WO2002017127A2

WO2002017127A2 - Method and device for the correlation analysis of data series

Info

Publication number: WO2002017127A2
Application number: PCT/EP2001/009710
Authority: WO
Inventors: Werner Terhalle
Original assignee: Genprofile Ag
Priority date: 2000-08-23
Filing date: 2001-08-22
Publication date: 2002-02-28
Also published as: AU2001285898A1; DE10041325A1; WO2002017127A3

Abstract

A method for processing data series, comprising a number of data points in a predetermined sequence of positions is disclosed, comprising the following steps: determination of correlation values for all doublets, triplets, or n-tuplets of positions in a set of data series on the basis of a pre-determined correlation scale, determination of position weightings from the correlation values for each position of the data series, determination of groups of positions correlated together in the data series, the position weightings of which are not zero and are different from a given threshold value and preparation of derived data series, formed from data in the correlated positions. A correlation device for the conversion method is also disclosed.

Description

Method and device for correlation analysis of data sequences

The invention relates to methods for processing data sequences, in particular for ---- correlation analysis of data sequences in order to detect positions of correlated data in different data sequences, such as. B. methods for compressing data sequences, for identifying significant positions in data sequences and / or for classifying data sequences by means of correlation analyzes, devices for carrying out the methods and applications of the methods.

In all areas of research and technology, data is collected that is in the form of symbols with technical meaning (e.g. alphabets made up of numbers, letters, names of substances or system states, or the like) information about a technical structure, a chemical reaction, a biological system, a physical state or the like. The data are generally obtained in a specific order, which results, for example, from a chronological order, a geometric arrangement or even a numerical system parameter. Data sequences can be one-dimensional (e.g. time series of measured values, biological substance frequencies). However, they can also be multidimensional: this is obviously the case with gray-scale matrices in image processing, but also, for example, with DNA sequences. The latter become multidimensional data sequences if you store their structural parameters for each nucleic acid. The amount of data available is constantly growing due to expanding measurement and storage options. For example, there is extensive biologically relevant information in genetic engineering. onen in the form of data sequences, e.g. B. as DNA sequences, protein sequences, encoded environmental data, encoded phenotypes, band patterns of a gel electrophoretic analysis, haplotypes, or combinations thereof. There is an interest in methods to separate the more important data from the less important data or to classify the data according to given criteria. This is important both for the effective handling of the data in data processing systems (storage requirements, computing times and the like) and for the evaluation of the data (pattern recognition, acquisition of new system parameters or the like). Especially in bioinformatics, the relevant positions and / or groups of positions and their association with external characteristics or environmental conditions of the biological system under consideration are to be recognized in data sequences of symbols with biological significance. There is a particular interest in characterizing the behavior of complex systems, to which several data sequences, e.g. B. with respect to internal system states and external system conditions. So far, no effective methods for processing data sequences of complex systems, in particular for recording correlations between significant positions in the data sequences, have been available.

Conventional methods for analyzing and classifying data sequences are based on an examination of the data only in positions and a calculation based on it. Such conventional techniques are described, for example, in MJ Bishop et al. in "DNA and Protein Sequence Analysis" Oxford 1997. However, they are unable to recognize the meaning of positions in the data sequences if this only results from the context of one or more other positions, which may be far apart in the data sequence, and therefore lead through neglect or even embezzlement of such positions based on the distinction between important and unimportant positions data compression and classification to erroneous results.

The object of the invention is to provide improved methods for examining data sequences which are distinguished in particular by the fact that the data can be processed and possibly reduced not only with high effectiveness, but in such a way that errors are avoided which are based on non-consideration of existing ones Dependencies between positions in the data strings are based. In particular, the improved method should also enable reliable classification of data. The object of the invention is also to provide devices for implementing the methods and new applications.

These tasks are solved with methods, computer program products and devices with the features according to patent claims 1, 14 and 15, respectively. Advantageous embodiments and applications of the invention result from the dependent claims.

The basic idea of the invention is to determine relationships or interactions (interdependencies) between individual positions of different data sequences by means of a correlation analysis with the following steps. First of all, a correlation value is determined in the entirety of all data sequences for all pairs of positions with a predetermined correlation measure. The data sequences can be understood as vectors, the components of which are formed by the data. The correlation measure for determining the respective correlation value is applied to all component pairs. In order to be able to assess the correlation values ascertained with regard to their significance, system-related reference values or possibly correlation values or representative reference values obtained therefrom are used for the comparison. The investigation Depending on the application, simulation correlation or reference values occur one or more times before or after the pairwise correlation values are determined. By comparing the correlation values, in particular with the reference values belonging to the corresponding position pairs, a simple threshold value procedure can be used to determine whether the respective correlation value or a position weighting value derived from it is so high that the associated data or positions of a group of correlated data or Positions are assigned or not. The steps mentioned can also be applied analogously to triples or higher n-tuples of positions.

Depending on the result of the threshold value method, a derived data sequence (at least) is generated for each data sequence, which is formed by the correlated positions of the output data sequences. On the basis of the comparison of the correlation values with the simulation correlation values or the representative reference values, more differentiated classifications can also be carried out within the groups of the correlated or non-correlated data.

The determination and evaluation of pairwise correlation values has the advantage that the further processing of the derived data sequences as well as the often time-consuming and costly generation of possibly further data sequences belonging to the considered data record can be limited to the relevant part of the data sequence depending on the point of interest of interest. The method according to the invention results in a data compression which saves storage and computing times as well as working time and costs. Furthermore, there is a particular advantage that associations between different positions can be determined between data sequences that belong to a system but contain completely different data types. For example, the data sequences can each have DNA sequences, relevant changes ^' old data and the associated phenotypes in a suitable coded form. The associations determined according to the invention provide relationships between groups of DNA positions, environmental influences and phenotypes and thus in turn new information as a starting point for an evaluation or change in the biological system under consideration.

The advantages mentioned play a role not only in the evaluation of biologically relevant data. There is generally a simplification and acceleration of work such. B. laboratory analysis of biological sequences, automated image recognition or the monitoring of technical systems, and the application-relevant interpretation of the data sequences. In complex technical systems, correlations between system states can be reliably recorded and used in relation to the control of process parameters or the issuing of warning signals. In addition to information processing in technical systems, preferred applications of the invention thus arise above all in molecular biology, medicine, biology, veterinary medicine, agriculture and ecobiology.

The invention also relates to a computer program product which is set up for compressing data sequences, capturing patterns in data sequences and / or capturing classes in data sequences according to the inventive method.

The invention further relates to a correlator device for. Processing of data sequences according to the inventive method. A correlator device comprises, in particular, a storage device for storing the data sequences to be processed, a computing device for determining correlation values, simulation correlation values and reference values, and a comparator device for evaluating the correction values. lation values and for recording the positions of correlated or non-correlated data.

Further details and advantages of the invention are illustrated below with the aid of a representation of the basic concept of the correlation analysis according to the invention, a method representation and an example. The explanation relates to the processing of biologically relevant information. However, the invention is not limited to this application, but can also be used in all other technical fields for processing data sequences.

Principles of the correlation analysis according to the invention

The inventive method is based on the following findings by the inventors. The individual positions of the considered set of data sequences are more or less "noisy". Some positions are occupied identically in (almost) all data sequences, while other positions are highly variable. The constant positions are unusable for the purpose of classifying or assigning different functional characteristics to the data sequences. Rather, the variable positions at which the data sequences to be classified do not match are to be considered. Here and in the following, the expression of the function is generally understood to mean a connection between data sequences and system conditions, which is generally interpreted causally in one direction or another. A change in the system conditions can cause a change in the measured values recorded in the data sequence. On the other hand, a change e.g. B. lead to a change in the phenotype in a gene sequence. The function expression in a suitably coded form can itself be part of the data sequence. Two fundamentally different qualities of the variability of a position in a data sequence can be distinguished. On the one hand, a position can be highly variable because a change in the staffing has no effect on the extent of the function. On the other hand, there can be a high degree of variability because the respective position is associated with different functional characteristics. Since the functional form of a data sequence is determined by the specific occupation of a combination of several, generally not adjacent positions, it can be assumed that the positions that are significant in connection with the function under consideration are occupied from one another and are correlatedly variable ("synchronous noise"), while the randomly rushing positions tend to be occupied independently of any other position.

The inventors have also found that the synchronous noise of the significant positions is not only limited to pairs of data, but also affects larger groups of data at certain positions. The method according to the invention is now directed towards quantifying the meaning of the individual positions in a set of data sequences which is related to a function under consideration and to subject the data sequences to compression, classification and / or prediction procedures on this basis. Data compression means that only the relevant positions or position groups are taken into account in the further processing of the data sequences.

The information obtained by the correlation analysis according to the invention can also be used directly for classification. The data sequences that have the same occupations (at least almost) at the positions of a group of highly dependent, noisy positions are combined into a subclass. Of the many theoretically possible c ^'hen cast at these positions come because of the interdependencies are only few, the respective sub-class characterizing pattern.

If the classification constructed in this way has the property that position sequences combined in a subclass do not differ or differ only slightly in terms of their function, a correlation with the function under consideration has been found, which also makes it possible to predict future system states with regard to the function under consideration , Are there additional data sequences in addition to the originally considered data sequences and do they have known positions at the excellent combinations? H. Occupations determined as part of the classification, these combinations of positions can be related to the corresponding function. Depending on the application, it can be provided that such predictions are validated by additional methods or information.

The technical application of the correlation analysis according to the invention results from the data compression, in which the important data positions are recognized and further processed in relation to a specific function, the pattern recognition or classification, in which combinations of position assignments at the identified important positions are determined, the relevant one Describe subclasses of the considered sequences, the association of patterns in the position sequences to the expressions of the considered functions and the prediction of functional expressions in new data sequences. Carrying out the correlation analysis according to the invention

Step 1: Provision of the data

In a first step, the data of interest are provided for the correlation analysis according to the invention, e.g. B. transferred to a correlator device. Depending on the application, the data are first measured or recorded, input into the correlator device via an interface, temporarily stored and compiled into data sequences. This sub-step is not absolutely necessary, the data sequences can already exist as measured value sequences, for example. The data sequences are then formatted to form a set of sequences which have corresponding data in the same position and which all have the same length. If the data initially provided lead to data sequences with different lengths, as can be the case, for example, with data sequences to describe a phenotype, gaps arise in the corresponding data sequence. For formatting, the gaps are filled or the corresponding positions in the other data sequences (e.g. gene sequences) are deleted. The filling is carried out, for example, with a separate "gap" or "gap" symbol, with the most frequent value at this position or - in the case of numerical data - with an average value.

The data sequences are possibly based on different symbol stocks or "alphabets" and are available, for example, in a stored form.

Step 2: Determination of correlation values and position weightings

Depending on the task, a problem-relevant method for calculating the dependencies between two positions different data sequences. The mutual dependencies (correlation values) in pairs are determined in a first sub-step by means of a correlation measure in accordance with the chosen method. Two correlation measures, namely the trans information and the predictability, are illustrated below as examples. However, the invention is not limited to these dimensions, but can be implemented with all methods which are generally suitable for characterizing associations or correlations between positions by specifying quantitative correlation values. Various such methods are known per se and are based, for example, on χ ² tests or algorithms known in textbooks.

(a) TransInformation

The transinformation is a correlation measure based on the Shannon "see entropy", which is known per se from information theory for characterizing the combination of two signals (see, for example, BH Rohling "Introduction to the information and coding theory", Stuttgart, 1995). The correlation value transinformation is formed as follows: If Ai is the alphabet for position i and Aj is the alphabet for position j, pi and pj the associated frequency distributions and i _{j is} the common frequency distribution of the two positions, then the transinformation T (ij) of positions i and j according to the following equation.

T (i, j) - ∑ p _y (a, b) log — l—,

aeATÄA _j Py (a, b)

The transinformation T is the sum of the entropies for the individual positions, minus the entropy of the position pair. In information theory, transinformation is a common measure for describing the mutual influence of two signals. It is minimal if intended positions are statistically independent, and maximum if both positions are equally distributed and are mutually unambiguous.

The correlation measure Transinformation provides a number for each pair of positions that describes the correlation. The correlation cannot be assessed from the quantitative value alone without additional information, since the size of T also depends on the number of symbols in the data sequences. The more symbols the alphabets contain, the larger T values occur. The assessment takes place in the third step (see below).

(b) predictability

Predictability is a newly developed, directed measure of correlations between different positions, which depends on whether in two considered positions one is derivable or predictable from the other. The correlation value predictability is a quantitative measure for the statement "if an a at position i, then a b at position j". The measure of predictability results from the following considerations. For each aeA _; let fi (a) eA _{j be} the most common "letter" at position j associated with a at position i. If there are several most common letters, one of them is chosen arbitrarily, since the result of the determination of the predictability does not depend on this selection from the most common letters. If N is the number of all data sequences and ni _j (a) the number of those data sequences among them that have an a at position i and a fij (a) at position j, then the predictability V (i, j) is Position j by position i given by the following equation.

Here, H (j) is the entropy H (j) = - 2 _j ( ^b ) ^lo gP _j ( ^D ) • ^Dif ≥ PredeterminationA _j is the proportion of the data sequences in which the Prediction of position j is correct, if one concludes from the knowledge of the occupation of position i that the most frequently associated occupation of position j.

Finally, in a further sub-step, position weightings are determined from the correlation values determined in pairs for all positions of the data sequences. For each position of the data sequences, all associated correlation values are subjected to a summation (synonymous with averaging) or a maximum formation, so that the position weighting results in each case as a quantitative parameter, which is output or stored in addition to the correlation values as a form of information compression. In this way those positions are heavily weighted, which - have on average for ^'all other positions a strong dependence and - - in the case of summing in the case of maximum generation - at least one other position.

Even after this step, depending on the application, the data sequence can be reduced by deleting all positions whose position weighting value is zero or so low that a correlation with other positions is ruled out. For this purpose, for example, a comparison is made with predetermined system-related reference values.

Step 3: ^" Determination of reference values for the statistical evaluation of the position weights

The quantitative values supplied with the correlation measure for characterizing the interdependency between positions can be in relation to their statistical significance can be evaluated by a simulation method. The implementation of the simulation method is not a mandatory feature of the invention. Depending on the application, this can be dispensed with if, for example, additional information about the system under consideration is available or if the determined correlations can be easily assessed as to whether it makes technical or biological sense in the system.

The simulation method u includes the generation of a large number of randomized reference data records (so-called "shuffles") - the reference data records each consist of the same number of data sequences as the data record under consideration, all have the same length as the given data sequences and result from them in the following way: If one imagines the individual data sequences of the given data record written line by line, the data within the columns, that is, the data at the same position, are randomly interchanged. Such intra-position swaps do not change the noise of the positions, but break up existing dependencies and possibly create new dependencies. As for step 2, the correlation measure is used for the quantitative evaluation of mutual dependencies for each reference data record. A large number of simulation correlation values result for all pairs of positions of each “shuffles” considered.

The maximum dependency that occurs between two positions is determined for each reference data record of the simulation process. Furthermore, the maximum position weighting is determined for each reference data record in accordance with the method selected for the given data record. The mean and variance of these two values, determined over all reference data sets, are used as representative reference values for later comparison with those for the data under consideration following calculated correlation values and position weightings output or saved.

Step 4: Acquisition of the positions of correlated data

In a first sub-step, dependency groups on positions are determined. For this purpose, the paired dependencies of the positions are compared with a predetermined threshold value. The threshold value is, for example (as is customary in decisions about statistical significance) the sum of the mean and variance of the maximum dependency in the reference data records determined in step 3. Alternatively, an application-dependent variable can be used as the threshold value, which is based on additional information, empirical values or the like. Correlated positions are preferably determined by forming dependency groups of the positions according to the following scheme.

Groups of positions whose paired dependencies on each other are all above the threshold are summarized as so-called cliques. If the majority of the correlation values are above the threshold value, but a small number of pairs of positions result in lower correlation values, the associated positions are combined in groups which are referred to as "near cliques". When defining an "almost clique", a second, lower threshold value can be taken into account as the minimum size for those correlation values which do not reach the threshold value for a clique. As the weakest form of a dependency group, positions that are only indirectly dependent on each other are summarized as "components". There is an indirect dependency of the positions i and q if there are positions j, k, ... q such that the position pairs (i, j), (j, k), ..., (p, q ) above each S ^'chwellwert lying correlation values have. However, a high correlation value for the position pair (i, q) does not necessarily have to be present.

For the purpose of shortening the data sequences and thus the data compression, all positions outside the dependency groups can be deleted (deleted). Then only the relevant data required for further processing remains.

In a further sub-step, the dependency groups are output or saved. The positions of the data sequences are assigned information that they belong to one of the dependency groups mentioned or not. Derived data sequences are formed that only include the correlated positions. Depending on the application, the derived data sequences are sent to an interface to another evaluation or diagnostic device, stored, displayed or otherwise displayed.

Step 5: Determination of sub-classes of the data sequences

Subclasses of the given set of data sequences are then determined on the basis of the dependency groups determined in step 4. The dependency groups form certain patterns, i. H. Combinations of positions. The subclasses and the patterns that characterize them within the data sequences are output or stored.

As a result, the number of data sequences relevant for further processing, display or evaluation has been reduced by selecting a representative data sequence for each subclass. Step 6: prediction

The prediction u summarizes the processing of one or more new data sequences in accordance with steps 1 to 5 and the comparison of the patterns determined for the new data sequences in step 5 with the patterns of the previously processed data sequences. If characteristic patterns match, then the respective positions of the new data sequences are assigned the subclass determined for the data sequences processed first, or the corresponding affiliation to this subclass is predicted.

Example Step 1:

The method according to the invention is explained using a constructed example. 16 position sequences of length 9 are considered, which are in position 8 above the alphabet "1,2,3 ...", in position 9 above the alphabet "+, -", otherwise via the alphabet "A, C, G, T" are formed. These are, for example, DNA sequences of length 7 with an environmental influence coded in the attached position 8 and a presence of a phenotypic property noted in position 9.

Position 1 2 3 4 5 6 7 8 9

Episode 1: G A A A A A A 3 +

Episode 2: T C A T C C A 3 +

Episode 3 A T A C T C G 2 -

Episode 4: A A A C A A G 2 +

Episode 5 C C A G C C T 1 -

Episode 6 G G A A G G A 3 +

Episode 7 C G A G G A T 1 -

Episode 8 G T A A T C A 3 +

Episode 9 G T A A T C A 3 +

Episode 10 A G A C G G G 2 +

Episode 11 C T A G T T T 1 -

Episode 12 T T A T T T A 3 -

Episode 13: C A A G A G T 1 -

Episode 14: G C A A C T A 3 +

Episode 15: T G A T G A A 3 -

Episode 16: ACACCTG 2 + 2 ^' . Step:

The pairwise dependencies between the positions are calculated as the correlation value TransInformation:

Pos.i Pos. J T (i, j) Pos.i Pos. J T (i, i)

1 2 0.0551 3 7 0.0000

1 3 0.0000 3 8 0.0000

1 4 1.3705 3 9 0.0000

1 5 0.0551 4 5 0.0551

1 6 0.0551 4 6 0.0551

1 7 1.0397 4 7 1.0397

1 8 1.0397 4 8 1.0397

1 9 0.4254 4 9 0.4254

2 3 0.0000 5 6 0.6943

2 4 0.0551 5 7 0.0169

2 5 1.3705 5 8 0.0169

2 6 0.6943 5 9 0.0418

2 7 0.0169 6 7 0.0169

2 8 0.0169 6 8 0.0169

2 9 0.0418 6 9 0.0091

3 4 0.0000 7 8 1.0397

3 5 0.0000 7 9 0.2636

3 6 0.0000 8 9 0.2636

A value of the transinformation of 0 means stochastic independence in the usual sense. This is particularly the case if one of the positions considered is constant, as is the data in position 3 here. The strongest dependencies in the example exist between positions 1 and 4 or between positions 2 and 5: While positions 2 and 5 are identical Positions 1 and 4 are mutually unambiguously determined - a "G" at position 1 is always connected with an "A" at position 4, a "T" with a " T ", an" A "with" C "and a" C "with a" G ".

Then follows the position weighting determined by totaling: Pos. Weight

1 4.0406

2 2.2506

3 0.0000

4 4.0406

5 2.2506

6 1.5417

7 3.4334

8 3.4334

9 1.4707

Positions 1 and 4 are of the greatest importance in terms of this weighting, since all other positions depend on them most on average.

3rd step:

The verification of the statistical relevance by means of simulation shows: 100 "shuffles" have an average maximum dependency of two positions on one another of 0.5941 with a variance of 0.0870; for the position pairs with a greater dependence than 0.5941 + 0.0870 = 0 , 6811 the statistical relevance is given.

4th step:

If one chooses the threshold value 0.5941 + 2 ^• 0.0870 = 0.7681, ie only considers those position pairs with a transinformation that is at least two variances larger than the expected maximum, two cliques are found: the group of positions 1,4,7,8 (two of these four positions each have a transinformation above the selected threshold) and the group of positions 2.5.

5. At positions 1,4,7,8 the following patterns occur within the set of position sequences: Position 1 4 7 8

Episode 1 G A A 3

Episode 2 T T A 3

Episode 3 A C G 2

Episode 4 A C G 2

Episode 5: C G T 1

Episode 6 G A A 3

Episode 7 C G T 1

Episode 8 G A A 3

Episode 9 G A A 3

Episode 10 A C G 2

Episode 11 C G T 1

Episode 12 T T A 3

Episode 13: C G T 1

Episode 14: G A A 3

Episode 15: T T A 3

Episode 16 A C G 2

This leads to the division of the quantity into four subclasses:

Subclass 1 (for model "GAA3"): Follow 1,6,8,9,14

Subclass 2 (for the pattern "TTA3"): episodes 2, 12, 15

Subclass 3 (for the pattern "ACG2"): episodes 3,4,10,16

Subclass 4 (for the pattern "CGT1"): episodes 5,7,11,13

It should be noted here that the classification according to the patterns occurring at positions 2.5 would have led to a different classification:

Subclass 1 (for the pattern "AA"): episodes 1,4,13

Subclass 1 (for the pattern "CC"): Follow 2,5,14,16

Subclass 1 (for the pattern "TT"): Follow 3,8,9,11,12

Subclass 1 (for the pattern "GG"): episodes 6,7,10,15

Optionally, the respective implied classification can be output for all position groups found in step 4, in order to then use additional information to decide which is most suitable in relation to the problem. It is also possible to construct a common partitioning - depending on the ^'objective about the coarsest partition that is finer than any found, or the finest among the coarser. 6th step:

Finally, for the sequence "GGAATTC3" not belonging to the original position sequences, a "+" is predicted for the function coded in position 9, that is to say the presence of the phenotypic property under consideration, since its pattern "GAA3" at positions 1,4,7 , 8 matches the pattern characterizing subclass 1 and each position sequence from this subclass has a "+" at position 9.

Correlation analysis device

A correlator device according to the invention comprises a formatting device for providing a plurality of data sequences of the same length, a computing device for determining the correlation values between all position pairs of the data sequences and the position weights derived therefrom, a comparator device for comparing the position weights with predetermined reference values and for determining correlated positions, and one Device for displaying, outputting or storing derived data sequences which are formed by the correlated positions. The various components of the correlator device are preferably controlled by a data processing system, e.g. B. implemented a computer.

Claims

claims

1. A method for processing data sequences, each comprising a number of data in a predetermined sequence of positions, with the steps:

Determination of correlation values for all pairs, triples or n-tuples of positions in a set of data sequences on the basis of a predetermined correlation measure,

Determination of position weightings from the correlation values for each position of the data sequences,

Detection of groups of correlated positions in the data sequences, the position weightings of which are not equal to zero and deviate from a predetermined threshold value, and

- Provision of derived data sequences, which are formed by data at the correlated positions.

2. The method of claim 1, wherein detecting the correlated positions comprises the following steps:

Determination of simulation correlation values for all pairs of positions in a large number of randomized reference data sets,

- Determination of representative reference values from the simulated reference data sets for the correlation values and position weights,

- determination of threshold values from the reference values,

- Assignment of the positions for which the position weighting of the data sequences to be processed is greater or smaller than the corresponding threshold value to a group of the correlated positions or a group of non-correlated positions.

3. The method according to claim 1 or 2, in which trans-information or prediction values are determined as correlation values.

4. The method according to claim 2 or 3, wherein the representative reference values by calculating statistical moments

(Expected value, variance, higher moments) and combinations of them or other mathematical functions from all or the respective maximum correlation values and position weights over all reference data sets.

5. The method of claim 4, wherein the group of correlated positions is divided into subgroups in which all pairwise correlations or the majority of all pairwise correlations or all pairwise indirect correlations of the positions exceed the threshold values.

6. The method according to any one of the preceding claims, in which the data sequences and / or the data sequences derived from them according to one of the preceding claims are subjected to a classification and / or a pattern recognition.

7. The method according to any one of the preceding claims, wherein the provision of the data sequences and / or their derivations comprises storing, displaying or sending to an interface of a data processing device.

I

8. The method according to any one of the preceding claims, in which the data sequences to be processed are formatted in such a way that the same number of positions is given in each data sequence.

9. The method according to any one of the preceding claims, wherein the data sequences over the same alphabet or different

Alphabets are formed.

10. The method according to claim 9, wherein the alphabets of the individual positions of the data sequences encode biological substances, properties of biological substances, nucleic acids, amino acids, structural parameters, expressions of phenotypic characteristics and / or expressions of environmental characteristics.

11. The method according to claim 1, wherein the data sequences include gene sequences, nucleic acid sequences, amino acid sequences, band patterns of gel electrophoretic analyzes, haplotypes, encoded phenotypes, encoded environmental data or combinations thereof.

12. The method according to any one of the preceding claims, wherein the alphabets comprise groups of system parameters of a control system, measured values and / or image values.

13. The method according to any one of the preceding claims, in which the derived data sequences are provided as an input variable for a prediction or diagnosis method.

14. Computer program product which is set up for compression of data sequences, detection of patterns in data sequences and / or detection of classes in data sequences according to a method according to one of the preceding claims.

15. Correlator device comprising:

a formatting device for providing a large number of data sequences of the same length,

a storage device for intermediate storage of the data sequences to be processed,

a computing device for determining the correlation values between all position pairs of the data sequences and the position weights derived therefrom, a comparator device for comparing the position weights with predetermined threshold values and for determining correlated positions, and

- An output and / or storage device for output or storage of derived data sequences, which are formed by the correlated positions.

16. Correlator device according to claim 15, which is set up to carry out the steps of a method according to one of claims 1 to 14.

17. Correlator device according to claim 15 or 16, which is formed by a data processing system.

18. Use of a method, a computer program product or a device according to one of the preceding claims for detecting correlations between positions in data sequences.