KR100839221B1 - Method for integrated analysis of microarray - Google Patents

Abstract

A microarray integrated analysis method is disclosed.

According to the present invention, when the experimental design is selected among the above items for an item of the upper menu including experimental design, standardization, estimation and test, cluster analysis, and classification, the experimental design of any one of dye exchange, reference design, or loop design is performed. Displaying an interface for inputting a selection interface and a design parameter to be used for the selected experimental design, and generating an annova design matrix for the selected experimental design using the input design parameters, wherein the standardization is selected from the items When the standardization result is generated by sequentially performing a single slide standardization, a single batch standardization, and a multiple slide standardization using the input slide information, when the estimation and the test are selected among the items, By performing estimation and testing methods Selecting a significant gene in a microarray experiment; if the cluster analysis is selected among the items, performing a predetermined cluster analysis using the standardization result and displaying a cluster analysis result generated by the graph and among the items When the classification is selected, candidate genes are selected by using the standardization result and the ratio between the variation between treatment groups and the variation within the treatment group, and a predetermined classification is performed on the candidate genes to be interested in the design parameters. Predicting the group, and outputting a misclassification rate.

According to the present invention, by processing the entire process of analyzing the microarray data in one integrated system, it is possible to perform a systematic statistical analysis of the microarray scanning image data by sharing the database, the optimal statistics according to the situation of each experiment By applying the analytical method, the false positive and false negative error rates due to the inappropriate method can be minimized and the reliability of the research results can be increased, and the analysis is easy by providing a convenient and user-friendly interface.

Description

{Method for integrated analysis of microarray}

1A and 1B are conceptual diagrams of experimental design methods applied to the present invention.

2 is a flowchart of a microarray integrated analysis method according to the present invention.

FIG. 3 illustrates an example of a screen configuration of an experimental design that is a subprogram for FIG. 2.

4A to 4F illustrate an example of a screen configuration of standardization, which is a subprogram for FIG. 2.

5A to 5N illustrate an example of a screen configuration of an option input of estimation and test, which is a subprogram for FIG. 2, and an example of an execution result graph.

6A to 6O illustrate an example of a screen configuration of an option input of cluster analysis, which is a subprogram for FIG. 2, and an example of an execution result graph.

7A to 7C illustrate an example of a screen configuration of an option input of classification, which is a subprogram for FIG. 2, and an example of an execution result graph.

The present invention relates to a DNA chip, and more particularly, to a microarray integrated analysis method.

As with other biological experiments, microarray experiments should have a design priority, which depends not only on the purpose of the experiment but also on the analytical method. The purpose of the experiment could be to search for genes with significant differences between the two groups, and to develop a biomarker for population selection on the basis of the four groups of the Small Round Blue Cell Tumor. The purpose may be to determine the type. It can also be used in research or diagnostics to broadly analyze differences in gene expression using microarrays. Alternatively, the purpose of the experiment may be to determine the interaction size of three factors of age, sex, and gene type in Drosophila transcripts, or to identify variation of gene expression between and within a certain species. Searching for genes that respond when two anticancer drugs are administered at the same time may be the purpose of the study. Of course, the design of the experiment and the method of analysis vary depending on the purpose of the study.

Due to the nature of the experiment, the microarray goes through several steps, and there is room for experiment error at every step. Accumulated errors at various stages are one of the factors that make statistical analysis difficult, but even more embarrassing to statisticians is that only a few of the thousands of genes have to be selected. Is composed of only a few dozen so-called 'big p, small n' problem. Where p is the number of explanatory variables, n is the number of observations, and traditionally, statisticians have mainly dealt with 'large n, small p' data, but microarray data make up 'large p, small n' data. Doing. Thus, statistical analysis of microarray experimental data presents new challenges for statisticians.

Conventional microarray experimental analysis method requires a separate program for each analysis method, data compatibility is difficult for each program, the analysis data can not be managed systematically, false positives due to inappropriate application of the method and There is a problem that cannot increase the false negative error rate.

Therefore, the technical task of the present invention is to share a database to perform systematic statistical analysis on the microarray scanning image data, to minimize false positive and false negative error rates due to the application of inappropriate methods and to increase the reliability of the results. In addition, it is possible to process the entire process of microarray data analysis in one integrated system to provide an easy microarray integrated analysis method.

In order to achieve the above technical problem, the present invention, if the experimental design is selected among the items of the upper menu consisting of experimental design, standardization, estimation and testing, cluster analysis and classification, dye exchange, reference design or loop Displays an interface for selecting an experimental design of any one of the designs and an interface for inputting design parameters to be used for the selected experimental design, and generates an annova design matrix for the selected experimental design using the input design parameters And generating the standardization result by sequentially performing a single slide standardization, a single batch standardization, and a multiple slide standardization using the input slide information when the standardization is selected among the items. If the standardization result Selecting a significant gene in a microarray experiment by performing a predetermined estimation and testing method; when the cluster analysis is selected among the items, a cluster analysis result generated by performing a predetermined cluster analysis using the standardization result Displaying a graph and selecting the classification among the items, selecting a candidate gene using the standardization result and the ratio between the variation between treatment groups and the variation within the treatment group, and selecting a predetermined classification for the candidate gene. It provides a microarray integrated analysis method comprising the step of predicting the target group of interest by the design parameters, and outputting a misclassification rate.

The microarray integrated analysis method according to the present invention relates to software for statistically analyzing image data generated by performing block indexing and spot indexing on a microarray scanning image. Microarray integrated analysis method according to the present invention is largely divided into five areas. That is, experimental design, standardization, estimation and testing, cluster analysis, and classification.

First, the experimental design provides dye exchange, reference design, loop design, and analytical methods.

Standardization provides single slide standardization, dye exchange standardization, multiple slide standardization and single slide batch method. Single slide normalization is possible with average standardization, intensity dependent standardization, print tip standardization, and scale normalization method, and multiple slide normalization is possible with multiple scale normalization and quantile normalization. Estimation and testing provide methods for estimating and testing significant genes. Cluster analysis is a method of classifying significant genes into groups. Finally, classification is possible with classical classification methods (DLDA, DQDA) and Tree method.

Hereinafter, with reference to the drawings will be described a preferred embodiment of the present invention. However, embodiments of the present invention illustrated below may be modified in various other forms, and the scope of the present invention is not limited to the embodiments described below.

The following is a description of the experimental design. Dye deflection constitutes one of several reasons for standardization. However, because standardization does not completely eliminate dye deflection for every spot on every slide, the experimenter recommends dye replacement experiments. Common designs used in cancer research include (common) reference design, block design, and loop design. Reference design is synonymous with the common reference design described above and uses the same internal reference sample for each array to control the possible variations between spots. This is represented as a diagram in FIG. 1A. The simple design of the ring design is as follows in FIG. Microarray experiments can be conducted with a loop design to compare the magnitude of longitudinal and vertical variation in fish.

2 is a flowchart of a microarray integrated analysis method according to the present invention.

The microarray integrated analysis method according to the present invention is assumed to be implemented on a network system including a computer system capable of processing a large database, a server and a client, or a system having similar processing power.

First, the items of the upper menu composed of experiment design, standardization, estimation and test, cluster analysis, and classification are output to the screen (step 210).

If an experimental design is selected among the items (step 221), an interface to select one of the dye designs, the reference design or the loop design, and the design to be used for the selected experimental design The interface for inputting parameters is displayed to designate design parameters according to the user's parameter input (step 231). Next, an ANOVA design matrix for the selected experimental design is generated using the input design parameters (step 232).

On the other hand, if standardization is selected among the items (step 222), the slide information is input, and standardization results are generated by sequentially performing single slide standardization, single batch standardization, and multiple slide standardization using the input slide information (step 242). . The generated standardized results are stored in a file so that they can be read in the process of estimation, test, cluster analysis and classification (step 243).

On the other hand, if estimation and testing are selected among the items (step 223), the standardization result is read (step 251). Next, with respect to the read standardization result, a predetermined estimation and assay method is performed to select a significant gene in a microarray experiment (step 252). In this case, the probability of erroneous discovery according to the selected gene may be calculated according to the analysis method performed. At this time, the predetermined estimation methods are Newton's method, B-statistic method, Probability of adjustment (ADP), Significant Analysis of Microarray (SAM) and EBIRical Bayes Analysis. of Microarray (EBAM)). Next, data such as estimated genetic information and probability of erroneous discovery (if computed) are stored in a database for reference in another analysis process (step 253).

On the other hand, if the cluster analysis of the item is selected (step 224), the standardization result is read (step 261). Next, a predetermined cluster analysis is performed using the read standardized result to display the generated cluster analysis result as a graph (step 262). At this time, the predetermined cluster analysis includes hierarchical clustering, k-means clustering, principal component analysis (PCA), self-organizing map (SOM), and true Geneshaving and the like. Next, the cluster analysis result is stored in a database for reference in another analysis process (step 263).

On the other hand, if the classification is selected among the items, the standardization result is read (step 271).

Next, the ratio between the variation between treatment groups and the variation within the treatment group is calculated using the read standardized result, and the candidate gene is selected using the calculated ratio of the variation (step 272). In addition, a predetermined classification is performed on the selected candidate genes to predict a target group of interest based on the design parameter, and the misclassification rate is output according to the selected parameter (step 272). In particular, the group classified by the genes significant for each treatment group, that is, the group of interest and the misclassification rate are output (step 272). In this case, the interested group is a group in which the disease to be found as a result of the experiment is expressed among the test sets used in the experiment, and information for identifying which group having a gene is associated with a specific disease such as liver cancer. In this case, the predetermined classification includes a classical classification method (DLDA, DQDA) and a tree method (or a decision tree). Next, the group information and misclassification rate in which significant genes are classified are stored in a database for reference in another analysis process (step 273).

Finally, it is determined whether all the analysis is completed (step 280). This process (280) may include a process of outputting a message requesting the user to select whether to terminate all processes or to perform another analysis in succession, and receiving a selection result from the user. At this time, if the user selects the end of all the analysis is terminated all processes, otherwise repeat the above steps (210-280).

The following describes ANOVA analysis software for dye-swap, reference and loop experimental data.

The input data format is configured as shown in Table 1 below.

Heat Variable name Remarks col 1 gene id Character string without space to distinguish gene col 2 R Expression of treatment stained with red dye (Cy5) log (foreground red-background red) col 3 G Expression of treatment stained with green dye (Cy3) log (foreground green-background green) col 4 flag The gene where the error occurred is 1, otherwise 0

The output data format is configured as shown in Table 2 below.

Heat Variable name Data type Remarks col 1 Gene integer Give each gene its own number col 2 intensity double Merge R and G values into one column col 3 Array integer 1 to array number col 4 Dye integer 1,2 col 5 Variety integer 1 to Variety

In FIG. 3, Experimental Design is selected from the left menu and one of dye-swap, reference, and loop is selected from the submenu that appears.

At this time, Experiment Type selects experiment design, Array file is Array file list, Number of Arrays is the number of arrays to be used for analysis, Array Replication is repeat between arrays, Multiple Spotting is repeat within array, RNA Species List is processed ( Variety) List, Array information is input processing information for each array, Name of the ANOVA Design Matrix is file name to be saved, Generate a Design Matrix is Data set generation button, Test sample is selection of processing to compare, Boostrap Replications is Bootstrap repetition The number of operations, Significance Level is the significance level, and RUN is the ANOVA run button.

First, the experimental design is selected from dye-swap, reference design, and loop design.

Array file is an array file list. Each file is an array data (gene id, red intensity, green intensity, and flag). Array file is a list display function.

Next, enter the number of arrays to be used for analysis in Number of Arrays. Only part of the tested array can be used, so enter the number of arrays to use for analysis. Array Replication also checks if there are inter-array repetitions in the experiment, such as repeating a dye swap experiment twice. Multiple Spotting checks if a gene is repeatedly dipped the same number of times (two or more times) during chip fabrication. In the RNA Species List, enter the treatments used in the experiment. For example, if you see the difference in expression over time, enter as time0, time1, time2, time3, etc., and if you see the difference according to the drug dosage, enter as 0mg, 10mg, 20mg, 30mg.

Name of the ANOVA Design Matrix is the file name where the design matrix will be stored.

Generate a Design Matrix is a button for generating a data set, an Nova design matrix. Enter all the experiment information and select it to create a dataset.

The test sample is a button to select the processing to be compared, the Boostrap Replications is the number of bootstrap repetition operations, and the Significance Level represents a significance level.

When the dye swap experiment is repeated three times in order to reduce the error of the experiment, a total of six arrays are generated as shown in Table 1c.

Red (Cy5) Green (Cy3) Array1 Variety2 Variety1 Array2 Variety2 Variety1 Array3 Variety1 Variety2 Array4 Variety2 Variety1 Array5 Variety1 Variety2 Array6 Variety1 Variety2

The following is a description of standardization.

Standardization is a process to remove systematic variation prior to performing statistical analysis after microarray experiment. The standardization method implemented in the system is as follows.

,

로 하는 것보다

로 변환하여 하는 장점은 절대적 강도는 마이크로어레이의 고유한 스팟-스팟(spot-spot)변동에 의해서 교락(confound)될 수 있기 때문에 여러 슬라이드에서 절대적 강도보다

으로 분석하는 것이 더 안정적이기 때문이다. 그래서 표준화도 각 절대적 강도의 로그비를 보정하는데 중점을 두는 것이다.Normalization in cDNA microarray experiments is used to correct for the ratios of Cy5 and Cy3. For cDNA microarray analysis, each

,

Than to do it

The advantage of converting to is that absolute strength can be confounded by the unique spot-spot variation of the microarray.

This is because the analysis is more stable. So standardization also focuses on correcting the logarithmic ratio of each absolute intensity.

Single slide normailzation is one of the first steps in analyzing a single microarray slide, and it is a step in advance to determine which of the methods is best used in a single batch for later use. In other words, the ratio of Cy3 and Cy5 in one slide.

] [

]이다.The input file format is [gene_id] [Cy5] [Cy5 Background] [Cy3] [Cy3 Background], where gene_id consists of letters and numbers, Cy5, Cy5 Background, Cy3, and Cy3 Background consists of numbers, and data is tabbed. Should be separated. In this case, the result file format is [geneid] [

] [

]to be.

If one of the analysis targets that appears when a single slide menu is selected is selected, a screen as shown in FIG. 4A is displayed. The screen consists of a Base Info table and a 'Statistical Analysis' table that contains information for analysis.

Analysis starts by selecting 'Execute' in the 'Standardization Input File' item of the 'Statistical Analysis' table. To perform the following analysis method, the software can be configured to execute Java Web start.

4B is a screen for inputting options for analysis. A single slide supports four standardization methods: Mean, Intensity, Print-tip, and Scale. Choose one of these methods and choose 'ok' below to begin the analysis. At this time, if you input Print-tip or scale among the analysis methods, you can configure the software to activate the 'Number of Tip' item and 'Number of genes in block' option.

Global standardization assumes that the strength of Cy5 and that of Cy3 are constant multiples. Intensity dependent normalization is a case where it is assumed that there is an A-dependent relationship between the strength of Cy5 and the strength of Cy3. Print-tip normalization is a case where there is a print-tip-group effect. Scale normalization is a method of correcting variance between print tip groups after normalizing print tips.

4C is a screen illustrating an analysis result according to the option of FIG. 4B. This result is a standardized image of the inputted data. The screen displays the distribution of each gene and the lowess line, shown by a yellow line.

Dye exchange during normalization is a method of comparing two experimental results by comparing only Cy3 and Cy5 under the same conditions. Dye-Swap normalizatoin is a method of standardization in dye swapped experiments.

] [

]의 형식이다.As an input file, dye exchange will receive two files of the form: These two files are the result of experimenting with only Cy3 and Cy5 reversed under the same conditions. The input file format is [gene_id] [Cy5] [Cy5 Background] [Cy3] [Cy3 Background], where gene_id must be a letter or number Cy5, Cy5 Background, Cy3, Cy3 Background numbers or data separated by tabs. The analysis file is [geneid] [

] [

].

4D is a single dyeswap analysis scene. The dye exchange analysis can be configured to select the two slides to be used for analysis among the slides displayed on the screen, and select the execute button, that is, 'execute' to generate a result file.

Preferably, the software can be configured to be selected by selecting the slide on the left side and then selecting the '->' arrow in the middle, and selecting '<-' after the slide on the right side.

A single batch of standardization is a subprogram that standardizes multiple slides in the same way, based on the results analyzed in a single slide. The resulting file is then automatically transferred to the list of multiple slide analysis methods. Single slide batch standardization is the ability to process all slides at once using the same standardization method.

] [

] ....이다. 여기서 M 값은 선택한 슬라이드 들의 M값이며 각 슬라이드에 따른 M값이 연속으로 나타나게 된다.The input file format is the same as a single batch dye exchange standardization, where multiple input files are input. The analysis result file is shown in the following format. That is, [geneid] [

] [

] ....to be. The M value is the M value of the selected slides, and the M value of each slide is displayed continuously.

4E is a single batch analysis screen. Single batch assay screens have an interface similar to single dye exchange. However, there is no limit to the number of slides that can be selected. Simply select the slide you want to apply and run the program.

A single batch should specify the standardized analysis method. This is the same method used for a single slide. At this time, global standardization is a case where the strength of Cy5 and the strength of Cy3 are assumed to be a constant multiple. Intensity dependent normalization is a case where it is assumed that there is an A-dependent relationship between the strength of Cy5 and the strength of Cy3. Print-tip normalization is a case where there is a print-tip-group effect. Immediately above the slide selection portion of the screen, when selecting a print tip, a portion for inputting additional information appears on the right side of the selection portion as shown in FIG. 4E.

During normalization, multiple slides are a way of creating practical data for later cluster analysis, estimation, testing, and classification. It works with the resulting file from the previous single batch. Multiple slide normalization is a method of matching the variance or displacement of a slide in a repeated slide.

The input file format receives the file resulting from a single batch. The analysis result file is also identical to the output format of a single batch.

4F is a multiple slide analysis screen. Select one of Multiscale and Quantile on the screen and select 'execute' to execute.

There are two analysis options. First, multi-scale normalization is a way to adjust the variance of each slide. Rank normalization is a method of normalizing by adjusting the displacement value of each slide. If you choose 'execute', the result file is created.

Multiple slides are a way in which the order of the slides affects the analysis results. Therefore, the user can directly change the information of the slide in the material. This allows you to select 'modify' and activate an editing program such as Excel to edit and save.

The following is an explanation of estimation and testing.

In order to estimate and test, the direction of analysis should be set. This part is divided into two parts. First, it is the estimation part which statistics are used to rank genes for expression, and the second is the test part that determines the level of confidence in the genes found by giving the corresponding statistics.

The B statistic (the Bayesian B statistic by Ronnstedt et al.) Is a statistic that estimates post-log-odds for each gene that is expressed differently. Like the S statistic, the B statistic is a method in which t statistics are given a certain penalty. In particular, significant penalties are applied to data with high t-values. The log-odd rank via the B statistic can be obtained with the bayesian program.

You can rank all possible statistics using the traditional t-statistic with the ADP program, the S-statistic with the SAM and EBAM, and the B-statistic with the Bayesian program. In particular, the user inputs a representative number of ratios of S statistics in EBAM to calculate the ranking.

The adjusted p-value is: The family-wise error rate refers to the probability that at least one of the differently expressed genes will be false, regardless of which composition of the gene is actually expressed differently. There are quite a few versions of the family-wise error rate estimation method, and the most conservative method is the probability of adjustment. Probability of adjusted oil can be obtained from the ADP program.

FDR is significantly differently expressed and is defined as the expected value of the error rate among the selected genes and means "the rate of misdetection." Tusher's SAM method calculates the expected order of each gene by random permutation and estimates the FDR based on the difference from the actual gene. FDR can be obtained from the SAM program.

You can estimate FDR through EBAM in much the same way as SAM. However, since there is a difference in the S statistic used, it induces an FDR result different from SAM. In this method, the FDR estimate is usually set at a posterior probability of 0.9 and the investigator may choose to empirically reverse the posterior probability selection, taking into account the number of FDRs and rejected genes. FDRs for the various S statistics can be obtained from the EBAM program.

Newton's method of estimation and testing is a subprogram that estimates significant genes in single slide microarray experiments. The input file format is as follows: [gene name] [A (_)] [M (_)].

The analysis file contains a list of genes whose log posterior odds are greater than zero, a list of genes whose log posterior odds are less than or equal to zero, a log posterior odds value for plotting contours using maximum and minimum values, and a Chen graph. (95%, 99%) Contains four intercept values.

In the execution result graph of FIG. 5A, the black contours represent contours with changes in posterior odds of 1: 1, 10: 1, and 100: 1, and the light blue lines represent 95% and 99% confidence intervals of the Chen method. The red dots represent genes with log posterior odds greater than zero, and the green dots represent genes with log posterior odds less than or equal to zero.

B-statistic during estimation and testing is a subprogram that estimates significant genes in repeated microarray experiments. The data format of the input file is tab separated as shown in Table 2 below.

 gene1 1.269 0.277 -0.343 0.659 gene2 0.606 0.838 -0.405 1.241 gene3 0.852 0.627 1.918 -1.077 gene4 -0.065 -0.570 -0.446 0.364 ...

The analysis result includes the mean M (_) value of each gene and the log posterior odds value of each gene.

In the option input screen of FIG. 5B, Data is in * .txt format, Number of rows is the number of genes, Number of replication is the number of slides, and P is the probability that any gene is significantly expressed.

5C is an example of a result graph according to the option of FIG. 5B.

At this time, red represents a significant gene (gene whose log posterior odds is greater than zero), and green represents an insignificant gene (gene whose log posterior odds is less than zero).

ADP is a subprogram that calculates the probability of significant genetic adjustment in four repeated microarray experiments. Input file formats are separated by tabs or spaces, as in the case of B-statistics.

The analysis results include 10 types of adjusted probability and 3 types of significant probability and provide various information related to ADP implementation.

In the option input screen of 5d, open data is in text format, response type represents four types of experiments (1 group, 2 groups, 3 groups or more, response format), and Number of permutations is random permutation, Number of bootstraps is the bootstrap repeat count, Seed Numbers is the initial value to generate random random numbers, Number of groups is the number of iterations of each group, Quantitative response file is the response format specification file for experiments with response format, p-value The probability of 10 types of adjustment is shown, the Run Result Capture shows the execution process, the Raw p-value shows the significance of random permutation and bootstrap, and the Tmp Outfile Capture shows the output of the Tmp file.

5E is an example of an execution result graph according to the option of FIG. 5D. In this case, blue represents a raw p-value and red represents an adjusted p-value.

In FIG. 5E, the software can be configured to view a screen as shown in FIG. 5F by searching for a significant gene with a mouse.

In estimation and testing, SAM is a subprogram that searches for significant genes and computes FDR in four repeated microarray experiments. Input file formats are separated by tabs or spaces, as in the case of B-statistics.

The analysis results provide significant gene number according to Delta value and FDR based on three kinds of selection criteria, provide various information related to execution of SAM and initial Delta value, and SAM) information, adjusted SAM diagram information for different Delta values, significant gene lists for final Delta values, and Q-values.

In the option input screen of FIG. 5G, Import data is data in text format, Response type is 4 types of experiment types (1 group, 2 groups, 3 or more groups, response format), Number of permutations is random permutation, and Seed Numbers is the initial value for generating random random numbers, Number of groups is the number of repetitions of each group, Quantitative response file is the response format file for the experiment with response format, Gene Observed score is the drawing of Gene and Observed score, Gene Expected The score plots the Gene and Expected scores, the Gene plot plots the denominator (standard error) and the Expected score of the SAM statistics, and the SAM plot plots the Observed and Expected scores. FDR plot provides Mean FDR, Median FDR, 90th percentile FDR plot, and FDR table list presents FDR result file.

5H is a screen illustrating an FDR calculated according to the option of FIG. 5G. Delta can be determined by looking at the FDR of FIG. 5H.

5I illustrates an example of a SAM plot according to the option of FIG. 5G. At this time, the Delta value may be selected through the input delta value. FIG. 5J illustrates an example of a SAM plot according to Delta input according to FIG. 5H or 5I. In this case, the red part represents a positively significant gene and the blue part represents a negatively significant gene.

Among the estimation and testing, EBAM is a subprogram that calculates significant genes and calculates FDR using Empirical Bayesian method in four repeated microarray experiments. Input file formats are separated by tabs or spaces, as in the case of B-statistics.

The analysis results provide a significant number of genes according to the posterior probabilities and FDRs based on three selection criteria, provide various information and initial posterior probabilities related to the execution of EBAM, and EBAM) provides information on drawings, EBAM drawings for other posterior probabilities, posterior probabilities and spline basis values by logit regression analysis, and provides significant gene lists and Q-values for final posterior probabilities. do.

In the option input screen of FIG. 5K, the data in the import data is in text format, the response type is 4 types of experiment types (1 group, 2 groups, 3 groups or more, response format), and the spline df inputs the degrees of freedom of the spline (usually 5), Furge Factor Percentile is the position value of the standard error (value of 0 ~ 1) to correct the standard error, Number of permutations is the random number of iterations, Seed is the initial value that generates random random numbers, and Number of groups is each group. The number of iterations of the input, the quantitative response file is the response format specification file for the experiment with the response format, the Standard Z score is the Gene and the uncorrected Z score plot, the Expression score is the Gene and the corrected Z score plot, and the Numerator: Denominator is The denominator (standard error) and molecular plot of the EBAM statistic, the natural spline LOGIT prob .. is the LOGIT result plot, and the EBAM plot is the plot between the Expected score and the posterior probability. FDR plot provides Mean FDR, Median FDR, 90th percentile FDR plot, and FDR table list presents FDR result file.

5L is an output screen of the FDR calculated according to the option of FIG. 5K. At this time, the post-probability may be determined by looking at the output FDR.

5M is an example of an EBAM) plot according to the option of FIG. 5K. In this case, the posterior probability value may be selected through the posterior probability setting. At this time, the red line shows the EBAM plot, the blue line shows the unadjusted EBAM plot, and the green part shows the gene's expression score.

When the post probability is input according to FIG. 5L or 5M, the software may be configured to output an EBAM plot according to the input post probability.

FIG. 5N is an example of a significant gene list according to the posterior probability input according to FIG. 5L or 5M.

The following is a description of cluster analysis.

There are two types of hierarchical clustering: the merging method and the partitioning method. The merging method is to look at all the objects in each cluster and then gather similar clusters. The partitioning method is to tie all the objects together into one cluster and then split the group. At this time, the similarity of the objects is represented by a tree structure drawing.

K-means clustering is the allocation of objects to a predetermined number of clusters (k), which defines the initial values of the clusters and assigns each individual to the determined initial cluster until it is optimized according to the specified criteria. After reallocation, the final cluster is determined. We assign objects based on the center of the cluster. We can use mean (k-means) or median (k-medoids) as the center.

Self Organizing Map (SOM) is a type of neural network model similar to k-means clustering, but the positional neighboring groups in the resulting group of genes show a similar pattern than the separated groups. For the analysis, the user has to decide in advance the node type and the number of grid matrices. Preferably, a rectangular and hexagonal node can be defined.

Principal Component Analysis (PCA) transforms multidimensional data that is complex in structure and correlated with each other to produce independent artificial variables called principal components, which can explain many of them with several significant principal components. This is suitable for simplifying or summarizing data, searching for outliers or clusters. Principal components can be computed from correlation or covariance matrices.

Geneshaving method calculates "Super gene" as the first principal component and then cuts the least relevant genes (lower 10%) until only the last gene remains. This procedure produces a chain of genetic blocks, where each cluster is configured to maximize variance between clusters and minimize variance within clusters. When using this method, genes can be assigned to one or more clusters rather than just one cluster.

Selecting the hierarchical clustering menu performs a hierarchical clustering of genes and samples and provides a dendrogram (tree) as a result.

The data to be used for analysis should be in the form of a tab-delimited text file (* .txt). As shown in the example below, each gene is input in the row direction, and experiments (samples and arrays) are input in the column direction. The first column contains the name of the gene and the first row contains the name of the experiment (sample, array).

In FIG. 6A, uploaded data is designated as Import data.

The linkage method selects a measure that defines the distance between clusters. In this case, as a measure, the average linkage defines an average of distances between all possible cases by taking one object in each cluster as the distance between two clusters. Complete linkage defines the longest distance as the distance between two clusters among the distances between two entities in each cluster. Single linkage defines the shortest distance between two clusters within two clusters.

Similarity measures measure the similarity between two entities. correlation (uncentered) uses Pearson's correlation coefficient. correlation (centered) uses Pearson's correlation coefficient with the mean of the two entities as zero. absolute correlation (uncentered) uses the absolute value of the correlation (uncentered) measure. absolute correlation (centered) uses the absolute value of the correlation (center) measure. Euclidean distance uses Euclidean distance.

In the dendrogram of FIG. 6B, gene clustering is shown to the left of the screen, and experimental (sample, array) clustering results are shown to the bottom of the screen. You can see the data value based on the rightmost color palette.

If you select the tree of the part you want to enlarge, a new window appears as shown in FIG. Gene name and experiment name are shown. On this screen, you can configure the software to resize pictures and text.

In the screen of FIG. 6C, when a gene name to be searched is inputted, the software can be configured to display an enlarged view of only the gene in a new window.

k-means clustering performs k-means clustering on genes or samples and provides clustering of genes or samples according to a given k number along with graphs.

The data to be used for analysis should be in the form of a tab-delimited text file (* .txt). As shown in the example below, each gene is input in the row direction, and experiments (samples and arrays) are input in the column direction. The first column contains the name of the gene and the first row contains the name of the experiment (sample, array).

In the screen of FIG. 6D, data to be analyzed is designated in the Import data portion. Select whether to cluster genes on the target or cluster them for the experiment. Select Gene for clustering genes or Experiments for clustering tests. Specify the number of clusters you want in Number of clusters. In the centroid definition, the average definition method is chosen. K-means averages the centers of clusters. K-medoids set the midpoint of the cluster as the median. For Number of iteration, specify the maximum number of iterations to run the analysis.

6E is an example of a result graph according to FIG. 6D. This is a graph of the average of the profiles for each cluster.

If one cluster is selected, a screen as shown in FIG. 6F appears and detailed information can be known. The left side shows the names of the genes belonging to this cluster and the right side shows a graph of their profiles. Select one of the left gene (experimental) names and the profile will appear on the right. At this time, the software can be configured to specify several genes at once.

In the Self Organizing Map (SOM), a Self Organizing Map (SOM) analysis of a gene or a sample is performed, and a clustering result of a gene or a sample according to a given number and type of nodes is provided along with a graph.

The data to be used for analysis should be in the form of a tab-delimited text file (* .txt). As shown in the example below, each gene is input in the row direction, and experiments (samples and arrays) are input in the column direction. The first column contains the name of the gene and the first row contains the name of the experiment (sample, array).

In the screen of FIG. 6G, the uploaded data is designated in the Import data portion. Select whether to target gene clustering or clustering for the experiment. Select Gene for clustering genes or Experiments for clustering experiments. In Number of rows, specify the number of rows you want. In Number of columns, specify the number of columns you want. In the figure, the number of rows and columns is set to 3 assuming a 3 * 3 SOM. For Number of iteration, specify the maximum number of iterations to run the analysis. For Initial learning rate, specify the initial learning rate, which is a value between 0 and 1. In Radius, specify the size of the neighbors in the map. The size of the neighbor must be greater than one. The neighborhood function specifies the type of kernel function. Specify Bubble or Gauss. Topology sets the shape of the grid. Specifies a grid of rectangular or hexagonal shapes.

6H is an example of a result graph according to FIG. 6G. This is a graph of the average of the profiles for each cluster.

In FIG. 6H, when one cluster is selected, a screen as shown in FIG. 6I is displayed to know detailed information. The left side shows the names of the genes belonging to this cluster and the right side shows a graph of their profiles. Select one of the left gene (experimental) names and the profile will appear on the right. At this time, the software can be configured to specify several genes at once.

Principal Component Analysis (PCA) performs principal component analysis on genes or samples and provides the results with graphs.

The data to be used for analysis should be in the form of a tab-delimited text file (* .txt). Each gene is input in the row direction and the experiment (sample, array) is input in the column direction. The first column contains the name of the gene and the first row contains the name of the experiment (sample, array).

In the screen of Fig. 6J, the uploaded data file is displayed in [Import data]. In the selection bar next to [PC based on], specify which data values to obtain eigenvalues and eigenvectors when performing PCA analysis. Use matrix includes R [Correlation] matrix, V [Covariance] matrix and S [SSCP] matrix.

6K is an example of a PCA graph according to FIG. 6J. This figure plots each gene with the first principal component and the second principal component as axes. At the bottom of the window, the eigenvalues and ratios of each of the first and second principal components are shown. When you place the mouse pointer on each point, the name of the corresponding gene appears and the coordinate values of 1 axis and 2 axis on the right side are displayed. To see specific genes, you can drag the region with the mouse. As illustrated in FIG. 6L, the software may be configured to display a list of the corresponding genes and a profile in a new window.

If you want to see the profile of any one of these genes, select the desired gene in the 'gene list' window, and the gene is displayed in red in the 'Gene plot' window and the 'Gene Profile' window. At this time, several genes may be selected.

Figure 6m is a screen for the experiment that appears when the View Experiment plot is selected and interpreted in the same manner as Figure 6l for the gene.

Geneshaving runs gene shaving analyzes on genes and presents the results with graphs.

The data to be used for analysis should be in the form of a tab-delimited text file (* .txt). Each gene is input in the row direction and the experiment (sample, array) is input in the column direction. The first column contains the name of the gene and the first row contains the name of the experiment (sample, array).

In the screen of FIG. 6N, the uploaded data file is displayed in [Data file]. In [Number of clusters], specify the desired number of clusters. For Number of permutation, choose the number of permutations needed to estimate the gap. 6N illustrates that the number of clusters is 3 and the number of permutations is 5.

FIG. 6O is an example of a result graph screen according to FIG. 6N. Three graphs are provided for each cluster.

The graph in the first column shows the experiment and gene expression in the first cluster. Experiments on the horizontal axis and genes on the vertical axis. The order of experiments is arranged in order of average expression level of genes in each cluster. The variance ratio (% vari = VB / VT * 100), the inter-group intra-group dispersion ratio (VB / VW), the inter-group variance (VB) and total variance (VT) are provided together. The higher the dispersion ratio, the more clusters can be interpreted. Preferably, double-selecting the graph may provide an enlarged graph in a new window.

The second graph is a method for determining the number of genes corresponding to each cluster, that is, cluster size, and the variance ratio (VB / VT *) in the data (blue graph) and permutation data (red graph). 100). Permutation data is used to calculate the variance that happens by chance when there is no relationship between genes and experiments. The size of the cluster is determined when the difference between this value and the variance ratio in the given data is greatest. In the program, the number of genes in each cluster is determined by the automatically calculated size according to this value. In this example, 18 genes in the first cluster, 2 in the second cluster, and 6 genes in the third cluster were allocated. If you select the graph, you can see it in a separate window. In this example, the total number of genes is 1500, and the horizontal axis represents numbers from 0 to 1500.

The third graph gives the same information as the second graph and shows the gap statistics (difference between the variance ratio between the actual data and the permutation data) according to the cluster size. Therefore, the point where this value is greatest is set as the size of each cluster.

The following is a description of classification.

The purpose of classification analysis (or discriminant analysis) in the analysis of microarray data is to find a significant gene that best distinguishes each of the already defined treatments (eg samples from normal and carcinoma tissues).

The classical discriminant methods are as follows. The most typical traditional classification method, FLDA (Fisher's Linear Discriminant Analysis) is a method that reduces the data in one dimension (ie, linear combination) to most clearly show the difference between the processing of data spread over multiple dimensions. If the variance is different for each treatment, that is, the data spread in each treatment group is different, a good method that can be used is DLDA (Diagonal Linear Discriminant Analysis). Finally, when the variance-covariance of the data shows a quadratic form, it is called DQDA (Diagonal Quadratic Discriminant Analysis).

The Decision Tree method finds one candidate gene that best describes the process, divides the data according to the criteria (stretches), and then selects the most significant candidate gene from each of the sequentially divided data. Selecting and stretching branches is one of the most useful ways to sort the whole data by taking the process of continuously stretching branches according to certain rules. Tree drawings provide results very similar to those of hierarchical clustering, which has the advantage of allowing users to intuitively understand the results and are used in various fields as well as microarray data.

Variable Selection is as follows. The process of preselecting candidate genes that are expected to be significant is essential for the classification analysis of microarray data. The variable selection method applied to the present invention is a method (BSS / WSS) using a ratio of variation between treatment groups (Within SS) and variation within treatment groups (Between SS). That is, it is a method of selecting candidate genes with a large difference in numerical values between treatment groups and similar numerical values in the same treatment group.

Classification analysis of microarray data is performed using the standardized data as follows. First, missing value correction. Depending on the analysis being performed, missing values should be summarized in advance. In the case of decision trees, analysis is possible without correction of missing values, but the rest of the methods are excluded from the analysis if the missing values are not corrected. Second, candidate gene selection using variable selection. Candidate genes expected to be significant by BSS / WSS are selected. The number of candidate genes is smaller than the total number of observations. Third is Cross Validation (CV). Classification analysis requires an evaluation tool to determine how accurate the results obtained are. The most intuitive is the misclassification rate. The misclassification rate refers to the proportion of individuals misclassified according to the calculated discriminant results. Generally, the data is divided into n equal parts and the classification analysis model is made into n-1 pieces, and the other one is misclassified. Calculate the rate. In this case, the data necessary for constructing the model is called a training set, and the data for evaluation is called a test set. In general, n-fold cross validation refers to a process obtained by repeating the above misclassification rate n times. Therefore, cross validation should be performed for the evaluation of classification analysis. Finally, the choice of analysis method. The analysis is carried out through each method.

The BSS / WSS gene selection method is a method of selecting candidate genes that are expected to be significant by using a ratio of variation between treatment groups (Within SS) and variation within treatment groups (Between SS).

The input file format can be configured as shown in Table 3a below.

[Gene name] [class] [class] [class] [class] [class] [class] [class] [class] one3 1 1 1 1 1 2 2 3 3 GENE0000 -0.549 -0.16135 -0.1183 -0.07699 -0.41637 -0.53994 -1.88314 -1.26389 GENE0001 -0.65766 -0.16635 -0.19691 -0.12792 -0.32658 -0.38762 -1.55486 -0.75601 GENE0002 -0.75556 -0.24525 -0.08846 -0.13201 -0.50722 -0.44909 -1.64757 -0.70395 GENE0003 -0.57216 -0.2546 -0.044085540.1 0.66926 -1.63925 -1.08307 GENE0004 -0.40403 -0.08962 -0.07734 -0.13194 -0.04299 -0.34925 -1.49124 -0.57431 GENE0005 -0.46961 0.04602 0.00438 0.03012 -0.26229 -0.41479 -1.41275 -0.55466 GENE0006 -0.49563 -0.05849 -0.0730.383 -0.31024 GENE0007 -0.51306 -0.01078 0.01036 -0.15065 -0.4402 -0.4574 -1.2619 -0.76481

The format of the analysis result file can be configured as shown in Table 3b below.

[Gene name] [BSS / WSS] GENE0000 7.252297 GENE0001 2.127878 GENE0002 1.411586 GENE0003 4.922214 GENE0004 2.015062 GENE0005 1.860405 GENE0006 0.877066 GENE0007 3.195392 GENE0008 4.270624 GENE0009 1.747254 GENE0010 6.200477

In FIG. 7A, Number of Classes means the number of groups.

The classical method of classification (or discriminant analysis) in the analysis of microarray data is to find a significant gene that best distinguishes each treatment (eg, samples from normal and carcinoma tissues). The discriminant analysis program implemented according to the present invention is implemented as a classical method such as (D) LDA-linear discrimination analysis and (D) QDA-secondary discrimination analysis as cross validation (CV).

The input file format requires two train data and test data, and the input format can be configured as shown in the following Table 4a.

 [index1] [index2] ... [indexN] [class] 32.2 43.2 ... 54.3 1 65.4 72.3 ... 24.3 2

The analysis result file shows Misclassification Error and the group of each genes classified. In case of linking with CV, only misclassification rate is displayed.

On the option input screen, DLDA is to run the Diagonal Linear Discriminant Analysis method, DLDA (CA) is to perform cross check to evaluate the model of Diagonal Linear Discriminant Analysis method, DQDA is to run the Quadratic Discriminant Analysis method, DQDA (CA) is to perform cross validation to evaluate the model of the Quadratic Discriminant Analysis method.

Decision tree is a subprogram that creates a decision tree. Decision trees are analytical methods that classify decision rules and classify interest groups into several subgroups or perform predictions.

The input form can be configured as shown in Table 4b below. Each gene should be separated by a space or a comma, and the last column should contain the class (target variable) to be predicted. class must be an integer starting at 1.

 32.2 43.2 ... 54.3 1 65.4 72.3 ... 24.3 2  1st Gene 2nd Gene ..... class

In the option input screen of FIG. 7B, the uploaded data is designated in the Import data portion. Numer of class is the number of classes and Number of Genes is the number of genes. It represents the number of columns in the data except the class (column number-1). N-fold cross validation divides data into N when examining cross validity. Pruning is performed when the tree is made unnecessarily large according to the pruning rate. For example, if the pruning rate is specified as 0.3, only 70% of the data will be used to build the tree, and 30% of the data will be used for pruning.

The result of FIG. 7C is the case of three classes. In the first category, the 1671st gene is diverging from baseline 59.8. This first branch will find all the first classes. The second branch occurs to the right of the first branch (the 1167th gene is greater than 59.8). The 1727th gene diverges at 324.8. The circle of the results of the tree indicates where the branching occurs, and the number in [] below the circle indicates the gene where the branching occurs. The bounding value is listed with an inequality sign on the bottom right. Where there are no more branches, they are represented by squares. The numbers in circles and squares represent the distribution of each class.

When the present invention is implemented in software, a high-level menu displaying experimental design, standardization, estimation and testing, cluster analysis and classification, and other interfaces can be implemented using Java Server Pages (JSP).

When the present invention is implemented in software, each component, experimental design, standardization, estimation and testing, cluster analysis and classification, can be implemented as separate Java applets sharing a database. In addition, the single slide standardization, dye exchange standardization, multiple slide standardization, and single slide placement method included in the standardization can be implemented as separate Java applets. Similarly, Newton's, B-statistic's, Probability of Probability (ADP), SAM, and EBAM's analysis methods included in estimation and testing can be implemented as separate Java applets. In addition, hierarchical cluster analysis, k-means cluster analysis, principal component analysis, self organization chart, and true shaving included in the cluster analysis may be implemented as separate Java applets.

The invention can be implemented via software. When implemented in software, the constituent means of the present invention are code segments that perform the necessary work. The program or code segments may be stored on a processor readable medium or transmitted by a computer data signal coupled with a carrier on a transmission medium or network.

The computer-readable recording medium includes all kinds of recording devices in which data is stored which can be read by a computer system. Examples of computer-readable recording devices include ROM, RAM, CD-ROM, DVD ± ROM, DVD-RAM, magnetic tape, floppy disks, hard disks, optical data storage devices, and the like. The computer readable recording medium can also be distributed over network coupled computer devices so that the computer readable code is stored and executed in a distributed fashion.

Although the present invention has been described with reference to one embodiment shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and variations may be made therefrom. However, such modifications should be considered to be within the technical protection scope of the present invention. Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

As described above, according to the present invention, by processing the entire process of microarray data analysis in one integrated system, it is possible to perform a systematic statistical analysis of the microarray scanning image data by sharing the database, the situation of each experiment By applying the optimal statistical analysis according to the method, it is possible to minimize false positive and false negative error rate due to improper application and to increase the reliability of the research results, and it is easy to analyze by providing a convenient and user-friendly interface.

Claims (7)

When the experimental design is selected from the items of the upper menu including experimental design, standardization, estimation and test, cluster analysis, and classification, an interface for selecting any one of dye exchange, reference design, or loop design and the selected experimental design Displaying an interface for inputting design parameters including the number of arrays to be used in the array, whether to repeat between arrays, whether to repeat within the array, and a processing list; And Generating an ANOVA design matrix for the selected experimental design using the input design parameters. The method of claim 1, wherein when the standardization of the items is selected, A microarray integrated analysis method comprising generating a standardization result by sequentially performing single slide standardization, single batch standardization, and multiple slide standardization using input slide information. The method of claim 2, Displaying an interface for selecting any one of average, intensity, print tip, or scale normalization methods when performing the single slide normalization, and normalizing the single slide with the selected method; And Micro dye array analysis method comprising the step of performing the standardization by comparing the results of the experiment by reverse Cy3 and Cy5 in the same conditions for the dye exchange experiment. 4. The method of claim 3, wherein if the estimation and the test of the items are selected, Estimating a significant gene in a microarray experiment by performing an analysis of either the Newton analysis or the B-statistic analysis on the standardization result; With respect to the standardization result, one of the analysis of the SAM or the EBAM is performed to search for a significant gene in a repeated microarray experiment, and to determine the probability of erroneous detection according to the searched gene. Calculating; And Performing normalization (ADP) analysis on the standardization result to estimate a significant gene in a microarray experiment and calculating a probability of the adjusted oil of the significant gene in the repeated microarray experiment, and displaying the probability of the adjusted oil. Microarray integrated analysis method characterized in that. The method of claim 3, wherein when the cluster analysis of the items is selected, Providing an interface for selecting any one of hierarchical clustering, k-means clustering, self-organization analysis, principal component analysis, or gene shaving analysis for genes or samples for genes or samples; If a hierarchical clustering is selected, displaying a dendogram as a result of the clustering analysis; Displaying an average graph of the profiles for each cluster as a result of the cluster analysis when any one of the k-means cluster analysis or the histogram analysis is selected; If the principal component analysis is selected, displaying a graph in which the respective genes are plotted with the first principal component and the second principal component as the cluster analysis results; And When the true shaving analysis is selected, displaying a graph indicating the degree of experiment and gene expression in each cluster, a graph indicating actual data and permutation data, and a graph indicating gap statistics according to the size of the cluster as the cluster analysis result. Microarray integrated analysis method comprising a. The method of claim 3, wherein when the classification of the items is selected, A candidate gene having a large variation between the standardization result and the treatment group and a small variation in the treatment group is selected, the candidate gene is classified, the target group of interest is predicted by the design parameter, and a misclassification rate is output. and, The classification may be any one of linear discrimination analysis, secondary discrimination analysis, cross-check of linear discrimination analysis, or cross-check of secondary discrimination analysis. The method of claim 6, Predicting a target group of interest based on the design parameter and outputting a misclassification rate And a step of classifying a decision rule into a plurality of sub-groups and generating a decision tree with respect to the standardization result.

Images