KR20200107841A - Method for identifying disease phenotype based on combined score of significant gene expression signatures from transcriptome sample of patients - Google Patents

Abstract

The present invention relates to a disease phenotype identification method for determining a disease in a computing device operated by at least one processor. According to the present invention, the disease phenotype identification method comprises the steps of: collecting gene expression data of peripheral blood mononuclear cells (PBMCs) of patients and gene expression data of PBMCs of normal people from a microarray database; evaluating the similarity between the collected gene expression data, and selecting genes that show a significant difference in expression between the gene expression data of the patients and the gene expression data of the normal people; identifying genes that are overexpressed and suppressed for disease state compared to normal state among selected genes; determining a reference disease prediction gene score based on expression values of the overexpressed genes and expression values of the suppressed-expressed genes; and if a disease prediction gene score calculated based on a gene expression value in a received random sample is higher than the reference disease prediction gene score, determining the random sample as a disease sample.

Description

METHOD FOR IDENTIFYING DISEASE PHENOTYPE BASED ON COMBINED SCORE OF SIGNIFICANT GENE EXPRESSION SIGNATURES FROM TRANSCRIPTOME SAMPLE OF PATIENTS}

The present invention relates to a method for discriminating a disease phenotype based on a score of a combination of significant gene expression indicators in a patient transcript sample.

Blood tissues circulate in the body to exchange and transport substances between tissues, and immune cells that make up blood interact with various tissues in the body at a molecular level. For this reason, blood tissue has been used as a major target tissue for inferring the phenotype and drug responsiveness of a specific disease using gene expression indicators in the blood, and discovering meaningful disease diagnosis and gene therapy biomarkers. In addition, blood sample collection by the patient is performed with minimal invasiveness, and processing can be performed through relatively simple steps. Due to these advantages, gene expression data of patient blood tissues and constitutive immune cells were collected and analyzed for various diseases.

However, the existing process of discovering disease predictive markers based on blood cell gene expression data does not have a scale that can be evaluated by collectively scoring a specific disease phenotype from different blood sample data sets, and the reproducibility of disease-normal discriminant markers. There is a problem with this falling.

Existing studies have found genes with significant expression differences of about several hundred between disease-normal phenotypes for disease markers shared by single or multiple diseases, analyzing the mechanism and function of the marker, and applying it to drug re-creation. I have been discussing the possibilities.

However, in the previous studies, qualitative analysis was conducted on genes that were significant in a specific data set.

Most of the results have not been built into a valid scoring model in a form applicable to independent patient samples from different populations. In actual clinical practice, it has not reached the level applied to judge disease status or specific phenotype.

Due to such a problem in the analysis of blood tissue disease omics, it is difficult to construct an effective predictive model that can be generalized for a specific disease phenotype.

The problem to be solved by the present invention relates to a method of discriminating a disease phenotype by using gene expression data of peripheral blood mononuclear cells (PBMCs), which are major immune cell groups constituting blood.

According to one feature of the present invention, a method for determining a disease phenotype is a method for determining a disease phenotype in a computing device operated by at least one processor, and a peripheral blood mononuclear cell (PBMC) of a patient from a microarray database. ), and collecting gene expression data of peripheral blood mononuclear cells of normal people, and evaluating the similarity between the collected gene expression data to make significant expression between the gene expression data of the patient and the gene expression data of the normal person. Selecting genes showing differences, identifying genes that are overexpressed and suppressed for disease states compared to normal conditions among the selected genes, based on expression values of overexpressed genes and expression values of suppressed-expressed genes, Determining a disease prediction gene score, and if the disease prediction gene score calculated based on the gene expression value in the received random sample is higher than the reference disease prediction gene score, discriminating the random sample as a disease sample. Include.

Between the step of selecting and the step of confirming, the step of determining a disease discrimination marker combination by sequentially combining the selected genes, the step of confirming, from among genes included in the disease discrimination marker combination Genes that are overexpressed and suppressed can be identified.

The step of determining the disease discrimination marker combination may include expanding the gene combination by adding another gene when the disease discrimination performance of any gene combination is increased compared to the previous gene combination, and the disease discrimination performance is increased compared to the previous gene combination. Otherwise, it may include determining the corresponding gene combination as a disease discrimination marker combination.

The determining of the disease discrimination marker combination includes calculating a disease discrimination performance for the random combination of genes, and the calculating of the disease discrimination performance comprises each disease discrimination model calculated through cross-validation. The arithmetic mean value of each discrimination accuracy can be calculated as the disease discrimination performance of the arbitrary gene combination.

The determining step includes calculating a difference between the geometric mean value of the patient's gene expression values for the overexpressed genes and the geometric mean value of the patient's gene expression values for the suppressed-expressed genes as the patient's disease prediction gene score. Step, calculating the difference between the geometric mean value of gene expression values of the normal person for the overexpressed genes and the geometric mean value of gene expression values of the normal person for the suppressed-expressed genes as a disease predicted gene score of the normal person, the patient Calculating the disease prediction gene score of the plurality of patients based on gene expression data of peripheral blood mononuclear cells, and selecting the lowest disease prediction gene score from among the calculated disease prediction gene scores, the disease prediction gene score of the normal person Calculating the gene expression data of the peripheral blood mononuclear cells of a plurality of normal individuals as a target, selecting the highest disease predicting gene score among the calculated disease predicting gene scores, and the lowest disease predicting gene score and the highest disease predicting gene score It may include the step of determining the average score of the reference disease prediction gene score.

In the step of determining, the level of the expression distribution of the gene expression data in the arbitrary sample is matched with the expression distribution of the gene expression data used for calculating the reference disease prediction gene score, and then the gene expression value in the arbitrary sample is based The disease predicted gene score calculated by may be compared with the reference disease predicted gene score.

According to another feature of the present invention, a disease phenotype determination method is a method of determining a disease phenotype in a computing device operated by at least one processor, comprising peripheral blood mononuclear cells of a patient collected from a microarray database. PBMC) and the gene expression data of the peripheral blood mononuclear cells of a normal person, calculating a reference disease prediction gene score, a disease prediction gene score calculated based on the gene expression data in any sample received Comparing the reference disease prediction gene score with the reference disease prediction gene score, when the disease prediction gene score calculated for the random sample has a value higher than the reference disease prediction gene score, determining the random sample as a disease sample, And determining the random sample as a normal sample if the disease prediction gene score calculated for the random sample has a value lower than the reference disease prediction gene score.

The step of calculating the reference disease predicted gene score includes gene expression values of normal individuals and patients for overexpressed genes compared to gene expression data of normal individuals among gene expression data collected from the microarray database, and suppressed expression. Based on the gene expression values of the normal person and the patient for the genes, the reference disease prediction gene score is calculated, and the disease prediction gene score calculated for the random sample is the overexpression among genes in the random sample. It can be calculated based on the gene expression values for each of the genes and the suppressed-expressed genes.

The calculating may include evaluating similarity between gene expression data collected from the microarray database, and selecting genes representing a significant difference in expression between gene expression data of a patient and gene expression data of a normal person, and For the genes, it may include calculating the reference disease prediction gene score.

After the selecting step, selecting one gene from among the selected genes as a starting gene, generating disease discrimination marker combination candidates by adding the remaining genes one by one to the starting gene, and each of the disease discriminating marker combination candidates Estimating a disease discrimination performance for, wherein the reference disease prediction gene score may be calculated using a disease discrimination marker combination candidate having a highest value for the disease discrimination performance.

After the estimating step, determining a candidate for a primary disease discrimination marker combination having the highest value of the disease discrimination performance, adding genes not included in the primary disease discrimination marker combination candidate among the selected genes one by one Generating secondary disease discrimination marker combination candidates, selecting one secondary disease discrimination marker combination candidate having the highest disease discrimination performance among the secondary disease discrimination marker combination candidates, and selected secondary disease discrimination marker When the combination candidate has increased disease discrimination performance compared to the primary disease discrimination marker combination candidate, genes not included in the primary disease discrimination marker combination candidate among the selected genes are added one by one to provide tertiary disease discrimination marker combination candidates. If the selected secondary disease discrimination marker combination candidate does not increase the disease discrimination performance compared to the primary disease discrimination marker combination candidate, the selected secondary disease discrimination marker combination candidate is used as the reference disease prediction gene score. It may further include determining the object to be calculated.

The disease discrimination performance is a plurality of disease discrimination generated by selecting random genes from among genes included in the disease discrimination marker combination candidate, separating them into training data and verification data, and learning the genes separated by the training data multiple times. After applying the genes classified by the verification data to the model, it may be calculated as an average value of the disease determination accuracy calculated through cross-validation.

According to an embodiment of the present invention, it is possible to obtain a gene combination that is simpler and more effective than a conventional diagnosis method through analysis of gene expression data of a patient based on a machine learning technique and various sample combinations.

In addition, it is possible to construct a disease discriminating marker combination in a form in which the over-fitting phenomenon is reduced in a limited number of transcript data through cross-validation in various sample combinations.

In addition, it is possible to increase diagnostic efficiency and accuracy by discovering optimal disease markers in peripheral blood tissues of lupus patients, and provide an objective and simplified diagnostic technique compared to existing lupus phenotype-based diagnostic scores.

In addition, the biomarker is related to lupus diagnosis, and finally, clinicalization and industrialization can be achieved through the development of a disease diagnosis chip or small-scale microarray through only the gene profile in the processed blood sample.

1 is a block diagram of an apparatus for discovering a disease phenotype gene expression marker according to an embodiment.
2 shows the operation of the disease phenotype-related gene selection unit 110 of FIG. 1.
3 shows the operation of the disease determination performance estimating unit 120 of FIG. 1.
4 shows the operation of the disease determination gene combination selection unit 130 of FIG.
5 shows the operation of the disease prediction gene score calculation unit 140 of FIG. 1.
6 shows the operation of the disease determination unit 150 of FIG. 1.
7 is a graph showing an experiment result applying an embodiment of the present invention.
8 is a block diagram showing a hardware configuration of an apparatus for discovering a disease phenotype gene expression marker according to another embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art can easily implement the embodiments of the present invention. However, the present invention may be implemented in various different forms, and is not limited to the embodiments described herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and similar reference numerals are assigned to similar parts throughout the specification.

Throughout the specification, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary.

In addition, terms such as "... unit", "... group", and "... module" described in the specification mean a unit that processes at least one function or operation, which can be implemented by hardware or software or a combination of hardware and software. I can.

Expressions described in the singular in this specification may be interpreted as the singular or plural unless an explicit expression such as "one" or "single" is used.

In the present specification, the same reference numbers refer to the same elements regardless of the drawings, and "and/or" includes each and all combinations of one or more of the mentioned elements.

The devices described in the present invention are composed of hardware including at least one processor, a memory device, a communication device, and the like, and a program that is combined with the hardware and executed is stored in a designated place. The hardware has a configuration and capability to implement the method of the present invention. The program includes instructions for implementing the operation method of the present invention described with reference to the drawings, and executes the present invention by combining it with hardware such as a processor and a memory device.

In the present specification, a marker is a marker that can detect a change in a living body with specific molecular information derived from a protein or gene.

Phenotype refers to the physical, chemical, and clinical aspects of the expression of a specific genotype.

Transcript expression profile refers to the analysis of genes expressed or transcribed from the genome. Gene expression pattern is one of the important determinants of cell phenotype and cell function. Transcript expression profiles have been used to analyze the cause and effect of disease.

1 is a block diagram of an apparatus for discovering a disease phenotype gene expression marker according to an embodiment.

Referring to FIG. 1, an apparatus for discovering disease phenotype gene expression markers (hereinafter, referred to as “excavation apparatus”) 100 performs disease phenotype discrimination based on a score of a significant gene expression index combination of a patient transcript sample.

The discovery device 100 operates as at least one processor, and includes a disease phenotype-related gene selection unit 110, a disease identification performance estimation unit 120, a disease identification gene combination selection unit 130, and a disease prediction gene score calculation unit ( 140) and a disease determination unit 150.

The disease phenotype-related gene selection unit 110 analyzes the expression statistics of the transcript expression profile of peripheral blood mononuclear cells (hereinafter referred to as "PBMC") between the patient and the normal person. Select visible genes. That is, a gene that is significantly expressed with the patient's PBMC is selected as a disease phenotype-associated gene.

The disease discrimination performance estimating unit 120 estimates disease discrimination performance for the gene selected by the disease phenotype-related gene selecting unit 110.

The disease discrimination gene combination selection unit 130 selects a disease phenotype gene expression marker among the disease phenotype-related genes selected by the disease phenotype-related gene selection unit 110 to increase the ability to discriminate between phenotypes.

The disease discrimination gene combination selection unit 130 infers the gene expression characteristics linked to the disease phenotype using gene expression data of the patient PBMC, and then combines the heuristic selection technique and the Monte-Carlo cross test from the discovery data set to perform specific discrimination. As described above, the optimal combination of disease discrimination markers that can distinguish between disease and normal can be selected.

The disease prediction gene score calculation unit 140 calculates a disease prediction gene score for the disease phenotype gene expression marker selected by the disease determination gene combination selection unit 130. The disease prediction gene score calculation unit 140 generates a disease prediction score model capable of best discriminating the selected optimal disease discrimination marker gene from the training data set.

The disease determination unit 150 determines whether there is a disease by comparing a gene expression value of a sample with a disease predicted gene score. If the gene expression value has a higher value than the disease predicted gene score, the disease determination unit 150 determines it as a disease sample, and if it has a low value, it determines it as a normal sample.

The disease determination unit 150 may verify the performance of the disease-specific blood tissue gene expression marker by applying the reference disease prediction gene score generated by the disease prediction gene score calculation unit 140 to the verification data set.

For each operation of the disease phenotype-related gene selection unit 110, the disease determination performance estimation unit 120, the disease determination gene combination selection unit 130, the disease prediction gene score calculation unit 140, and the disease determination unit 150 It will be described in detail with reference to FIG. 6 in FIG. 2.

2 shows the operation of the disease phenotype-related gene selection unit 110 of FIG. 1.

Referring to FIG. 2, the disease phenotype-related gene selection unit 110 selects a normal sample and a disease sample from the microarray database 200 (S101).

Microarrays are a collection of gene fragments arranged on a DNA (DeoxyriboNucleic Acid) chip, and are a tool that can broadly analyze gene expression. The DNA chip is an orderly arrangement of cDNA or oligo DNA encoding genes on glass. The cDNA or oligo DNA arranged on the DNA chip is used as a type of probe.

The microarray database 200 stores transcript microarray gene expression data of PBMCs having disease conditions and normal conditions. For example, the microarray database 200 may be a database of Gene Expression Omnibus (GEO). Typically, transcript data refers to the sum of all expressed RNAs.

A plurality of samples are registered in the microarray database 200. Here, the sample is a genetic data set containing gene expression data. Gene expression data are transcript microarray data of PBMCs having disease and normal conditions.

Samples of the microarray database 200 are divided into normal samples and disease samples. A normal sample is a genetic data set containing gene expression data extracted from PBMCs of a normal person. The disease sample refers to a genetic data set including gene expression data extracted from PBMCs of a patient.

The disease phenotype-related gene selection unit 110 maps the probe serial number of gene expression data included in each of the disease sample and the normal sample selected in step S101 with the gene ID (S103). Here, as the gene ID, Entrez Gene ID may be used. Entrez Gene is a database of the National Center for Biotechnology Information (NCBI) for gene specific information.

When a plurality of probes correspond to one gene ID, the disease phenotype-related gene selection unit 110 designates the average of the gene expression values of each of the probes as the representative expression value of the corresponding gene ID. Do (S105).

The disease phenotype-related gene selection unit 110 normalizes the value of the representative expression value designated for each gene ID through step S105 (S107). The disease phenotype-related gene selection unit 110 may remove differences between samples and conditions in the gene expression data by performing quantile normalization (S107).

The disease phenotype-related gene selection unit 110 performs significance evaluation on the representative expression values specified in the gene ID of the normal sample and the representative expression values specified in the gene ID of the disease sample (S109). In addition, the disease phenotype-related gene selection unit 110 selects at least one gene ID showing a significant difference in expression between the disease sample and the normal sample (S111). At this time, the t-test is used as the significance evaluation method.

The disease phenotype-related gene selection unit 110 may perform "evaluation of significance between two conditions (normal/disease) of genes" in order to correct a problem occurring in a multi-statistical test process. For example, it is possible to evaluate the significance of the gene by performing the process of adjusting the false discovery rate of "Benjamini-Hochberg". At this time, a gene ID having a corrected p-value of "0.05" or less may be selected.

3 shows the operation of the disease determination performance estimating unit 120 of FIG. 1.

Referring to FIG. 3, the disease discrimination performance estimation unit 120 receives candidate gene expression data from the disease phenotype-related gene selection unit 110 (S201).

Here, the candidate gene expression data is composed of a disease phenotype-associated gene list consisting of the gene ID selected in step S111 of FIG. 2, representative expression values of the genes under normal conditions, and representative expression values under disease conditions. Here, the representative expression values are normalized values in step S107 of FIG. 2.

The disease discrimination performance estimation unit 120 randomly selects each gene data set corresponding to each disease phenotype-related gene selected in step S111 of FIG. 2 and separates it into training data and verification data (S203). That is, the disease discrimination performance estimating unit 120 randomly selects a disease gene data set including a gene ID and a normal gene data set and separates them into training data and verification data. At this time, diseases included in each of the training data and verification data The ratio between the genetic data set and the normal genetic data set can be made constant. For example, the ratio between the disease gene data set and the normal gene data set may be a ratio of 2:1.

The disease discrimination performance estimating unit 120 is a "disease discrimination model" that can determine a disease phenotype-related gene and a normal phenotype-related gene by learning a gene expression pattern between a disease sample and a normal sample by applying a machine learning algorithm to the training data set. To generate (S205).

In this case, the disease discrimination performance estimating unit 120 may learn the gene expression pattern of the training data set and information on whether the sample in which the expression pattern is displayed is a disease sample or a normal sample. For example, the gene expression pattern of the training data set may indicate an increase or decrease in expression in a patient's PBMC among a plurality of probes showing a difference in expression between a normal person's PBMC and a patient's PBMC. .

According to an embodiment, as a machine learning algorithm, a support vector machine (collectively referred to as “SVM”) may be used. SVM is applied according to the C-classification type, and a Radial Basis Function (RBF) can be applied.

The disease discrimination performance estimating unit 120 applies the gene expression values of the verification data set to the disease discrimination model generated in step S205 to infer whether the pattern of each gene expression value represents a normal sample or a disease sample, It is determined whether the inference result is consistent with the normal condition or disease condition labeled with the gene expression value. Then, the discrimination result is scored to calculate discrimination accuracy (S207).

The disease discrimination performance estimating unit 120 applies the gene expression values of each of the normal samples and the disease samples in the verification data set to the disease discrimination model generated in step S205 to determine whether the pattern of each gene expression value represents a normal sample or disease. Infer whether it represents a sample. And by comparing the inference result with the actual label, the discrimination accuracy (A) of each disease discrimination model is calculated. Here, a label such as "normal" or "disease" is given to the gene expression value. The calculation formula of the discrimination accuracy (A) is as follows.

Here, N _A is the number of samples obtained by accurately inferring whether the gene expression value of the verification data set is a normal sample or a disease sample by applying the gene expression value of the verification data set to the disease discrimination model generated at N times (1≦N). That is, N _A is the number of samples in which the inference result and the actual label match.

N _T is the total number of samples applied to the disease discrimination model generated in the Nth round (1≦N), that is, the total number of samples included in the verification data set.

The disease discrimination performance estimating unit 120 determines whether steps S203 to S207 are repeated N times as many as a designated number of cross-validations (S209). If the number of cross-validations has not been reached, it starts again from step S203.

Step S209 includes estimating the power of tens of thousands to hundreds of thousands of times according to the amount of information collected and the number of iterations specified, and the disease discrimination performance estimating unit 120 may utilize a parallelization calculation technique using distributed computing for high-speed calculation. .

In this way, the disease discrimination performance estimating unit 120 performs steps S205 and S207 for the N "training data set-validation data set" combinations selected through N random sub-sampling (S203). To estimate N discrimination accuracy.

At this time, the N "training data set-validation data set" combinations are randomly extracted from the genetic data set of genes related to the disease phenotype selected in step S111 of FIG. 2, so that the cross-validation is Monte Carlo Cross-validation. It corresponds to the method.

Steps S203 to S209 are repeated N times for each candidate gene to estimate N discrimination accuracy. The disease discrimination performance estimating unit 120 determines the arithmetic mean value of the N discrimination accuracy as the final disease discrimination performance of the candidate genes (S211). As such, the disease discrimination performance is calculated for each of the at least one candidate gene ID selected in step S111 of FIG. 2 through steps S203 to S211.

4 shows the operation of the disease determination gene combination selection unit 130 of FIG.

Referring to FIG. 4, the disease determination gene combination selection unit 130 receives candidate gene expression data from the disease phenotype-related gene selection unit 110 (S301). The candidate gene expression data is the same as step S201 of FIG. 3.

The disease discrimination gene combination selection unit 130 designates one candidate gene randomly selected from the genes included in the input candidate gene expression data (hereinafter, referred to as "candidate gene") as a starting gene of the disease discrimination marker combination candidate. (S303).

The disease determination gene combination selection unit 130 generates disease determination marker combination candidates by adding one of the remaining candidate genes to the starting gene (S305).

The disease determination gene combination selection unit 130 estimates the disease determination performance through steps S203 to S211 of FIG. 3 for each of the disease determination marker combination candidates generated in step S305 (S307). As described in step S307, the disease determination performance is estimated through steps S203 to S211 of FIG. 3.

The disease determination gene combination selection unit 130 selects a disease determination marker combination candidate having the largest disease determination performance estimated in step S307 (S309).

The disease determination gene combination selection unit 130 determines whether the disease determination performance of the disease determination marker combination candidate selected in step S309 is higher than the disease determination performance of the previous disease determination marker combination candidate (S311).

If increased, the disease determination gene combination selection unit 130 starts again from step S305. That is, steps S305 to S311 are performed by adding one by one candidate genes that are not included in the candidate disease identification marker combination selected in step S309.

Meanwhile, if it is determined that there is no increase in step S315, the disease discrimination gene combination selection unit 130 determines the disease discrimination marker combination candidate generated in step S309 as the optimal disease discrimination marker combination (S313).

In this way, the disease determination gene combination selection unit 130 tests while adding genes that are not in the disease determination marker combination candidate one by one, and selects an optimal disease determination marker combination candidate composed of genes having the best disease determination performance among them. It uses the best-first search, ie "heuristic selection technique".

The above steps S303 to S313 will be described as an example.

Assume that candidate genes selected in step S111 of FIG. 2 are “A, B, C, D”. At this time, when the disease determination gene combination selection unit 130 designates "A" as a starting gene (S303), sequentially generates "AB", "AC", and "AD" (S305), and The disease discrimination performance is estimated (S307). The disease discrimination performance of a single gene "A" is calculated through FIG. 3. The disease determination gene combination selection unit 130 acquires disease determination performance for each of "A-B", "A-C", and "A-D" generated in step S305 through the disease determination performance estimation unit 120. For example, the disease discrimination performance for "A-B" is estimated by performing steps S203 to S211 using normal/disease samples including "A" and normal/disease samples including "B".

The disease discrimination gene combination selection unit 130 may select a disease discrimination marker combination candidate, for example, "A-C" having the highest discrimination performance among them (S309).

The disease discrimination gene combination selection unit 130 determines whether the disease discrimination performance of "A-C" is higher than the disease discrimination performance of the previous combination (S311).

At this time, since the previous combination corresponds to the starting gene, if the disease discrimination performance of "A-C" is increased than the disease discrimination performance of "A", steps S305 to S311 are repeated. That is, the disease determination gene combination selection unit 130 sequentially generates "ACB" and "ACD" by adding genes (B, D) not included in "AC" (S305), and diseases for each of them The discrimination performance is estimated (S307).

The disease discrimination gene combination selection unit 130 selects a disease discrimination marker combination candidate having a higher disease discrimination performance value from among "A-C-B" and "A-C-D" (S309). For example, if "A-C-B" is selected (S309), it is determined whether the disease discrimination performance of "A-C-B" is increased than that of "A-C" by comparing the disease discrimination performance of "A-C-B" and "A-C" (S311).

For example, if the disease discrimination performance of "A-C-B" is not increased than that of "A-C", "A-C" is determined as an optimal disease discrimination marker combination candidate (S313).

On the other hand, if the disease discrimination performance of "A-C-B" is higher than that of "A-C", steps S307 to S311 are repeated from the process of adding a gene "D" not included in "A-C-B" (S305).

After this process, if the disease discrimination performance of "A-C-B-D" is increased compared to "A-C-B", the disease discrimination gene combination selection unit 130 determines "A-C-B-D" as the optimal disease marker combination (S313).

In this way, after performing steps S303 to S317 for candidate genes "A, B, C, D", an optimal disease marker combination is determined (S313).

If there are two or more candidate disease marker combinations selected in step S309, for example, there may be a case where the disease discrimination performance of "A-C-B" and "A-C-D" are all equal to 1. In this case, the disease discrimination performance estimation unit 120 calculates the discrimination accuracy by performing steps S203 to S207 of FIG. 3 for normal/disease samples each including "A", "C", and "B", and The discrimination accuracy is calculated by performing steps S203 to S207 using normal/disease samples each including "A", "C", and "D". The disease determination gene combination selection unit 130 performs a t-test between the disease sample and the normal sample for the calculated determination accuracy. And one combination in which the most significant disease-normal score difference appears, for example, one of "A-C-B" or "A-C-D" is selected.

5 shows the operation of the disease prediction gene score calculation unit 140 of FIG. 1.

Referring to FIG. 5, the disease prediction gene score calculation unit 140 checks whether overexpression and suppression expression for disease states compared to normal for disease genes included in the optimal disease discrimination marker combination derived in step S313 of FIG. 4. Confirm (S401).

The disease prediction gene score calculation unit 140 applies step S109 of FIG. 2 to a disease sample and a normal sample including disease genes included in the optimal combination of disease determination markers. Through this, based on the sign of the t-statistic of the 2-sample t-test between the disease sample and the normal sample, a positive result is determined as overexpression in a disease sample, and a negative result is determined as suppressive expression in a disease sample.

The disease prediction gene score calculation unit 140 includes a geometric mean of the overexpressed gene expression values identified in step S401 for each of the disease samples and normal samples including disease genes included in the optimal disease discrimination marker combination, and suppression expression gene expression. The geometric mean of the values is calculated, and the difference is determined as a disease prediction gene score of each disease sample and a normal sample that is a control group of the disease sample (S403).

The disease prediction gene score calculation unit 140 determines the average value of the lowest predicted gene score of the disease sample and the highest predicted gene score of the normal sample among the disease predicted gene scores determined for each disease sample and normal sample as the final disease predicted gene score (S405). ).

The above steps S401 to S405 will be described by way of example.

Assume that the optimal combination of disease discrimination markers determined in FIG. 4 is “A-B-C-D-E”, the overexpressed genes are “A, B, C” and the suppressed-expressed genes are “D, E”. Assuming that the disease sample including all of "A, B", "C", "D", and "E" is "S1", "S2", "S3", "A, B", "C", " Assume that normal samples including both D" and "E" are "S4" and "S5".

The disease prediction gene score calculation unit 140 includes the geometric mean (A1) of the overexpressed gene expression values for each of the gene expression values "A", "B", and "C" of "S1" and "S1". The geometric mean (A2) of the respective gene expression values for D", "E", that is, the inhibitory expression gene expression values is calculated. In addition, the disease prediction gene score calculation unit 140 calculates the difference between the geometric mean A1 and the geometric mean A2 as a disease prediction gene score of “S1”. In this way, disease prediction gene scores for each of "S1", "S2", "S3", "S4", and "S5" are calculated (S403).

The disease prediction gene score calculation unit 140 includes the lowest disease prediction gene score among disease prediction gene scores of disease samples "S1", "S2" and "S3", and diseases of normal samples "S4" and "S5". Among the predicted gene scores, the average score of the highest disease predicted gene score is determined as the final disease predicted gene score (S405). For example, if "S1" is the lowest and "S5" is the highest, the average score of the disease predictor gene score of "S1" and the disease predictor gene score of "S5" is finally the threshold disease predictor gene score. Is determined by

6 shows the operation of the disease determination unit 150 of FIG. 1.

6, the disease determination unit 150 normalizes the transcript expression data of a random patient (S501). Here, normalization may perform quantile normalization as described in FIG. 2.

The disease determination unit 150 makes the gene expression distribution of the normalized transcript expression data of any patient equal to the expression distribution of genes included in the optimal disease determination marker combination selected by the disease determination gene combination selection unit 130 ( S503). To this end, the disease determination unit 150 calculates the ranks of gene expression values of the normalized randomized patient's transcript expression data, respectively. At this time, the Spearman rank correlation evaluation method for the basal-like gene expression center may be used.

The disease determination unit 150 is an expression value in the disease gene data set and the normal gene data set (hereinafter referred to as “excavation data set”) each including genes selected in step S111 of FIG. 2 according to the calculated position. Quintile normalized) replaces gene expression values of transcript expression data of any patient (S503).

At this time, if the transcript expression data of any patient does not contain a specific gene included in the discovery data set, the disease determination unit 150 replaces the specific gene except for the specific gene.

The disease determination unit 150 calculates a disease prediction gene score through the process of FIG. 5 on the transcript expression data in which the distribution of the discovery data set and the gene expression value is the same in step S503 (S505). In other words, among the genes in the transcript expression data, the difference between the geometric mean of the expression values of each of the genes identified as overexpression in step S401 and the geometric mean of the expression values of each of the genes identified as suppressed expression is used as the disease prediction gene score. Calculate.

Referring to the example of Figure 5, when the overexpressed gene "A, B, C" and the suppressed-expressed gene "D, E" includes transcript expression data, the expression of each of "A, B, C" The difference between the geometric mean of the values and the geometric mean of the expression values of "D, E" is calculated as a disease predictive gene score (S505).

The disease determination unit 150 compares the disease predicted gene score calculated in step S505 with the reference disease predicted gene score calculated in step S405 of FIG. 5 (S507) to determine whether it has a relatively high score (S509).

At this time, if it is determined that the disease determination unit 150 has a high score in step S509, it is determined as a disease (S511). On the other hand, if it is determined that the disease determination unit 150 has a low score in step S509, it determines that it is normal (S513).

Meanwhile, Lupus is a chronic autoimmune disease in which symptoms are expressed in various connective tissues in the body. The onset can occur in various ages and sexes, but the incidence of the onset is high especially in women of childbearing age between 15 and 44 years of age.

Lupus symptoms appear when autoimmune antibodies attack connective tissue. Since lupus symptoms progress from skin and joints to organs of the body including lungs, kidneys, and heart, specific and accurate diagnostic techniques are required for timely treatment.

The diagnostic criteria for existing lupus patients include 11 clinical phenotype-based diagnostic criteria (1997, Hochberg, Arthritis Rheum) established in 1997 by the American College of Rheumatology (ACR), and 6 ACR criteria. It has been achieved through the criteria of the International Systemic Lupus Erythematosus International Collaborating Clinics (SLICC) established in 2012 by adding molecular immunological criteria (2012, Petri et al., Arthritis Rheum).

However, these criteria have a problem in that they are simply summed up of various clinical characteristics with the same weight without considering the clinical characteristics even though the clinical characteristics are not uniform with the progression of lupus.

In addition, among each phenotypic diagnosis item, it shows a high false-positive rate in the diagnosis process of early patients with few corresponding items and in the process of discrimination from other immune diseases, so procedures such as additional biopsy of lesions of internal organs must be accompanied to confirm the disease. It has limitations.

Due to these existing problems, in particular, in lupus disease, an analysis of gene expression characteristics specific to the disease state has been made based on the difference in gene expression and significance correlation module between the disease and the normal state using transcript data. Based on this, the possibility of molecular diagnosis and discovery of therapeutic targets was continuously presented.

Accordingly, FIGS. 1 to 6 described above may be applied to various autoimmune diseases, but in particular, the following experimental examples were obtained by applying them to lupus.

<Experimental Example>

In order to discover a PBMC sample-based lupus discrimination marker, lupus patient transcript microarray data as shown in Table 1 below were collected.

Microarray
Serial Number platform Number of patient samples Number of normal samples GSE8650 GPL96/GPL97 38 21

For the purpose of independently selecting the lupus discrimination gene of FIGS. 2 to 4 and constructing the model and evaluating the lupus discrimination performance for the random samples of FIGS. 5 to 6, the lupus patient transcript microarray data is randomly 2:1 Divided by the ratio of, respectively, the excavation data set for lupus discrimination gene selection and model construction (lupus patient 25, normal 14) and independent verification data set (lupus patient 13, normal 7) were used separately. Here, the discovery data set includes training data and verification data used to generate a disease discrimination model. The independent verification data set is used to derive lupus discrimination performance.

As a result of applying the series of processes shown in FIG. 2 to a data set of a patient sample and a data set of a normal sample within the excavation data set of Table 1, genes showing a significant difference in expression between lupus patients and normal as shown in Table 2 were derived. .

P-value Number of genes <0.001 502 <0.01 1386 <0.05 2743

As a result of selecting genes with P<0.05 as significant genes, and selecting candidate genes for combination of disease discrimination markers through the process of FIGS. 3 and 4 based on these, disease discrimination performance of a genetic data set including a total of 234 genes This value of 1 (100%) was expressed to derive the optimal lupus discrimination marker combination.

Table 3 shows the top 10 gene combinations with the largest difference in score calculated through the process of FIG. 5 for genes included in the 234 optimal lupus discrimination marker combinations.

At this time, the disease prediction gene score of the lupus sample containing these genes was calculated by applying the process of FIG. 5 to 234 genes, and the difference between these scores by calculating the disease prediction gene score of the normal sample containing these genes. The top 10 gene combinations are indicated in the largest order, and for each gene combination, the constituent gene list, the p-value of the t-test between the disease-normal score, and the score for the disease-normal classification are specified.

Gene list Number of genes P-value Score for disease-normal classification RUSC1, IFI44, GM2A 3 3.49E-19 9.038 DGCR8, IFI44, GM2A 3 4.30E-19 8.764 AAR2, IFI44, PNPLA2, GM2A 4 6.14E-19 9.369 GAPBA, IFI44, PITPNC1, ATP6V0C 4 1.20E-18 1.640 SMG7, IFI44, GM2A 3 1.33E-18 8.713 PFKM, IFI44, UBA7, GM2A 4 1.37E-18 9.234 LRCH4, IFI44, GM2A 3 1.42E-18 9.314 TMEM147, IFI44, MTX1 3 1.60E-18 9.751 CDS2, IFI44, PPP6R1, GM2A 4 1.77E-18 9.204 SDF4, IFI44, GM2A 3 2.04E-18 9.291

To further explain Table 3, for example, the P-values of RUSC1, IFI44, and GM2A are calculated by applying RUSC1, IFI44, and GM2A to step S109 of FIG. 2. The disease-normal classification reference score of the overexpressed genes RUSC1, IFI44, and GM2A is the reference disease prediction gene score calculated by applying RUSC1, IFI44, and GM2A to FIG. 5.

Referring to Table 3, it was confirmed that the optimal lupus discrimination marker combination with the highest disease-normal classification criterion score was composed of three genes {RUSC1, IFI44, GM2A}.

In order to test the lupus disease discrimination performance of the verification data set through the disease-normal classification criterion score of {RUSC1, IFI44, GM2A}, the process of FIG. 6 is applied to the independent verification data set to derive lupus discrimination performance of random patients As a result, accuracy 0.9, sensitivity 0.9231, and specificity 0.8571 were found. This result is comparable to the standard in terms of sensitivity and specificity when compared with the diagnostic performance of the sensitivity (=0.932) and specificity (=0.84) of the existing clinically used lupus systemic lupus international collaborating clinic (SLICC). It showed a level of diagnostic performance.

In addition, PBMC gene expression data of three different autoimmune diseases as shown in Table 4 were collected in order to confirm whether the PBMC-linked gene combination according to an embodiment of the present invention is a gene pattern that generally appears in other autoimmune diseases.

Microarray serial number platform Disease name Number of patient samples Number of normal samples GSE15573 GPL6102 Rheumatoid arthritis (RA) 18 15 GSE3365 GPL96 Ulcerative colitis (UC) 26 42 GSE3365 GPL96 Crohn's disease (CD) 59 42

As described above, scores based on the lupus discrimination model were calculated in the same manner as described above for the collected PBMC gene expression data of three different autoimmune diseases, and the results of comparing them are shown in FIG. 7.

7 is a graph comparing disease prediction gene scores of three other autoimmune diseases compared to lupus disease.

Referring to FIG. 7, the horizontal axis (Class) is a category of systemic lupus crythematesus (SLE), rheumatoid arthritis (RA), ulcerative colitis (UC), and Crohn's disease (CD), which are autoimmune diseases in Table 4. Represents. Each color represents a legend for each disease (SLE, RA, UC, CD). The vertical axis (Sample.score) is the lupus discrimination score in each disease patient calculated based on the patient sample of each disease and the lupus optimal discrimination gene combination.

In FIG. 7, disease prediction gene scores calculated for samples of three autoimmune diseases were compared with disease prediction gene scores calculated for samples of lupus disease through a t-test.

As a result of comparing the p-values between each group, p=0.028 between SLE-RA, p<5.21E-15 between SLE-UC, and p<2.2E-16 between SLE-CD were derived.

As a result of comparison, it can be seen that for all three diseases, the disease predictive gene score in the lupus disease sample was significantly higher based on p<0.05. Therefore, when compared to other diseases, the lupus discrimination marker combination and the optimal disease discrimination prediction score according to an embodiment of the present invention indicate the possibility of a specific diagnosis for lupus disease.

As described above, for other autoimmune diseases, the lupus discrimination prediction score according to the embodiment of the present invention shows the performance for effective lupus-specific diagnosis, and when the criteria for distinguishing between diseases are newly devised, diagnosis between other diseases and lupus Showed the possibility of.

Meanwhile, FIG. 8 is a block diagram showing a hardware configuration of an apparatus for discovering a disease phenotype gene expression marker according to another embodiment of the present invention.

Referring to FIG. 8, the discovery device 100 includes a disease phenotype-related gene selection unit 110, a disease determination performance estimation unit 120, a disease determination gene combination selection unit 130, and a disease prediction described in FIGS. 1 to 7. The gene score calculation unit 140 and the disease determination unit 150 may execute a program including instructions described to execute the operation of the present invention in the computing device 200 operated by at least one processor. have.

The hardware of the computing device 200 may include at least one processor 201, a memory 203, a storage 205, and a communication interface 207, and may be connected through a bus. In addition, hardware such as an input device and an output device may be included.

The computing device 200 may be equipped with various software including an operating system capable of driving a program.

The processor 201 is a device that controls the operation of the computing device 200, and may be various types of processors that process instructions included in a program. For example, a CPU (Central Processing Unit) or a Micro Processor Unit (MPU) ), MCU (Micro Controller Unit), GPU (Graphic Processing Unit), etc. The memory 203 may load a corresponding program such that instructions described to perform the operation of the present invention are processed by the processor 201. The memory 203 may be, for example, read only memory (ROM), random access memory (RAM), or the like. The storage 205 may store various types of data and programs required to perform the operation of the present invention. The communication interface 207 may be a wired/wireless communication module.

The embodiments of the present invention described above are not implemented only through an apparatus and a method, but may be implemented through a program that realizes a function corresponding to the configuration of the embodiment of the present invention or a recording medium on which the program is recorded.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (12)

A method of determining a disease phenotype in a computing device operated by at least one processor,
Collecting gene expression data of peripheral blood mononuclear cells (PBMC) of patients and gene expression data of peripheral blood mononuclear cells of normal people from the microarray database,
Evaluating the similarity between the collected gene expression data, selecting genes that show a significant difference in expression between the gene expression data of the patient and the gene expression data of the normal person,
Identifying genes that are overexpressed and suppressed for disease states compared to normal states among selected genes,
Based on the expression values of the overexpressed genes and the expression values of the suppressed-expressed genes, determining a reference disease prediction gene score, and
If the disease prediction gene score calculated based on the gene expression value in the received random sample is higher than the reference disease prediction gene score, determining the random sample as a disease sample
Containing, disease phenotype determination method. In claim 1,
Between the step of selecting and the step of checking,
Further comprising the step of sequentially combining the selected genes to determine a disease identification marker combination,
The step of confirming,
A method for determining a disease phenotype to identify genes that are overexpressed and suppressed from among genes included in the disease discrimination marker combination. In paragraph 2,
The step of determining the disease discrimination marker combination,
If the disease discrimination performance of any gene combination increases compared to the previous gene combination, expanding the gene combination by adding another gene, and
If the disease discrimination performance does not increase compared to the previous gene combination, determining the corresponding gene combination as a disease discrimination marker combination
Containing, disease phenotype determination method. In paragraph 3,
The step of determining the disease discrimination marker combination,
Comprising the step of calculating disease discrimination performance for the arbitrary combination of genes,
The step of calculating the disease determination performance,
A disease phenotype discrimination method for calculating an arithmetic mean value of each discrimination accuracy for each disease discrimination model calculated through cross-validation as a disease discrimination performance of the random gene combination. In claim 1,
The determining step,
Calculating the difference between the geometric mean value of the patient's gene expression values for the overexpressed genes and the geometric mean value of the patient's gene expression values for the suppressed-expressed genes as the patient's disease prediction gene score,
Calculating a difference between a geometric mean value of gene expression values of a normal person for the overexpressed genes and a geometric mean value of gene expression values of a normal person for the suppressed-expressed genes as a disease prediction gene score of a healthy person,
Calculating the disease prediction gene score of the patient based on gene expression data of peripheral blood mononuclear cells of a plurality of patients, and selecting the lowest disease prediction gene score from among the calculated disease prediction gene scores,
Calculating the disease prediction gene score of the normal person based on gene expression data of peripheral blood mononuclear cells of a plurality of normal people, and selecting the highest disease prediction gene score from among the calculated disease prediction gene scores, and
Determining an average score of the lowest disease predicted gene score and the highest disease predicted gene score as the reference disease predicted gene score
Containing, disease phenotype determination method. In claim 1,
The determining step,
After matching the level of the expression distribution of the gene expression data in the arbitrary sample to the expression distribution of the gene expression data used to calculate the reference disease predicting gene score, the disease predicting gene calculated based on the gene expression value in the arbitrary sample Comparing the score and the reference disease prediction gene score, disease phenotype determination method. A method of determining a disease phenotype in a computing device operated by at least one processor,
Calculating a reference disease prediction gene score based on the gene expression data of the patient's peripheral blood mononuclear cells (PBMC) collected from the microarray database and the gene expression data of the peripheral blood mononuclear cells of the normal person,
Comparing the disease prediction gene score calculated based on the gene expression data in any received sample with the reference disease prediction gene score,
If the disease prediction gene score calculated for the random sample has a higher value than the reference disease prediction gene score, determining the random sample as a disease sample, and
If the disease prediction gene score calculated for the random sample has a value lower than the reference disease prediction gene score, determining the random sample as a normal sample
Containing, disease phenotype determination method. In clause 7,
The step of calculating the reference disease prediction gene score,
Among the gene expression data collected from the microarray database, gene expression values of normal and patient for overexpressed genes, and gene expression values of normal and patient for suppressed-expressed genes compared to normal human gene expression data On the basis of, calculate the reference disease prediction gene score,
The disease prediction gene score calculated for the random sample,
A method for determining a disease phenotype, which is calculated based on gene expression values for each of the overexpressed genes and the suppressed-expressed genes among the genes in the arbitrary sample. In clause 8,
The calculating step,
Evaluating the similarity between the gene expression data collected from the microarray database to select genes that show a significant difference in expression between the gene expression data of the patient and the gene expression data of the normal person, and
Calculating the reference disease prediction gene score for the selected genes
Containing, disease phenotype determination method. In claim 9,
After the step of selecting,
Selecting one of the selected genes as a starting gene,
Adding the remaining genes one by one to the starting gene to generate disease discriminating marker combination candidates, and
Further comprising estimating disease discrimination performance for each of the disease discrimination marker combination candidates,
The reference disease prediction gene score,
A method for determining disease phenotypes, which is calculated using a candidate disease determination marker combination having the highest value of the disease determination performance. In claim 10,
After the estimating step,
Determining a candidate primary disease discrimination marker combination having the highest value of the disease discrimination performance,
Generating secondary disease discrimination marker combination candidates by adding one by one genes that are not included in the primary disease discrimination marker combination candidate among the selected genes,
Selecting one secondary disease discrimination marker combination candidate having the highest disease discrimination performance among the secondary disease discrimination marker combination candidates,
When the selected secondary disease discrimination marker combination candidate has increased disease discrimination performance compared to the primary disease discrimination marker combination candidate, genes not included in the primary disease discrimination marker combination candidate among the selected genes are added one by one to 3 Generating primary disease discrimination marker combination candidates,
When the selected secondary disease discrimination marker combination candidate does not increase the disease discrimination performance compared to the primary disease discrimination marker combination candidate, the selected secondary disease discrimination marker combination candidate is determined as a target for calculating the reference disease prediction gene score. step
Further comprising a disease phenotype determination method. In clause 11,
The disease determination performance,
A plurality of disease discrimination models generated by selecting random genes from genes included in the disease discrimination marker combination candidate and separating them into training data and verification data, and learning the genes separated by the training data multiple times as the verification data A method for determining disease phenotype, which is calculated as the average value of the disease determination accuracy calculated through cross-validation after applying the separated genes.

Images