JP4840597B2 - Drug discovery multi-target screening device - Google Patents

Description

  The present invention provides an efficient screening apparatus for screening compounds against two or more target proteins in the field of drug discovery.

  A method of screening a compound against one target protein in drug discovery has been established. A major pharmaceutical company uses a system that automatically performs experiments using an apparatus called high-throughput screening (HTS).

  Recently, there has been a movement to search for compounds having activity against two or more targets for one compound, and when trying to examine the activity against multiple target proteins in the above system, all compounds included in the compound library are searched. After investigating the activity against the target protein, the results are analyzed together.

  According to this method, all the compounds contained in the library are examined for activity against individual target proteins, which requires a great deal of time and expense.

  Therefore, in order to improve the screening efficiency, a learning method is used to search in advance for compounds that may act on the target protein on a computer and narrow down candidates. For example, as a method for improving screening efficiency, a drug discovery screening compound selection method using an active learning method has been developed (Non-patent Document 1).

  The active learning method is a learning method in which a learner (computer) actively selects training data to improve prediction accuracy. This is one of the mining techniques for learners to actively select learning data. By selecting the data to be experimented, learning, and predicting unexperimental data, it is possible to repeatedly select the data to be experimented from the unexperienced data. Therefore, sufficient prediction performance can be realized with a small number of experiments. For example, it is described in the description “Active Learning Overview” by Abe and Nakamura described in Japanese magazine “Information Processing”, Vol. 38, No. 7, pp. 558-561 published in 1997.

  Non-Patent Document 1 is attracting attention as a drug discovery screening method that improves the efficiency of learning by active learning and efficiently discovers compounds active against a specific protein from a vast variety of compounds.

However, even in screening (<Non-Patent Document 1>) using the so-called in silico technique, in which the activity of a compound selected by a computer is measured and the target compound is found out, the activity against individual target proteins is examined, and the results are obtained. In terms of matching, it is essentially the same as the above method.
Proceedings of the 32nd Symposium on Structure-Activity Relationship "Drug Screening Based on Active Learning Method Using Descriptor Sampling Method"

  As described above, when screening for compounds that have activity on multiple target proteins, regardless of whether or not a learning method is used, after searching for compounds that have activity on individual target proteins, It has been done to select compounds with activity.

  Therefore, when searching for a compound having activity on a plurality of target proteins simultaneously, the conventional screening method has a significantly low search efficiency. That is, it cannot be said that the conventional apparatus is sufficient as a system for simultaneously viewing the activity of a plurality of targets.

  Therefore, the present invention provides an efficient screening method in searching for compounds having activity with respect to a plurality of target proteins.

  The present invention also provides an apparatus that enables efficient screening in searching for compounds having activity with respect to a plurality of target proteins.

The system according to the present invention comprises:
A system for identifying a compound having activity against two or more target proteins,
A compound library in which a plurality of compounds are arranged at predetermined positions;
A protein library including a plurality of identical plates each having two or more target proteins independently arranged at predetermined positions;
A reaction apparatus for reacting a test compound and a target protein in each reaction region of the plate;
A detection device for detecting a reaction between the test compound and the target protein in the reaction region;
And a control means for assigning a test selection order to the compounds in the compound library.

The screening method according to the present invention comprises:
In a screening method for compounds having activity on two or more target proteins,
Measuring the activity of at least one of the two or more target proteins for compounds sequentially selected from the group of compounds,
A screening method characterized by selecting a compound to be measured next by a learning method or a similarity calculation method.

The screening method according to the present invention comprises:
In a screening method for compounds having activity on two or more target proteins,
Measuring the activity of at least one of the two or more target proteins for compounds sequentially selected from the group of compounds,
Calculate the similarity between the active compound and the individual compounds in the compound library,
This screening method is characterized in that the next compound to be tested is selected based on the similarity calculation result.

  According to the present invention, further efficiency can be achieved when screening a plurality of target proteins.

  FIG. 1 shows a configuration in the present invention.

  This system includes a compound library 101 storing compounds, a protein library 102 storing target proteins, a compound protein mixing device 103 for mixing the compound and the target protein, and a reaction device for performing the reaction after mixing. 104, an activity measurement device 105 as a detection device for receiving the plate after reaction and measuring the activity, a control device 106 as a control means for controlling the entire device, and an experiment based on information from the control device It comprises a compound pick-up device 107 as a distribution means for carrying out the compound to be carried out. Here, the compound protein mixing device 103, the reaction device 104, the activity measuring device 105 and the pickup device 107 are described as separate and independent devices for the sake of simplification of description. There is no need to provide it, and it is also possible to configure as a single experimental apparatus having a function of mixing a compound with a target protein, a function of performing a reaction, a function of measuring activity, and a function of carrying out a compound.

  The compound library 101 and the control device 106 are connected by wiring, and information on the compounds included in the compound library is transmitted from the compound library management device to the control device 106. The compound library 101 and the compound pickup device 107 are connected so that the compounds included in the compound library can be stored in the compound pickup device 107. The protein library 102 and the compound protein mixing apparatus 103 are connected so that the plate storing the target protein can move between them. The compound protein mixing device 103 and the reaction device 104 are connected so that the mixed plate can be moved to the reaction device 104. The reaction device 104 and the activity measuring device 105 are connected so that the plate after the reaction can move to the activity measuring device 105. The activity measuring device 105 and the control device 106 are connected so as to provide activity measurement information of the compound whose activity has been measured. The control device 106 and the compound pickup device 107 are connected so that the selected compound to be tested next determined by the control device 106 is transmitted to the compound pickup device 107. The compound protein mixing apparatus 103 and the compound pickup apparatus 107 can put the selected compound carried out from the compound library 101 by the compound pickup apparatus 107 into the well of the plate in which the target protein provided from the protein library 102 is stored. Are tied together.

  Here, information on compounds included in the compound library may be stored in the control device 106 in advance.

  The compound library and the protein library can be constructed using an apparatus having a usual configuration. For example, a solution of each compound is placed in an ampoule or vial at a position specified by the position information, and a required amount is distributed to a reaction area composed of a plate well or the like by a dispensing device such as a pipetter. That's fine.

  Next, the operation of each apparatus will be described with reference to FIG.

  First, it is assumed that a user arranges a plurality of target proteins in the protein library 102 and arranges compounds whose activity is measured in the compound library 101 (step 201).

  There are no particular restrictions on the order in which the compounds are arranged in the compound library 101. In addition, compound information is simultaneously stored in a compound library management device or the like. What is stored as compound information is a representation of the compound in a learnable form, such as a structure descriptor or a physicochemical constant of the compound. In addition, compound information is stored together with position information in the compound library.

  As a description method of the position information, for example, numbering may be performed sequentially from the end of the plate, or it may be expressed by a set of a plate number and a number inside the plate.

  The protein library 102 is not limited within one plate, but must be arranged in the same order between the plates. That is, the precondition is that the same protein is injected into the first hole in any plate.

  Next, the compound contained in the compound library 101 is sent to the compound pickup device 107, and the information on the compound is sent to the control device 106 together with its position information (step 202).

  Next, the control device 106 learns the stored compound information, selects the compound to be tested next, and transmits the information to the compound pickup device 107 (step 203).

  At this time, as a means for selecting a compound, a method of selecting a compound having a high positive example score as a result of learning (mining), or a similarity with a compound having activity is calculated by a similarity calculation method such as a Tanimoto coefficient. Thus, there is a method of selecting compounds in descending order of similarity. Moreover, it is also possible to select the compound to be measured next by using an active learning method as described in Non-Patent Document 1.

  The learning method used is, for example, learning input / output data using some expression such as a decision tree, decision list, neural network, naive Bayes, Bayesian network, genetic algorithm, regression analysis, support vector machine, etc. Any algorithm can be used. The score is a numerical value of the likelihood of a positive example of each unknown data. For example, the larger the value, the higher the possibility of being a positive example.

  Next, in the compound protein mixing device 103, the selected compound sent from the compound pickup device 107 is mixed in individual wells in which the target proteins sent from the protein library 102 are stored. The mixed plate with the selected compound and target protein is sent to the reactor 104 (step 204).

  Next, in order to examine the activity of the selected compound with respect to the target protein in the reaction apparatus 104, the reaction is performed under conditions corresponding to each target protein. At this time, the reaction conditions of the individual target proteins arranged in the plate may be the same or different. The plate after completion of the reaction is sent to the activity measuring device 105 (step 205).

  The activity measuring device 105 measures the activity of the selected compound against individual target proteins. The obtained activity measurement information is transmitted to the control device 106 together with the position information of the selected compound. Here, the activity measurement information includes, for example, presence / absence of activity, a measured value of activity, and the like.

  Next, the control device 106 accumulates and stores the activity measurement information received from the activity measurement device 105. At this time, as a method of storing the activity information of the compound for each target protein, for example, the information of the measured value of the activity is stored as it is, or when a certain threshold is set and the activity is higher than the threshold, 1 is set. In this case, there is a method of handling as 0.

  In addition, with regard to the experimental results of a selected compound for a plurality of target proteins, for example, the results of individual target proteins are individually memorized, or when activity is recognized in at least one protein, all proteins If no activity is observed in, the value can be stored as 0. The activity measurement information is stored in the control device 106 in association with the position information of the selected compound.

  Next, learning and similarity calculation are performed based on the compound information accumulated in the control device 106 and the newly acquired activity measurement information of the selected compound, and as a result, an experiment is performed among the compounds with high scoring results. The compounds to be tested next are selected in the order from none. Then, the position information of the newly selected compound is sent to the compound pickup device 107 (step 207). After sending the position information of the new selected compound, the processing from Step 203 to Step 207 is repeated until the user ends the processing.

  In addition, although the structure of this invention was demonstrated based on the example by the method by which target protein is accommodated in the well of a plate and puts a test compound in there and reacts, it is not limited to this example in particular. For example, a test tube or a vial may be used as a container for storing the target protein, or a test tube or a vial may be used as a container for reaction, and a compound solution or a protein solution is mixed in the test tube. Can do.

  In the description of the configuration of the present invention, the active learning method is used as a method for selecting the compound whose activity is to be measured next. However, the learning method used in the present invention is particularly limited to the active learning method. Instead, a normal learning method or similarity calculation method can be used.

  EXAMPLES Hereinafter, although an Example is given and this invention is demonstrated more concretely, this invention is not limited to these Examples at all.

Example 1
In Example 1, the compounds contained in 4962 commercially available libraries are used as the compound library, and two receptors, Gaba and Benzodiazepine, are used as the target proteins, and the constitution of the present invention is used to search for compounds having activity in both target proteins. It was confirmed how efficiently the search can be performed when a screening apparatus having the above is used.

  Of the compounds included in the compound library used, 39 compounds have activity only for the Gaba receptor, 6 compounds have activity only for the Benzodiazepine receptor, and compounds that have activity on both Gaba and Benzodiazepine receptors. Are included. In addition, 44 compounds having activity with respect to the Gaba receptor and 11 compounds having activity with respect to the benzodiazepine receptor are included.

  When the proposed apparatus is adapted for the above purpose, it is as follows.

  There are 4962 compounds in the compound library 101, and the plate for storing the target protein stored in the protein library 102 has at least two wells, and each well stores an individual target protein. ing. The well becomes a reaction region, and a compound for measuring the activity is injected and mixed to perform the reaction. Since one plate is used for one kind of compound, a plurality of identical plates in which individual target proteins are stored in wells at predetermined positions are prepared.

  The compound pickup device 107 has a system configuration in which 96 compounds are selected simultaneously. That is, 96 compounds are selected by the compound pick-up device 107, and each compound is injected into the well of the plate containing 96 individual target proteins by the compound protein mixing device 103 and mixed.

  Inside the control device 106, an active learning method (see “Discovery Science and Data Mining”, ISBN4-320-12018-3) using a decision tree as a subordinate learning machine was used to determine a compound to be subjected to the next experiment. As a result of the active learning, the test selection order was determined in the order in which the scores were judged higher from the compounds not yet tested, and the compounds were selected.

The learning data is divided into 10 parts, one of the data sets starts learning as activity known data, the other 9 data sets are treated as activity unknown data, and act on both receptors from the activity unknown data. It was decided to identify the compound to be. The data is divided into (i) G
Inactive compounds for both aba and benzodiapine receptors are divided into 10 parts, (ii) compounds having activity in both Gaba and benzodiazepine receptors are divided into 10 parts, and (iii) compounds having activity only in Gaba receptor are provided in 10 parts Dividing, (iv) After dividing the compound having activity only on the benzodiazepine receptor into 10 parts, the data in 10 parts were prepared by combining 1 part in (i) to (iv). In addition, the part which cannot be divided correctly was distributed appropriately.

  The evaluation was performed based on how early all 5 compounds having activity against both target proteins can be identified. The presence or absence of activity of the selected compound was confirmed by information registered in the database without actually measuring the activity.

There are the following four types of expression methods of activity stored in the control device 106.
(A) When only compounds having activity at both receptors of Gaba and Benzodiazepine are set to 1 and other compounds are expressed as 0 (B) Only compounds having activity at the Gaba receptor are set to 1, and other compounds are When expressed as 0 (C) When a compound having activity at the benzodiazepine receptor is 1, only when the other compound is expressed as 0 (D) Only a compound having activity at one of the receptors of Gaba and benzodiapine Is expressed as 1 and other compounds are expressed as 0. (A) directly identifies compounds acting on both receptors, (B) acts on both receptors using compounds acting on Gaba receptors as clues. When identifying a compound, (C) is a compound that acts on the Benzodiazepine receptor. When identifying a compound that acts on both receptors as a key, (D) corresponds to identifying a compound that acts on both receptors using a compound acting on either receptor as a clue.

  In the case of (A) and (C), if the number is divided into 10, a data set that does not include data having a value of 1 appears in the initial stage of learning, and the performance cannot be accurately measured. Therefore, learning is started from one theoretically valid data set divided into five, and data is selected from the remaining four data sets.

Under such a problem setting, the number of compounds required to identify all active compounds for both protein receptors is:
(A) is a 1473 compound,
(B) 976 compounds,
In (C), 880 compounds,
In (D), it was 784 compounds.

  When an assay experiment is performed as it is with HTS as in the conventional method, it is necessary to experiment on all compounds, and therefore it is necessary to experiment on all 4962 compounds. When searching only for compounds having activity against two targets from the beginning (in the case of (A)), after searching for compounds that act on 1473 compounds and Gaba receptors, compounds that also act on Benzodiapine receptor from among them When searching for 976 compounds, searching for compounds that act on the benzodiazepine receptor and then searching for compounds that also act on the gaba receptor, 880 compounds, compounds that act on either the gaba receptor or the benzodiapine receptor In this case, it was possible to search for 784 compounds, and to show the effectiveness of the apparatus of the present invention.

(Example 2)
The control device 106 can use not only the active learning method as described above but also a normal learning method as means.

  In this case, as a processing method in the control device 106, ranking is performed in the order of scores that have characteristics of a compound having activity against the target protein as a result of learning. Experiments were performed when using a normal learning method with the same system configuration as above.

  As a subordinate learning machine, ensemble learning using decision trees was performed. As typical ensemble learning, there are, for example, bagging and boosting.

  As an evaluation method at that time, the number of compounds having activity at both receptors in the first 96 compounds selected after learning was compared.

  As a result, the following results were obtained.

In the case of (A) 0 compound in the case of (B) 1 compound in the case of (C) 2 compound in the case of (D) As a result, (D) is searching for the most active compounds at an early stage. Thus, it was found that the learning method in the control device 106 is not only the active learning method but also the normal learning method is effective.

(Example 3)
The control device 106 can use not only a learning method but also a similarity calculation method such as a Tanimoto coefficient.

  In this case, as a processing method in the control device 106, the similarity between the compound having high activity and the compound contained in the compound library 101 is calculated, and ranking is performed in descending order of the similarity. As shown by the performance of the related compound selection method in <Non-Patent Document 1>, the effectiveness of such a method has been shown to have the same excellent effect if not as much as the active learning method.

Example 4
The proposed method can be easily extended not only when searching for two targets but also when searching for compounds that act simultaneously on three or more targets. In the expansion method, the number of targets in the protein library 102 is set to 3 or more, and the control device 106 applies 1 to a compound whose activity is recognized in at least one target, and a compound whose activity is not recognized for any target. This is because it may be expressed as 0. By expressing in this way, the problem setting is the same as that of the above-described apparatus, so that it can be said that the same effect is obtained by learning or similarity search.

It is a block diagram of an example of the system which concerns on this invention. It is a figure which shows the flow of an example of the screening method which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 101: Compound library 102: Protein library 103: Compound protein mixing apparatus 104: Experimental apparatus 105: Activity measuring apparatus 106: Control apparatus 107: Compound pick-up apparatus

Claims (5)

A system for identifying a compound having activity against two or more target proteins,
A compound library in which a plurality of compounds are arranged at predetermined positions;
A protein library including a plurality of identical plates each having two or more target proteins independently arranged at predetermined positions;
A reaction apparatus for reacting a test compound and a target protein in each reaction region of the plate;
A detection device for detecting a reaction between the test compound and the target protein in the reaction region;
Control means for assigning a test selection order to the compounds in the compound library ;
A dispensing means for removing a test compound from the compound library and dispensing it into the reaction region of the plate;
Have
The control means instructs the distribution means to distribute the test compound selected based on the test selection order to a new analysis plate,
The control means selects a compound selected based on the test selection order using the position information of the compound in the compound library using the position information of the compound in the compound library, and is a compound at the selected position. Command distribution to a new plate,
The test selection order is determined by learning or similarity calculation using detection data from the detection device and information on each compound in the compound library,
As a method for expressing the activity for the target protein when performing the learning or similarity calculation, a value indicating that the target protein has activity with respect to at least one target protein, and a value indicating that the target protein does not have activity The system according to claim 1, wherein the system is expressed by binary values. In a screening method for compounds having activity on two or more target proteins,
Measuring the activity against at least one of the two or more target proteins with respect to compounds sequentially selected from the compound group included in the compound library ,
The compound to be measured next is selected by the learning method or the similarity calculation method ,
As a method of expressing the activity against the target protein when performing learning or similarity calculation, a value indicating that it has activity against at least one target protein, and a value indicating that it does not have activity against any target protein A screening method characterized in that the expression is performed using the binary value . The screening method according to claim 2 , wherein the learning method selects a compound to be measured next by active learning. 4. The screening method according to claim 2 , wherein the learning machine of the learning method is a decision tree algorithm. The similarity between the individual compounds included in the target protein at least one also compound with said compound library have a activity in the process is calculated,
The screening method according to any one of claims 2 to 4 for selecting a compound that next experiment based on the calculation result of the similarity.

Images