KR101951727B1 - Computing method for predicting multiple kinds of cancers, and computing apparatus using the same - Google Patents

Abstract

The present invention relates to a computing method for multi-cancer determination, comprising the following steps: making a binary classification of diagnostic biomarker group information of k number of cancers and determining whether the diagnostic biomarker information of k number of cancers corresponds to cancer; if a result of the binary classification of the diagnostic biomarker information of k number of cancers is determined to be a cancer, using and analyzing the diagnostic biomarker information of k number of cancers and determining which cancer corresponds to the diagnostic biomarker information of k number of cancers. The computing method for multi-cancer determination is to improve the reliability of the determination of a cancer by using a statistical determination method which uses a biomarker group for a multi-cancer diagnosis.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention [0001] The present invention relates to a computing method for multi-arm discrimination,

The present invention relates to a computing method for multi-cancer discrimination and a computing apparatus using the same. More particularly, the present invention relates to a method of binarizing information of k cancer diagnostic biomarkers, K cancer diagnostic biomarker information is analyzed using a neural network to determine whether the information of k cancer diagnostic biomarkers is k < RTI ID = 0.0 > The present invention relates to a computing method for discriminating a multi-arm that discriminates which of cancer is cancer, and a computing apparatus using the same.

Tumor metastasis is an important part of cancer-related deaths in patients with solid tumors, with a portion of the tumor falling off and moving through the blood to other parts of the body. Currently, a common method for diagnosing cancer is biopsy, in which a part of the tissue is removed from the initial metastatic state. However, it is not easy to determine the exact biopsy site. On the other hand, the liquid biopsy method, which has been attracting attention recently, is to detect a tumor cell in a biological sample by collecting a biological sample derived from the patient's body such as blood or urine, It is thought that it is not only useful for early detection and diagnosis, but also for measuring progress.

A biomarker is an index that can detect changes in the body using proteins, nucleic acids, and metabolites contained in biological samples.

For example, conventional biomarker techniques for early diagnosis of lung cancer include those which essentially include A1AT, IGF-1, RANTES and TTR as disclosed in Korean Patent No. 10-1463588.

However, there is a limit to the diagnosis of cancer with only a single biomarker, and a cancer diagnosis method using a complex biomarker that improves the diagnostic sensitivity and specificity of cancer by using two or more biomarkers in combination is currently used in the technical field .

However, when such a complex biomarker is used to identify a specific cancer, the specific biomarker may not only indicate an indicator for a specific cancer but may also indicate an indicator for other cancer, A false discrimination may occur.

 For example, in the diagnosis of adenocarcinoma using a complex biomarker, the result of discrimination using the information of the complex biomarker can be discriminated as adenocarcinoma. Actually, the cancer of the patient is squamous cell carcinoma ).

In addition, when biomarker information of each of the complex biomarkers is analyzed to discriminate the cancer, different results may be derived depending on the discriminator.

The present invention aims at solving all of the above problems.

Another object of the present invention is to accurately discriminate the type of cancer using a group of biomarkers for diagnosis of multiple cancers.

It is another object of the present invention to make it possible to discriminate a multi-cancer using a biomarker group for multi-cancer diagnosis.

Another object of the present invention is to accurately discriminate types of cancer in a statistical discrimination method using a biomarker group for diagnosis of multiple cancers.

It is another object of the present invention to improve the reliability of cancer discrimination by a statistical discrimination method using a biomarker group for multi-cancer diagnosis.

In order to accomplish the objects of the present invention as described above and achieve the characteristic effects of the present invention described below, the characteristic structure of the present invention is as follows.

According to an embodiment of the present invention,

(A) when information on a group of k cancer diagnostic biomarkers is inputted, the computing device binarily classifies the information of the k cancer diagnostic biomarkers, Determining whether the information of the diagnostic biomarker is cancer or cancer; And (b) if it is determined that the information on the k cancer diagnostic biomarker groups is cancerous as a result of the binary classification, the computing device analyzes the k cancer diagnostic biomarker group information using a neural network, Determining whether the information of the group of cancer diagnosis biomarkers corresponds to which of the k cancers; A method for computing a multi-arm is provided.

According to an embodiment of the present invention, there is provided a computing device for discriminating a multi-arm, comprising: a communication unit for acquiring information on k cancer diagnostic biomarker groups; And (i) binarizing the information of the k cancer diagnostic biomarker groups to determine whether the information of the k cancer diagnostic biomarker groups is cancer or cancer, and (ii) The information of the k cancer diagnostic biomarker group is analyzed using the neural network, and when the information of the k cancer diagnostic biomarker group is analyzed as a result of the binary classification, A processor for performing a process of determining which cancer corresponds; A computing device for multi-cancer determination is provided.

In addition, a computer readable recording medium for recording a computer program for executing the method of the present invention is further provided.

The present invention can accurately discriminate types of cancer by using a group of biomarkers for diagnosis of multiple cancers.

In addition, the present invention makes it possible to discriminate a multi-cancer using a biomarker group for diagnosis of multiple cancers.

Further, the present invention makes it possible to accurately determine the type of cancer in the statistical discrimination method using the biomarker group for diagnosis of multiple cancers.

In addition, the present invention can improve the reliability of cancer discrimination by a statistical discrimination method using a biomarker group for multi-cancer diagnosis.

FIG. 1 schematically shows a computing apparatus for discriminating a multi-arm according to an embodiment of the present invention,
2 schematically shows a computing method for discriminating a multi-arm according to an embodiment of the present invention,
3 schematically illustrates a model of an expert network for binary classification in a method for computing a multi-arm according to an embodiment of the present invention,
FIG. 4 is a schematic view illustrating a method of selecting an expert in an expert network for binary classification in a multi-arm computing method according to an embodiment of the present invention,
5 is a schematic diagram of a neural network model for multiclassification in a computing method for identifying a multi-arm according to an embodiment of the present invention.

The following detailed description of the invention refers to the accompanying drawings, which illustrate, by way of illustration, specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the present invention are different, but need not be mutually exclusive. For example, certain features, structures, and characteristics described herein may be implemented in other embodiments without departing from the spirit and scope of the invention in connection with an embodiment. It is also to be understood that the position or arrangement of the individual components within each disclosed embodiment may be varied without departing from the spirit and scope of the invention. The following detailed description is, therefore, not to be taken in a limiting sense, and the scope of the present invention is to be limited only by the appended claims, along with the full scope of equivalents to which such claims are entitled, if properly explained. In the drawings, like reference numerals refer to the same or similar functions throughout the several views.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, so that those skilled in the art can easily carry out the present invention.

FIG. 1 schematically illustrates a computing apparatus 1000 for identifying a multi-arm according to an embodiment of the present invention. Referring to FIG. 1, a computing apparatus 1000 for discriminating a multi-arm includes a communication unit 100, Processor 200 may be included.

The communication unit 100 performs the function of acquiring the information of the k cancer diagnostic biomarker groups, and calculates the index values of the respective biomarkers from the k cancer diagnostic biomarker kits composed of the respective biomarkers for k cancer diagnosis Receiving the respective biomarker index values received by the user or receiving the respective biomarker index values from an external device measuring the respective biomarker index values.

Processor 200 classifies the information of the k cancer diagnostic biomarker clusters obtained in communication unit 100 to determine whether the information of k cancer diagnostic biomarker clusters is cancer or cancer, When the information of cancer diagnosis biomarker group is judged to be cancer as a result of binary classification, information of k cancer diagnostic biomarker group information is analyzed by using neural network, and information of k cancer diagnostic biomarker group is classified into any one of k cancer The process of determining whether or not the process corresponds to the process shown in FIG.

A method for discriminating a multi-arm according to an embodiment of the present invention using a computing apparatus 1000 for discriminating a multi-arm according to an embodiment of the present invention will now be described.

2, the computing device 1000 acquires k cancer diagnostic biomarker group information through the communication unit 100 (210), and the processor 200 of the computing device 1000 transmits the k cancer diagnostic biomarker group information (K) of the k cancer diagnostic biomarker group information obtained from the cancer diagnostic biomarker 100 to determine whether the information of the k cancer diagnostic biomarkers is cancer or cancer (220).

In this case, the k cancer diagnostic biomarker group information includes information of each biomarker for diagnosing k cancer. The information includes proteins, nucleic acids, and proteins contained in biological samples derived from the patient's body such as blood and urine, It can be an indicator value for cancer diagnosis using metabolites and the like. In addition, the k cancer diagnostic biomarker group information may include information such as age, sex, and medical history of the patient.

The processor 200 of the computing device 1000 may perform a binary classification to determine whether cancer is cancer or not using the k cancer diagnostic biomarker group information using a mixture of experts A decision tree, a tree ensemble, a random forest, a Bayesian network, a support vector machine, a neural network, and a logistic regression At least one of logistic regression may be used. Hereinafter, the method of mixture of experts is mainly described.

That is, the processor 200 of the computing device 1000 determines whether the patient holds at least one of the k cancers using the index values of each of the k cancer diagnostic biomarkers affected by the multi-cancer Do.

For example, referring to FIG. 3, the processor 200 of the computing device 1000 may compare k cancer diagnostic biomarker group information x with k learned expert networks 221 (221-1, 221-2, ..., And 221-k, respectively, so as to cause the learned k expert networks 221 to deduce whether each of the learned experts is cancer. Specifically, the first expert network 221-1 analyzes k cancer diagnostic biomarker group information to deduce whether it is a first cancer, and the second expert network 221-2 analyzes k cancer diagnostic biomarker group information Information is analyzed to infer whether the second cancer is a cancer, and the kth expert network 221-k deduces whether the k cancer diagnostic biomarker group information is k cancer.

The processor 200 of the computing device 1000 then calculates the gating functions g1, g2, and g3 of the gating network 222 for which the contribution to each of the k expert networks has been determined corresponding to the k cancer diagnostic biomarker group information x, ..., gk) to recalculate the speculative values of each of the k expert networks (221) and determine whether the information (x) of the k cancer diagnostic biomarker groups is cancer, using the re- (Y).

At this time, the processor 200 of the computing device 1000 can use the soft max algorithm to classify whether or not the information of the k cancer diagnostic biomarker groups is cancer using the re-computed result values.

의 확률로 j번째 엑스퍼트 네트워크가 선택되고 출력 y는

의 확률로 선택됨을 알 수 있다.The binary classification model of FIG. 3 can be expressed by the following Equation 1, and when the input value x is given

, The jth expert network is selected and the output y is

Is selected as a probability.

[Equation 1]

In Equation (1), g (-) is a gating function of the gating network 222, which determines which expert network to use.

At this time, the gating function can be expressed by the following equation (2).

&Quot; (2) "

의 조건을 가진다.And the gating function

.

(A) e2 is used for the input value 0 and the output is [0, 1, 0], (b) e2 and e3 are used for the input value 0.5, and the output is [0, 0.5, 0.5].

2, when the processor 200 of the computing device 1000 determines that the cancer is a result of binary classification of the information of the k cancer diagnostic biomarker groups, the k cancer diagnostic biomarkers Using the neural network, the information of the k cancer diagnostic biomarkers can be used to determine which of the cancer is cancer (230).

5, when the k cancer diagnostic biomarker group information determined as a cancer as a result of binary classification is input to the input layer 231, the processor 200 of the computing device 1000 determines 1 connection strength Wj is used to calculate k cancer diagnostic biomarker information and input into the learned hidden layer 232 composed of a plurality of neurons and the first connection strength Wj is calculated from the learned hidden layer 232, Is calculated using a hidden layer activation function. At this time, the hidden layer activation function may use one of sigmoid function (), hyperbolic tangent function (tanh), and rectified linear unit function (ReLU). In addition, the hidden layer 232 may be formed as a single layer or may be formed in multiple layers, but it is assumed in FIG. 5 that the hidden layer 232 is a single layer. Then, the learned second connection strength Wk is applied to the computed value of each neuron of the learned hidden layer 232 and input to the output layer 233, and in the output layer 233, And the k cancer diagnostic biomarker group information can be used to discriminate the cancer of the patient from any of k cancer. At this time, the classification for each of the k arms can use the soft max algorithm.

를 i번째 입력 변수,

을 m번째 히든 레이어의 뉴런,

를 k번째 출력 뉴런이라 하며, 입력 레이어(231)의 차원 수를 p, 히든 레이어(232)의 뉴런 수를 M이라 하고, 출력 레이어(233)의 뉴런 수를 K라고 하면, 도 5의 모형은 다음의 수학식 3과 같이 나타내어질 수 있다.In the neural network model of Fig. 5,

Is the i-th input variable,

The neuron of the mth hidden layer,

The number of neurons of the input layer 231 is denoted by p, the number of neurons of the hidden layer 232 is denoted by M, and the number of neurons of the output layer 233 is denoted by K, Can be expressed by the following equation (3).

&Quot; (3) "

In the above example, k cancer diagnostic biomarker group information is used to determine whether cancer is through binary classification through the learned expert network and the learned neural network, and in the case of cancer, In this section, we explain how to determine which cancer is cancer, and briefly explain how to learn Expert Network and Neural Network.

First, the learning method of the expert network will be described. The processor 200 of the computing device 1000 inputs k cancer diagnostic biomarker group information for training to each of k expert networks before training, In order to infer whether the biomarker group information is cancerous or not, the gating function of the gating network determined in correspondence with the k cancer diagnostic biomarker group information for training, The inference values are recomputed, and the information of the k cancer diagnostic biomarker groups for training is classified using the recalculated result values. Then, the parameters of the k expert networks before learning and the parameters of the gating network before learning are learned by referring to the result of classifying.

At this time, any one of the gradient ascent, the EM algorithm, and the quasi-Newton method, which maximizes the log likelihood, can be used to learn the parameters of the k expert networks before learning and the gating network before learning have.

이며,

이므로, 수학식 1에서

로 로지스틱 모형을 사용할 수 있다.That is, in Equation 2,

Lt;

Therefore, in Equation 1,

A logistic model can be used.

이며, 추정하고자 하는 모수는

이다.here,

, And the parameter to be estimated is

to be.

Next, a description will be given of how the processor 200 of the computing device 1000 learns the neural network. When k pieces of cancer diagnostic biomarker group information for training are input to the input layer before learning, the first connection strength before being learned , The information of k cancer diagnostic biomarker group for training is computed and input to the hidden layer before learning which is composed of a plurality of neurons and the value calculated by using the first connection strength before being learned in the hidden layer before learning, And a second connection strength before being learned is applied to an operation value in each neuron of the hidden layer before being learned and input to the output layer before being learned, do. Then, the class value is compared with the reference value corresponding to the k cancer diagnostic biomarker group information for training to obtain the loss value, and the first connection strength, learning The parameters of the hidden layer before learning, and the parameters of the output layer before learning are learned.

의 웨이트(weight) 행렬과 b(M 차원), c(K 차원)의 바이어스 벡터이며, 손실 함수를 크로스 엔트로피 손실 함수로 하고, 이것을 최소화하도록 백프로퍼게이션 알고리즘을 사용하여 모수를 업데이트한다.That is, in Equation 3, the parameter to be estimated is

And a bias vector of b (M dimension) and c (K dimension), respectively. The loss function is a cross entropy loss function, and the parameter is updated using a backprojection algorithm to minimize the loss function.

In the above description, a method of learning k cancer diagnostic biomarker group information for each training network and a neural network is described. Alternatively, k cancer diagnostic biomarker group information is used, but the final result of the multiclassification result Can be used to simultaneously learn neural networks and expert networks.

In the above description, the multi-arm is discriminated by learning and learning parameters using the same computing device. Alternatively, the multi-arm may be learned through a separate learning device, and the learned result may be used in a computing device for multi-arm discrimination .

The embodiments of the present invention described above can be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, and the like, alone or in combination. The program instructions recorded on the computer-readable recording medium may be those specially designed and constructed for the present invention or may be those known and used by those skilled in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules for performing the processing according to the present invention, and vice versa.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, Those skilled in the art will appreciate that various modifications, additions and substitutions are possible, without departing from the scope and spirit of the invention as disclosed in the accompanying claims.

Therefore, the spirit of the present invention should not be construed as being limited to the above-described embodiments, and all of the equivalents or equivalents of the claims, as well as the following claims, I will say.

Claims (22)

A method of computing a multi-arm,
(a) when the information of the k cancer diagnostic biomarker group is inputted, the computing device binarily classifies the information of the k cancer diagnostic biomarker groups so that the information of the k cancer diagnostic biomarkers is either cancerous or cancerous Judging whether or not it is a non-volatile memory; And
(b) if it is determined that the information on the k cancer diagnostic biomarker groups is cancerous as a result of the binary classification, the computing device analyzes the k cancer diagnostic biomarker group information using a neural network, Determining whether the information of the cancer diagnosis biomarker group corresponds to which of the k cancers;
, ≪ / RTI &
In the step (a)
Wherein the computing device inputs the k cancer diagnostic biomarker group information to each of k learned expert networks to deduce whether the k cancer diagnostic biomarker group information is cancer, Computing the respective inference values of each of the k expert networks using a gating function of the gating network in which the contribution to each of the k expert networks is determined in response to the group information, Cancer Diagnosis Classification of whether the information of the biomarker group is cancer,
In the step (b)
The computing device calculates the k cancer diagnostic biomarker information using the learned first connection strength when the k cancer diagnostic biomarker group information determined to be cancer in the step (a) is input to the input layer Wherein the hidden layer includes a plurality of neurons and a plurality of neurons, and wherein the hidden layer includes a plurality of neurons, a learned hidden layer having a plurality of neurons, Wherein the second connection strength is applied to the computed value in the output layer and is input to the output layer, and the classifying is performed for each of the k arms in the output layer. delete The method according to claim 1,
Wherein the computing device uses a soft max algorithm to classify whether the information of the k cancer diagnostic biomarker groups is cancer using the re-computed result value. Way. The method according to claim 1,
Before the step (a)
(a01) The computing device inputs k cancer diagnostic biomarker group information for training into each of the k expert networks before learning, so as to deduce whether the k cancer diagnostic biomarker group information for training is cancer Computing recursive values of each of the k expert networks using the gating function of the gating network in which the contribution to each of the k expert networks is determined corresponding to the k cancer diagnostic biomarker group information for training, Classifying the information of the k cancer diagnostic biomarker groups for training into cancer using the calculated result values, referring to the classified result values, and determining parameters of the k expert networks before being learned And learns parameters of the gating network before being learned Computing method for determining a multi-arm. 5. The method of claim 4,
Wherein the computing device includes a gradient ascent, an EM algorithm, and a quasi-Newton method for maximizing a log likelihood in learning the parameters of the k prior expert networks and the gating network before the learning, The method comprising the steps of: The method according to claim 1,
In the step (a)
The binary classifications may include a decision tree, a tree ensemble, a random forest, a Bayesian network, a support vector machine, a neural network, , And logistic regression. ≪ Desc / Clms Page number 20 > delete The method according to claim 1,
Wherein the hidden layer activation function is one of a sigmoid function (?), A hyperbolic tangent function (tanh), and a rectified linear unit function (ReLU). The method according to claim 1,
Wherein the classification for each of the k arms uses a soft max algorithm. The method according to claim 1,
Before the step (b)
(b01) when the computing device inputs the k pieces of cancer diagnostic biomarker group information for training into the input layer before learning, the k pieces of cancer diagnostic biomarker group information for training Calculating a value calculated using the first connection strength before being learned in the hidden layer before the learning by using a hidden layer activation function, Classifies the k arrays by inputting them to the output layer before being learned by applying the second connection strength before being learned to arithmetic values in each neuron of the hidden layer before being learned, And a result corresponding to the k cancer diagnostic biomarker group information for training Values of the first connection strength before learning, the parameters of the hidden layer before the learning, and the learning value of the learning layer before learning are obtained through back propagation using the loss value, Wherein the parameters of the output layer before the calculation are learned. The method according to claim 1,
Wherein the hidden layer is formed as a single layer or a multi-layered structure. A computing device for determining a multi-arm,
a communication unit for acquiring information on the k cancer diagnostic biomarker group; And
(i) binarizing the information of the k cancer diagnostic biomarker groups to determine whether the information of the k cancer diagnostic biomarker groups is cancer or cancer; and (ii) Wherein the k cancer diagnostic biomarker group information is analyzed using a neural network to determine whether the information of the k cancer diagnostic biomarker group is the one of the k cancer cells A processor for performing a process of determining whether the subject is cancerous;
, ≪ / RTI &
The processor comprising:
In the process (i), the k cancer diagnostic biomarker group information is input to each of k learned expert networks to deduce whether the k cancer diagnostic biomarker group information is cancer, and the k cancer diagnostic biomarker group information Computing the respective speculative values of the k expert networks using a gating function of the gating network in which the contribution to each of the k expert networks is determined corresponding to the biomarker group information, k Cancer Diagnosis Classifies the information of the biomarker group as cancer,
In the process (ii), when the k cancer diagnostic biomarker group information determined to be cancer in the process (i) is input to the input layer, the k cancer diagnostic biomarker group information Calculating a value calculated using the first connection strength in the learned hidden layer by using a hidden layer activation function, and calculating a value of the learned hidden layer using the hidden layer activation function, Wherein the second connection strength learned in each neuron is applied to the output layer, and the classification is performed for each of the k arms in the output layer. delete 13. The method of claim 12,
Wherein the processor uses a soft max algorithm to classify whether the information of the k cancer diagnostic biomarker group is cancer using the re-computed result value. . 13. The method of claim 12,
The processor comprising:
Before the process (i), the k cancer diagnostic biomarker group information for training is input to each of the k expert networks before being learned, so as to deduce whether the k cancer diagnostic biomarker group information for training is cancer Computing recursive values of each of the k expert networks using the gating function of the gating network in which the contribution to each of the k expert networks is determined corresponding to the k cancer diagnostic biomarker group information for training, Classifying the information of the k cancer diagnostic biomarker groups for training into cancer using the calculated result values, referring to the classified result values, and determining parameters of the k expert networks before being learned The process of learning the parameters of the gating network before learning Wherein the step of determining the number of multi-arms is further performed. 16. The method of claim 15,
Wherein the processor is configured to calculate parameters of the k prior expert network and the gating network before the learning by using any one of a gradient ascent, an EM algorithm, and a quasi-Newton method that maximizes a log likelihood Wherein the one of the plurality of arcs is used. 13. The method of claim 12,
The processor comprising:
In the process (i), the binary classification may be a decision tree, a tree ensemble, a random forest, a Bayesian network, a support vector machine ), A neural network, and a logistic regression. ≪ RTI ID = 0.0 > [0002] < / RTI > delete 13. The method of claim 12,
Wherein the hidden layer activation function is any one of a sigmoid function (?), A hyperbolic tangent function (tanh), and a rectified linear unit function (ReLU). 13. The method of claim 12,
Wherein the classification for each of the k arms utilizes a soft max algorithm. 13. The method of claim 12,
The processor comprising:
Wherein when the k cancer diagnostic biomarker group information for training is input to the input layer before learning (ii) before the step (ii), the first k-th cancer biomarker group information for training Calculating a value calculated using the first connection strength before being learned in the hidden layer before the learning by using a hidden layer activation function, Classifies the k arrays by inputting them to the output layer before being learned by applying the second connection strength before being learned to arithmetic values in each neuron of the hidden layer before being learned, Corresponding to the k cancer diagnostic biomarker group information for training A second connection strength before being learned, a parameter of a hidden layer before being learned, and a second connection strength before learning are obtained through back propagation using the loss value, And a step of learning parameters of the output layer before the calculation of the parameter of the multi-arm. 13. The method of claim 12,
Wherein the hidden layer is formed as a single layer or a multi-layered structure.

Images