KR20190053695A - Breast cancer prognosis prediction method and system based on machine learning using next generation sequencing - Google Patents

Abstract

A method for predicting a breast cancer prognosis based on machine learning using next-generation sequencing comprises the steps of: measuring an expression level of a target gene using RNA sequencing data of a tissue of a subject, by a computer device; inputting the expression level of the target gene into an artificial neural network provided in advance, by the computer device; and estimating a breast cancer prognosis for the subject based on an output value of the artificial neural network, by the computer device, wherein the artificial neural network is provided in advance to have expression levels of a plurality of samples as an input value and to output a result according to a recurrence score of the Oncotype DX for the plurality of samples.

Description

TECHNICAL FIELD The present invention relates to a method and a system for predicting breast cancer prognosis based on machine learning using a next-generation base sequence analysis.

The technique described below relates to a technique for predicting breast cancer prognosis using gene expression data.

Various studies are under way to predict the prognosis of the tumor. For example, in the field of breast cancer, techniques for predicting breast cancer prognosis through genetic analysis have been developed. Oncotype DX ^® and Mammaprint ^® , which are used as the standard for predicting breast cancer prognosis, are all based on RT-PCR (Real-time PCR).

Korean Patent Publication No. 10-2012-0079295

RT-PCR-based assays have limitations that make it difficult to simultaneously analyze many genes in terms of cost and efficiency. The technique described below is to provide a technique for estimating the prognosis of breast cancer by analyzing the gene expression amount by the next-generation sequencing (NGS) technique.

A method for predicting breast cancer prognosis based on the next generation nucleotide sequence analysis comprises the steps of: measuring the expression level of a target gene using RNA sequencing data of a tissue of a subject, wherein the computer device measures the expression level of the target gene A step of inputting the input data into an artificial neural network provided in advance, and the computer device estimating a breast cancer prognosis for the subject based on the output value of the artificial neural network.

A machine learning-based breast cancer prognosis prediction system using next generation nucleotide sequence analysis includes a client apparatus for storing RNA gene data of a tissue of a subject and RNA sequencing data for NGS (next-generation sequencing) -based sequencing of the gene data And an analysis server for estimating a breast cancer prognosis for the subject based on an output value obtained by inputting the expression amount of the target gene into a neural network prepared in advance and measuring the amount of expression of the target gene .

The artificial neural network has a target gene expression amount of a plurality of samples as an input value and is provided in advance to output a result based on a recurrence score of Oncotype DX for the plurality of samples.

The technique described below allows NGS techniques to rapidly analyze target gene sequences at a lower cost than RT-PCR based techniques. The technique described below enables rapid and accurate prediction of breast cancer prognosis using a machine learning model that is learned by the amount of target gene expression and the recurrence score of Oncotype DX.

Figure 1 is an example of a flowchart for a machine learning-based breast cancer prognosis prediction method using next-generation sequencing.
Figure 2 is an example of a target gene.
Figure 3 is an example of the data quality of a target gene based on NGS.
4 is an example using gene expression data of a specific target exon region.
5 is an example of verifying the stability of a method using a specific target exon region.
6 is an example of a process of normalizing RNA sequence data.
Figure 7 is an example of an artificial neural network structure for machine learning.
FIG. 8 shows an example of a machine learning-based breast cancer prognosis prediction system using next-generation nucleotide sequence analysis.

The technique described below relates to a technique for predicting breast cancer prognosis using gene expression data. A machine learning model is used as a tool to predict breast cancer prognosis. The machine learning model is studied using the recurrent score (RS) of Oncotype DX ^{® and} Oncotype DX ^® . The Oncotype DX is briefly described first.

Oncotype DX is an analysis tool developed by Genomic Health. Oncotype DX is an assay that measures and analyzes the activity of 21 different genes in breast cancer tissues to determine the likelihood of recurrence of breast cancer and the effectiveness of chemotherapy. Oncotype DX calculates RS based on 16 genes and 5 reference genes. The formula for calculating RS is as follows. RS = + 0.47 x HER2 Group Score - 0.34 x ER (Estrogen) Group Score + 1.04 x Proliferation Group Score + 0.10 x Invasion Group Score + 0.05 x CD68 - 0.08 x GSTM1 - 0.07 x BAG1. Here, each item refers to a gene group according to functional division. The HER2 Group (2 genes), ER Group (4 genes), Proliferation Group (5 genes) and Invasion Group (2 genes) contain multiple genes. CD68, GSTM1, and BAG1 are each an individual gene. The RS score has a value ranging from 0 to 100 points. For example, if the RS score is low, the recurrence rate is low and the effect of chemotherapy is likely to be low. Conversely, if the RS score is high, recurrence rate is high and chemotherapy is likely to be effective.

As described above, Oncotype DX detects gene activity based on RT-PCR, but the technique described below uses the NGS technique. The techniques described below sequence RNA and determine gene expression levels based on the NGS technique. The NGS technique is also briefly described.

NGS technology has the ability to multiplex hundreds of thousands of responses simultaneously, allowing sequencing to even a small sample volume. NGS uses clonal amplification, massively parallel sequencing, and new nucleotide sequencing methods that differ from Sanger's method and mechanism of action, although the specific application technique is somewhat different according to the commercialized technology. Brief introduction of commercialization technology. In 2007, Roche launched the 454 GSO improved FLX model sequencer with 454 Cooperation. Illumina released Genome Analyzer HiSeq in 2006, and Applied Biosystems in 2007 released SOLiD in turn. All three platforms have adopted a massively parallel mass sequencing technique, which eliminates complex library construction and cloning processes, adopted clone amplification technology, and can process large amounts of data at once. By using cyclic sequencing, Sequencing by synthesis was used to determine the nucleotide sequence to exclude the complicated electrophoresis process. We also use a shotgun method to arrange short readings on a computer to find duplicates and complete the entire sequence.

The technique described below predicts the prognosis of breast cancer based on the amount of gene expression (RNA expression level) for a sample extracted from a specific tissue (breast tissue). Hereinafter, an example of a process of preparing a sample and extracting RNA from a specific operation will be described first.

Selection of breast cancer patients and preparation of test tissue

1) Select formalin-fixed paraffin-embedded (FFPE) blocks among the surgical tissues of hormone receptor positive and lymph node metastasis-negative breast tumors.

- Select pathologist's H & E staining slides to identify blocks. At this time, it is necessary to select the block in which the target tumor is definitely present. It is preferable that the area of the tumor in the cross section is the largest, and the necrosis in the tumor tissue is little or absent.

2) Prepare 10 non-dyed slides 10 μm thick.

RNA Extraction Protocol from FFPE Tissue

1) RNA Extraction Kit: One of the two commercial kits listed below can be used.

① Ambion RecoverAll ^™ Total Nucleic Acid Isolation Kit for FFPE

② QIAGEN RNeasy FFPE Kit

2) Preparation of Wash Solution

- Add 42 mL of 100 mL ethanol to Wash 1 -> Wash 1

- Wash 2/3 with 48 mL of 100 mL ethanol -> Wash 2/3

3) Deparaffin (preparation: tissue, 100% xylene, 100% ethanol, heating block 50 ° C, pipette, vortex mixer, centrifuge)

① Preparation of tissue: Prepare 4-8 pieces of 10 μm thick paraffin slices cut from paraffin block, total 40-80um. If the size of the tumor in the section is less than 40 mm2, then all 8 sections are used. Make sure that only the tumor is inside the slice, and place the prepared slice in a 1.5 mL tube.

② Add 1 mL of 100% xylenes into the tissue, mix with a vortex mixer and shortly centrifuge. Allow the paraffin to dissolve by placing it at 50 ° C for 3 minutes. (If it does not melt, repeat this process again)

③ Centrifuge at maximum speed for 2 minutes to make lumps. If not firmly attached, add 2 minutes of centrifugation. Discard the xylen without breaking the chunk.

④ Wash xylene

a. Add 1 mL of 100% ethanol to the sample and mix with a vortex mixer. (It becomes cloudy.)

b. Centrifuge at maximum speed for 1 minute at room temperature to make lumps.

c. Remove the ethanol while not losing the lumps.

d. Repeat steps a-c once.

e. After briefly centrifuging, remove as much of the remaining ethanol as possible without touching the lumps as much as possible.

⑤ Dry at room temperature for 15-45 minutes.

4) Proteolysis (Preparation: Heat block 50 ℃ & 80 ℃, Protease (protease) is removed from freezer and dissolved at room temperature)

Add 200 μl of Digestion Buffer and 4 μl of Protease to each sample. At this time, gently shake to mix well.

② Place the specimen in a heating block at 50 ° C (protease activation temperature) for at least 15 minutes until it becomes completely transparent

③ After that, place in the heating block for 15 minutes at 80 ℃ (protease deactivation temperature). At this time, we keep the time exactly.

④ If not dissolved, add 4 μl of protease and repeat the above procedure (2 & 3).

5) Nucleic Acid Separation (Preparation: Isolation Additive / Ethanol mixture and all other reagents)

① Isolation Additive / Ethanol mixture manufacturing

- Isolation Additive 240 μl + 100% Ethanol 500 μl = Total 790 μl

- Store in a 50 mL tube after manufacture.

(When preparing a large number of specimens, prepare 5% more than quantitative.)

② Mix the prepared Isolation Additive / Ethanol mixture in a tube containing 790 μl of each sample using a pipette.

③ Mixture filtration

a. Insert the filter cartridge into the tube provided in the kit.

b. Place 700 μl of the mixture prepared in step 2 on the filter and close the lid.

c. Centrifuge at 10,000 rpm for 30 seconds.

d. Discard the filtered solution and place the filter in the same tube.

e. If necessary (if the mixture is not sufficiently filtered), perform one more centrifugation to filter the mixture into the filter.

④ Wash 1

a. Add 700 μl of Wash 1 to the filter cartridge

b. Centrifuge at 10,000 rpm for 30 seconds.

c. Discard the filtered solution and place the filter in the same tube.

⑤ Wash 2/3

a. Add 500 μl of Wash 1 to the filter cartridge

b. Centrifuge at 10,000 rpm for 30 seconds.

c. Discard the filtered solution and place the filter in the same tube.

d. Centrifuge once more at 10,000 rpm to remove any remaining solution.

6) RNA isolation and purification (preparation: DNase (DNA degradation enzyme) and Nuclease (nucleolytic enzyme) are removed from the freezer and dissolved)

① RNA isolation

a. Preparation of DNase mixture: 6 占 퐇 of 10X DNase Buffer + 4 占 퐇 of DNase + 50 占 퐇 of Nuclease free water = 60 占 퐇

b. Add 60 μl of the DNase mixture to the center of each filter cartridge.

c. Close the lid and leave for 30 minutes at room temperature of 22-25 ° C

② Wash 1

a. Add 1 700 μl of Wash into the filter cartridge and leave it at room temperature for 30-60 seconds

b. Centrifuge at 10,000 rpm for 30 seconds.

c. Discard the filtered solution and place the filter in the same tube.

③ Wash 2/3

a. Wash 2/3 500 μl into the filter cartridge.

b. Centrifuge at 10,000 rpm for 30 seconds.

c. Discard the filtered solution and place the filter in the same tube.

d. Repeat a-c once more.

e. Centrifuge at 10,000 rpm for 1 minute.

④ Addition and Storage of Elution Solution

a. Place the filter cartridge in a new tube.

b. Add 60 μl Elution Solution to the center of the filter

c. Close the lid and leave for 1 minute.

d. Centrifuge at full speed for 1 minute, discard the filter, and store the filtered solution at -20 ° C or less.

Hereinafter, the process of predicting breast cancer prognosis using RNA extracted from a sample tissue is described. Figure 1 is an example of a flowchart for a machine learning-based breast cancer prognosis prediction method using next-generation sequencing. First, sequencing is performed on the RNA sample (110). RNA sequencing can be performed in a variety of ways. RNA can be sequenced using a variety of commercial kits and commercial solutions. One example is described.

Targeted RNA-sequencing

1) Remove the ribosomal RNA from the total RNA using KAPA Stranded RNA-Seq kit with RiboErase (KK8483, KAPABIOSYSTEMS) kit.

2) cDNA is prepared from mRNA and an additional process is performed to complete the cDNA NGS library. Solution-based hybridization capture is performed using cDNA library, hybridization solution, and target capture probe.

3) Library amplification is performed by amplifying a part of the obtained product.

4) The amount of RNA expression can be predicted based on the sequencing depth data of the target region generated by sequencing the final product (using Illumina's kit).

When NGS-based RNA sequencing is performed, the RNA sequence in the commercial program is generated as constant digital data. In addition, the commercial program can calculate the expression amount of each RNA gene using the sequencing result. Therefore, the process of analyzing the RNA sample and the analysis result are performed by a computer device. Therefore, it is presumed that the following computer apparatus performs prediction of breast cancer prognosis using RNA expression amount. The computer device generates 120 expression volume data for the sample RNA.

On the other hand, the analysis can be performed by selecting a gene (hereinafter referred to as a target gene) that is related to the prognosis of breast cancer without analyzing the entire gene of the sample RNA. The published gene data was used to determine the target gene. Genetic data from subjects who were positive for estrogen receptor and not metastasized to lymph nodes were used. Public data utilized are GSE2034, GSE2990, GSE3494, GSE4922, GSE6532, GSE7390 and GSE12093.

The correlation between each gene and Oncotype DX RS (recurrence score) was analyzed based on public data. Pearson and Spearman techniques, which are representative correlation analysis techniques, were used. The genes with an average correlation coefficient of 0.5 or more with Oncotype DX RS (recurrence score) in each open data were selected. 135 genes were selected as shown in the table below.

GSE2034 GSE2990 GSE3494 GSE4922 GSE6532 GSE7390 GSE12093 Total (> 0.5) Oncotype DX RS 184 298 267 292 168 77 56 135

In addition, 16 additional genes used in the Oncotype DX RS calculation were further screened. As a result, the target gene used 149 genes. Of course, the target gene may experimentally select other gene combinations. However, the target gene is determined to be highly correlated with Oncotype DX RS.

Figure 2 is an example of a target gene. Figure 2 shows all 149 genes. In Fig. 2, the shaded gene is a gene associated with a cell cycle regulating cell division. In Fig. 2, the gene indicated by the solid line circle is a gene related to a mechanism of regulating cell division involved in the p53 signal pathway. The gene indicated by the dotted circle in FIG. 2 is a gene that regulates the DNA replication process. The gene indicated by the solid line in Fig. 2 is a gene involved in the cell cycle and the p53 signaling pathway. In Fig. 2, the dotted rectangle is a gene involved in the cell cycle and DNA replication.

As described above, a computer device based on NGS measures the expression amount of a target gene. Figure 3 is an example of the data quality of a target gene based on NGS. Fig. 3 is an example of a graph showing the quality of the target RNA sequenced data. For the data quality verification, the deviation between the measured gene expression level and the value measured by total-transcript sequencing was confirmed. A total of 84 panel genes were subjected to target sequencing and total - transcript sequencing for 10 RNA samples, respectively, and expression levels were calculated and Pearson correlation coefficients were measured. As a result, a high correlation of 0.85 or more was confirmed. Therefore, the results of the target RNA sequencing used in the experiment are interpreted to have a data quality similar to that of the full - transcript RNA sequencing, since the target RNA sequencing and the total - transcript RNA sequencing results are highly correlated.

A specific exon region commonly expressed in each sample can be used without calculating the total gene expression amount for the sample. In this case, the process of calculating the gene expression amount can be performed more quickly. A conserved exon panel commonly used for the sample (hereinafter referred to as CE method)

4 is an example using gene expression data of a specific target exon region. 4 shows the expression state of the gene exons to a plurality of samples (A to D). A region (target region) that is commonly expressed for a plurality of samples can be determined, and the amount of gene expression can be determined based on the region. Since the difference in expression level of individual gene subtype transcripts is not considered in the method using the whole gene region (hereinafter referred to as WG method), there may be a deviation in the gene expression amount measurement due to the difference in the expression ratio of subtype transcripts in each patient. Since the CE method uses only the region shared by the subtype transcripts, measurement deviations occurring in the WG method do not occur, so that it is possible to measure the expression amount more stably. Stability refers to the degree to which the measured value changes during repeated measurements.

Furthermore, the stability of CE method was verified by using sample data. 5 is an example of verifying the stability of a method using a specific target exon region. FIG. 5A is an example for explaining a process for verifying the stability of the CE system. The replicate data is a repeated measurement for a single sample. Therefore, the error between the repeatedly measured data corresponds to the measurement error. A total of 18 target RNA sample sequencing data were generated (target RNA sequencing I and target RNA sequencing II) by performing 2 repeated iterations on each of 9 RNA samples extracted from tumor cell lines. 5 (B) is an experimental result showing the stability between the measured value of the WG system and the measured value of the CE system. Eight out of 9 samples appear to be more stable (Pearson factor is higher) in comparison between replicate data of CE method than WG method. The average Pearson coefficient for nine samples is also higher for the CE method than for the WG method, and the standard deviation between samples is also small. Therefore, it is interpreted that CE method is more stable than WG method.

Returning to the description of FIG. 1, the computer apparatus can regularly post-process and normalize the expression amount data of the generated target gene (130). First, an example of the post-processing that can be applied will be described. Postprocessing and normalization correspond to the process of processing digital data (file) regularly.

Targeted RNA-sequencing Post-treatment

1) Remove the read

Leads that do not meet the lead quality criteria are removed (e.g., the average quality is above 20 and the average quality is below 2 and the base is below 5%)

Use the Trimmomatic (0.33) program to remove the inserted index sequence during the sequencing process.

2) Align sequenced leads to the reference genome (Align sequenced reads to the reference genome)

 Use the STAR aligner program to locate the references based on the reference genome (hg19) of the sequenced leads and sort them together with the SortedByCoordinate option.

3) Gene expression amount calculation

Use the cufflinks program to calculate the amount of gene expression and transcript expression from the sorted lead information. The expression level can be calculated as the value of FPKM (Fragments Per Kilobase of exon per Million fragments mapped). The amount of expression calculated for each gene is generated in the gene.fpkm_tracking file, and the expression amount calculated for each transcript can be generated as isoforms.fpkm_tracking file.

In order to precisely align the sequence and measure the expression level, the preprocessing process removes the bad quality leads from the sequencing leads and inserts an index sequence (inserted in the sequencing process) that may remain at the end of each lead Can be removed. For pre-processed leads, use the STAR program to determine the position of each lead on the reference genome. The confirmed location information is generated in the BAM file format, and the BAM file can be used to calculate the expression level of each gene and transcript using the Cufflinks program.

The computer device may normalize the generated data (130).

Targeted RNA-sequencing Expression information Normalization

It is known that the "Trimmed Mean of M-value (TMM)" technique used in R package edge R (Robinson et al. Bioinformatics 2010) among the conventional normalization techniques is the most stable. The computer device is capable of designing a pipeline that automatically extracts normalized gene expression information from the target RNA sequencing data generated by mounting the edgeR package, a commercial package.

The sequencing data generated using NGS technology is mapped to the reference genome using the usage alignment software (e.g., RNA-STAR). The mapping results can be used to count the number of sequences from each gene, which is a direct estimate of the amount of gene expression.

The normalization pipeline receives the processed data in the BAM file format after the mapping is completed. The mapped data can be computed as a normalized expression value that can be compared between samples by a series of software packages HTseq-count and edgeR in the pipeline.

6 is an example of a process of normalizing RNA sequence data. 6 is an example of a process of normalizing data for two different samples in Fig. First, the sample data is input with mapped data representing the amount of gene expression. Computer devices use HTseq-count to calculate gene expression levels. Each sample has a different library size (Sample 1 is 100 sequences, Sample 2 is 300 sequences). The computer device then normalizes the expression level by applying edgeR. 6 is an example of normalizing the expression amount of a sample on the basis of a library size of 100. FIG.

The computer device inputs the gene expression amount data into a machine learning model prepared in advance (140). The gene expression data input to the machine learning model is normalized data. The machine learning model is learned in advance by the amount of target gene expression and Oncotype DX RS. For example, a machine learning model can be judged to be high risk (high likelihood of recurrence) when Oncotype DX RS is 25 or more, and low risk if less than 25.

Figure 7 is an example of an artificial neural network structure for machine learning. 7 is an example of an artificial neural network among machine learning models. An artificial neural network can utilize any one of a variety of types. The artificial neural network of Fig. 7 is a fully connected network. It uses a hidden node for each step. In order to prevent over-fitting of learning data, it is possible to apply batch normalization to each layer. Batch normalization is the process of forcibly distributing the activation values for the results that pass through the hierarchy.

As described in the machine learning model, artificial neural networks are also learned in advance by using data on a plurality of samples. In other words, an artificial neural network model is generated using Oncotype DX RS of a target gene (sample) using an input value of a target gene expression amount for a plurality of samples.

Now, the data of a particular subject can be entered into the neural network to predict the breast cancer prognosis of the subject. The value input to the previously learned artificial neural network is the expression amount of the target gene of the subject. The output value of artificial neural network may be the result of breast cancer prognosis prediction based on Oncotype DX RS. The output value of the artificial neural network can be output as a quantitative result that is equal to a specific score. Or the output value of an artificial neural network may be a qualitative evaluation result such as high risk or low risk.

Although FIG. 7 illustrates the artificial neural network as an example, various other machine learning models can be used to estimate the prognosis of breast cancer. For example, a machine learning model may utilize a variety of approaches such as genetic algorithms, support vector machines, Bayesian networks, and the like.

The computer device predicts (150) the breast cancer prognosis for the currently entered sample (subject) based on the results of the machine learning model. For example, the computer device may have a certain score in the result value of the machine learning model. In this case, the computer device compares the output value of the artificial neural network with a preset reference value, and if the output value is greater than or equal to a specific value, it can be determined that the current sample (subject) has a high risk. On the contrary, the computer device can judge that the output value of the artificial neural network is lower than the reference value.

FIG. 8 shows an example of a machine learning-based breast cancer prognosis prediction system using next-generation nucleotide sequence analysis. Figure 8 (A) is an example of a system 200 implemented in a network. The breast cancer prognosis prediction system 200 includes a client device 210 and an analysis server 220. Further, the breast cancer prognosis prediction system 200 may include a model DB 230. The analysis server 220 corresponds to the above-described computer apparatus.

The client device 210 is a device that provides data on a subject. The client device 210 transmits data (gene data) about gene expression of the subject to the analysis server 220. The data transmitted by the client device 210 may be data representing a gene sequence.

The analysis server 220 measures the expression level of the target gene using RNA sequencing data that has been subjected to next-generation sequencing (NGS) -based sequencing on the gene data, and determines the expression level of the target gene And estimates the breast cancer prognosis for the subject based on the output value. The analysis server 220 may perform the data post-processing and the normalization process described above. The analysis server 220 uses a machine learning model provided in advance. The machine learning model may be stored in a separate model DB 230. The process of analyzing data by the analysis server 220 and predicting the prognosis of breast cancer is as described above.

8 (B) is an example of a computer device 300 for predicting breast cancer prognosis. The computer device 300 shown in Fig. 8 (B) may be the analysis server 220 described above. The computer device 300 refers to a device such as a PC, a notebook, a smart device, or a server. The computer device 300 includes an input device 310, a computing device 320, a storage device 330, and an output device 340.

The input device 310 receives the gene data of the subject. The gene data refers to data or gene sequences related to the expression of the target gene. The input device 310 is a device for inputting subject data to the computer device 300 through communication or a separate storage device. Further, the input device 310 may be an interface device (a keyboard, a mouse, a touch screen, or the like) that receives clinical data of the subject directly through the computer device 300.

The storage device 330 is a device that stores the above-described machine learning model. The storage device 330 may store genetic data of the subject transmitted from the input device 310. The storage device 330 may be a device equipped with software for analyzing gene data to predict breast cancer prognosis.

The computing device 320 measures the amount of gene expression for the input gene data, inputs the gene expression amount into the machine learning model, and estimates the subject's breast cancer prognosis based on the output value of the machine learning model.

The output device 340 is a device that outputs information on the prognosis in a predetermined form. The output device 340 includes at least one of a display device, an output device for outputting a document, and a communication device for communicating information about the prognosis to another device.

Further, the machine learning-based breast cancer prognosis prediction method using the next generation nucleotide sequence analysis as described above can be implemented as a program (or application) including an executable algorithm that can be executed in a computer. The program may be stored and provided in a non-transitory computer readable medium.

A non-transitory readable medium is a medium that stores data for a short period of time, such as a register, cache, memory, etc., but semi-permanently stores data and is readable by the apparatus. In particular, the various applications or programs described above may be stored and provided on non-volatile readable media such as CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM,

The present embodiment and drawings attached hereto are only a part of the technical idea included in the above-described technology, and it is easy for a person skilled in the art to easily understand the technical idea included in the description of the above- It will be appreciated that variations that may be deduced and specific embodiments are included within the scope of the foregoing description.

200: Thyroid Cancer Prognosis Prediction System
210: Client device
220: Analysis server
230: Model DB
300: computer device
310: input device
320:
330: Storage device
340: Output device

Claims (12)

Measuring the expression level of a target gene using RNA sequencing data of a tissue of a subject;
Inputting an amount of expression of the target gene into an artificial neural network provided in advance; And
The computer device estimating a breast cancer prognosis for the subject based on an output value of the artificial neural network,
Wherein the artificial neural network has a target gene expression amount of a plurality of samples as an input value and outputs a result based on a recurrence score of oncotype DX for the plurality of samples, Based learning - based breast cancer prognostic method. The method according to claim 1,
Wherein the computer device is a machine learning-based breast cancer prognosis predicting method using a next-generation nucleotide sequence analysis that measures an expression level of the target gene using a next-generation sequencing (NGS) technique. The method according to claim 1,
The target gene includes 16 genes used in the onco type recurrence point calculation and an additional gene having a correlation value with the recurrence score of at least a reference value, wherein the additional gene is a sample in which ER (estrogen) receptor is positive and lymph node metastasis is absent A method for predicting breast cancer prognosis based on machine learning using next generation nucleotide sequence analysis by performing correlation analysis with the recurrence score for each of the genes and selecting the genes with correlation coefficient higher than the reference value. The method according to claim 1,
Wherein the RNA sequencing data comprises a conserved exon region that is commonly expressed in a plurality of sample data. The method according to claim 1,
Wherein the computer apparatus normalizes the expression level of the target gene using a trimmed mean of M-value (TMM) technique and inputs a normalized expression amount to the artificial neural network. The method according to claim 1,
The method of predicting breast cancer prognosis based on machine learning using next generation nucleotide sequence analysis, wherein the artificial neural network includes three fully hidden hidden layers and batch normalization is applied to each layer. The method according to claim 1,
The computer device
A method for predicting breast cancer prognosis based on machine learning using next generation nucleotide sequence analysis, wherein the subject is at a high risk and the prognosis is estimated when the output value is above a reference value and the subject is at low risk when the output value is below a reference value. A computer-readable recording medium on which a program for executing a machine learning-based breast cancer prognostic prediction method using the next generation nucleotide sequence analysis according to any one of claims 1 to 7 is recorded on a computer. A client device for storing RNA gene data of a subject tissue; And
The amount of expression of the target gene is measured using RNA sequencing data that has been subjected to next-generation sequencing (NGS) based on the gene data, and the expression amount of the target gene is input to a neural network provided in advance And an analysis server for estimating a breast cancer prognosis for the subject based on an output value output from the analysis server,
Wherein the artificial neural network has a target gene expression amount of a plurality of samples as an input value and outputs a result based on a recurrence score of oncotype DX for the plurality of samples, Machine Learning Based Breast Cancer Prognosis Prediction System Using. 10. The method of claim 9,
The target gene is a next-generation sequencing analysis that includes genes having a correlation coefficient of at least a reference value by performing Correlation Analysis with the recurrence score for each gene of a sample having positive ER (estrogen) receptor and having no lymph node metastasis Machine learning based breast cancer prognosis prediction system. 10. The method of claim 9,
Wherein said RNA sequencing data is a machine learning based breast cancer prognostic prediction system using next generation sequencing comprising a conserved exon region commonly expressed in a plurality of sample data. 10. The method of claim 9,
The analysis server may include a machine learning-based breast cancer prognostic prediction using a next generation nucleotide sequence analysis that normalizes the expression level of the target gene using a trimmed mean of M-value (TMM) technique and inputs a normalized expression amount into the artificial neural network system.

Images