JP2008100918A - Similarity calculation processing system, processing method and program of the same - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a similarity calculation processing system capable of accurately extracting compounds, for example, which are not similar to each other as a whole but partially similar, and to provide a processing method and program of the same. <P>SOLUTION: The similarity calculation processing system includes: a training data memory device 101 for storing inputted training data; a prediction data memory device 102 for storing data for prediction; at least one descriptor selecting device 103 for extracting a part of descriptors of the input training data and of the prediction data; a similarity calculating device 104 provided corresponding to each of the descriptor selecting device 103 for calculating a similarity between the training data and the prediction data; and a calculation result collecting device 105 for calculating a similarity from the calculation results by the similarity calculating device 104 by each of the data stored in the prediction data memory device. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a similarity calculation processing system for similar compounds, a processing method thereof, and a program. More specifically, the present invention relates to a similarity calculation processing system, a processing method, and a program effective when searching for similar compounds in the field of drug discovery. .

  The similarity calculation method is widely known as a method for searching for similar data, and is used in various situations.

  For example, in the compound estimation method disclosed in Patent Document 1, in order to estimate a compound having pharmacological activity or toxicity of an unknown cell, it is a method of estimation from gene information of a known cell similar to gene information of an unknown cell.

  Further, the compound alignment method disclosed in Patent Document 2 is a method for determining the alignment of compounds by calculating and calculating the similarity of alignment between compounds from the three-dimensional structure information.

  The common structure extraction device disclosed in Patent Document 3 is a method of finding corresponding common points in two different three-dimensional structures, and thereby calculating the similarity of the entire structure.

  Here, as a measure for measuring the similarity of compounds, for example, the Tanimoto coefficient is a well-known similarity calculation method. By preparing a plurality of such descriptors using 0 and 1 bits indicating the presence or absence of a certain structure, a compound can be expressed by a bit string of 0 and 1. The Tanimoto coefficient for such a bit string of 0, 1 is defined as follows. The closer the value is 1, the more similar the compounds are. Is done.

  Assume that the expressions of the two compounds are defined as those shown in the following formula 1.

  Assuming that the equation 1 is defined, the Tanimoto coefficient is defined as the following equation 2.

  In the above equation 2, “and” and “or” represent bit operations, respectively, and are defined as the following equation 3.

  In the field of drug development, etc., it is necessary to search for a compound having activity against a target from a compound library containing hundreds of thousands to millions of compounds. In such a case, for example, the similarity is calculated based on the information on the active compound described in the literature information such as patents, and the assay experiment is performed only on the compound having a high similarity. By doing so, it becomes possible to find active compounds with high probability without experimenting with all the compounds contained in the compound library, which is one of in silico screening techniques.

  As a conventional similarity calculation method, a method using a distance function (Euclidean, binary, etc.) or a method using a method such as the Tanimoto coefficient as described above has been used even if the distance axiom is not satisfied. When using the distance function, data having a short distance is considered as data with high similarity, and when using the Tanimoto coefficient, data closer to 1 is considered as data with high similarity.

JP 2006-107395 A JP 2006-134252 A JP-A-7-287717 Byung-Je Sung et al: Nature, vol 425, pp. 98-102, September (2003).

  The conventional similarity calculation method and apparatus are methods for calculating the similarity by capturing the overall features as described in Patent Documents 1, 2, and 3.

  When screening a compound, it can be said that the characteristics of the compound are similar if the overall characteristics of the compound are similar, but the overall characteristics do not necessarily have to be similar, and proteins and drugs are key and keyholes. If a compound has a certain characteristic, it can also act on a protein, in which case it does not necessarily have to be similar overall. In such a case, if the similarity calculation method as described above is used, it may not be possible to pick out active compounds that are not similar but partially similar.

  Therefore, a technical problem of the present invention is to provide a similarity calculation processing system, a processing method thereof, and a program capable of accurately extracting, for example, compounds that are partially similar but partially similar. is there.

  According to the present invention, storing input training data, storing data for prediction, extracting a part of descriptors of the input training data and prediction data, A similarity calculation processing method comprising: calculating similarity between training data and the prediction data; and calculating a similarity for each of the stored prediction data based on the calculation result of the similarity Is obtained.

  In addition, according to the present invention, in the similarity calculation processing method, a similarity calculation processing method characterized in that similarity calculation using a Tanimoto coefficient is performed for the similarity calculation.

  According to the present invention, there is provided the similarity calculation processing method according to any one of the similarity calculation processing methods, wherein the similarity calculation outputs a maximum value from the calculated similarities. can get.

  According to the present invention, in any one of the similarity calculation processing methods, in calculating the similarity, an average value and a variance are calculated by aggregating values corresponding to the stored prediction data. A similarity calculation method characterized in that a final numerical value is output using a function of two values is obtained.

  According to the present invention, in any one of the similarity calculation processing methods, the descriptor is extracted a plurality of times, and the same number of random numbers is generated between the plurality of extraction operations. A similarity calculation processing method is obtained.

  According to the present invention, in any one of the similarity calculation processing methods, the calculation of the similarity may include outputting a final numerical value by using a function of a distance from a predetermined plane. A characteristic similarity calculation processing method is obtained.

  According to the present invention, the training data storage device for storing the input training data, the prediction data storage device for storing the data for prediction, the input training data and the prediction At least one descriptor selection device for extracting a part of a descriptor of data, and a similarity calculation device provided corresponding to each descriptor selection device and calculating the similarity between the training data and the prediction data And a calculation result aggregating device for calculating the similarity for each data stored in the prediction data storage device. The similarity calculation processing system is obtained. .

  In addition, according to the present invention, in the similarity calculation processing system, the similarity calculation processing system is characterized in that the similarity calculation device performs similarity calculation using a Tanimoto coefficient.

  According to the present invention, in any one of the similarity calculation processing systems, the similarity calculation apparatus outputs a maximum value from the calculated similarities. Is obtained.

  Further, according to the present invention, in any one of the similarity calculation processing systems, the calculation result aggregation device calculates an average value and a variance by aggregating values corresponding to data stored in the prediction data storage device. A similarity calculation processing system characterized in that a final numerical value is output using a function of the two values is obtained.

  According to the present invention, in any one of the similarity calculation processing systems, each of the descriptor selection devices generates the same number of random numbers between the descriptor selection devices. A processing system is obtained.

  Further, according to the present invention, in any one of the similarity calculation processing systems, the calculation result totaling device outputs a final numerical value by using a function of a distance from a predetermined plane. A characteristic similarity calculation processing system is obtained.

  According to the present invention, the training data storage means for storing the input training data, the prediction data storage means for storing the data for prediction, the input training data and the prediction At least one descriptor selecting means for extracting a part of the descriptor of the data, and a similarity calculating means for calculating the similarity between the training data and the prediction data provided corresponding to each of the descriptor selecting devices And a calculation result aggregating means for calculating a similarity for each data stored in the prediction data storage device. The similarity calculation processing program is obtained. .

  Further, according to the present invention, in the similarity calculation processing program, a similarity calculation processing program is obtained in which the similarity calculation means performs similarity calculation using a Tanimoto coefficient.

  According to the present invention, in any one of the similarity calculation processing programs, the similarity calculation means outputs a maximum value from the calculated similarities. Is obtained.

  Further, according to the present invention, in any one of the similarity calculation processing programs, the calculation result aggregation means calculates an average value and a variance by aggregating values corresponding to data stored in the prediction data storage device. Thus, a similarity calculation processing program characterized in that a final numerical value is output using a function of the two values is obtained.

  Further, according to the present invention, in any one of the similarity calculation processing programs, each of the descriptor selection means generates the same number of random numbers between the descriptor selection means. A processing program is obtained.

  According to the present invention, in any one of the similarity calculation processing programs, the calculation result totaling device outputs a final numerical value by using a function of a distance from a predetermined plane. A characteristic similarity calculation processing program is obtained.

  In the present invention, for example, it is possible to provide a similarity calculation processing system, a processing method thereof, and a program capable of accurately extracting compounds that are partially similar but partially similar.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a diagram showing a schematic configuration of a similarity calculation processing apparatus according to an embodiment of the present invention. Referring to FIG. 1, a similarity calculation processing device 100 includes a training data storage device 101 for storing input training data, a prediction data storage device 102 for storing data for prediction, A description selection device 103 for extracting a part of the descriptor of the input training data and prediction data, and a similarity calculation device 104 that calculates the similarity between the training data and the prediction data and outputs a numerical value with the highest similarity. And a calculation result totaling device 105 that calculates the average and variance for each data stored in the prediction data storage device 102 and calculates the similarity based on the results calculated by each similarity calculation device 104. Has been.

  In FIG. 1, descriptor selection devices 103a, b,... (Hereinafter collectively shown as 103) are attached to the respective similarity calculation devices 104a, b,. , Each is expressed to work independently.

  In the sense that they operate independently, it is not always necessary to install a plurality of similarity calculation devices 104 and descriptor selection devices 103, and one descriptor selection device 103 and similarity calculation device 104 are installed. A configuration in which the same processing is sequentially repeated may be employed.

  The similarity calculated by each similarity calculation device 104 is received by the calculation result totaling device 105, and an average and variance are calculated therein, and the similarity is calculated from these indices.

  FIG. 2 is a diagram illustrating a specific configuration of the descriptor selection device 103. Referring to FIG. 2, a descriptor extracting apparatus 201 that randomly calculates a number corresponding to a descriptor for selecting from input training data and prediction data, and a training that receives a descriptor corresponding to the calculated number. A data conversion device 202 that selects data from the predicted data and converts the data is configured.

  Next, the operation of the similarity calculation processing apparatus 100 of FIG. 1 will be described with reference to FIG. FIG. 3 is a diagram for explaining the operation of the similarity calculation processing apparatus 100 of FIG.

  Referring to FIG. 3, training data and prediction data are input to the descriptor selection device (step 301).

  The input training data is stored in the training data storage device 101, and the input prediction data is stored in the prediction data storage device 102 (step 302).

  The input training data and the prediction data need to be described in the same order using the same descriptor. In this description, one data corresponds to one row, but it does not necessarily have such a data format, and one data may correspond to one column.

  When one piece of data corresponds to a column, in the following description, “row” must be read as “column”, and “column” must be read as “row”. is there.

  The training data and the prediction data stored in the training data storage device 101 and the prediction data storage device 102 are sent to the descriptor selection device 103, respectively.

  Here, the internal operation of the descriptor selection device 103 will be described with reference to FIG. 2 again. Referring to FIG. 2, for example, it is assumed that data represented by the following equation 4 is sent from the training data storage device 101 and the prediction data storage device 102.

  Since the representation format of the training data and the prediction data is the same, one of them is shown in the above example. Actually, the same processing is performed on the training data and the prediction data input simultaneously. In this example, only one of the processes is shown for simplicity.

  Assume that the descriptor extraction device 103 generates a random number represented by the following equation 5 for the data input to the descriptor selection device 103 (step 303).

  The random number value is sent to the data converter 202. Data input to the descriptor selection device 103 is also sent to the data conversion device 202. In the data conversion device 202, the first, third, and fifth columns are extracted from the data input corresponding to the similarity selected by the descriptor extraction device 201, and are expressed by the following equation (6). Data is output (step 304).

  The data is output to each similarity calculation device 104. In this example, the same column as the value output by the descriptor extraction device 103 is output with priority on understanding, but it is not always necessary to deal with such a case. For example, for a numerical value i ((i + 5 ) Divided by the number of columns + 1) columns may be selected. Further, it is not necessary to correspond one column to one value, and processing such as extracting a plurality of columns may be performed. That is, this correspondence may be one-to-one, one-to-many, many-to-one, or many-to-many.

  Data corresponding to the training data and the prediction data is output from the descriptor selection device 103 and sent to the similarity calculation device 104. For simplicity, these output data are called training data A and prediction data A, respectively. The similarity calculation device 104 calculates the similarity from the training data A for each piece of input prediction data A (step 305). That is, the similarity of the number included in the training data A is calculated for one piece of the prediction data A. The data having the highest similarity among the plurality of values is output. Since one degree of similarity is output for one piece of data included in the prediction data A, when it is accumulated, prediction values corresponding to the number of data are output. These processes are performed independently in the installed similarity calculation device 104. It is not always necessary to install a plurality of devices in the sense that they are performed independently, and a single similarity calculation processing device may be installed and processed sequentially.

  Next, the similarity corresponding to the input prediction data output from the similarity calculation device 104 is sent to the calculation result totaling device 105. The calculation result totaling apparatus 105 performs totalization for each prediction data from the plurality of similarities calculated (step 305). Here, the aggregation calculates the average value and the variance. In this description, the average value, variance, and description are described for the sake of simplicity. However, the average value and the variance are not necessarily required, and any function such as a constant multiple of the average value may be used. The same applies to the dispersion. In the case of variance, further replacement is possible, and it is possible to search for the maximum and minimum values and use the difference or use a function of the difference. In total, the average and variance of the similarity are calculated for the predicted data. Next, a function corresponding to one value from those indices is used to output corresponding to one value. As the function, for example, the horizontal axis represents the average value, the vertical axis represents the dispersion, and the distance from a certain point to a certain straight line can be used as the function. If a straight line that is far enough parallel to the vertical axis is taken, the average value of the similarity is output as the final result. Similarly, if a straight line that is sufficiently separated as the straight line and is not parallel to the axis is taken, a function using the mean and the variance is obtained, and a value using the function is output.

  Next, the effect of the similarity calculation processing device according to the embodiment of the present invention will be specifically described with reference to FIG.

  The effect of the present invention is that it becomes easy to find partially similar data that was difficult to find by the conventional similarity calculation method.

  As training data, the data obtained by replacing the structure of the three compounds described in Non-Patent Document 1 with a bit string and the evaluation data obtained by replacing the structure of about 16000 compounds commercially available with a bit string were input.

  In the descriptor extracting apparatus 201, a random number that takes a value of (1 to the number of columns) is generated.

  In addition, the same number of random numbers is generated between the descriptor selection devices 103.

  In the data converter 202, a column corresponding to the output random number is taken out.

  The similarity calculation device 104 calculates the similarity using the Tanimoto coefficient, and the calculation result totaling device 105 calculates the average and variance, and the function shown in the following equation 7 is used.

  The above formula 7 shows the distance to the plane of x + 10 * y-10. The result calculated at that time is shown in FIG.

  The horizontal axis in FIG. 4 indicates the numerical value output by the present apparatus. The closer the value is to 0, the higher the ranking is, and the closer the value is to 1, the lower the ranking is. The vertical axis represents the number of active compounds included in the ranks so far.

  The two lines in FIG. 4 indicate the result of the index according to the present invention on the upper side and the result of the conventional method in which the lower side is ranked by the Tanimoto coefficient.

  It can be seen that the method of the present invention contains a larger amount of the active compound in the higher rank portion than the conventional method. If some of the commercially available compounds are selected using such ranking, the active compound is positioned higher by using the present invention, and it is possible to select more active compounds. Show.

  As described above, the present invention is effective when searching for similar compounds in the field of drug discovery, and is applied to fields such as pharmaceuticals.

It is a figure which shows schematic structure of the similarity calculation processing apparatus by embodiment of this invention. It is a figure which shows the specific structure of the descriptor selection apparatus of FIG. It is a figure used for operation | movement description of the similarity calculation processing apparatus 100 of FIG. The effect of the similarity calculation processing apparatus according to the embodiment of the present invention will be specifically described.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Similarity calculation processing apparatus 101 Training data storage apparatus 102 Predictive data storage apparatus 103 Descriptor selection apparatus 104 Similarity calculation apparatus 105 Calculation result totaling apparatus 201 Descriptor extraction apparatus 202 Data conversion apparatus

Claims (18)

  Storing input training data; storing data for prediction; extracting a portion of a descriptor of the input training data and prediction data; the training data and the prediction data And calculating a similarity for each of the stored prediction data, and calculating a similarity.   2. The similarity calculation processing method according to claim 1, wherein the similarity calculation is performed by calculating a similarity using a Tanimoto coefficient.   3. The similarity calculation processing method according to claim 1, wherein the similarity calculation outputs a maximum value from the calculated similarities.   The similarity calculation processing method according to any one of claims 1 to 3, wherein the similarity is calculated by calculating an average value and a variance by aggregating values corresponding to the stored prediction data, A similarity calculation processing method characterized by outputting a final numerical value using a function of the two values.   5. The similarity calculation processing method according to claim 1, wherein the descriptor is extracted a plurality of times, and the same number of random numbers is generated between the plurality of extraction operations. Similarity calculation processing method.   6. The similarity calculation processing method according to claim 1, wherein the similarity is calculated by outputting a final numerical value by using a function of a distance from a predetermined plane. A similarity calculation processing method characterized by the above.   A training data storage device for storing input training data, a prediction data storage device for storing data for prediction, and a part of descriptors of the input training data and prediction data. At least one descriptor selection device for extracting, a similarity calculation device that is provided corresponding to each of the descriptor selection devices and calculates the similarity between the training data and the prediction data, and the similarity calculation device includes: A similarity calculation processing system comprising: a calculation result aggregating device for calculating a similarity for each data stored in the predicted data storage device.   8. The similarity calculation processing system according to claim 7, wherein the similarity calculation device performs similarity calculation using a Tanimoto coefficient.   9. The similarity calculation processing system according to claim 7 or 8, wherein the similarity calculation device outputs a maximum value from the calculated similarities.   The similarity calculation processing system according to any one of claims 7 to 9, wherein the calculation result aggregation device calculates an average value and a variance by aggregating values corresponding to data stored in the prediction data storage device. A similarity calculation processing system characterized by calculating and outputting a final numerical value using a function of the two values.   The similarity calculation processing system according to any one of claims 7 to 10, wherein each of the descriptor selection devices generates the same number of random numbers between the descriptor selection devices. Degree calculation processing system.   12. The similarity calculation processing system according to claim 7, wherein the calculation result totaling apparatus outputs a final numerical value by using a function of a distance from a predetermined plane. A similarity calculation processing system characterized by that.   Training data storage means for storing input training data, prediction data storage means for storing data for prediction, and part of descriptors of the input training data and prediction data At least one descriptor selecting means for extracting, similarity calculating means provided corresponding to each of the descriptor selecting devices, for calculating the similarity between the training data and the prediction data, and the similarity calculating device A similarity calculation processing program, comprising: a calculation result totaling unit that calculates a similarity for each data stored in the prediction data storage device based on the calculated result.   14. The similarity calculation processing program according to claim 13, wherein the similarity calculation means performs similarity calculation using a Tanimoto coefficient.   15. The similarity calculation processing program according to claim 13, wherein the similarity calculation means outputs the maximum value from the calculated similarities.   The similarity calculation processing program according to any one of claims 13 to 15, wherein the calculation result aggregation means calculates an average value and a variance by aggregating values corresponding to data stored in the prediction data storage device. A similarity calculation processing program characterized by calculating and outputting a final numerical value using a function of the two values.   The similarity calculation processing program according to any one of claims 13 to 16, wherein each of the descriptor selection means generates the same number of random numbers between the descriptor selection means. Degree calculation processing program.   The similarity calculation processing program according to any one of claims 13 to 17, wherein the calculation result aggregation device outputs a final numerical value by using a function of a distance from a predetermined plane. A similarity calculation processing program characterized by that.

Images