TW201339310A - Gene set for predicting post-surgery recurrence or metastasis risk in cancer patients and method thereof - Google Patents

Abstract

The present invention provides a gene set for predicting post-surgery recurrence or metastasis risk in cancer patients and a method of using said genes to predict post-surgery recurrence or metastasis risk in cancer patients, especially in liver cancer. The precision rate of predicting post-surgery recurrence in patients with liver cancer according to the present invention is 76.47%; and the precision rate of predicting post-surgery metastasis in patients with liver cancer is 82.61%. Thereby, the best successive treatment method can be evaluated by precisely predicting the recurrence and metastasis risk in patients with liver cancer.

Description

Gene combination and method for predicting risk of postoperative recurrence or metastasis in cancer patients

The invention relates to a gene combination and a method for predicting the risk of postoperative recurrence or metastasis of a cancer patient, in particular to a gene combination and a method for predicting the risk of postoperative recurrence or metastasis of a cancer patient by using the concept of gene-gene interaction, in particular to liver cancer.

Hepatocellular carcinoma (HCC), referred to as liver cancer, is one of the most common cancers in Taiwan. At the same time, liver cancer is also among the best in the world in terms of the incidence and mortality of various types of cancer. According to statistics, about 1 million people worldwide suffer from liver cancer every year, and 5 to 1 million people die from liver cancer. In terms of the region, liver cancer is better developed in developing countries, mainly in East Asia and Africa. More than women, about four to one.

According to the National Health Service of the Department of Health, the incidence of liver cancer in Taiwan is 74.3 per 100,000 people suffering from liver cancer. Liver cancer is the leading cause of all cancer deaths. About 3,700 patients die each year because of liver cancer. The status, causes and risk factors of liver cancer are very important for early diagnosis and tracking.

The causes of liver cancer are mainly divided into two categories: exposure to chemical carcinogens and infection with hepatitis. Chemical carcinogens include male hormones, female hormones, alcohol, general environmental pollutants such as carbon tetrachloride, DDT, dioxin, etc., and in particular, the storage of cereals caused by xanthine fungus caused by xanthine toxin, After long-term use, it is also one of the important causes of liver cancer. According to epidemiological and animal experiments, it has been determined that chronic hepatitis B and hepatitis C virus infection are the most important risk factors for the formation of liver cancer. Bisley found that hepatitis B positive carriers have the opportunity to develop liver cancer. It is 98 times higher than the negative one. At the same time, scholars have found that the incidence of liver cancer in the general population in Taiwan is 10-30 people per 100,000 people, but male hepatitis B patients have 200-812 cases per 100,000 population. The unfortunate development of patients to the stage of cirrhosis, the annual incidence in Taiwan will increase to 1000-5000 cases, so how to avoid and reduce hepatitis virus infection and treatment of hepatitis is an important public health issue.

The most common treatments for liver cancer include surgical resection, transcatheter arterial embolization, percutaneous ethanol intratumor injection, cryopreservation, and transplantation. Among these therapies, only surgical resection and organ transplantation are the most effective. However, most patients cannot be removed at the time of disease discovery because of poor liver function (more than 75% of patients with potentially chronic liver disease), bilateral liver disease, or extrahepatic metastases. Whether a liver cancer patient is suitable for surgical resection depends on the stage of liver cancer and liver function, and only about 15% of liver cancer patients are suitable for surgical resection. If liver cancer cannot be removed, the prognosis is poor and the average survival is only a few months. Even in patients undergoing surgical resection, the 5-year recurrence rate is as high as 63%. The 5-year survival rates of some cancers such as liver cancer, lung cancer, esophageal cancer and gastric cancer are no more than 30%, and the 5-year survival rate of liver cancer is lower than that of other cancers, only 5% to 6%.

Because the recurrence rate is very high (~63%) within five years after surgical resection of liver cancer, the patient will relapse after surgery and the doctor will consider taking liver or medication. However, if the frequency of patient visits is not high, it may be found that hepatic therapy cannot be performed at the time of recurrence. Although there are some clinical features that can be used to judge the probability of short-term recurrence within two years, there is no other way to judge the probability of long-term recurrence. Therefore, if there is a prognostic product that can judge the long-term recurrence rate of liver cancer, you can inform the patient 4 How high is the recurrence rate in ~5 years, on the one hand, it can remind patients to check back regularly during this period, on the other hand, they can arrange for liver replacement or medication treatment as soon as possible.

At present, most post-surgical treatments for cancer are usually based on the experience of physicians to determine whether patients need further active treatment, but this method often results in insufficient judgment accuracy due to different personal experience values. However, cancer is often caused by complex pathogenic causes, and clinically used biochemical indexes (such as AFP or CA-125) or instrumental examinations (such as computed tomography, ultrasound, or magnetic resonance imaging) to track patients after surgery. Whether recurrence is not positive enough, in addition to questioning the correct rate of these biochemical values, usually also represents the extent to which cancer tumors have grown enough to be discovered, even when they have spread. At present, it has been found in international clinical research that gene maps can be used for the prediction of some diseases. The advantage of using gene map prediction is that it can be used to predict the incidence of diseases even if the mechanism of action of the disease is unclear. Therefore, the present invention utilizes specific gene expression profiling to predict the probability of recurrence or metastasis of liver cancer, so that patients can receive treatment early and prolong their survival time.

In order to accurately predict the risk of liver cancer recurrence and metastasis to solve the above problems, it is an object of the present invention to provide a gene combination for predicting the risk of postoperative recurrence of liver cancer patients, which is the nucleic acid sequence shown in SEQ ID NOS: 1 to 33. Any combination of at least two genes.

In one embodiment of the present invention, the above gene combination is the nucleic acid sequence shown in SEQ ID NOS: 1 to 6, 8 to 10.

Using the above combination of genes, the following steps can be used to predict the risk of postoperative recurrence in liver cancer patients:

(1) Obtaining a tumor tissue and a surrounding normal tissue of a liver cancer patient;

(2) analyzing a first gene combination and a second gene combination of the tumor tissue and the surrounding normal tissue, wherein the first gene combination is composed of the nucleic acid sequences shown in SEQ ID NOS: 1 to 3, and The second gene combination is a gene combination II (a) composed of the nucleic acid sequences shown in SEQ ID NOS: 4 to 6, or a gene combination composed of the nucleic acid sequences shown in SEQ ID NOS: 4 to 5, 7. (b) dividing the gene expression in the gene combination into the tumor tissue and the surrounding normal tissue to obtain the expression ratio of each gene in the combination of the genes, wherein the nucleic acids shown in SEQ ID NOS: 1 and 4 The sequence is a shared gene, and a first gene combination interaction value and a second gene combination interaction value are calculated from the remaining genes in the gene combination;

(3) when the first gene combination interaction value is lower than a first threshold value and the second gene combination interaction value is higher than or equal to a second threshold value, the patient is determined to have a high risk of recurrence, wherein The first gene combination interaction value is multiplied by the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 1 and SEQ ID NO: 2, and the nucleic acid sequences shown in SEQ ID NO: 1 and SEQ ID NO: As a result of multiplication of the gene expression ratio, the interaction value of the second gene combination II(a) is multiplied by the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 4 and SEQ ID NO: 5 plus SEQ ID NO: 4 multiplied by the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 6, the interaction value of the second gene combination II (b) is the nucleic acid sequence shown in SEQ ID NO: 4 and SEQ ID NO: 5. The gene expression ratio is multiplied together with the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 4 and SEQ ID NO: 7.

Another object of the present invention is to provide a combination of genes for predicting the risk of postoperative metastasis in a liver cancer patient, which is any combination of at least two of the nucleic acid sequences shown in SEQ ID NO: 28, 34-42.

In one embodiment of the present invention, the aforementioned gene combination is the nucleic acid sequence shown in SEQ ID NO: 28, 34 to 38.

Using the above combination of genes, the following steps can be used to predict the postoperative risk of liver cancer patients:

(1) Obtaining a tumor tissue and a surrounding normal tissue of a liver cancer patient;

(2) analyzing the tumor tissue and one of the surrounding normal tissues, the fifth gene combination is composed of the nucleic acid sequences shown in SEQ ID NO: 28, 34 to 35, and the gene in the fifth gene combination is in the tumor tissue and The gene expression intensity of the surrounding normal tissues is divided, and the gene expression ratio of each gene is obtained, wherein the nucleic acid sequence shown by SEQ ID NO: 34 is a shared gene, and the nucleic acid sequence shown in SEQ ID NOS: 35 and 28 is calculated. Five gene combination interaction values;

(3) When the fifth gene combination interaction value is lower than a fifth threshold value, the patient is determined to have a high risk of metastasis; wherein the fifth gene combination interaction value is SEQ ID NO: 34 and SEQ ID NO: 35 The gene expression ratios of the nucleic acid sequences shown are multiplied together with the gene expression ratio of the nucleic acid sequences shown in SEQ ID NO: 34 and SEQ ID NO: 28.

It is still another object of the present invention to provide a method for obtaining a gene predicting a prognosis of a cancer patient, comprising:

(1) taking one of a plurality of cancer patients to take a tissue sample;

(2) obtaining a gene expression value for each gene of the extracted tissue sample;

(3) selecting a plurality of gene pairs with strong gene-gene interaction according to a type of prognosis, and classifying the genes into a plurality of gene clusters according to a plurality of shared genes, wherein the shared gene systems are Genes that are identical between pairs; and

(4) In each of the gene clusters, at least a group of genes is numerically multiplied according to the degree of gene interaction, and then a decision tree model is used to generate a prediction according to the prognostic type. The prognostic type of gene.

In one embodiment of the present invention, the gene expression value of the above gene combination is a ratio of the expression of the mRNA transcribed from each gene of the tumor of the liver cancer patient and the surrounding tissue or the amount of the polypeptide translated from the gene.

In another embodiment of the present invention, the prognosis type is recurrence or metastasis; and the gene pair with strong gene-gene interaction in step (3) is a gene satisfying formula (1):

-log ₁₀ ( p ( v _A × v _B )) C _th -log ₁₀ ( p ( v _A ))-log ₁₀ ( p ( v _B )) (1).

Through the gene combination of the present invention and the method for predicting the prognosis of liver cancer patients, the accuracy rate of predicting postoperative recurrence of liver cancer patients is 76.47%, and the accuracy of predicting postoperative metastasis of liver cancer patients is 82.61%, which shows that the present invention is The combination of genes and methods for predicting postoperative prognosis in patients with liver cancer can provide an effective method for predicting the risk of postoperative recurrence or metastasis in patients with liver cancer.

Through the disclosure of the present specification, any skilled person skilled in the art can use the prior art to make a biochip or microarray analysis of the gene combination of the present invention, and predict the liver cancer patient using the test results of the biochip or microarray analysis. Risk of recurrence or metastasis.

The embodiments of the present invention are further described in the following description, and the description of the present invention and the following examples are intended to illustrate the invention and provide further explanation of the scope of the patent application of the present invention.

definition

The "shared gene" referred to in the present invention is the same gene among all gene pairs in a gene cluster.

The "pairing gene" referred to in the present invention is a gene paired with a gene-sharing gene; in the gene combination, the "pairing gene" is the remaining gene in the gene combination except the shared gene.

The "interaction value" referred to in the present invention is the sum of the gene expression ratios of the shared genes and the gene expression ratios of the first to n paired genes, respectively, where n is the total number of paired genes calculated.

It is an object of the present invention to provide a combination of genes that interact with each other, including any of the genes listed in Tables I and III, wherein the sequence numbers of the genes are listed in Tables 2 and 4, respectively.

The present invention also encompasses any of the sub-columns of the same column listed in Table 1, for predicting whether a liver cancer patient has relapsed or not relapsed within 7.3 years. A preferred sub-combination of the nine gene formations is shown in Table 5.

A further object of the present invention is to provide a genetic composition comprising 11 genes as shown in Table 3 for predicting a condition in which a liver cancer patient has metastasized or not metastasized within 7.3 years.

The present invention also encompasses any of the sub-columns of the same column listed in Table 3 for predicting whether a liver cancer patient has metastasized or not metastasized within 7.3 years. A preferred combination of the six gene formations is shown in Table 6.

In the present invention, we propose a novel model for detecting gene-gene interaction. The following examples classify genes into gene clusters based on shared genes and developmental algorithms, and process the gene clusters with a decision tree to find a distinction between recurrence and non-recurrence of liver cancer. Transferred or untransferred genes, and the association of the aforementioned genes with each other.

In the embodiment of the present invention, "recurrence" is defined as recurrence of liver cancer in the liver from 12 months to the last tracking date after surgery, and if there is no recurrence of liver cancer during the follow-up period, it is "no recurrence"; It is defined as a liver cancer cell that has spread to tissues other than the liver during the entire tracking period. If the liver cancer cells are not spread during this period, it is "not transferred". In addition, the aforementioned tracking period refers to the time from the postoperative period to the last tracking date of the patient, and the period of the tracking period of the patient used in the embodiment of the present invention is 7.3 years.

Example 1: Method of screening genes 1-1 Patient conditions and sample collection

The study was approved by the Human Body Testing Committee of the Taiwan University Hospital and the subject consents for all subjects have been obtained. Patients with hepatocellular carcinoma (hereinafter referred to as liver cancer) are divided into two groups: a training set and a test set. Both groups were confirmed by pathological analysis of liver cancer, and all surviving patients were followed for at least four years. Each patient obtained a specimen of the tumor and surrounding normal tissue, and the tissue examination system was obtained by surgically removing the cancer tissue or by thick needle sectioning. There were 45 patients with liver cancer in the training group. The average age was 52.4±12.5 years old. All of them were infected with hepatitis B (HBV) and mild fibrosis in the liver (METAVIR grade was F1-F2), in which liver cancer recurred. There were 21 patients with liver cancer who relapsed, and 17 patients did not relapse. In terms of liver cancer metastasis, 12 patients had distant metastases and 33 patients did not metastasize. Among them, 5 patients had recurrence and metastasis. On the other hand, there were 24 patients with liver cancer in the test group, 2 were neither HBV nor HCV, 12 were HBV, 12 were HCV, and 2 of them were HBV and HCV. Among them, there were 18 males with an average age of 60.8±8.2 years, and the average age of 6 females was 66.8±10.5 years old. For the 24 patients with liver cancer in this test group, 12 patients had recurrence of liver cancer, and 5 patients had no recurrence. In terms of liver cancer metastasis, 5 patients had liver cancer metastasis and 18 patients did not metastasize.

In the embodiment of the present invention, the data of the patient in the training group is used as the training of the algorithm of the present invention to obtain a gene that may predict the prognosis of liver cancer, and then verified by the patient data in the test group.

1-2 RNA isolation and microarray

Liver frozen tissue sections were homogenized with TRIzol reagent (Invitrogen, Cat. no.: 15596026, Life Technologies). Total RNA was extracted using the After Tri-reagent RNA clean-up kit (Favorgen, Cat. no.: FAATR001) according to the manufacturer's standard protocol. The amount of RNA was measured by spectrophotometer and the RNA size was measured by RNA Nano LabChips on an Agilent 2100 Bioanalyzer. Each sample produced an RNA Integrity Number (RIN).

In ^{Amino Allyl MessageAmp TM II aRNA amplification Kit} (Ambion, Austin, TX) amplified RNA, this amplification method is based on the T7 promoter [Van Gelder RN, von Zastrow ME , Yool A, Dement WC, Barchas JD, Eberwine JH. Amplified RNA synthesized from limited quantities of heterogeneous cDNA. Proc Natl Acad Sci US A. 1990 Mar; 87(5): 1663-7]. This experiment uses two versions of Human (HOAv4 & HOAv5), using different fluorescent staining methods. The training group included HOAv4 (45 patients × 2 tissues × 3 hybridization tests = 270 wafer results) and HOAv5 (35 patients × 2 tissues × 1 hybridization test = 70 wafer results) data results, HOAv5 patients included Within the HOAv4 patient population, and the HOAv5 data was only used to select stable genes, but not for the calculation of gene expression values and the establishment of a decision tree model. The test group used only HOAv4 (24 patients x 2 tissues x 3 hybridization tests = 144 wafer results). The HOAv4 assay uses 20 μg of Cy5-calibrated aRNA according to Fong [Fong S et al., Molecular mechanisms underlying selective cytotoxic activity of BZL101, and extract of Scutellaria barbata towards breast cancer cells. Cancer Biol Ther. 2008 Apr: 7(4) :577-86] ARNA fragmentation is performed on the wafer. The HOAv5 assay used a TotalPrep RNA amplification kit (Ambion) commercially available product, and 10 μg of biotinylated aRNA was fragmented and hybridized to the wafer, followed by subsequent staining with Cy3-Streptavidin (PA43001, GE Healthcare). Finally, the array was scanned with a DNA microarray scanner (Agilent Technologies, Santa Clara, US) and used as GenePix Flu 4.0 signal intensity was captured by Pro 4.0 (Molecular Devices, Sunnyvale, CA).

1-3 Data collection and processing

For each gene in the HOAv4 data array, the expression intensity value is calculated from the difference between the foreground median intensity at 635 nm and the background median intensity at 635 nm. Yes, and the HOAv5 data is calculated at a wavelength of 532 nm. For each tissue sample, the sample with a correlation coefficient below 0.9 was removed from the replicates, and the remaining replicate samples were combined with the gene expression intensity values by the median method. The gene expression ratio is obtained by dividing the gene expression intensity of the tumor tissue with the surrounding normal tissue, and then the sample and the standard deviation of the gene expression ratio are sampled by z-score normalization. Regularization between the two. Finally, we selected 6,505 genes with a gene expression intensity of at least 500 and a correlation of at least 0.4 on HOAv4 wafers with HOAv5 wafers. According to the number of samples recommended by Tsai et al. (2005) [Tsai CA et al. (2005). Sample size for gene expression microarray experiments. Bioinformatics, 21(8): 1502-8], our training group (each group) Number of samples The correct rate of 12 and the number of genes = 6,505) can reach more than 0.99, or the expected value of the false positive is 1, and the expected value of the averaged effect (standardized effect size) is 2, the family test force The specificity of (family-wise power) at 90% was 0.90. In addition, the above-mentioned values are further improved by repeating the test using the technique of the wafer and the gene expression ratio or permitting.

Example 2: 2-1 Detection of gene-gene interaction and construction of gene clusters

Recent research indicates that gene-gene interactions may play a more important role than a single gene in a variety of complex diseases. We propose to detect gene-gene interactions by identifying the product of the gene expression values of the two genes in the microarray between the case (relapse or metastasis) and the control group (non-recurring or non-metastatic) than the individual genes. The gene expression value may be a single gene expression intensity of a single tissue, or a gene expression ratio in which the tumor tissue is divided by the intensity of the gene expression of the surrounding normal tissue. In one embodiment of the present invention, the gene expression ratios of genes A and B are set to v _A and v _{B , respectively} . The above ideas can be presented by:

-log ₁₀ ( p ( v _A × v _B )) C _th -log ₁₀ ( p ( v _A ))-log ₁₀ ( p ( v _B )) (1)

p ( v ) is the Student's T test value between the case and the control group, and C _th is a normal number. It was found that the gene pair (1) had a high correlation between the case group and the control group, and the z-score normalization method used in the pre-processing step allowed us to distinguish that the two interaction genes were highly positively correlated ( Synergy) or highly negative correlation (antagonism). In addition, many interaction gene pairs share a shared gene and can be used for classification of groups of subjects. We aggregate these gene pairs to form gene clusters, or permit them to clarify some disease mechanisms.

2-2 Building a Decision Tree Based on Gene Clusters

Assuming that each gene cluster constructed represents a known mechanism, the next step will determine which mechanism is most influential for relapse/non-recurrence or metastasis/non-metastasis. Here we use the easy-to-interpret decision tree model to build prediction models. We first sort the gene pairs in the gene cluster according to the degree of significance ( sig _ diff ). The formula for the degree of significance is as follows:

Sig _ diff =log ₁₀ ( p ( v _A ))+log ₁₀ ( p ( v _B ))-log ₁₀ ( p ( v _A × v _B )) (2)

The values of the ratios of the gene expression ratios of the first n genes in a gene cluster are summed to represent their hidden mechanisms. This approach is to maintain a high discrimination grouping effect to reduce the non-specific noise effects of a single gene pair. When constructing a decision tree, the variables of high discrimination are taken into consideration, so the ordering of variables will affect the resulting decision tree structure. We rank these variables by the number of genes contained in the gene cluster, as larger gene clusters may play a more important role in the underlying mechanism of decision making (prognostic types, such as recurrence or metastasis). Then, multiple decision trees are constructed by using different degrees of reproducible genes and gene pairs in the gene cluster. Finally, the final decision tree is determined by the 0.632 bootstrap [Efron B. Estimating the error rate of a prediction rule: improvement on cross-validation. J. Am. Stat. Assoc. 1983; 78, 316-331]. The 0.632 bootstrap method is a statistical method suitable for small sample numbers [Braga-Neto UM and Dougherty ER. Is cross-validation valid for small-sample microarray classification? Bioinformatics. 2004 Feb 12; 20(3): 374-80.]. The algorithmic logic for determining the appropriate gene pairs in each gene cluster and constructing the decision tree is as follows:

Step 1: The wafer experiment of Example 1.2 was selected by screening in Example 1.3 to select the results of the gene expression ratio of the HOAv4 and HOAv5 wafers of the qualified 23 patients, and the statistical correlation between the gene expression ratio and HOAv4 and HOAv5 was calculated. And select all genes whose correlation is greater than the threshold (ie, corr_thres , here greater than or equal to 0.4);

Step 2: Select the qualified gene group that meets step 1 from the HOAv4 data of 45 patients in the training group, and further select the gene pair that conforms to formula (1);

Step 3: Group the gene pairs by sharing genes to construct a gene cluster. Assume that the number of gene clusters has a total of noc;

Step 3.1: The gene clusters are ordered by the number of genes contained in the gene cluster, assuming that the number of gene pairs in each gene cluster is nopi, where i is the gene cluster number, i=1, 2, ... noc;

Step 3.2: The gene in each gene cluster is subjected to a descending order of the gene-gene interaction significance degree obtained by the formula (2);

Step 4: Pick a decision tree with one of the smallest average expected errors from the following decision trees.

Step 4.1: construct an input matrix M by using the pre-p (p=1~the largest number of gene pairs in each gene cluster) gene pair in each gene cluster;

Step 4.1.1: Calculate the array v(vector) data, where v=[v ₁ ,v ₂ ,...,v _c ,...,v _noc ] ^T , and v _c =[v _c,1 ,v _c,2 ,...,v _c,i ,...,v _{c,min(p,nopi)} ] ^T ,v _c,i = (The gene expression ratio of the kth gene pair in a gene cluster is multiplied). According to the above logic, an array v is established for each patient;

Step 4.1.2: The above array v is arranged in a straight line to generate an input matrix M;

Step 4.2: construct a gene cluster using the matrix M and the prognostic type classification;

Step 4.3: Calculate the average expected error of the constructed decision tree using the 0.632 bootstrap method.

2-3 proportion of patients with uneven proportion of patients

Our training group included 12 patients with metastases and 33 patients with no metastases. The proportion of patients with such small sample size and uneven proportion (the number of patients who have not metastasized is almost three times the number of patients transferred) may result in mismatching of gene pairs and poor prediction of classification models [Gustavo EAPA Batista, Ronaldo C. Prati, and Maria Carolina Monard. A study of the behavior of several methods for balancing machine learning training data. In Special issue on learning from imbalanced datasets, volume 6, pages 20-29, 2004.]. To reduce misclassification of gene pairs, random under-sampling is used in pattern establishment [Ricardo Barandela, Rosa M. Valdovinos, J. Salvador Sanchez, and Francesc J. Ferri. The imbalanced training sample problem : Under or over sampling? Lecture Notes in Computer Science, 3138: 806-814, 2004]. Therefore, 10 sets of test data were generated from the original test group data, each test data contained 12 transfer patients and 12 untransferred patients were randomly selected from the original 33 untransferred patients. A search for a gene pair satisfying the above formula (1) is searched for in each test data, and if the gene pair appears in at least five sets of test data, it is retained and used to construct a gene cluster. The median value of each of the qualified genes for the 10 significant degree values (calculated according to the above formula (2)) calculated from the 10 sets of test data is the final significant degree value of the pair.

In addition to addressing the uneven proportion of data, we also adjust the cost matrix in the training decision tree [Japkowicz N, Class Imbalances: Are we Focusing on the Right Issue?, In Proceedings of the ICML'03 Workshop on Learning From Imbalanced Data Sets, 2003] to improve the accuracy of classification results. The cost matrix is a two-dimensional square matrix M _c , where M _c ( i , j ) is the cost of classifying a subject as type i into type j. In this prediction model, the cost matrix is set to Mc=[0 33; 12 0], because in this case, the liver cancer metastasis classification is less than the non-metastasis classification, so if a metastasis patient is divided into no When transferring, a penalty weight is given to reduce the probability of such classification errors.

Example 3: Classification and outcome of patients with recurrent and non-recurrent liver cancer 3-1 Construction and selection of gene clusters

Gene is only taken into account for the strong interaction, C _th embodiment in the formula (1) of the embodiment of the present invention is set to 1.5. The p- value indicating the multiplication of the two gene expression ratios (Student's t-test) is at least 10 ^1.5 = 31.62 times smaller than the multiplicative value of the individual p values of the two gene expression ratios. There are 2,416 gene pairs that match this setting.

3-2 Decision Tree for Constructing Recurrence

The decision tree determined by the minimum classification error calculated by the 0.632 bootstrap method is as shown in the first figure, and the selected genes in the gene cluster are shown in Table 5. When the number of gene clusters is 541 ( noc = 541), the best decision tree is selected. When doing the decision tree training, each gene cluster is calculated by the input of the previous two gene pairs ( nop = 2). However, the optimal decision tree uses only three gene clusters as shared genes: EXOC6 (SEQ ID NO: 1), PDDC1 (SEQ ID NO: 4) and TCEAL1 (SEQ ID NO: 8) gene clusters.

The classification accuracy of this decision tree in the HOAv4 training group is 100%. After applying this decision tree to the test group, the classification accuracy rate was 76.47% (13/17*100%), representing 13 patients with 17 liver cancer patients were accurately classified; the sensitivity was 0.83 (10/) 12), 10 patients with 12 recurrent liver cancers were correctly classified as relapse; specificity was 0.60 (3/5), representing 3 patients with no recurrence and 3 were correctly classified as non-recurrent. In the training group and the test group, this decision tree has high sensitivity and specificity, which can be confirmed as a method for effectively predicting the recurrence or non-recurrence of liver cancer.

The sequences of each gene in Table 5 are listed in the sequence listing: EXOC6 (SEQ ID NO: 1), ZNF692 (SEQ ID NO: 2), PTGDS (SEQ ID NO: 3), PDDC1 (SEQ ID NO: 4), MYBL2 ( SEQ ID NO: 5), GNG13 (SEQ ID NO: 6), TCEAL1 (SEQ ID NO: 8), BANK1 (SEQ ID NO: 9), and Clord54 (SEQ ID NO: 10).

From the first figure, it can be seen that the classification of the first node consisting of EXOC6, ZNF692, and PTGDS and the second node consisting of PDDC1, MYBL2, and GNG13 has a prediction accuracy of 94% in the training group. The prediction accuracy of the test group was 59%. After the classification of the third node, the prediction accuracy of the training group reached 100%, while the prediction accuracy of the test group was 76.47%.

In addition, other decision trees can be obtained using the same method, that is, the decision tree of the gene composition shown in Table 1, each column represents a decision tree, and the sequence numbers of the genes are listed in Table 2, and the gene combinations shown in Table 1 are shown. Can also get a good prediction of the risk of liver cancer recurrence.

Example 4: turn Classification and results of patients with or without metastatic liver cancer 4-1 Selection and construction of gene clusters

After adjusting the data of the training group with uneven proportion and the C _{th of the} formula (1) was set to 1.5, 347 gene pairs were obtained in the transferred data.

4-2 Decision tree for constructing transfer or not

The decision tree determined by the minimum classification error calculated by the 0.632 bootstrap method is shown in the second figure, and the genes selected in the gene cluster are shown in Table 6. When the number of gene clusters is 13 ( noc = 13), the best decision tree is selected. When doing the decision tree training, each gene cluster is calculated by the input of the previous two gene pairs ( nop = 2). However, the best decision tree uses only two gene clusters as shared genes: BMP4 (SEQ ID NO: 10) and CYP2E1 (SEQ ID NO: 13).

The classification accuracy of this decision tree in the HOAv4 training group was 98% (44/45*100%), representing 44 patients with 45 liver cancer patients being accurately classified; the sensitivity was 0.92 (11/12). The 10 patients with liver cancer who represented 12 metastases were correctly classified as metastasis; the specificity was 1.0 (33/33), indicating that all 33 patients who did not metastasize were correctly classified as non-metastasis. After applying this decision tree to the test group, the classification accuracy rate was 82.61% (19/23*100%), representing 19 patients with 23 liver cancer patients being accurately classified; the sensitivity was 0.40 (2/). 5), 2 patients with liver cancer representing 5 metastases were correctly classified as metastasis; specificity was 0.94 (17/18), representing 17 patients with no metastasis were correctly classified as non-metastasis. In the second picture, four misclassified patients (represented by gray bottom) included two male metastatic patients, one female metastatic patient, and one female non-metastatic patient. In addition, a misclassified male metastatic patient is a liver cancer patient with HBV infection, and the remaining misclassified patients are HCV-infected liver cancer patients. This decision tree has high sensitivity in the training and test groups, so it can be an effective way to distinguish between cancer patients with or without metastasis.

The sequences of each gene in Table 6 are listed in the sequence listing: BMP4 (SEQ ID NO: 10), C5orf32 (SEQ ID NO: 11), DTX3 (SEQ ID NO: 12), CYP2E1 (SEQ ID NO: 13), YARS ( SEQ ID NO: 14) and CLDN4 (SEQ ID NO: 15).

It can be seen from the second figure that the first node consisting of BMP4 , C5orf32 , and DTX3 has a prediction accuracy of 91% in the training group and 74% in the test group. After classification, the prediction accuracy rate in the training group was 98%, while the prediction accuracy of the test group was 82.61%.

In addition, the same method can be used to obtain other decision trees, namely the decision tree of the gene composition shown in Table 3. Each column represents a decision tree. The sequence numbers of these genes are listed in Table 4, and the gene combinations shown in Table 3 are shown. Can also get a good prediction of liver cancer metastasis risk results.

It can be seen from the above embodiments that the method of the present invention calculates the gene combination and method for predicting the prognosis of liver cancer patients by the training group, wherein the accuracy of predicting the recurrence of liver cancer patients is 76.47%, and predicting the postoperative metastasis of liver cancer patients. The accuracy rate is 82.61%. This result shows that the combination of the resulting decision tree with the gene combination of the present invention can provide an effective method for predicting the risk of postoperative recurrence or metastasis of liver cancer patients.

In addition, biochip or microarray analysis can be made by the above-mentioned obtained gene combination using conventional techniques, and the test results of the biochip or microarray analysis can be used to predict the risk of postoperative recurrence or metastasis of liver cancer patients.

While the embodiments of the present invention are disclosed as described above, it is not intended to limit the scope of the present invention, and those skilled in the art can make some modifications and retouchings without departing from the spirit and scope of the present invention. The scope of protection is subject to the definition of the scope of the patent application attached.

<110> Hualian Biotechnology Co., Ltd.

<120> A genetic combination and method for predicting the risk of postoperative recurrence or metastasis in cancer patients

<160> 57

<210> 1

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> EXOC6

<400> 1

<210> 2

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> ZNF692

<400> 2

<210> 3

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> PTGDS

<400> 3

<210> 4

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> PDDC1

<400> 4

<210> 5

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> MYBL2

<400> 5

<210> 6

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> GNG13

<400> 6

<210> 7

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> ROBO3

<400> 7

<210> 8

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> TCEAL1

<400> 8

<210> 9

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> BANK1

<400> 9

<210> 10

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> C1orf54

<400> 10

<210> 11

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> BANP

<400> 11

<210> 12

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> LRRC14

<400> 12

<210> 13

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> LOC730427

<400> 13

<210> 14

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> ELOVL7

<400> 14

<210> 15

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> KIAA1522

<400> 15

<210> 16

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> EVI2B

<400> 16

<210> 17

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> MBOAT1

<400> 17

<210> 18

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> CENPO

<400> 18

<210> 19

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> LMBR1L

<400> 19

<210> 20

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> SPOCK2

<400> 20

<210> 21

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> GNL3L

<400> 21

<210> 22

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> DCTN5

<400> 22

<210> 23

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> ZNF8

<400> 23

<210> 24

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> STK32C

<400> 24

<210> 25

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> ZNF480

<400> 25

<210> 26

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> CHAF1A

<400> 26

<210> 27

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> PPP1R13L

<400> 27

<210> 28

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> DTX3

<400> 28

<210> 29

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> CDCA4

<400> 29

<210> 30

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> NCOR2

<400> 30

<210> 31

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> SLAMF8

<400> 31

<210> 32

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> PTPRO

<400> 32

<210> 33

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> DYNC2H1

<400> 33

<210> 34

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> BMP4

<400> 34

<210> 35

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> C5orf32

<400> 35

<210> 36

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> CYP2E1

<400> 36

<210> 37

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> YARS

<400> 37

<210> 38

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> CLDN4

<400> 38

<210> 39

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> WISP2

<400> 39

<210> 40

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> LIMK2

<400> 40

<210> 41

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> GAL3ST4

<400> 41

<210> 42

<211> 60

<212> DNA

<213> Artificial sequence

<220>

<223> AK1

<400> 42

The first graph is a decision tree generated by recurrence or non-recurrence classification of liver cancer, and the results of classifying patients in the training group and the test group using this decision tree.

The second graph is the decision tree generated by the liver cancer metastasis or non-metastasis classification, and the results of classifying the patients in the training group and the test group using this decision tree.

Claims (25)

A gene combination for predicting the risk of postoperative recurrence of a liver cancer patient is any combination of at least two of the nucleic acid sequences shown in SEQ ID NOS: 1 to 33. The gene combination according to claim 1, wherein the gene combination is the nucleic acid sequence shown in SEQ ID NOS: 1 to 6, 8 to 10. The combination of genes as described in claim 1 is for biochip, microarray or real-time quantitative polymerase technology analysis. A method for predicting the risk of postoperative recurrence of liver cancer patients from a liver cancer tissue comprises: (1) obtaining a tumor tissue and a surrounding normal tissue of a liver cancer patient; (2) analyzing the tumor tissue and one of the surrounding normal tissues. a combination of a gene and a second gene combination, wherein the first gene combination is composed of the nucleic acid sequences shown in SEQ ID NOS: 1-3, and the second gene combination is represented by SEQ ID NOS: 4-6 a gene combination II(a) consisting of the nucleic acid sequence, or a gene combination II(b) consisting of the nucleic acid sequences shown in SEQ ID NOS: 4 to 5, 7, and the genes in the combination of the genes are in tumor tissues and The gene expression intensity of the surrounding normal tissues is divided, and the expression ratio of each gene in the combination of the genes is obtained, wherein the nucleic acid sequences shown in SEQ ID NOS: 1 and 4 are shared genes, and the remaining genes in the combination of the genes are calculated. a gene combination interaction value and a second gene combination interaction value; and (3) when the first gene combination interaction value is lower than a first threshold value and the second gene combination interaction value is higher than or equal to one When the second threshold is depreciated, A patient identified as having a high risk of recurrence, wherein the first gene combination interaction value is multiplied by the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 1 and SEQ ID NO: 2 plus SEQ ID NO: 1 and The gene expression ratio multiplication result of the nucleic acid sequence shown by SEQ ID NO: 3, the interaction value of the second gene combination II (a) is the gene of the nucleic acid sequence shown by SEQ ID NO: 4 and SEQ ID NO: 5. The performance ratio is multiplied together with the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 4 and SEQ ID NO: 6, and the interaction value of the second gene combination II (b) is SEQ ID NO: 4 The gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 5 is multiplied by the gene expression ratio of the nucleic acid sequence shown in SEQ ID NO: 4 and SEQ ID NO: 7. The method of claim 4, wherein the step (2), if the gene combination II (a) is selected for analysis, further analyzing a third gene combination, the third gene combination is: SEQ ID NO: The nucleic acid sequence consisting of the nucleic acid sequence represented by the nucleic acid sequence shown in 8 to 10, the nucleic acid sequence represented by the nucleic acid sequence shown by SEQ ID NOS: 13 to 15, and the nucleic acid sequence shown in SEQ ID NO: 19 and Gene combination III(d) composed of the nucleic acid sequence shown in the gene combination III(c) and SEQ ID NOS: 20 to 21, and the gene combination III(e) composed of the nucleic acid sequences shown in SEQ ID NOS: 22-24 Or the gene combination III(f) consisting of the nucleic acid sequences shown in SEQ ID NOS: 32 to 33, wherein the nucleic acid sequences shown in SEQ ID NOS: 8, 13, 19, 20, 22 and 32 are shared genes, and The third gene combination interaction value is calculated by the remaining genes in the gene combination; when the first gene combination interaction value is higher than or equal to a first threshold value, and the third gene combination III(a) interaction value is higher than Or equal to the third threshold or the third gene combination III(b)~III(f) interaction value is lower than the third threshold, then it is judged as high recurrence Patient of risk. The method of claim 4, wherein the step (2) if the gene combination II (b) is selected for analysis, further analyzing a fourth gene combination, the fourth gene combination is: SEQ ID NO: The nucleic acid sequence consisting of the nucleic acid sequence shown in 11 to 12, the nucleic acid sequence shown in the nucleic acid sequence shown in SEQ ID NO: 16 to 18, and the nucleic acid sequence shown in SEQ ID NO: 25 to 27. Gene combination IV(d) composed of the gene combination IV(c), the nucleic acid sequence shown in SEQ ID NOS: 28 to 30, or the gene combination IV (e) composed of the nucleic acid sequences shown in SEQ ID NOS: 31 and 2 Wherein the nucleic acid sequences set forth in SEQ ID NOS: 11, 16, 25, 28 and 31 are shared genes, and a fourth gene combination interaction value is calculated from the remaining genes in the gene combination; when the first gene combination is interacted When the action value is higher than or equal to a first threshold value and the fourth gene combination interaction value is lower than the fourth threshold value, the patient is determined to have a high risk of recurrence. The method according to any one of claims 4 to 6, wherein the gene expression intensity of the gene combination is determined based on the amount of mRNA transcribed from each gene or the amount of polypeptide translated from the gene. The method of any one of claims 4 to 6, wherein the liver cancer tissue is selected from surgically resected cancer tissue or obtained by thick needle sectioning. A combination of genes for predicting the risk of postoperative metastasis in a liver cancer patient is any combination of at least two of the nucleic acid sequences set forth in SEQ ID NO: 28, 34-42. The gene combination according to claim 9, wherein the gene combination is the nucleic acid sequence shown in SEQ ID NO: 28, 34 to 38. The combination of genes as described in claim 9 is for biochip, microarray or real-time quantitative polymerase technology analysis. A method for predicting the risk of postoperative metastasis of a liver cancer patient from a liver cancer tissue comprises: (1) obtaining a tumor tissue and a surrounding normal tissue of a liver cancer patient; (2) analyzing the tumor tissue and one of the surrounding normal tissues. The gene combination is composed of the nucleic acid sequences shown in SEQ ID NO: 28, 34 to 35, and the gene expression in the fifth gene combination is divided by the gene expression intensity of the tumor tissue and surrounding normal tissues to obtain the gene expression of each gene. a ratio, wherein the nucleic acid sequence set forth in SEQ ID NO: 34 is a shared gene, a fifth gene combination interaction value is calculated with the nucleic acid sequences set forth in SEQ ID NOS: 35 and 28; and (3) when the fifth gene is The combination interaction value is lower than a fifth threshold value, and is determined to be a patient with high risk of metastasis; wherein the fifth gene combination interaction value is a gene expression of the nucleic acid sequence shown by SEQ ID NO: 34 and SEQ ID NO: The ratio is multiplied together with the result of multiplication of the gene expression ratio of the nucleic acid sequence shown by SEQ ID NO: 34 and SEQ ID NO: 28. The method of claim 12, wherein the step (2) further analyzes a sixth gene combination, wherein the sixth gene combination is a combination of the nucleic acid sequences represented by the SEQ ID NOS: 36-38 Gene combination VI(c) consisting of the nucleic acid sequence represented by VI(a), SEQ ID NO: 39-40, or the nucleic acid sequence represented by SEQ ID NOS: 41-42, wherein SEQ ID NO: The nucleic acid sequence shown in 36, 39 and 41 is a shared gene, and a sixth gene combination interaction value is calculated from the remaining genes in the gene combination; when the fifth gene combination interaction value is higher than or equal to one When the five thresholds are depreciated and the sixth gene combination interaction value is lower than a sixth threshold, the patient is determined to have a high risk of metastasis. The method of any one of claims 12 to 13, wherein the expression amount of the gene combination is determined based on the amount of mRNA transcribed from each gene or the amount of polypeptide translated from the gene. The method of any one of claims 12 to 13, wherein the liver cancer tissue is selected from a surgically resected cancer tissue or obtained by a thick needle section. A method for obtaining a gene predicting prognosis in a cancer patient, comprising: (1) obtaining a tissue sample from one of a plurality of cancer patients; and (2) obtaining a gene expression for each gene of the extracted tissue sample Gene expression value; (3) According to a prognostic type, select a plurality of gene pairs with strong gene-gene interaction, and classify the genes into a plurality of gene clusters according to a plurality of shared genes, wherein The shared gene lines are the same genes among the pair of genes; and (4) in each of the gene clusters, the gene multiplication values of at least one set of genes are sequentially added according to the significant degree of gene interaction, and then Based on this prognostic type, a decision tree model is used to generate genes that predict the type of prognosis. The method of claim 16, wherein the prognosis type is relapse or metastasis. The method of claim 16, wherein the extracting tissue is selected from a surgically resected cancer tissue or obtained by a thick needle section. The method of claim 16, wherein the cancer is liver cancer. The method of claim 16, wherein the extracted tissue is selected from the tumor tissue or/and the surrounding normal tissue. The method of claim 16, wherein the gene exhibits a numerical value, a single gene expression intensity of a single tissue, or a tumor tissue is divided by the intensity of the gene of the surrounding normal tissue. Gene expression ratio. The method of claim 16, wherein the step (2) further comprises a data standardization process, the data normalization being z-score normalization. The method of claim 16, wherein the gene-gene interaction of the step (3) is a gene that satisfies the formula (1): -log ₁₀ ( p ( v _A × v _B ) ) C _th -log ₁₀ ( p ( v _A ))-log ₁₀ ( p ( v _B )) (1) wherein the C _th value is greater than or equal to 1.5. The method of claim 16, wherein the degree of interaction of the step (4) is calculated using the formula (2): sig_diff = log ₁₀ ( p ( v _A )) + log ₁₀ ( p ( v _B ))-log ₁₀ ( p ( v _A × v _B )) (2). The method of claim 16, wherein the step (4) further selects an appropriate decision tree using a 0.632 bootstrap.