[0140] Purification of transformation products:

[0141] (1) Place the EpiArt DNA Columns into the Collection Tubes.

[0142] (2) Add 600 μL of E-Binding Buffer to the adsorption column, then add the converted reaction product to the adsorption column and gently invert the column 8-10 times to mix the reaction solution and E-Binding Buffer completely.

[0143] (3) Centrifuge at 12,000 rpm (13,400 × g) for 30-60 seconds, discard the filtrate, and return the adsorption column to the collection tube.

[0144] (4) Add 500 μL of E-Wash Buffer (ethanol added) to the adsorption column, centrifuge at 12,000 rpm (13,400 × g) for 30-60 seconds, discard the filtrate, and return the adsorption column to the collection tube.

[0145] (5) Add 500 μL of E-Desulphonation Buffer to the adsorption column, incubate at room temperature (15-25°C) for 15 min, centrifuge at 12,000 rpm (13,400 × g) for 30-60 sec, discard the filtrate, and return the adsorption column to the collection tube.

[0146] (6) Add 500 μL of E-Wash Buffer (ethanol added) to the adsorption column, centrifuge at 12,000 rpm (13,400 × g) for 60 sec, discard the filtrate, and return the adsorption column to the collection tube.

[0147] (7) Repeat step 6.

[0148] (8) Centrifuge the empty column at 12,000 rpm (13,400 × g) for 2 min.

[0149] (9) Transfer the adsorption column to a new 1.5 mL centrifuge tube, open the lid for 2 min to dry thoroughly, add 10-20 μL of E-Elution Buffer to the center of the adsorption column membrane, let it stand at room temperature for 1-2 min, centrifuge at 12,000 rpm (13,400 × g) for 2 min, and collect the DNA filtrate.

[0150] (10) Discard the adsorption column.

[0151] (11) The concentration and purity of the converted DNA were measured using a Nanodrop spectrophotometer.

[0152] (12) Store DNA at -20°C. For long-term storage, store at -70°C.

[0153] 6.6 Methylation Molecular Clock PCR

[0154] Genomic DNA was modified with bisulfite. During the modification process, unmethylated cytosine was converted to uracil, while methylated cytosine remained unchanged. The KAPA2G Robust HotStart ReadyMix PCR Kit (Roche, China) was used to amplify methylated primers with fecal DNA from bisulfite-converted pks-positive samples. The primer sequence sizes are shown in the table.

[0155] The PCR reaction system is as follows:

[0156]

[0157]

[0158] The PCR reaction procedure is as follows:

[0159]

[0160] Finally, the quality of the PCR product was detected by agarose gel electrophoresis with a 2% gel concentration: 5 μL of the PCR product was added to the agarose gel for electrophoresis, and the gel was imaged after electrophoresis. The sample with the same size as the target gene band was selected for sequencing analysis.

[0161] 566 cancer driver gene mutation assay, the extracted fecal DNA was sent to a sequencing company for whole exon sequencing, and the returned raw data was analyzed;

[0162] Experimental data collection and pretreatment:

[0163] Through experiments and sequencing, a plurality of characteristics were obtained through bioinformatics analysis means, wherein the methylation level of the methylation molecular clock was identified by BS-Seeker2, and the mutation rate of the cancer driver gene and the mutation rate of the pks characteristic were obtained by GATK bioinformatics analysis;

[0164] Experimental data collection:

[0165] After the above experiments, we collected the data of 25 characteristics of the 52 samples, the target variable is the binary variable in the group column, 0 represents normal people, and 1 represents colorectal cancer patients, and the 25 characteristic variables include the following table contents:

[0166]

[0167]

[0168]

[0169] Through the above statistics, the 25 characteristic data sets of each of the 52 samples were obtained, and then the data sets were processed according to the machine learning model construction process;

[0170] Data standardization:

[0171] Due to the significant difference in the dimension and numerical range of multi-omics characteristics (such as methylation rate, mutation frequency, and age), standardization is needed to eliminate the dimension effect and ensure that the model treats each feature fairly.

[0172] The Z-score standardization method is used, the formula is Z=(X-μ) / σ, wherein X is the original data, μ is the feature mean, and σ is the feature standard deviation. The data is standardized to eliminate the dimension effect, so that different feature data has the same scale, and the stability and accuracy of the model training are improved.

[0173] Missing value processing:

[0174] For missing values in the data, mean filling or interpolation method is used to handle, the method for filling missing values is multiple filling or chain equation filling (MICE) multiple filling method. The core idea is to generate multiple possible filling values to form several complete data sets, then analyze these complete data sets and integrate the analysis results to get the final analysis conclusion. It regards each feature with missing values as the dependent variable and other features as the independent variables. A prediction model is constructed using these independent variables to predict the missing values. The code also designs a maximum of ten iterations for this process, and each iteration uses updated filling values. By default, this missing value filling method uses Bayesian ridge regression as the base estimator. This method relies on the original complete data column as a reference to fill the data column with missing values, and tries to ensure the statistical relationship between features. If a feature value is missing for a sample, the mean of that feature in other samples is calculated to fill it. Or use interpolation method to estimate the missing value according to the feature values of adjacent samples, to ensure data integrity and avoid affecting subsequent analysis.

[0175] Data splitting:

[0176] The preprocessed data is divided into 70% training set and 30% validation set. The training set is used for model training to learn the relationship between data features and colorectal cancer risk. The validation set is used to evaluate the model performance to ensure the model has good generalization ability.

[0177] Feature selection:

[0178] Feature selection is a key step in machine learning feature engineering. Its core goal is to identify and retain the most predictive feature subset. This process quantifies the correlation between each feature and the target variable or calculates the feature importance score, and then filters out irrelevant and redundant features. This dimensionality reduction process can improve model accuracy, enhance generalization ability, reduce computational complexity, and improve model interpretability.

[0179] From a practical application perspective, feature selection can extract truly predictive features, simplifying the model structure and helping to understand the underlying rules behind the data. It also effectively alleviates the overfitting problem caused by the "curse of dimensionality".

[0180] Recursive feature elimination based on random forest:

[0181] The random forest algorithm is used to calculate the importance score of each feature. The features are sorted according to the score, and a threshold is set to gradually remove features with importance scores below the threshold. After each removal, the random forest model is retrained, and the performance (such as AUC value) on the validation set is used to determine whether to continue removing. The process continues until the model performance no longer improves, and the remaining features are selected as key features.

[0182] Decision tree screening:

[0183] A decision tree model is constructed based on information gain or Gini impurity. During the construction process, the features that maximize information gain or reduce Gini impurity the most are selected for splitting. The splitting of each feature in the decision tree is analyzed, and the features that have a greater impact on the classification results and are located at the upper layer of the decision tree are screened out as important features.

[0184] Recursive feature elimination based on XGBoost:

[0185] The XGBoost algorithm is used to calculate feature importance. Similar to the recursive feature elimination method of random forest, starting from all features, the feature with the lowest importance is eliminated each time, the XGBoost model is retrained and the performance on the validation set is evaluated. This process is continued until the model performance no longer improves, and the filtered feature set is obtained.

[0186] Comprehensive screening results:

[0187] like Figure 3 As shown in the figure, by comparing the results obtained by the three feature screening methods, features that are considered important in multiple methods, such as related genes in the methylation molecular clock (ZNF454 and GLRA3, etc.), some indicators related to cancer driver gene mutations (number of CDG mutation genes in SNV / 566, etc.) and pks-related features, are selected as the key features for the final model construction, removing redundant features, reducing model complexity, and improving model efficiency and accuracy.

[0188] Model construction:

[0189] Hyperparameter tuning of machine learning models is a key step in improving model performance. Its essence is to achieve optimal performance on specific tasks through systematic parameter optimization. Understanding the basic principles and methods of parameter tuning is an important foundation for machine learning practice. Unlike the automatic learning parameters during model training, hyperparameters must be manually set and directly affect the model's structure and learning process. An effective parameter tuning strategy can not only significantly improve the model's prediction accuracy and generalization ability, but also avoid problems such as overfitting and underfitting.

[0190] The parameters of the four models used in this study have different meanings. During the parameter adjustment process, a parameter range will be set. Within this range, the model will automatically learn and find the optimal parameter value.

[0191] Random Forest Model:

[0192] Parameter settings:

[0193]

[0194] Model training: Using the screened key features and corresponding sample labels (healthy or colorectal cancer patients) as input, a random forest model is trained on the training set. The model uses a bootstrap sampling method to extract samples with replacement from the training set to construct multiple decision trees. A feature subset is randomly selected at each node split, and the prediction results of multiple trees are finally integrated by voting or averaging.

[0195] Support Vector Machine Model:

[0196] Parameter settings:

[0197]

[0198]

[0199] Model training: Map the training set's feature data into a high-dimensional space (if using a nonlinear kernel function) and find the hyperplane with the largest margin to distinguish healthy samples from colorectal cancer samples. By adjusting the parameter C and the kernel function, the model achieves better classification results on the training set.

[0200] XGBoost Model

[0201] Parameter settings:

[0202]

[0203]

[0204] Model training: XGBoost sequentially builds tree models. Each new tree learns the prediction errors (residuals) of all previous trees. During the training process, the model performance is optimized by performing a second-order Taylor expansion on the loss function and setting regularization terms, so that the model better fits the training data.

[0205] Artificial neural network model:

[0206] Parameter settings:

[0207]

[0208] Model training: Construct a neural network structure consisting of an input layer, a hidden layer, and an output layer. The input layer receives preprocessed feature data, the hidden layer performs feature conversion on the data through an activation function, and the output layer generates the final prediction result. During the training process, the backpropagation algorithm is used to calculate the error and update the weights, and the network parameters are continuously adjusted to gradually reduce the loss function value of the model on the training set.

[0209] Model Evaluation:

[0210] Evaluation metric selection:

[0211] The AUC value (area under the curve) is used as the main evaluation index, the closer the AUC value is to 1, the higher the prediction accuracy of the model; at the same time, auxiliary indicators such as accuracy (the proportion of correctly predicted samples in the total number of samples), recall rate (the proportion of actual positive samples correctly predicted as positive samples), and F1 value (an index that considers accuracy and recall rate) are used to comprehensively evaluate the performance of the model.

[0212] Evaluation process:

[0213] The trained random forest, support vector machine, XGBoost, and artificial neural network models are evaluated using the validation set data. The feature data of the validation set is input into the model to obtain the prediction results, which are compared with the actual labels to calculate the AUC value, accuracy, recall rate, and F1 value, etc. From the results Figure 4 of the three groups of scientific models, the AUC values of different models are different, such as xgb-XGBoost AUC = 0.841, rf-XGBoost and dc-XGBoost AUC = 0.921, etc. By analyzing these indicators, the performance of each model on the validation set can be clearly compared, and the advantages and disadvantages of each model can be determined.

[0214] Model parameter tuning:

[0215] Parameter tuning method selection:

[0216] Methods such as grid search, Bayesian optimization, or genetic algorithm are used for model parameter tuning. For example, using grid search, multiple parameters of the model are combined for exhaustive search, and the performance of the model under each parameter combination is evaluated on the validation set to select the optimal parameter combination. Random search randomly samples from the parameter space for evaluation, which is relatively efficient. Bayesian optimization builds a probability model of the objective function and intelligently searches for the optimal parameters based on historical results. Genetic algorithm simulates the biological evolution process and finds the best parameter combination through selection, crossover, and mutation operations.

[0217] Parameter tuning implementation:

[0218] Take the random forest model as an example. If the AUC value of the model on the validation set does not reach the expected value in the initial evaluation, use grid search to optimize parameters such as n_estimators, max_depth, min_samples_split, and min_samples_leaf. Set the value range of the parameters, such as n_estimators = [50, 100, 150, 200, 250, 300], max_depth = [3, 5, 7, 10, 15, 20], etc. Train and evaluate each parameter combination, and select the parameter combination that maximizes the AUC value on the validation set as the optimized parameters. Similar parameter tuning operations are performed on support vector machines, XGBoost, and artificial neural network models. Adjust the parameters until the performance of the model on the validation set reaches the optimal value.

[0219] Model building:

[0220] As shown in Figure 4 : The results of model evaluation and parameter tuning are shown in the ROC curves of the three groups and the ROC curves of the single group. We can see that the highest value and the average value of the AUC values of all single groups (Marker1-Marker3) on the 12 models are not more advantageous than those of the three groups (three Marker combinations). At the same time, there are some sharp fluctuations in the ROC curves of the single groups, and the AUC values obtained may show different effects on different models, making the AUC values obtained by this single group method unreliable and unable to perform well on multiple models. It may also imply that the prediction results obtained by the single group method may overfit on some models, leading to some AUC values being artificially high. Compared with the ROC curves of the single groups, the ROC curves of the three groups show excellent results on multiple models and do not have large fluctuations in the results on multiple models. The prediction results obtained are more reliable.

[0221] Model application:

[0222] The actual fecal sample data is processed according to the data preprocessing steps, and the key features are extracted. The processed feature data is input into the final established intestinal cancer risk assessment model, and the model outputs the risk prediction results of the sample suffering from intestinal cancer. According to the prediction results, individuals can be divided into low-risk, medium-risk, and high-risk groups, providing important reference for early screening and prevention of intestinal cancer, and assisting doctors in developing personalized diagnosis and treatment plans.

[0223] The prediction feature rules of rf-XGBoost and dc-XGBoost models are described:

[0224] 1. Feature threshold analysis of rf-XGBoost reveals the following key rules that significantly affect the prediction probability of the positive class (target class):

[0225] Feature influence ranking (in descending order of influence strength):

[0226] (1) Age-influence intensity: 0.0111, direction: positive;

[0227] Normal range: [-0.8520-0.9764];

[0228] Critical threshold: 0.2005 (feature range: -1.5416-1.4675);

[0229] High value effect: increase the probability of positive class, low value effect: reduce the probability of positive class;

[0230] When the feature value is greater than the threshold, the average change in predicted probability is: -0.0054;

[0231] When the feature value is less than the threshold, the average change in predicted probability is: -0.0165;

[0232] (2) Total number of CDG mutations in ID18 / total INDELs - impact strength: 0.0104, direction: negative;

[0233] Normal range: [-0.3856-0.1496];

[0234] Critical threshold: -0.1361 (feature range: -0.3856-0.4043);

[0235] The effect of high value is to reduce the probability of positive class, and the effect of low value is to increase the probability of positive class;

[0236] When the feature value is greater than the threshold, the average change in predicted probability is: -0.0020;

[0237] When the feature value is less than the threshold, the average change in predicted probability is: 0.0084;

[0238] (3) SBS88 / SNV-influence intensity: 0.0091, direction: positive;

[0239] Normal range: [-0.5282-1.2033];

[0240] Critical threshold: 0.2892 (feature range: -1.0496-2.1300);

[0241] High value effect: increase the probability of positive class, low value effect: reduce the probability of positive class;

[0242] When the feature value is greater than the threshold, the average change in predicted probability is: 0.0122;

[0243] When feature value < threshold, average change in predicted probability: 0.0031;

[0244] (4) GLRA3 total methylation rate influence intensity: 0.0074, direction: negative;

[0245] Normal range: [-0.7497-0.6997];

[0246] Critical threshold: -0.2378 (feature range: -0.7574-2.5338);

[0247] High value effect: decrease positive class probability, low value effect: increase positive class probability;

[0248] When feature value > threshold, average change in predicted probability: -0.0064;

[0249] When feature value < threshold, average change in predicted probability: 0.0010;

[0250] GLRA3 highest methylation rate influence intensity: 0.0063, direction: negative;

[0251] Normal range: [-0.7485-1.1526];

[0252] Critical threshold: -0.3125 (feature range: -1.1198-1.9479);

[0253] High value effect: decrease positive class probability, low value effect: increase positive class probability;

[0254] When feature value > threshold, average change in predicted probability: -0.0010;

[0255] When feature value < threshold, average change in predicted probability: 0.0054;

[0256] (6) ZNF454 highest methylation rate influence intensity: 0.0060, direction: positive; Normal range: [-1.4013-0.7574];

[0257] Critical threshold: 0.7574 (feature range: -1.8997-0.7574);

[0258] High value effect: increase positive class probability, low value effect: decrease positive class probability;

[0259] When feature value > threshold, average change in predicted probability: 0.0038;

[0260] When feature value < threshold, average change in predicted probability: -0.0022;

[0261] (7) ZNF454 total methylation rate influence strength: 0.0041, direction: positive;

[0262] Normal range: [-1.1750-1.4044];

[0263] Critical threshold: -0.1222 (feature range: -1.1750-3.8259);

[0264] High value effect: increase the probability of positive class, low value effect: reduce the probability of positive class;

[0265] When the feature value is greater than the threshold, the average change in predicted probability is: -0.0044;

[0266] When the feature value is less than the threshold, the average change in predicted probability is: -0.0085;

[0267] (8) ID18 / INDEL-influence intensity: 0.0025, direction: positive;

[0268] Normal range: [-0.9245-0.5441];

[0269] Critical threshold: -0.6148 (feature range: -1.2481-1.7601);

[0270] High value effect: increase the probability of positive class, low value effect: reduce the probability of positive class;

[0271] When the feature value is greater than the threshold, the average change in predicted probability is: 0.0036;

[0272] When the feature value is less than the threshold, the average change in predicted probability is: 0.0011;

[0273] (9) Number of CDG mutant genes in INDEL / 566 - Impact strength: 0.0012, direction: negative;

[0274] Normal range: [-0.9942-1.7153];

[0275] Critical threshold: -0.1767 (feature range: -1.0739-4.6081);

[0276] The effect of high value is to reduce the probability of positive class, and the effect of low value is to increase the probability of positive class;

[0277] When the feature value is greater than the threshold, the average change in predicted probability is: -0.0009;

[0278] When the feature value is less than the threshold, the average change in predicted probability is: 0.0004;

[0279] (10) Number of mutations in CDG genes in SNV / 566 - Impact strength: 0.0003, Direction: positive;

[0280] Normal range: [-0.5634 -1.4703];

[0281] Critical threshold: -0.5634 (feature range: -0.5634 -4.2362);

[0282] High value effect: increase positive class probability, low value effect: decrease positive class probability;

[0283] When the feature value > threshold, the average change of prediction probability: -0.0056;

[0284] When the feature value < threshold, the average change of prediction probability: -0.0059;

[0285] (11) Number of mutations in CDG genes in INDEL / INDEL - Impact strength: 0.0000, Direction: negative;

[0286] Normal range: [-0.8054 -1.0963];

[0287] Critical threshold: -2.2365 (feature range: -2.2365 -2.3400);

[0288] High value effect: decrease positive class probability, low value effect: increase positive class probability;

[0289] When the feature value > threshold, the average change of prediction probability: 0.0012;

[0290] When the feature value < threshold, the average change of prediction probability: 0.0012;

[0291] (12) Number of mutations in CDG genes in SNV / SNV - Impact strength: 0.0000, Direction: negative;

[0292] Normal range: [-0.8880 -0.6257];

[0293] Critical threshold: -1.1912 (feature range: -1.1912 -2.3952);

[0294] High value effect: decrease positive class probability, low value effect: increase positive class probability;

[0295] When the feature value > threshold, the average change of prediction probability: -0.0032;

[0296] When the feature value < threshold, the average change of prediction probability: -0.0032;

[0297] Prediction rule summary:

[0298] The following are the features that have the greatest impact on the prediction probability of the positive class and their threshold rules:

[0299] (1) When age > 0.2005, the probability of being predicted as a positive class increases;

[0300] Normal range: [-0.8520-0.9764];

[0301] (2) When the total number of mutations of CDG in ID18 / total INDEL < -0.1361, the probability of being predicted as a positive class increases;

[0302] Normal range: [-0.3856-0.1496];

[0303] (3) When SBS88 / SNV > 0.2892, the probability of being predicted as a positive class increases;

[0304] Normal range: [-0.5282-1.2033];

[0305] (4) When GLRA3 total methylation rate < -0.2378, the probability of being predicted as a positive class increases;

[0306] Normal range: [-0.7497-0.6997];

[0307] (5) When GLRA3 highest methylation rate < -0.3125, the probability of being predicted as a positive class increases;

[0308] Normal range: [-0.7485-1.1526];

[0309] When ZNF454 highest methylation rate > 0.7574, the probability of being predicted as a positive class increases;

[0310] Normal range: [-1.4013-0.7574];

[0311] (7) When ZNF454 total methylation rate > -0.1222, the probability of being predicted as a positive class increases;

[0312] Normal range: [-1.1750-1.4044];

[0313] (8) When ID18 / INDEL > -0.6148, the probability of being predicted as a positive class increases;

[0314] Normal range: [-0.9245-0.5441];

[0315] (9) When the number of CDG mutation genes in INDEL / 566 < -0.1767, the probability of being predicted as a positive class increases;

[0316] Normal range: [-0.9942 -1.7153];

[0317] (10) When CDG number in SNV / 566 > -0.5634, the probability of predicting positive class increases;

[0318] Normal range: [-0.5634 -1.4703];

[0319] 2. The feature threshold analysis of dc-XGBoost reveals the following key rules:

[0320] Feature influence ranking (in descending order of influence strength):

[0321] (1) SBS88 / SNV - Influence strength: 0.0163, direction: positive;

[0322] Normal range: [-0.5282 -1.2033];

[0323] Critical threshold: 0.2892 (feature range: -1.0496 -2.1300);

[0324] High value effect: increase positive class probability, low value effect: decrease positive class probability;

[0325] When the feature value > threshold, the average change of prediction probability: 0.0177;

[0326] When the feature value < threshold, the average change of prediction probability: 0.0013;

[0327] (2) age - Influence strength: 0.0093, direction: positive;

[0328] Normal range: [-0.8520 -0.9764];

[0329] Critical threshold: 0.2005 (feature range: -1.5416 -1.4675);

[0330] High value effect: increase positive class probability, low value effect: decrease positive class probability;

[0331] When the feature value > threshold, the average change of prediction probability: -0.0247;

[0332] When the feature value < threshold, the average change of prediction probability: -0.0340;

[0333] (3) ID18 / total INDELs in CDG - Influence strength: 0.0082, direction: negative;

[0334] Normal range: [-0.3856 -0.1496];

[0335] Critical threshold: -0.2192 (feature range: -0.3856-0.4044);

[0336] The effect of high value is to reduce the probability of positive class, and the effect of low value is to increase the probability of positive class;

[0337] When the feature value is greater than the threshold, the average change in predicted probability is: -0.0130;

[0338] When the feature value is less than the threshold, the average change in predicted probability is: -0.0048;

[0339] (4) ID18 / INDEL-influence intensity: 0.0054, direction: positive;

[0340] Normal range: [-0.9245-0.5441];

[0341] Critical threshold: -0.6148 (feature range: -1.2481-1.7601);

[0342] High value effect: increase the probability of positive class, low value effect: reduce the probability of positive class;

[0343] When the feature value is greater than the threshold, the average change in predicted probability is: 0.0122;

[0344] When the feature value is less than the threshold, the average change in predicted probability is: 0.0068;

[0345] (5) Number of CDG mutant genes in SNV / 566 - Impact strength: 0.0022, direction: negative;

[0346] Normal range: [-0.5634-1.4703];

[0347] Critical threshold: 1.2048 (feature range: -0.5634-4.2362);

[0348] The effect of high value is to reduce the probability of positive class, and the effect of low value is to increase the probability of positive class;

[0349] When the feature value is greater than the threshold, the average change in predicted probability is: -0.0008;

[0350] When the feature value is less than the threshold, the average change in predicted probability is: 0.0015;

[0351] Summary of prediction rules:

[0352] The following are the features and their threshold rules that have the greatest impact on the predicted probability of the positive class:

[0353] (1) When SBS88 / SNV>0.2892, the probability of predicting the positive class increases;

[0354] Normal range: [-0.5282-1.2033];

[0355] (2) When age>0.2005, the probability of predicting the positive class increases;

[0356] Normal range: [-0.8520-0.9764];

[0357] (3) When the total number of CDG mutations in ID18 / total INDELs is <-0.2192, the probability of predicting the positive class increases;

[0358] Normal range: [-0.3856-0.1496];

[0359] (4) When ID18 / INDEL>-0.6148, the probability of predicting the positive class increases;

[0360] Normal range: [-0.9245-0.5441];

[0361] (5) When the number of CDG mutation genes in SNV / 566 < 1.2048, the probability of predicting the positive class increases;

[0362] Normal range: [-0.5634-1.4703]

[0363] Model prediction mechanism:

[0364] According to the above-mentioned influence of different feature data on the prediction results, the model will comprehensively evaluate the probability of the positive class based on the features it obtains, that is, the probability of cancer. When the comprehensively calculated probability of cancer is less than 50%, the model evaluates the sample as a negative sample, that is, a normal sample. If it is greater than 50%, it is judged as a cancer sample. We can use the probability of predicting it as a positive class to evaluate the risk of cancer. For example, in the validation set of the rf-XGBoost model, the P20 sample is actually a cancer sample. The predicted positive score obtained by the model evaluation is 0.878, which means that it is predicted to be a cancer patient with a probability of 87.8%. In the normal sample P5, the model predicts a positive score of 0.248, which means that the model predicts that the sample is a negative sample with a probability of 75.2%. According to the constructed model, we will call each of the three omics models for sample prediction, use the model's AUC value ratio for weighting, and comprehensively predict the model results. The calculation formula is as follows:

[0365] There are 11 models in total, one of which has an AUC value lower than 0.7, so this model is removed. The AUC value of each model is AUC k , (where k = 1, 2, ..., 11), then the weight w of the kth model k , can be calculated as:

[0366]

[0367] in: is the sum of the AUC values ​​of all models; w k is the weight of the k-th model, satisfying all w k The total is 1.

[0368] Calculate the combined probability:

[0369] P(comprehensive)=w1*P1+w2*P2+…+w 11 *P 11

[0370] Risk levels are divided according to the comprehensive probability P(comprehensive):

[0371]

[0372] The following supplementary figures are the experimental and theoretical basis of the present invention:

[0373] Figure 4 This is the electrophoresis result of detecting pks+ Escherichia coli samples in fecal samples from different sources using pks genomic island-specific primers. A+, B+, N+, and Q+ are positive controls for clb A, clb B, clb N, and clb Q primers, respectively. PV1, PV5, and PC9 are the sample numbers. PV1 and PV5 are from healthy individuals, and PC9 is from a colorectal cancer patient. The electrophoresis diagram shows that the four primer pairs can screen out positive samples in fecal DNA. The amplified products of these samples have been sent to Shanghai Biotech for sequencing, confirming that they are the gene sequences of the PKS island.

[0374] Figure 6 and Figure 7 This figure shows the electrophoresis results of detecting pks+ Escherichia coli samples in fecal samples from different sources using a mixture of pks genomic island-specific primers clbA and clbN. Multiplex detection 2+ is a positive control for the mixed primers clbA and clbN, and numbers such as PV5 are sample numbers. The figure shows that multiple primers can be used to amplify multiple target bands with high specificity.

[0375] Figures 8-12 The methylation levels of the five methylated molecular clocks selected by the inventors in their previous studies were significantly linearly correlated with age. Figure 8 It can be seen that with the increase of age, the methylation level of each methylation molecular clock increases, and there is a relatively obvious positive correlation.

[0376] Meanwhile, the previous inventors isolated normal tissue blocks and tumor tissue blocks from the intestinal tissue of tumor patients, respectively determined the methylation degree of the five methylation molecular clocks, and the results are as follows, as can be seen from the figure, the methylation degree of different parts of the intestinal tissue in the same tumor patient has obvious difference, the methylation degree of the molecular clock of the tumor part is obviously higher than that of the normal tissue part, which shows that under normal circumstances, the methylation molecular clock of the cells of the human intestinal tract should be in a low methylation state, but due to the malignant proliferation of tumor cells, the methylation degree is greatly increased, and this result lays a theoretical foundation for judging the proliferation degree of the intestinal tissue by the methylation degree of the present application.

[0377] We extracted the fecal DNA of healthy people and tumor patients, after bisulfite conversion, PCR amplification and sequencing, it was verified that the methylation molecular clock ZNF454 of healthy people was significantly lower than that of tumor patients, which was consistent with the conclusion obtained by using tumor tissue in the previous stage, at the same time, it was also verified that the methylation degree of the fecal sample could be determined, so the fecal tissue could be used instead of tumor tissue to implement a non-invasive detection method, such as Figure 6 The comparison chart of ZNF454 gene methylation state of healthy people and intestinal cancer patients, wherein the abscissa represents the CpG site, and the ordinate represents the methylation degree, PV6 is a healthy person, and PC9 and PC12 are intestinal cancer patients.

[0378] The inventors previously sequenced the genome of tumor tissue of tumor patients, and found that the methylation degree of the methylation molecular clock had a relatively obvious increasing relationship with the number of mutations of cancer driver genes Figure 13 ), and as the number of mutations of cancer driver genes increased, the methylation degree also increased, so according to this result, it can be known that the methylation degree of the molecular clock has an influence relationship with the mutation of the cancer driver gene.

[0379] (1) Early detection and diagnosis: This multi-omics evaluation method can comprehensively consider various biomarkers in the intestinal tract, including genes, microorganisms, and epigenetic characteristics (such as methylation), through a non-invasive method (such as fecal sample analysis), risk assessment can be performed in the early stage of intestinal cancer, thereby greatly improving the early detection rate of intestinal cancer.

[0380] (2) Improve patient management: Through regular non-invasive detection, doctors can timely adjust treatment plans to ensure the best treatment effect for patients.

[0381] (3) Guide the development of treatment strategies: By understanding the genetic, microbial, and epigenetic characteristics of patients, doctors can develop more personalized treatment strategies for patients, such as targeted treatment for specific cancer driver genes, or microecological regulation treatment according to the imbalance of intestinal microbial communities.

[0382] Economic effect:

[0383] (1) Reduce medical costs: Traditional colonoscopy is costly and requires professional medical equipment and personnel operation. The present application greatly reduces the detection cost and reduces the burden on the medical system through non-invasive fecal sample analysis.

[0384] (2) Improve screening coverage: Due to the simple, non-invasive and low-cost detection method of the present application, it can be widely used in large-scale screening, improve the coverage of colorectal cancer screening, and reduce the high medical expenses caused by the treatment of advanced colorectal cancer.

[0385] Technical effects:

[0386] (1) Multi-omics joint analysis: The present application combines the information of intestinal pks+ E. coli markers, cancer driver genes and methylation molecular clock at multiple levels by designing three different Marker models and combining clinical samples, realizing comprehensive evaluation of colorectal cancer risk. This multi-omics joint analysis method can more accurately reflect the biological characteristics of colorectal cancer, improve the accuracy and sensitivity of diagnosis.

[0387] (2) Non-invasive detection: Compared with traditional invasive detection methods such as colonoscopy, the present application uses a non-invasive and economical method to collect and analyze patients' fecal samples to predict colorectal cancer risk. This method not only reduces the pain and discomfort of patients, but also reduces the risk in the detection process.

[0388] (3) Improve the accuracy of risk assessment: Combined with multiple biomarkers for risk assessment, it can provide more comprehensive and accurate information. Compared with traditional single biomarkers, multi-omics assessment method can more accurately identify high-risk groups of colorectal cancer, reducing the possibility of missed diagnosis and misdiagnosis.

[0389] Finally, it should be noted that: the above only for the preferred examples of the present application, and not for limiting the present application, although the present application has been described in detail with reference to the foregoing examples, for those skilled in the art, it still can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. Based on PKs + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: This includes extraction of fecal DNA samples, primer design and screening, triple specificity validation, non-invasive testing, and multi-omics quantitative scoring models; Based on the detection of pks+ Escherichia coli in feces, pks+ Escherichia coli can synthesize the cytotoxin colicin, which induces DNA double-strand breaks and chromosomal instability in host cells, causing cell cycle arrest and ultimately leading to cell death.

2. The pks-based method according to claim 1 + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: The operation of extracting stool DNA samples includes collecting stool samples from different sources and extracting genomic DNA from the stool using a stool DNA extraction kit.

3. The pks-based method according to claim 1 + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: Design and screening of primers For the design of primers for the PKS gene island, primers for four genes already available in the literature were selected, namely clbA, clbB, clbN, and clbQ. PCR amplification was performed using the genomic DNA of the pks-positive bacterium, Escherichia coli Nissle1917, and the synthesized primers. If a bright and single target band can be amplified, it indicates that the primers have high specificity and are used for subsequent screening of PKS-positive samples. Design five methylation molecular clock primers, namely TFAP2B, GLRA3, CHAT, TCTEX1D1, and ZNF454. Use A375 cell line DNA and the designed primers for PCR amplification. If a bright and single band is generated, it indicates that the primers are highly specific and can be used. Otherwise, continue to optimize the primers.

4. The pks-based method according to claim 1 + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: The triple specificity validation was performed by a single biomarker susceptible to interference from diet and intestinal inflammation.

5. The pks-based method according to claim 4 + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: Triple specificity validation: Detection of the positive rate of pks+ Escherichia coli in feces of healthy people, patients with intestinal cancer and other intestinal diseases: Marker 1: The extracted fecal DNA and PKS gene primers were used for PCR amplification. The PCR amplification products were run on agarose gel electrophoresis and compared with DS TM 2000 DNA marker to compare the band size. If the band size is consistent with the target product and is specific, the sample is sent for sequencing to determine the gene sequence of the sample. After sequencing verification, it is determined whether the amplified product is the target gene. Screening of cancer driver gene mutation signatures, SBS88 / ID18 mutation enrichment analysis distinguishes oncogenic from non-oncogenic mutations: Marker 2: Sequencing of 566 cancer driver genes in fecal DNA. In addition to counting characteristic mutations caused by PKS gene islands, the ratio of various mutations in cancer driver genes is also counted. The characteristic mutations caused by PKS+ E. coli are named SBS88 and ID18. If the mutation occurs in a cancer driver gene, the risk of cancer increases. Methylation molecular clock age correction eliminates age-related interference based on the methylation baseline of healthy people, improving overall specificity: Marker 3: The fecal DNA of the positive sample after bisulfite conversion was PCR amplified with the methylated molecular clock primers, and the transformed A375 cell line DNA was used as a positive control. TM 2000 DNA marker to compare the band size. If the band size is consistent with the target product and is specific, second-generation sequencing is performed to further determine the methylation status of the sample's methylation molecular clock and detect the methylation level of Marker3 methylation molecular clock. The number of cell divisions in a normal and healthy individual is positively correlated with age. If the methylation level is abnormally high, there is a risk of cancer.

6. The pks-based method according to claim 1 + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: The non-invasive test requires the collection of stool samples, which can be completed by sampling at home and transporting at room temperature.

7. The pks-based method according to claim 1 + A multi-omics non-invasive colorectal cancer risk assessment method based on intestinal bacteria, cancer driver genes, and methylation molecular clocks is characterized by: The multi-omics quantitative scoring model integrates the pks+ positive rate, driver gene mutation load, and methylation deviation index to output low, medium, and high risk stratification, providing a basis for individualized intervention.

Images