Cancer Prediction Model Establishment System and Methods, Cancer Prediction System

Abstract

This invention provides a cancer type prediction model establishment system and method. The cancer type prediction model establishment system includes a first-layer learning model establishment unit and a second-layer first-learning model establishment unit. The first-layer learning model establishment unit uses machine learning technology to establish a first-layer learning model based on the first copy number variation, the first cancer type, and the first sex of each of multiple first learning specimens; and uses the second sex and second copy number variation of each of multiple second learning specimens as input to the first-layer learning model, causing the first-layer learning model to output multiple first output cancer types for each of the second learning specimens. The second-layer first-learning model establishment unit uses machine learning technology to establish a second-layer first-learning model based on these first output cancer types and the second cancer types of each of the second learning specimens.

Description

Technical Field

[0001] This disclosure relates to a cancer prediction model establishment system and method, and a cancer prediction system. Background Technology

[0002] Cancer can occur in many organs of the body, such as the liver, kidneys, gastrointestinal tract, and brain. Early detection and treatment through regular physical examinations can improve treatment outcomes. Therefore, developing cancer detection technologies is one of the goals of companies in this field. Summary of the Invention

[0003] Therefore, this disclosure proposes a cancer type prediction model establishment system and method, and a cancer type prediction system, which can improve the aforementioned known problems.

[0004] This disclosure presents an embodiment of a cancer type prediction model establishment system. The cancer type prediction model establishment system includes a first-layer learning model establishment unit and a second-layer first-learning model establishment unit. The first-layer learning model establishment unit uses machine learning techniques to establish a first-layer learning model based on the first copy number variation, the first cancer type, and the first sex of each of multiple first learning specimens; and uses the second sex and second copy number variation of each of multiple second learning specimens as input to the first-layer learning model, causing the first-layer learning model to output multiple first output cancer types for each of the second learning specimens. The second-layer first-learning model establishment unit uses machine learning techniques to establish a second-layer first-learning model based on these first output cancer types and the cancer types of the second specimens.

[0005] This disclosure also discloses a cancer type prediction system. The cancer type prediction system includes a storage unit and a prediction unit. The storage unit stores the aforementioned first-layer learning model and second-layer first-learning model. The prediction unit inputs the copy number variation of the specimen and a specimen gender into the first-layer learning model to obtain a first predicted cancer type for the specimen; determines whether the specimen gender is a second gender; and, if the specimen gender is a second gender, inputs the first predicted cancer type into the second-layer first-learning model to obtain a second predicted cancer type for the specimen.

[0006] This disclosure also discloses a method for establishing a cancer type prediction model. The method includes the following steps: using machine learning techniques, establishing a first-layer learning model based on the first copy number variation, cancer type, and sex of each of multiple first learning specimens; using a second sex and second copy number variation of each of multiple second learning specimens as input to the first-layer learning model, the first-layer learning model outputs multiple first output cancer types for each of the second learning specimens; and using machine learning techniques, establishing a second-layer first learning model based on these first output cancer types and a second cancer type of the second learning specimens.

[0007] To provide a better understanding of the above and other aspects of this disclosure, specific embodiments are described below in conjunction with the accompanying drawings: Attached Figure Description

[0008] Figure 1 A functional block diagram of a cancer prediction model establishment system according to an embodiment of the present disclosure is shown.

[0009] Figure 2 A functional block diagram of a cancer prediction system according to an embodiment of the present invention is shown.

[0010] Figure 3 Draw Figure 1 The flowchart illustrates the method for establishing a cancer prediction model.

[0011] Figure 4 Draw Figure 2 The flowchart of the cancer prediction method of the cancer prediction system. Detailed Implementation

[0012] Please refer to Figure 1 The figure illustrates a functional block diagram of a cancer type prediction model building system 100 according to an embodiment of the present disclosure. The cancer type prediction model building system 100 includes a first-layer learning model building unit 110, a second-layer first learning model building unit 120, and a second-layer second learning model building unit 130. The first-layer learning model building unit 110, the second-layer first learning model building unit 120, and / or the second-layer second learning model building unit 130 are, for example, physical circuits formed using semiconductor processes. In one embodiment, at least two of the first-layer learning model building unit 110, the second-layer first learning model building unit 120, and the second-layer second learning model building unit 130 may be integrated into a single unit. In one embodiment, at least one of the first-layer learning model building unit 110, the second-layer first learning model building unit 120, and the second-layer second learning model building unit 130 may be integrated into a controller or a processor.

[0013] like Figure 1 As shown, the first-layer learning model building unit 110 is used to: (1) use a machine learning technique to calculate the first copy number variation (CNV) V of each of a plurality of first learning subjects P1 (unlabeled). P1 First specimen cancer type (learning about the type of cancer the specimen belongs to) C1 P1 and primary gender (gender of the learner's specimen) S P1 (1) Establish the first-layer learning model M1; (2) Use the second gender S of each of the multiple second learning subjects P2 (unlabeled) P2 and the second copy number variation V P2 As input to the first-layer learning model M1, the first-layer learning model M1 outputs multiple first-output cancer types C2 for each second-learning specimen P2. P2 The second-layer first learning model building unit 120 is used to employ machine learning techniques, based on these first output cancer types C2 of each second learning specimen P2. P2 and the second specimen cancer type C1 P2 A second-layer first learning model M21 is established. In this embodiment, the first-layer learning model M1 is a hybrid learning model (e.g., not limited to gender), thus avoiding learning (training) errors caused by some cancer types with small sample sizes.

[0014] like Figure 1 As shown, in one embodiment, the first-layer learning model building unit 110 is further configured to: use machine learning techniques to determine the fourth copy number variation V of each of the plurality of healthy specimens P4 (unlabeled). P4 Health Category H P4 and the fourth gender S P4 A first-layer learning model M1 is then established. Thus, the established first-layer learning model M1 further includes the health category H. P4 It can be determined that it belongs to the health category H. P4 The sample to be tested. Fourth gender S P4 Regardless of gender, the copy number of the genomic fragment in healthy sample P4 did not show any abnormal increase or decrease (no mutated genes), i.e., its fourth copy number variation V... P4 That's normal.

[0015] In addition, the establishment of the second-layer first learning model M21 can be based on the first output cancer type C2. P2 and the second specimen cancer type C1 P2 In addition, it can also be established based on other specimen information, such as the age of the learning specimen. For example, such as Figure 1 As shown, the second-layer first learning model building unit 120 can establish the first output cancer type C2 based on each second learning specimen P2.P2 Second specimen cancer type C1 P2 and age G P2 Establish the second layer first learning model M21.

[0016] like Figure 1 As shown, besides being able to be based on the second gender S P2 In addition to establishing the second-layer first-learning model M21, other learning models can also be established based on other genders. For example, the first-layer learning model establishment unit 110 is further used to: use the third gender S of each of multiple third-learning subjects P3 (unlabeled) P3 and third copy number variation V P3 As input to the first-layer learning model, the first-layer learning model M1 outputs multiple second-output cancer types C2 for each third-learning specimen P3. P3 The second-layer second learning model building unit 130 is used to employ machine learning techniques, based on these second output cancer types C2 of each third learning specimen P3. P3 and the third specimen cancer type C1 P3 A second-layer second learning model M22 is established.

[0017] In addition, the establishment of the second-layer second learning model M22 can be based on the second output cancer type C2. P3 In addition, it can also be established based on other specimen information, such as the age of the learning specimen. For example, such as Figure 1 As shown, the second-layer second learning model building unit 130 can establish the second learning model based on these second output cancer types C2 of each third learning specimen P3. P3 Third specimen cancer type C1 P3 and age G P3 A second-layer second learning model M22 is established.

[0018] Furthermore, the output cancer types in this paper (e.g., the first output cancer type C2) P2 and / or the second exported cancer type C2 P3 For example, it can be represented by a probability type. Table 1 below lists the multiple first output cancer types C2 output by the first-layer learning model M1. P2 The probability. Table 1 uses only 5 learning specimens as an example, but more than 5 can be used. As shown in Table 1 below, different numbers represent different learning specimens. These first output cancer types C2 P2 This refers to different types of cancer, such as liver cancer, lesion cancer, and breast cancer. Taking sample #1 as an example, it indicates liver cancer (first output cancer type C2). P2 The probability of developing C2 is 9.958%, the probability of developing breast cancer is 77.335%, and the probability of developing leukemia is 0.011%. The remaining primary cancer type is C2. P2Although not listed, each will have a probability value. The values ​​in Table 1 are merely examples and are not intended to limit the embodiments of the present invention. The interpretation of other numbers is similar to that of number #1, and will not be repeated here. The second-layer first learning model building unit 120 can build the second-layer first learning model M21 based on the probability information in Table 1. In addition, the second-layer second learning model building unit 130 can also build the second-layer first learning model M21 based on probability information similar to that in Table 1 (second output cancer type C2). P3 ), and establish the second layer second learning model M22.

[0019] Table 1

[0020]

[0021] Furthermore, the machine learning technique used in this paper is, for example, a Support Vector Machine (SVM). However, the embodiments of this invention are not limited to any particular type of machine learning technique; any technique that can learn the input information required by this paper and establish a cancer-specific learning model can be used as an application of the machine learning technique described herein.

[0022] Furthermore, the learning specimen used in this paper is, for example, circulating tumor cells (CTCs). Circulating tumor cells refer to tumor cells that have detached from tumor tissue and entered the bloodstream. Circulating tumor cells from different organs tend to carry specific types of gene mutations. Therefore, by detecting the specific types of gene mutations carried by circulating tumor cells, it is possible to infer the organ (cancer type) from which these circulating tumor cells originated. Using circulating tumor cells as a specimen has the following advantages: (1) It can be tested with a blood sample (no need for physical biopsy or radiation imaging), resulting in low cost and risk; (2) It is suitable for long-term monitoring of cancer recurrence; (3) It overcomes the heterogeneity of tumors; (4) It can predict distant metastasis early; and (5) It can quickly reflect the current tumor status.

[0023] Furthermore, copy number variation (CNV) is a phenomenon in which parts of the genome are repeated, and the number of repeats in the genome varies between individuals. Specifically, CNV is an event that affects the duplication or deletion of a large number of base pairs. Approximately two-thirds of the entire human genome may consist of repeats, and 4.8% to 9.5% of the human genome can be classified as CNV. An increase in the copy number of a genomic segment indicates a CNV abnormality.

[0024] Furthermore, the study samples used in this paper are, for example, taken from The Cancer Genome Atlas (TCGA), which records the cancer name, age, CNV, and sex of each sample. TCGA includes approximately 34 cancer names. In one embodiment, these 34 cancer names can be summarized or consolidated into, for example, 12 cancer types or fewer, such as brain cancer, esophageal cancer, lung cancer, kidney cancer, male cancers (cancer names include, for example, testicular cancer, prostate cancer), gynecological cancers (cancer names include, for example, ovarian cancer, cervical cancer, uterine cancer, endometrial cancer), liver cancer, bladder cancer, anterior mediastinal cancers (cancer names include, for example, thyroid cancer, thymic cancer), head and neck cancer, breast cancer, and / or gastrointestinal cancers (cancer names include, for example, colorectal cancer, rectal cancer, pancreatic cancer, stomach cancer), etc. Additionally, the first sex in this paper is S. P1 It is not limited to male or female; for example, the primary sex of the individuals belonging to these first learning specimens P1 can be a combination of male and female. Secondary sex differs from tertiary sex. Secondary sex S P2 and third gender S P3 Limited to males or females, for example, the second gender S of all second learning specimens P2. P2 It is male, and the third gender S belongs to all the third learning specimens P3. P3 It is female; or, for example, the second gender S of all second learning specimens P2. P2 She is female, and the third gender S belongs to all the third learning specimens P3. P3 He is male.

[0025] Please refer to Table 2 below. In this embodiment, the total number of samples used for the first-layer learning model M1, the second-layer first-learning model M21, the second-layer second-learning model M22, and the number of samples used for validation, when taken as 100%, means that the number of samples used for the learning models accounts for, for example, 80%, and the number of samples used for validation accounts for, for example, 20%. The total number of samples used for the first-layer learning model M1, the second-layer first-learning model M21, and the second-layer second-learning model M22, when taken as 100%, means that the number of samples used for the first-layer learning model M1 accounts for 80%, and the total number of samples used for the second-layer first-learning model M21 and the second-layer second-learning model M22 accounts for 20%. These sample numbers can be obtained from TCGA, hospitals, and / or government units, etc. The above 80%, 20%, etc., proportions are merely examples and are not intended to limit the embodiments of the present invention.

[0026] Table 2

[0027]

[0028] After obtaining the first-layer learning model M1, the second-layer first learning model M21, and the second-layer second learning model M22, these learning models can be used to obtain the predicted cancer type of at least one specimen. Further examples are given below.

[0029] Please refer to Figure 2 The diagram illustrates a functional block diagram of a cancer prediction system 200 according to an embodiment of the present invention. The cancer prediction system 200 includes a memory 210 and a prediction unit 220. The memory 210 and the prediction unit 220 are, for example, physical circuits formed using semiconductor processes. Specifically, the memory 210 may be a memory, which may be integrated into the prediction unit 220 or configured separately from the prediction unit 220. In one embodiment, the prediction unit 220 may be integrated into a controller or a processor.

[0030] Storage 210 is used to store the first-layer learning model M1 and the second-layer first-layer learning model M21. Prediction unit 220 is used to: (1) input the sample copy number variation V of the sample PT to be tested. PT and the sex of the specimen S PT The first layer of the learning model M1 is used to obtain the first predicted cancer type C1 of the test specimen PT. PT (2) Determine the sex of the test specimen PT S. PT Is it a second gender S? P2 ; and, (3). When the sex of the test specimen PT is the second sex S P2 Enter the first predicted cancer type C1 of the PT sample to be tested. PT In the second layer of the first learning model M21, the second predicted cancer type C2 of the test sample PT is obtained. PT .

[0031] The aforementioned specimen's gender S PT Not limited to male or female, in other words, the sex of the specimen S PT It can be either male or female.

[0032] Furthermore, storage 210 is used to store the aforementioned second-layer second learning model M22. Prediction unit 220 is further used to: (1) determine the sex S of the test specimen PT. PT Is it a third gender S? P3 ; and, (2). When the sex of the test specimen PT is S PT For the third gender S P3 When inputting the first predicted cancer type C1 of the PT sample, enter the sample's PT value. PT In the second layer, the second learning model M22 is used to obtain the second predicted cancer type C2 of the test sample PT. PT .

[0033] The following explains the second predicted cancer type C2 for multiple test samples.PT The accuracy of the results is shown in Tables 3-1 and 3-2 below, which list the second predicted cancer type C2 for multiple test samples. PT Data analysis with actual cancer types, such as Sensitivity (Sens), Specificity (Spec), Positive Predictive Value (PPV), and Negative Predictive Value (NPV). For esophageal cancer, due to its unique location (between the head and neck and the gastrointestinal tract), any prediction of head and neck cancer or gastrointestinal cancer is considered a correct prediction.

[0034] Table 3-1

[0035]

[0036]

[0037] Table 3-2

[0038]

[0039]

[0040] Sensitivity (Sens), specificity (Spec), positive predictive value (PPV), and negative predictive value (NPV) are expressed by the following formulas (1) to (4), where the definitions of parameters TP, FP, FN, and TN are explained in Table 4.

[0041] Sens=TP / (TP+FN)….(1)

[0042] Spec=TN / (FP+TN)….(2)

[0043] PPV=TP / (TP+FP)……(3)

[0044] NPV=TN / (FN+TN)….(4)

[0045] Table 4

[0046]

[0047]

[0048] Furthermore, in Tables 3-1 and 3-2 above, T1 (Top 1) refers to the following: For TP, when the actual situation is cancer A and the highest probability result in the test is also cancer A, the TP value increases by 1; when the actual situation is cancer A but the highest probability result in the test is not cancer A, the TP value increases by 0. T3 (Top 3) refers to the following: For TP, when the actual situation is cancer A and the top 3 highest probabilities in the test also include cancer A, the TP value increases by 1; when the actual situation is cancer A but the top 3 highest probabilities in the test do not include cancer A, the TP value increases by 0.

[0049] Based on the analysis data in Tables 3-1 and 3-2, the sensitivity (Sens) is higher than 90% for some types of cancer (such as gastrointestinal cancer, liver cancer, male cancer, brain cancer, breast cancer, and female cancer). In early screening applications, sensitivity should be given the highest priority.

[0050] As shown in Table 5 below, the accuracy of predicting cancer in men using the first-layer learning model M1 is 68.5%, while the accuracy of predicting cancer in women is 69.8%. Using the second-layer learning model, the accuracy of predicting cancer in men improves to 72.91%, while the accuracy of predicting cancer in women improves to 71.21%.

[0051] Table 5

[0052]

[0053] Please refer to Figure 3 Its illustration Figure 1 The flowchart shows the method for establishing a cancer prediction model system 100.

[0054] In step S 110, please also refer to Figure 1 The first-layer learning model building unit 110 uses machine learning technology, based on the first copy number variation V of each of the multiple first learning specimens P1. P1 First specimen cancer type C1 P1 and first gender S P1 A first-layer learning model M1 is established. In one embodiment, please also refer to... Figure 1 The first-layer learning model building unit 110 is further used to: use machine learning techniques to analyze the fourth copy number variation V of each of multiple healthy specimens P4 (unlabeled). P4 Health Category H P4 and the fourth gender S P4 Establish the first-layer learning model M1.

[0055] In step S120, please also refer to Figure 1 The second sex S of each of the multiple second learning specimens P2 P2and the second copy number variation V P2 As input to the first-layer learning model M1, the first-layer learning model M1 outputs multiple first-output cancer types C2 for each second-learning specimen P2. P2 .

[0056] In step S130, please also refer to Figure 1 The second-layer first learning model building unit 120 uses machine learning technology, based on these first output cancer types C2 of each second learning specimen P2. P2 and the second specimen cancer type C1 P2 A second-layer first learning model M21 is established. In one embodiment, the second-layer first learning model establishment unit 120 can further establish the model based on the age G of each second learning specimen P2. P2 Establish the second layer first learning model M21.

[0057] In step S140, please also refer to Figure 1 The second layer, second learning model building unit 130 uses multiple third learning subjects P3 to establish the third gender S of each subject. P3 and third copy number variation V P3 As the input to the first-layer learning model M1, the first-layer learning model M1 outputs the corresponding second output cancer type C2. P3 .

[0058] In step S150, please also refer to Figure 1 The second-layer second learning model building unit 130 uses machine learning technology, based on these second output cancer types C2 of each third learning specimen P3. P3 and the third specimen cancer type C1 P3 A second-layer second learning model M22 is established. In one embodiment, the second-layer second learning model establishment unit 130 can further establish the model based on the age G of each third learning specimen P3. P3 A second-layer second learning model M22 is established.

[0059] Please refer to Figure 4 Its illustration Figure 2 A flowchart of the cancer prediction method of the cancer prediction system 200.

[0060] In step S210, please also refer to Figure 1 The prediction unit 220 is input with the copy number variation V of the test sample PT. PT and the sex of the specimen S PT The first layer of the learning model M1 is used to obtain the first predicted cancer type C1 of the test specimen PT. PT .

[0061] In step S220, please also refer to Figure 1The prediction unit 220 determines the sex S of the test specimen PT. PT Is it a second gender S? P2 If the sex of the sample PT to be tested is S PT For the second gender S P2 If not, the process proceeds to step S230; otherwise, the process proceeds to step S240.

[0062] In step S230, please also refer to Figure 1 The prediction unit 220 inputs the first predicted cancer type C1 of the sample PT to be tested. PT In the second layer of the first learning model M21, the second predicted cancer type C2 of the test sample PT is obtained. PT .

[0063] In step S240, the prediction unit 220 determines the sex S of the sample PT to be tested. PT Is it a third gender S? P3 If the sex of the PT sample to be tested is S PT For the third gender S P3 The process proceeds to step S250.

[0064] In step S250, the prediction unit 220 inputs the first predicted cancer type C1 of the sample PT to be tested. PT In the second layer, the second learning model M22 is used to obtain the second predicted cancer type C2 of the test sample PT. PT .

[0065] In the embodiment, the sex of the specimen S PT Not a second gender S P2 That is, the third gender S P3 Therefore, step S240 can be omitted from the process.

[0066] In summary, this disclosure presents an embodiment of a cancer type prediction model building system, a cancer type prediction system, and a method for applying the same. The cancer type prediction model building system utilizes machine learning techniques to build a two-layer learning model. Using this two-layer learning model, the cancer type prediction system can easily and quickly predict the cancer type of a test specimen. In one embodiment, the first-layer learning model is, for example, a hybrid learning model (e.g., gender-neutral), thus avoiding learning (training) errors caused by small sample sizes for some cancer types. The second-layer learning model can predict cancer types specific to a particular gender.

[0067] In summary, although this disclosure has been presented above with reference to embodiments, it is not intended to limit the scope of this disclosure. Those skilled in the art to which this disclosure pertains can make various modifications and refinements without departing from the spirit and scope of this disclosure. Therefore, the scope of protection of this disclosure shall be determined by the claims.

Claims

1. A cancer type prediction model establishment system, characterized in that, include: The first layer of the learning model building unit is used for: Using machine learning techniques, a first-layer learning model is established based on the first copy number variation, cancer type, and first sex of each of the first learning specimens. The first sex of these first learning specimens is a combination of male and female. The second sex and second copy number variation of each of the multiple second learning specimens are used as inputs to the first layer learning model, so that the first layer learning model outputs multiple first output cancer types for each of the second learning specimens, and the probability of these first output cancer types being different cancer types; Using the third sex and third copy number variation of each of multiple third learning specimens as input to the first-layer learning model, the first-layer learning model outputs multiple second-output cancer types for each of the third learning specimens, where the probabilities of these second-output cancer types being different cancer types are also provided. The second layer, first learning model building unit, is used for: Using this machine learning technique, a second-layer first learning model is established based on the first output cancer type and the second cancer type of each second learning specimen. All the second genders of these second learning specimens are either male or female. The second layer, second learning model building unit, is used for: Using this machine learning technique, a second-layer second learning model is established based on the second output cancer types of each third learning specimen and the cancer type of the third specimen; In these second learning specimens, all of the third genders are either male or female, and the second gender is different from the third gender.

2. The cancer type prediction model establishment system as described in claim 1, characterized in that, The second layer, second learning model building unit is further used for: Based on the age of each third learning specimen, the second layer second learning model is established; The second layer first learning model building unit is further used for: Based on the age of each second learning specimen, the first learning model of the second layer is established.

3. A cancer type prediction system, characterized in that, include: Storage, used for: Store the first-layer learning model as described in claim 1, the second-layer first learning model, and the second-layer second learning model as described in claim 1; Prediction unit, used for: Input the copy number variation and sex of the sample to be tested into the first layer of the learning model to obtain multiple first predicted cancer types of the sample to be tested; Determine whether the sex of the sample to be tested is the second sex; Determine whether the sex of the sample to be tested is the third sex; when the sex of the sample to be tested is the third sex, input the first predicted cancer type of the sample to the second layer of the second learning model to obtain the second predicted cancer type of the sample; and When the gender of the specimen to be tested is the second gender, the first predicted cancer type of the specimen to be tested is input into the second layer of the first learning model to obtain the second predicted cancer type of the specimen to be tested.

4. A method for establishing a cancer type prediction model, characterized in that, include: Using machine learning techniques, a first-layer learning model is established based on the first copy number variation, first cancer type, and first sex of each of the multiple first learning specimens. The first sex of these first learning specimens is a combination of male and female. The second sex and second copy number variation of each of the multiple second learning specimens are used as inputs to the first layer learning model. The first layer learning model outputs multiple first output cancer types for each of the second learning specimens, and the probability of these first output cancer types being different cancer types. Using the third sex and third copy number variation of each of multiple third learning specimens as input to the first-layer learning model, the first-layer learning model outputs the corresponding third cancer type; and Using this machine learning technique, a second-layer first learning model is established based on the first output cancer types and the second cancer types of each second learning specimen; Using this machine learning technique, a second-layer second learning model is established based on these third cancer types; The second gender is different from the third gender.

5. The method for establishing as described in claim 4, characterized in that, The steps for establishing the second-layer second learning model based on these third cancer types further include: Based on the age of each third learning specimen, the second layer second learning model is established; The step of establishing the second-layer first learning model based on the first output cancer types of each second learning specimen and the cancer type of the second specimen further includes: Based on the age of each second learning specimen, the first learning model of the second layer is established.

Images