NL1013297C1 - Visualization of relationships in datasets. - Google Patents

Description

Title: Visualization of relationships in datasets

Database comparison shows whether there are any similarities or differences in the files. Usually, the mutual relationship between two sets of data 5 is compared and sometimes between several sets. This comparison is often a time-consuming activity (“data-mining”) where mistakes can easily be made. The invention described here concerns a technique for comparing two or more data sets by showing the differences and / or similarities between the sets in figures. This technique is important, for example, for comparing stock market data 10 (stock stocks) to determine whether there is a correlation between the value of shares of certain companies in different economic developments. Other examples are datasets from the: meteorology, cosmology, 15 mathematics, population studies, patient studies, sociological studies, physical determinations, 20 biotechnology,

High througput systems (the effect of large numbers of different substances are studied several times in different test systems. By combining the results obtained in the different test systems, more information about the substances to be tested is obtained).

25

Analysis of data files of DNA microarrays is further elaborated here. DNA microarray files are an example of biotechnology data files (other examples include dot blots, cDNA-AFLP, Northern blots, Southern blots, and protein arrays from the proteomics). The use of DNA microarrays is taking off in molecular genetics. In this technique, DNA fragments associated with certain genes are placed on a glass slide as small “spots” (the sequence of the DNA fragments and the function of the associated gene may be known after analysis). With certain equipment it is possible to make many copies of such a 1013297 2 glass slide (the position on the glass slide of the DNA spots is the same for all copies). Many thousands of DNA (or cDNA or RNA) spots can be placed on a small glass slide (for example one square inch). This makes it possible to demonstrate which genes are active in a particular tissue, under certain conditions, via certain standard DNA techniques (hybridization techniques with labeled probes). Briefly, this goes as follows: the DNA spots on the array are allowed to hybridize (connect) with DNA or RNA from the tissue to be tested. This DNA or RNA is labeled with a certain color (for example, yellow) and is also referred to as a probe. After hybridization with the probe, the intensity of the staining is proportional to the (relative) expression level of the gene. Two DNA or RNA probes can also be tested simultaneously. For example, the second probe comes from tissue that has been treated differently than the tissue from which the first probe is made. The second probe can be labeled with a different color (e.g. red). Spots on the microarray that are yellow or red after hybridization are expressed only in one of the tissues, and spots of intermediate color (orange-like colors) are expressed in both tissues. With optical equipment, the color of a certain fabric can be measured very precisely and the mutual expression difference in the treated and untreated tissue per spot can thus be determined. The differences in expression can be easily registered: genes that are expressed specifically enough under certain circumstances (for example from tissue after the administration of a certain drug) can be isolated and then further studied. There are different variants for applying DNA microarrays (cDNA microarrays, subtractive cDNA 25 microarrays, microarrays where EST, expressed sequence tags, are spotted on the glass slide, etc.). Because large numbers of genes are compared (often many thousands per array), large data files are created. These data files have hitherto been processed in a less sophisticated manner. For each spot, the change in expression is analyzed and stored under different conditions (via a spreadsheet or via a colored representation of the array, the color reflecting the change in gene expression). Because this method of comparison is very time-consuming (all thousands of spots have to be compared in detail) and because errors can easily arise, in practice, a predefined expression profile is often searched and the strongest exponent is selected 1 n 1 a 7 q 7 3 with the result that a lot of information is lost. Thus, in such a one-dimensional analysis, the interrelationships between the different expression patterns are often not clearly revealed. While this is actually valuable information that can be generated with micro arrays. Even if only a specific expression profile is searched for, it is very useful to know whether this profile can be placed in a larger context. For example, it may be important to know whether genes with a specific expression pattern are involved in multiple processes. This also reduces the chance that the profile found is the result of measurement errors.

10 The aim of the analysis program described here is to display the expression patterns in such a way that the interrelationships become fast, orderly and clearly visible. This is accomplished through a computer program that rearranges the spots of the original array. Here, the spots with a certain specificity associated with a treatment (eg specific gene expression after administration of a drug) are placed together in a figure that represents a virtual array. The spots expressed in another treatment are also grouped, and overlaps within these groups of genes are also visualized. As a result, links in gene expression with different treatments become immediately clear. Genes that are always expressed together and that form networks of gene expression, as it were, are also identified. The computer program in this technique is such that the information about the original position of the spots is preserved so that one can search back which gene (function and sequence) is specifically expressed. The new position of the spot in this virtual array is determined by the relative level of expression relative to the expression of the control genes or of genes expressed in other treatments. In a single (1 determination / 1 physiological condition) experiment, the results are easy to compare. However, if several interrelated conditions (eg a time series and / or a concentration series) are to be compared, the location of the spot in the virtual array also depends on the result of the other measurements within the specified series. The result of such an analysis is shown in Figure 1. In this figure, the data from a hypothetical experiment, the results of which are shown in Figure 2, are rearranged using the computer program. The groups of genes expressed in relation to a particular treatment are grouped in Figure 3. In this hypothetical experiment, gene expression was measured in diseased muscle tissue after administration of different concentrations of two drugs: drug 1 and drug 2. On the moment when the correct ratio of drug 1 and drug 2 is present, the tissue recovers.

If only drug 1 is present, lipid peroxidation occurs and if only drug 2 is present glycogen metabolism no longer takes place. In this hypothetical case, as a result of rearranging with the computer program, all genes expressed in all treatments (and control tissue that was untreated) are placed in the lower left center. These genes are probably 10 “household genes” and are not specifically related to the administration of a particular drug (or recovery from the disease). All genes specific for drug 1 administration are listed above these household genes. Genes specific for drug 2 are to the right of the household genes. Genes that are expressed at the optimal concentration of drug 1 and drug 2 and thus belong to the recovery of the disease are projected at the top right of the household genes. Genes that are expressed simply by administering a substance to a patient are placed at the bottom left of the housekeeping genes. Genes related to the side effects of drug 1 (supra-optimal concentrations of drug 1 in combination with sub-optimal concentration of drug 2 lead to lipid peroxidation) are shown on the far left 20 and genes expressed in reverse (supra-optimal concentrations of drug 2 in combination with sub-optimal concentration of drug 1 lead to loss of the ability to utilize glycogen) are projected below.

What these figures clearly illustrate is that rearranging the results in this way provides a much clearer picture of any relationships between the genes. Analysis without the technique described here will take a very long time and it is doubtful whether the relationships described here will be found because it will depend on the time available and the experience of the investigator performing the analyzes. In this hypothetical experiment, only genes whose expression increases relative to gene expression in the untreated tissue are shown. In a real experiment, genes will also be less expressed relative to this tissue. These differences can also be processed in the virtual array.

1013297 5

Analysis program

The algorithm is able to (virtually) rank the results obtained in experiments with microarrays. The microarrays consist of DNA spots containing copies of complete genes or parts of them. The experiments measure the expression of the genes present on the microarray. To measure this expression, a known mRNA pool isolated from tissue is converted into labeled cDNA. The labeled cDNAs can bind to complementary DNA in the spots. The expression level of a gene determines its share in the mRNA pool. This expression level can therefore be measured by determining the amount of label (of the cDNA) at a given spot.

The experiments start from tissues incubated under different conditions. These incubations are such that they have a certain relationship with each other (for example a concentration series of a certain substance and / or a time series). In the example, a hypothetical experiment 15 with two substances (drug 1 and drug 2) is shown. The function and relationship of the spotted DNA fragments can be determined from the virtual arrangement obtained.

Description of the data.

The data are expression levels measured on the microarrays (amount of label on a spot) in the above experiments. For the algorithm, these data are put in columns. Each column hereby represents the data obtained at a certain incubation condition (determined concentration of drug 1 or 2). A row represents the results of a spot under all 25 incubation conditions.

Data transformation

When sorting the data, the expression values are first classified into a limited number of levels, so each level represents an expression value between two preselected limits. At those spots 30 where the expression value is below the lowest limit value, the level value is assigned zero (low expression levels are assigned the value zero). The final result can be optimized by choosing the limit values for the different levels. For example, if down-regulation of genes 1013297 6 is also studied, the zero value can be assigned to a different level.

Sorting the data.

5 The spots in the different columns are clustered. This can be done as follows: A binary code is assigned to each spot. Each expression value of a particular column is assigned one binary number (0 or 1). One of the expression levels, for example expression level 0, is given the binary number 0 and all other expression levels are given the value 1. Thus, a row of numbers is formed which forms the binary code. All spots with the same binary code are clustered.

Placement of arrays in virtual arrangement.

Prior to the virtual arrangement, the arrays representing the different incubation conditions are placed in a certain order / position. This placement represents, for example, the trial design. In this case, the incubation with the lowest concentration of substance 1 on the left becomes the one with the highest (and at the bottom the lowest concentration of substance 2 and at the top the highest of substance 2 (as is usual when plotting figures).

20 Calculating the optimal position of the clusters on the virtual arrays.

The virtual arrangement of clusters within the arrays may be based on the following. All spots within a cluster have the same calculated position.

If a cluster is expressed at a high concentration of substance 1 and substance 2, then this cluster is also placed at the top right of all virtual arrays. In other words, the “X and Y * coordinates of the cluster (and the spots within this cluster) depend on the expression pattern, of the clustered spots on the different arrays. In addition to the sorting method based on the binary code given above (see “sorting”), it is also possible to sort on the basis of these X and Y coordinates, possibly supplemented by the square sum thereof (this may be less accurate, but 30 saves time. ).

Generation of the virtual arrays

After the X and Y coordinates have been calculated, spots can be placed on the virtual arrays. This is done cluster by cluster (all spots within a cluster 1013297 7 initially have the same X and Y coordinates). This can be done in order of ascending or descending X and / or Y value or ascending cluster size, etc. If the calculated position is already occupied when placing, the nearest empty position is chosen. By starting with the spot that has the highest total expression within a 5 cluster, the visualization will be clearer. If a new clustered calculated position is occupied, a different position is selected. This position can be, for example, the nearest position that is still available, but other placement strategies are also conceivable.

For a good overview of the virtual arrays (good delineation of different 10 clusters), it can be advantageous to make the virtual arrays larger than the original arrays. In this way, an equal number of virtual arrays are generated if there are originals. The expression values values on the virtual arrays are thereby determined by means of colors, grays or hatches that represent the expression level, visualized. In addition, the virtual spot also contains information about the original 15 spot (description of the DNA, place on the original arrays, etc.).

Clustering can be based on more than two dimensions, the visualization (placement of the spots and clusters) is two-dimensional and the expression level can be displayed as a third dimension (via color or shading etc.).

1013297

Claims (1)

The technology described here includes a method to visualize relationships between different datasets in a figure so that the interrelationships can be quickly analyzed and the chance of errors is small. 1013297