Mediator search method, mediator search program, mediator search apparatus, mediator, and mediator mixture

Abstract

To efficiently search for mediators. [Solution] A mediator search method is provided, which involves a computer learning training data that associates experimental values ​​of the electrical properties of a solution containing a mediator and cells with the mediator; estimating the electrical properties of each of a plurality of mediators not included in the training data based on the results of the learning; and adding the experimental values ​​of the electrical properties of at least some of the plurality of mediators whose electrical properties have been estimated, along with the mediators themselves, to the training data, repeating this process until the experimental values ​​of the electrical properties meet a criterion, thereby searching for a mediator whose experimental values ​​meet the criterion.

Description

[Technical Field]

[0001] The present invention relates to a mediator search method, a mediator search program, a mediator search apparatus, a mediator, and a mediator mixture. [Background technology]

[0002] Technologies are being explored that utilize electrochemical reactions between bacteria and electrode surfaces to detect and kill bacteria. These electrochemical reactions are mediated by redox molecules called mediators, which act as intermediaries in the redox reaction of bacteria. For example, if the mediator is a reducing molecule, a large amount of electrons are removed from the bacteria through the reduction reaction, resulting in a large current flow. By detecting this current, the presence of bacteria can be detected. On the other hand, if the mediator is an oxidizing molecule, the bacteria are killed when a large amount of electrons are injected into them through the oxidation reaction.

[0003] The electric current mediated by a mediator is also called the mediating current. A mediator with a large mediating current can efficiently detect and kill bacteria. Non-patent document 1 proposes a method for experimentally determining the mediating current for several commercially available mediators as a way to search for such mediators with large mediating currents.

[0004] However, the method described in Non-Patent Document 1 involves haphazardly measuring mediating currents in experiments with commercially available mediators, which is extremely inefficient in terms of the time, cost, and effort required for reagent preparation and data analysis. Furthermore, since the mediators being searched for are limited to a few commercially available types, finding mediators with large mediating currents among them relies on chance. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] O. Simoska, et al., “Understanding the properties of phenazine mediators that promote extracellular electron transfer in Escherichia coli,” Journal of The Electrochemical Society, Vol. 168, No. 2, p. 025503, 2021, IOP Publishing. [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] In one aspect, the present invention aims to efficiently search for mediators. [Means for solving the problem]

[0007] According to one aspect of the present invention, a mediator search method involves a computer learning training data that associates experimental values ​​of the electrical properties of a solution containing a mediator and cells with the mediator; estimating the electrical properties of each of a plurality of mediators not included in the training data based on the results of the learning; and adding the experimental values ​​of the electrical properties of at least some of the plurality of mediators whose electrical properties have been estimated, along with the mediators themselves, to the training data, repeating this process until the experimental values ​​of the electrical properties satisfy a criterion, thereby searching for a mediator whose experimental values ​​satisfy the criterion.

[0008] In the mediator search method described above, the electrical characteristics may also be the characteristics of the current flowing through the solution.

[0009] In the mediator search method described above, the electrical characteristic may be the maximum value of the time variation of the current.

[0010] In the mediator search method described above, the experimental values ​​of the electrical characteristics may be measured values ​​obtained by measuring each of the multiple wells holding the solution for each type of mediator.

[0011] In the mediator search method, some of the plurality of wells hold the solution containing the mediators of the same type, and the experimental value of the electrical property may be the average value of the measured values in the some wells.

[0012] In the mediator search method, the learning includes generating a function including a kernel function indicating the similarity between the mediators included in the learning data and mediators not included in the learning data, and returning an estimated value of the electrical property, and estimating the electrical property may be performed by estimating the electrical properties of each of the plurality of compounds not included as the mediators in the learning data among the plurality of compounds included in the database based on the function generated as a result of the learning.

[0013] In the mediator search method, in the database, the compound is associated with information indicating whether the compound is commercially available, and the computer further identifies the commercially available compounds among the plurality of compounds stored in the database, and the calculation of the estimated value may be performed on the identified compounds.

[0014] In the mediator search method, the computer may present the compounds in the order of the estimated electrical properties being good.

[0015] In the mediator search method, the computer further extracts the compounds whose electrical properties are higher than a predetermined rank as the some mediators, and adding to the learning data may be performed by adding the extracted some mediators and the experimental values of the electrical properties of the some mediators to the learning data.

[0016] In the mediator search method, the solution may include the mediator and Escherichia coli composed of the cells.

[0017] The above mediator search method may further include generating a predictive model that predicts the experimental values ​​of the electrical properties from the chemical properties of at least some of the mediators among the plurality of mediators, and identifying mediators as adjuvants in which the experimental values ​​of the electrical properties are greater than the predicted values ​​from the predictive model.

[0018] In the mediator search method described above, the electrical characteristics may also be the maximum value of the current flowing through the solution at a predetermined time.

[0019] In the mediator search method described above, the chemical properties may be properties related to the mediator's ability to permeate the cell membrane.

[0020] In the mediator search method described above, the chemical property may be the number of rotatable bonds or the number of hydrogen bond donors.

[0021] In the mediator search method described above, in generating the prediction model, a prediction model may be generated for each of several different chemical properties, and in identifying the mediator as an adjuvant, a mediator whose experimental value of the electrical properties is greater than all of the predicted values ​​from the prediction models for each of the several chemical properties may be identified as an adjuvant.

[0022] According to another aspect of the present invention, the mediator comprises thionine or chlorazole black E.

[0023] According to another aspect of the present invention, the mediator mixture comprises the mediator and cells.

[0024] According to yet another aspect of the present invention, the mediator is either BB24 (Basic Blue 24) or PYO (Pyocianine), and functions as an adjuvant.

[0025] According to yet another aspect of the present invention, the mediator search program causes a computer to perform a process of searching for mediators whose experimental values ​​satisfy the criteria by learning training data that associates experimental values ​​of the electrical properties of a solution containing mediators and cells with the mediators, estimating the electrical properties of each of a plurality of mediators not included in the training data based on the results of the learning, and adding the experimental values ​​of the electrical properties of at least some of the plurality of mediators whose electrical properties have been estimated, along with the mediators, to the training data, repeating this process until the experimental values ​​of the electrical properties satisfy the criteria.

[0026] According to yet another aspect of the present invention, the mediator search device includes a learning unit that learns learning data relating experimental values ​​of the electrical properties of a solution containing mediators and cells with the mediators; an estimation unit that estimates the electrical properties of each of a plurality of mediators not included in the learning data based on the results of the learning; an addition unit that adds experimental values ​​of the electrical properties of at least some of the plurality of mediators whose electrical properties have been estimated, along with the mediators, to the learning data; and a determination unit that determines whether the added experimental values ​​meet a criterion, wherein if the learning unit determines that the experimental values ​​do not meet the criterion, it performs the learning with the added learning data. [Effects of the Invention]

[0027] According to the present invention, mediators can be efficiently searched for. [Brief explanation of the drawing]

[0028] [Figure 1] Figure 1 is a diagram showing the configuration of the mediator search system according to the first embodiment. [Figure 2] Figure 2 is a schematic diagram showing the data structure of the training data according to the first embodiment. [Figure 3] Figure 3 is a schematic diagram showing the data structure of the compound database according to the first embodiment. [Figure 4]Figure 4 is a schematic diagram showing the data structure of the compound list according to the first embodiment. [Figure 5] Figure 5(a) is a schematic diagram of the electrochemical measuring apparatus according to the first embodiment, and Figure 5(b) is an enlarged cross-sectional view of the well plate according to the first embodiment. [Figure 6] Figure 6(a) is an enlarged top view of the bottom surface and surrounding area of ​​the well according to the first embodiment, Figure 6(b) is an enlarged top view of each electrode in the resin-covered portion, and Figure 6(c) is an enlarged bottom view of the bottom surface of the well plate according to the first embodiment. [Figure 7] Figures 7(a) and 7(b) are schematic diagrams showing the electrode reaction of the solution held in the well according to the first embodiment. [Figure 8] Figure 8 is a functional configuration diagram of the mediator search device according to the first embodiment. [Figure 9] Figure 9 is a schematic diagram of the function μc(x) generated in the first embodiment. [Figure 10] Figure 10 is a flowchart showing an example of the processing performed by the mediator search device according to the first embodiment. [Figure 11] Figure 11 schematically shows the compounds extracted when the flowchart in Figure 10 is applied to the initial training data. [Figure 12] Figure 12 is a box plot diagram obtained from the results in Figure 11. [Figure 13] Figure 13 shows the maximum mediating current values ​​for each compound in Figure 12, obtained by measuring them with an electrochemical measuring device. [Figure 14] Figure 14 is a box plot diagram obtained from the results in Figure 13. [Figure 15] Figure 15 is a diagram (part 1) obtained by investigating the correlation between the electrical characteristics and chemical properties of the mediator in the second embodiment. [Figure 16] Figure 16 is a diagram (part 2) obtained by investigating the correlation between the electrical characteristics and chemical properties of the mediator in the second embodiment. [Figure 17]Figure 17 is a diagram (part 3) obtained by investigating the correlation between the electrical characteristics and chemical properties of the mediator in the second embodiment. [Figure 18] Figures 18(a) to (c) show the relationship between the mediator concentration and the adjuvant effect, which was experimentally determined in the second embodiment. [Figure 19] Figure 19 is a functional configuration diagram of the mediator search device according to the second embodiment. [Figure 20] Figure 20 is a flowchart showing an example of the processing performed by the mediator search device according to the second embodiment. [Figure 21] Figure 21 is an example of a hardware configuration diagram of a mediator search device according to the first and second embodiments. [Modes for carrying out the invention]

[0029] (First Embodiment) Embodiments of the present invention will be described below with reference to the drawings. Similar elements are denoted by the same reference numerals, and their descriptions are omitted.

[0030] Figure 1 is a diagram showing the configuration of the mediator search system according to this embodiment. The mediator search system 1 is a system for searching for mediators with large mediating currents, and comprises an electrochemical measuring device 2 and a mediator search device 3.

[0031] The electrochemical measuring device 2 is a device that automatically measures the electrical properties of a solution containing mediators and bacteria. In this example, bacteria are used as the measurement target, but the cells that make up bacteria may also be used as the measurement target. Based on the measurement results of the electrochemical measuring device 2, the user generates training data 6 including those measurement results. Although Figure 1 illustrates a case where there is only one electrochemical measuring device 2, the number of electrochemical measuring devices 2 is not limited to this, and multiple electrochemical measuring devices 2 may be provided.

[0032] Figure 2 is a schematic diagram showing the data structure of the training data 6 according to this embodiment. As shown in Figure 2, the training data 6 is data that associates a "mediator" with the "maximum slope of the current (nA / s)" of the solution of that "mediator".

[0033] In this example, a fingerprint showing the structure of the "mediator" is used. The "maximum slope of the current (nA / s)" is an example of an experimental value of the electrical properties of a solution containing cells constituting bacteria, etc., and the mediator, and is the maximum value of the time change of the mediating current after the start of measurement. It is empirically known that the mediating current reaches a maximum after a certain amount of time has elapsed since the start of measurement, and then decreases. However, since the mechanism by which the mediating current exhibits such a time change has not been elucidated, it is uncertain whether the maximum value of the mediating current will be reproducible even with the same mediator. On the other hand, according to the inventor's experience, the time change amount defined as the time derivative of the mediating current shows a certain degree of reproducibility. Therefore, in this example, the maximum value of the time change of the mediating current is adopted as the electrical property of the solution.

[0034] The method for generating training data 6 is not particularly limited. For example, when the user inputs a CAS number that uniquely identifies each mediator into the search device 3, the search device 3 executes the Python library PubChemPy and generates the SMILES for that mediator. SMILES is a string that indicates the structure of the mediator. Furthermore, the search device 3 generates MACCS keys as fingerprints corresponding to the SMILES by executing the Python library RDkit.

[0035] The electrochemical measuring device 2 then automatically measures the electrical properties of the mediator solution corresponding to the MACCS Keys. Subsequently, the user associates the measured value, "maximum current slope (nA / s)," with the MACCS Keys and stores it in the training data 6. Note that while the fingerprints such as MACCS Keys are bit strings, in Figure 2, for clarity, the structure of each mediator is identified by strings such as "RF" and "FMN." These strings are sequences of letters that appear in the English spelling of the mediator's name.

[0036] Refer to Figure 1 again.

[0037] The mediator search device 3 is a computer such as a PC (Personal Computer) or a server. For example, the mediator search device 3 learns from training data 6 and, based on the results of that learning, estimates the electrical characteristics of multiple mediators not included in training data 6.

[0038] The mediators to be used for electrical property estimation are extracted from the compound database 4 by the mediator search device 3.

[0039] Figure 3 is a schematic diagram showing the data structure of the compound database 4 according to this embodiment. As shown in Figure 3, the compound database 4 is a database that associates the structure of at least one candidate compound for mediator with information indicating whether the compound is commercially available. An example of such a database is ZINC20.

[0040] The mediator search device 3 extracts commercially available compounds from the compound database 4 and estimates the electrical properties of those compounds as described above.

[0041] Refer to Figure 1 again. Furthermore, the mediator search device 3 presents the user with a compound list 7 showing multiple compounds whose electrical properties have been estimated.

[0042] Figure 4 is a schematic diagram showing the data structure of the compound list 7 according to this embodiment. As shown in Figure 4, the compound list 7 is a list that associates "compounds" with "ranks". In this example, the string described with reference to Figure 2 is used as the "compound". The "rank" indicates the rank when multiple compounds whose electrical characteristics have been estimated by the mediator search device 3 are arranged in order of their electrical characteristics from best to worst. For example, if the maximum value of the time change of the mediating current is used as the electrical characteristic as described above, the compound list 7 is a list in which the "compounds" are arranged in descending order of this maximum value.

[0043] Refer to Figure 1 again. The user will again use the electrochemical analyzer 2 to perform experiments on water-soluble compounds from among the multiple compounds included in compound list 7, and measure the electrical properties of the solution containing the compound and cells that make up bacteria, etc. After that, the user will add the electrical properties obtained from the measurement and the compound to training data 6, and repeat each of the above processes.

[0044] The mediator search system 1 searches for compounds with electrical properties that satisfy predetermined criteria by performing Bayesian optimization through repeated processing.

[0045] Figure 5(a) is a schematic diagram of the electrochemical measuring device 2 according to this embodiment. As shown in Figure 5(a), the electrochemical measuring device 2 comprises a well plate 10 and a measuring unit 11.

[0046] The well plate 10 is a resin plate in which multiple wells 12, each holding a solution containing mediators and bacteria, are arranged in a matrix when viewed from above. The number of wells 12 is not particularly limited, but in this example, 96 wells 12 are provided in the well plate 10. Each well 12 holds a solution of a different type of mediator.

[0047] The measurement unit 11 is a measurement circuit that simultaneously measures the electrical properties of the solution for each well 12. The electrical property to be measured is the maximum value of the time variation of the mediating current, as shown in the training data 6 (see Figure 2). Because measurements are performed simultaneously for each well 12 in this way, the electrochemical measuring device 2 can measure the electrical properties of solutions with different types of mediators in high throughput. The experimental values ​​of the electrical properties in the training data 6 (see Figure 2) are the measured values ​​obtained by the measurement unit 11 measuring the electrical properties of each well 12 in this manner.

[0048] Furthermore, some of the 96 wells 12 may hold solutions containing the same type of mediator. The measurement unit 11 may then acquire the average value of the measured electrical properties in some of the wells 12 as an experimental value, and the user may store this experimental value in the training data 6. For example, four of the 96 wells 12 may hold solutions containing the same type of mediator, and the average value of the electrical properties of those solutions across the four wells 12 may be used as the experimental value. This allows for the acquisition of experimental values ​​with reduced variation between wells 12, thereby improving the accuracy of the training data 6 using these experimental values.

[0049] Figure 5(b) is an enlarged cross-sectional view of the well plate 10 according to this embodiment. As shown in Figure 5(b), the well plate 10 has opposing upper surfaces 10a and lower surfaces 10b. A well 12 with a bottom surface 12a is formed on the upper surface 10a.

[0050] Figure 6(a) is an enlarged top view of the bottom surface 12a and its surroundings of the well 12 according to this embodiment. As shown in Figure 6(a), a counter electrode 13c, a reference electrode 13r, and a working electrode 13w are provided on the bottom surface 12a at intervals from each other. These electrodes 13c, 13r, and 13w are electrically connected to a measuring unit 11 (see Figure 5(a)), and the measuring unit 11 measures the electrical properties of the solution using the three-electrode method.

[0051] Furthermore, an insulating resin 14 is formed on the bottom surface 12a, covering a portion of each of the electrodes 13c, 13r, and 13w.

[0052] Figure 6(b) is an enlarged top view of each electrode 13c, 13r, and 13w in the portion covered by the resin 14. As shown in Figure 6(b), wiring 15 is connected to each electrode 13c, 13r, and 13w, and through holes 10c, 10r, and 10w are formed at the ends of each wiring 15. Each through hole 10c, 10r, and 10w is formed in the well plate 10 so as to penetrate from the bottom surface 12a to the lower surface 10b (see Figure 5(b)). Furthermore, each through hole 10c, 10r, and 10w is filled with the aforementioned resin 14, thereby preventing the solution held in the well 12 from leaking out through the through holes 10c, 10r, and 10w.

[0053] Figure 6(c) is an enlarged bottom view of the bottom surface 10b of the well plate 10 according to this embodiment. As shown in Figure 6(c), the bottom surface 10b is provided with a conductive pad 16c for the counter electrode, a conductive pad 16r for the reference electrode, and a conductive pad 16w for the working electrode. These conductive pads 16c, 16r, and 16w are electrically connected to each electrode 13c, 13r, and 13w (see Figure 6(a)) via a conductive film (not shown) formed on the inner surface of each through hole 10c, 10r, and 10w.

[0054] The measuring unit 11 (see Figure 5(a)) applies a predetermined potential to each electrode 13c, 13r, and 13w via a plurality of conductive pins (not shown) that contact each conductive pad 16c, 16r, and 16w, thereby measuring the electrical properties of the solution held in the well 12.

[0055] Figures 7(a) and 7(b) are schematic diagrams showing the electrode reaction of the solution held in well 12 according to this embodiment. Of these, Figure 7(a) is a schematic diagram when the mediator is a reducing molecule.

[0056] As shown in Figure 7(a), the solution contains mediator 21, a reducing molecule, and bacteria 20. While bacteria 20 are not particularly limited, the following example uses E. coli as the bacterium 20. In this case, mediator 21, the reducing molecule, absorbs electrons e from bacteria 20. - It steals the electrons e from the working electrode 13w.- This supplies current. Therefore, the current flows from the working electrode 13w toward the solution.

[0057] On the other hand, Figure 7(b) is a schematic diagram of the case where the mediator is an oxidizing molecule. In this case, the mediator 22, which is an oxidizing molecule, receives electrons e from the counter electrode 13c. - It steals the electrons e from bacteria 20. - This supplies current. Therefore, the current flows from the solution toward the counter electrode 13c.

[0058] Figure 8 is a functional configuration diagram of the mediator search device 3 according to this embodiment. As shown in Figure 8, the mediator search device 3 comprises a control unit 30, a storage unit 31, and a communication unit 32.

[0059] The memory unit 31 is a processing unit that stores the aforementioned learning data 6 (see Figure 2). The communication unit 32 is an interface for connecting the mediator search device 3 to the network 5 (see Figure 1).

[0060] The control unit 30 is a processing unit that controls each part of the mediator search device 3. As an example, the control unit 30 includes an acquisition unit 33, a determination unit 34, a learning unit 35, a first identification unit 36, an estimation unit 37, a list presentation unit 38, a second identification unit 39, an extraction unit 40, and an addition unit 41.

[0061] The acquisition unit 33 acquires the learning data 6 (see Figure 2) and stores it in the storage unit 31. For example, when a user instructs the mediator search device 3 to acquire the learning data 6 by operating an input device (not shown), the acquisition unit 33 acquires the learning data 6 previously created by the user via the network 5 (see Figure 1) or the device's media reader.

[0062] The determination unit 34 determines whether there is a mediator included in the learning data 6 whose experimental value of the electrical property of the solution satisfies the criterion. The criterion is stored in the storage unit 31 by the user in advance. For example, when the electrical property is the "maximum slope of current (nA / s)" as shown in FIG. 2, the user determines in advance the reference value of the "maximum slope of current (nA / s)" that can efficiently detect and sterilize bacteria, and stores it in the storage unit 31. Then, when the learning data 6 includes a mediator whose experimental value of the "maximum slope of current (nA / s)" is greater than or equal to the reference value, the determination unit 34 determines that the criterion is satisfied, and when it is not, the determination unit 34 determines that the criterion is not satisfied.

[0063] The learning unit 35 is a processing unit that learns the learning data 6. Hereinafter, the learning data 6 is represented as {x i , y i}. (i=1,…,N) However, x i is the structure of the mediator, and y i is the experimental value of the electrical property of the solution containing the mediator. Also, N is the number of experimental values included in the learning data.

[0064] When the structure of the mediator not included in the learning data 6 is represented by x, the learning unit 35 generates the mean μ c (x) and the variance σ c (x) of the estimated value of the electrical property of the mediator whose structure is represented by x by learning the learning data 6. The mean μ c (x) and the variance σ c (x) are functions estimated by Gaussian process regression, and are represented by the following equations (1) and (2), respectively.

[0065]

Equation

[0066]

Equation

[0067] However, in equations (1) and (2), I is the N×N identity matrix. Also, the y shown in bold in equation (1) is each y in the training data 6 as shown in equation (3) below. i This is a column vector whose elements are (i=1,…,N).

[0068]

number

[0069] Furthermore, the function k(x,x') in equation (2) is a kernel function that indicates the similarity between structure x and structure x', and is expressed by the following equation (4).

[0070]

number

[0071] The kernel function k(x,x') is a function whose value is closer to 1 the more similar the structures x and x' are. Using this kernel function k(x,x'), the function k(x) shown in bold in equation (1) can be expressed as shown in equation (5).

[0072]

number

[0073] Furthermore, K in equations (1) and (2) is an N×N matrix, and can be expressed using the kernel function k(x,x') as shown in equation (6).

[0074]

number

[0075] Note that σ in equations (1) and (2) and η in equation (4) are hyperparameters, which are pre-optimized by the learning unit 35 using an appropriate optimization algorithm.

[0076] The learning unit 35 performs the learning process using each value x included in the learning data 6, as shown in equations (1) to (6) above. i , y i Each function μ c (x), σ c This refers to generating (x). Each function μ c (x), σ c The algorithm for generating (x) is not particularly limited. In this embodiment, the learning unit 35 executes PHYSBO, a type of Python library, to generate each function μ c (x), σ c Generate (x).

[0077] Figure 9 shows the function μ generated in this embodiment. c This is a schematic diagram of (x). The curve shown by the solid line in Figure 9 represents the mean μ c (x) is shown. The black circles on the curve indicate experimental values ​​included in training data 6. The hatched region represents the variance σ. c (x) is shown.

[0078] Refer to Figure 8 again. The first identification unit 36 ​​refers to the compound database 4 (see Figure 3) and identifies commercially available compounds from among the multiple compounds stored in the compound database 4.

[0079] The estimation unit 37 estimates the electrical properties of multiple mediators not included in the training data 6 based on the learning results of the learning unit 35. In this example, the estimation unit 37 estimates the average value of the estimated electrical properties of compounds identified as commercially available in the compound database 4. For example, if the structure of a commercially available compound is represented by x, the estimation unit 37 estimates that the average value of the predicted electrical properties of that compound is μ c We estimate that (x) is true.

[0080] As shown in equations (1) and (5), the mean value μ c (x) represents the structure x of the compound included in training data 6. i The kernel function k(x) shows the similarity between (i=1,…,N) and the structure x of a compound not included in training data 6. iThis includes (x). Therefore, the electrical properties of compounds not included in training data 6 can be estimated according to their similarity to the compounds included in training data 6.

[0081] The list display unit 38 is a processing unit that displays the compound list 7 (see Figure 4). For example, the list display unit 38 displays the estimated μ c Compound list 7 is created by arranging the compounds in descending order of (x), and this list is displayed on a display device such as a liquid crystal display (not shown) and presented to the user.

[0082] Note that the list display unit 38 is μ c It is not necessary to include all compounds for which (x) was estimated in compound list 7. For example, the list display unit 38 displays μ c Of the multiple compounds for which (x) was estimated, only some compounds ranked higher than a predetermined rank may be included in compound list 7. The predetermined rank is not particularly limited and may be, for example, rank 50. This improves the readability of compound list 7, allowing users to quickly identify mediators with good electrical properties.

[0083] The second identification unit 39 identifies water-soluble compounds from among multiple compounds included in the compound list 7 by accessing a predetermined web server via the network 5. The web server stores information that associates a compound with information indicating whether that compound is water-soluble. Such information is stored on a web server accessible via a URL such as "https: / / pubchem.ncbi.nlm.nih.gov / ". The second identification unit 39 identifies water-soluble compounds by referring to this information.

[0084] The extraction unit 40 extracts compounds that have been identified as water-soluble and whose estimated electrical properties are higher than a predetermined rank. For example, if the number of compounds in the compound list 7 is 50 as described above, the extraction unit 40 may extract the top 10 compounds by setting the predetermined rank to "10th place".

[0085] The user then uses electrochemical measuring device 2 (see Figure 1) to measure the electrical properties of the extracted compound and obtains experimental values ​​of those electrical properties.

[0086] The additional unit 41 receives input from the user, including the extracted compounds and experimental values ​​of their electrical properties, and adds them to the training data 6.

[0087] Next, we will describe an example of the processing performed by the mediator search device 3.

[0088] Figure 10 is a flowchart showing an example of the processing performed by the mediator search device 3 according to this embodiment. This flowchart starts, for example, when the mediator search device 3 receives a user's instruction to start the search.

[0089] First, the acquisition unit 33 acquires the learning data 6 (see Figure 2) and stores it in the storage unit 31 (step S11).

[0090] Next, the determination unit 34 determines whether any of the mediators included in the learning data 6 meet the experimental values ​​for the electrical properties of the solution (step S12). If it is determined that the criteria are met (YES), the process ends.

[0091] On the other hand, if it is determined that the criteria are not met (NO) in step S12, the process proceeds to step S13.

[0092] In step S13, the learning unit 35 learns the learning data 6, thereby determining the functions μ in equations (1) and (2) described above. c (x), σ c Generate (x).

[0093] Next, the first identification unit 36 ​​refers to the compound database 4 (see Figure 1) and identifies a commercially available compound from among the multiple compounds stored in the compound database 4 (step S14).

[0094] Next, the estimation unit 37 estimates the electrical properties of the compound that has been identified as being commercially available (step S15). For example, if the structure of the compound that has been identified as being commercially available is represented by x, the estimation unit 37 estimates that the average value of the predicted electrical properties of that compound is μ c We estimate that (x) is true.

[0095] Next, the list presentation unit 38 displays the estimated mean value μ c A compound list 7 is created by arranging the compounds in descending order of (x), and this list is displayed on a display device such as a liquid crystal display (not shown) and presented to the user (step S16). For example, the list display unit 38 displays the average value μ c When the compounds are sorted in descending order of (x), compounds ranked 50th or higher are included in compound list 7 and presented to the user. This allows the user to identify compounds that are presumed to have good electrical properties.

[0096] Next, the second identification unit 39 accesses a predetermined web server via the network 5 to identify water-soluble compounds from among the multiple compounds included in the compound list 7 (step S17).

[0097] Next, the extraction unit 40 extracts compounds from among those identified as water-soluble that have an estimated electrical property value higher than a predetermined rank (step S18). For example, the extraction unit 40 extracts the top 10 compounds from among those identified as water-soluble in the compound list 7.

[0098] The extracted compound is guaranteed to be commercially available in step S14. Therefore, the user can actually obtain the extracted compound and obtain experimental values ​​of its electrical properties using the electrochemical measuring device 2.

[0099] Furthermore, since the compound extracted in step S18 is water-soluble, it can be assured that the user can dissolve the compound in an electrolyte to create a solution and then inject that solution into well 12 (see Figure 5(a)).

[0100] Subsequently, the addition unit 41 receives input from the user, consisting of the compound extracted by the extraction unit 40 and experimental values ​​of the electrical properties measured by the electrochemical measuring device 2 for that compound. The addition unit 41 then updates the learning data 6 by associating the compound with the experimental values ​​and adding them to the learning data 6 (step S19).

[0101] Subsequently, the process returns to step S11, and steps S11 to S19 are repeated until a mediator included in the training data 6 is found whose experimental values ​​for the electrical properties of the solution meet the criteria, and a "YES" result is determined in step S12.

[0102] This concludes the explanation of an example of the processing performed by the mediator search device 3.

[0103] According to the embodiment described above, the following processes are repeated: learning of the learning data 6 (step S13), estimation of electrical properties based on the learning (step S15), and addition of experimental values ​​of the compounds whose electrical properties have been estimated to the learning data 6 (step S19). Since the learning data 6 is updated in step S19 each time these processes are repeated, the likelihood of discovering compounds (mediators) whose experimental values ​​of the electrical properties of the solution meet the criteria increases. As a result, for example, it is possible to discover mediators that conduct a sufficiently large mediating current to efficiently detect or kill bacteria.

[0104] Furthermore, the experimental values ​​added to the training data 6 in step S19 are values ​​obtained by the electrochemical measuring device 2, which is capable of high-throughput measurement that simultaneously measures multiple wells 12 (see Figure 5(a)). Therefore, even if multiple compounds are extracted in step S18, the electrical properties of the solutions of those compounds can be measured simultaneously by the electrochemical measuring device 2, and compounds with electrical properties that meet the criteria can be discovered in a short time.

[0105] Next, we will describe an example. In this example, E. coli was used as the bacterium. In addition, as initial training data 6 before running the flowchart in Figure 10, we used data relating the following 11 compounds and their experimental electrical properties:

[0106] RF (Riboflavin), FMN (Flavin Mononucleotide), TB (Toluidine Blue), MB (Methylene Blue), BB24 (Basic Blue 24), NB (Nile Blue), AS (1-Acetylisatin), Isatin (Isatin), 4FHD (4-Fluoro-1H-indole-2,3-Dione), CV (Crystal Violet), Saf (Safranin).

[0107] Furthermore, the electrolyte used to dissolve these compounds consists of 2.5 g / L of NaHCO3, 0.08 g / L of CaCl2·2H2O, 1 g / L of NH4Cl, 0.2 g / L of MgCl2·6H2O, 10 g / L of NaCl, 0.5 g / L of Yeast Extract, and HEPES(C8H 18 Defined Medium (DM) containing N2O4S) at various concentrations of 7.2 g / L was used. The concentration of each compound in the solution was set to 100 μM. The concentration of E. coli in the solution was set to OD. 600 The value was set to =1.0. As mentioned above, the maximum slope of the mediating current (nA / s) was adopted as the electrical property of such a solution. The experimental value of the maximum slope of the mediating current (nA / s) was automatically measured using electrochemical measuring device 2, and the initial training data 6 was created by associating this value with each compound.

[0108] Figure 11 schematically shows the compounds extracted in step S18 when the flowchart in Figure 10 is executed on the initial training data 6.

[0109] The "Maximum slope" on the vertical axis of Figure 11 represents the experimental value of the maximum slope of the mediated current of the extracted compound, measured using electrochemical measuring device 2.

[0110] Furthermore, "Initial" refers to the compounds included in the initial training data 6 and their experimental values. "1st time," "2nd time," and "3rd time" refer to the compounds added in step S19 and their experimental values ​​when steps S11 to S19 in the flowchart of Figure 10 were performed once, twice, and three times, respectively. The number of times referred to as "1st time," "2nd time," "3rd time," etc., will also be called the number of searches below.

[0111] The compounds added to training data 6 in the first, second, and third explorations were as follows:

[0112] • 1st time Thio (Thionin), Rho6g (Rhodamine 6g), Lom (Lomefloxacin), Tiz (Tizanidine), Pix (Pixantrone), Clob (Clobenpropit), Etcr (Ethacridine (2-Ethoxy-6,9-diamino-acridine Lactate Monohydrate)), PPP (Propionylpromazine Hydrochloride).

[0113] • 2nd time AG3 (Acid Green3), CSS (Cefsulodin Sodium Salt), Asu (Asulam), MIF (MIF antagonist p425), NY (Nitrazine Yellow).

[0114] • 3rd time ChT (Chloramine T), CBE (Chlorazol Black E), 5Ana (5-Amino-1-naphthalenesulfonic acid), Fam (Famotidine).

[0115] As shown in Figure 11, in this example, "Thio" (thionine), represented by the chemical formula C8H8S, was discovered in a single search as the compound with the maximum gradient exceeding the initial level. "Thio" (thionine) is an example of a mediator discovered in this embodiment. Furthermore, a mixture of "Thio" (thionine) and E. coli is an example of a mediator mixture containing mediators and cells.

[0116] The maximum slope of "Thio" (thionine) is approximately 0.028 (nA / s). Therefore, if 0.028 (nA / s) is stored in the memory unit 31 as the reference value for the determination unit 34, the execution of the flowchart in Figure 10 can be completed in one search. However, in this example, in order to know which compounds are added in step S19 for each search, no specific reference value for the determination unit 34 was set, and the search was continued for two, three, and so on.

[0117] As a result, after three searches, "CBE" (Chlorazol Black E) was discovered, which had a greater maximum slope than "Thio". "CBE" has the chemical formula C 34 H 25 The compound is represented by N9Na2O7S2, and its maximum slope is approximately 0.029 (nA / s). "CBE" (Chlorazole Black E) is an example of a mediator discovered in this embodiment, and a mixture of "CBE" (Chlorazole Black E) and E. coli is an example of a mediator mixture.

[0118] Figure 12 is a box plot diagram obtained from the results in Figure 11. In Figure 11, the horizontal axis represents the number of searches, and the vertical axis represents the maximum slope. In the box plot, the circles represent data points, and the upper and lower whiskers represent the maximum and minimum values. The bottom edge of the box represents the data point that is in the 1 / 4 of the total data when the data is sorted in descending order. The top edge of the box represents the data point that is in the 3 / 4 of the total data when the data is sorted in descending order. Furthermore, the horizontal line within the box represents the median of the data, and the "x" marks represent the mean of the data.

[0119] As shown in Figure 12, the average value marked with an "x" increased with each subsequent search (1st, 2nd, and 3rd search), and the average value in the 3rd search exceeded the initial average value. This result confirms that the flowchart in Figure 10 is effective as a process for searching for mediators with good electrical characteristics.

[0120] Figure 13 shows the maximum mediating current values ​​for each compound in Figure 12, obtained by measuring them with the electrochemical measuring device 2. As shown in Figure 13, among the multiple compounds in the "initial," "first," "second," and "third" trials, the maximum mediating current value for Thio (thionine) was the largest.

[0121] Figure 14 is a box plot diagram obtained from the results in Figure 13. The meaning of each part of the box plot is the same as that explained in Figure 12, so the explanation is omitted here.

[0122] As shown in Figure 14, in this example, the average value, indicated by the × mark, tends to decrease as the number of searches increases. This is likely because the experimental value in training data 6 is the maximum slope (nA / s) of the mediating current, and therefore the learning unit 35 has not learned the maximum value of the mediating current.

[0123] (Second Embodiment) In this embodiment, the mediator search results from the first embodiment are used for adjuvant search.

[0124] The inventors of this application investigated the correlation between the electrical properties and chemical properties of the mediator. The results are shown in Figures 15 to 17.

[0125] Figures 15 to 17 show the results obtained by investigating the correlation between the electrical properties and chemical properties of mediators in this embodiment. Here, we investigated a total of 11 mediators: the top 10 mediators in terms of electrical properties from among the multiple mediators included in the training data 6 (see Figure 8) described in the first embodiment, and PYO (Pyocianine), a mediator not included in training data 6. For the electrical properties, we used the logarithm of the maximum value of the mediating current flowing through the solution over a predetermined time. The predetermined time was 51 hours, and the base of the logarithm was set to 10. In this case, the top 10 mediators in training data 6 after performing step S11 four times according to the flowchart in Figure 10 of the first embodiment were as follows:

[0126] BB24 (Basic Blue 24), Thio (Thionin), NY (Nitrazine Yellow), TB (Toluidine Blue), MB (Methylene Blue), NB (Nile Blue), FMN (Flavin Mononucleotide), RF (Riboflavin), Lom (Lomefloxacin), Saf (Safranin).

[0127] Figure 15 shows the case when redox potential is adopted as the chemical property. In Figure 15, the horizontal axis represents the redox potential, and the vertical axis represents the logarithm of the maximum current value. When a regression line 51 was generated to predict the maximum current value from the redox potential of each of the 11 mediators, the correlation coefficient was 0.77, indicating a positive correlation.

[0128] Figure 16 shows the case where the number of rotatable bonds is adopted as a chemical property. The number of rotatable bonds is the total number of rotatable bonds in the mediator. In Figure 16, the horizontal axis shows the number of rotatable bonds, and the vertical axis shows the logarithm of the maximum current value. When a regression line 52 was generated to predict the maximum current value from the number of rotatable bonds of each of the 11 mediators, the correlation coefficient was -0.69, indicating a negative correlation.

[0129] Figure 17 shows the case where the number of hydrogen bond donors is adopted as a chemical property. The number of hydrogen bond donors is the total number of hydrogen bond donors in the mediator. In Figure 17, the horizontal axis shows the number of hydrogen bond donors, and the vertical axis shows the logarithm of the maximum current value. When a regression line 53 was generated to predict the maximum current value from the number of hydrogen bond donors for each of the 11 mediators, the correlation coefficient was -0.75, indicating a negative correlation.

[0130] As shown in Figures 15-17, most mediators are located near regression lines 51-53, but some mediators are significantly outside of this range. For example, PYO and BB24 are located above both regression lines 52 and 53. The explanatory variables for each regression line 52 and 53, namely the number of rotatable bonds and the number of hydrogen donors, are both chemical properties related to the mediator's ability to permeate the cell membrane. For example, the more rotatable bonds a mediator has, the more flexible it becomes and the more various three-dimensional structures it can adopt. Therefore, it becomes difficult for the mediator to maintain a three-dimensional structure suitable for permeating the cell membrane, making it harder for the mediator to permeate the cell membrane. Also, the more hydrogen donors there are, the easier it is for the mediator to form strong hydrogen bonds with the highly electronegative atoms on the surface of the cell membrane. Therefore, the mediator tends to remain on the surface of the cell membrane, making it harder for it to permeate the cell membrane.

[0131] The current values ​​of mediators near regression lines 52 and 53 can be understood as being generated by a mechanism in which intracellular mediators permeate the cell membrane and donate electrons to extracellular electrodes. On the other hand, the large current values ​​of PYO and BB24, which are located above regression lines 52 and 53, cannot be explained by this mechanism alone and are thought to be caused by the cell's drug efflux pump.

[0132] The drug efflux pump is a protein that allows bacteria treated with antibiotics to expel the antibiotic from their cells. This pump is acquired by antibiotic-resistant bacteria. Because the bactericidal effect of antibiotics is very weak against resistant bacteria, the spread of resistant bacteria narrows treatment options for infectious diseases and poses a public health threat.

[0133] As mentioned above, the reason why larger current values ​​than predicted were obtained with mediators such as PYO and BB24 is thought to be because these mediators, which are the carriers of the current, are expelled from inside the cell to outside the cell by the drug efflux pump, and a large current is generated by the flow of these mediators. If this is correct, then administering mediators such as PYO and BB24 to bacteria along with antibiotics will cause these mediators to concentrate in the drug efflux pump. This makes it more difficult for the drug efflux pump to expel the antibiotic from the cell, and it is thought that the mediators function as drug efflux pump inhibitors (adjuvants) that inhibit the function of the drug efflux pump.

[0134] To verify this, the inventors of the present invention mixed an antibiotic, E. coli, and a mediator, and experimentally determined the adjuvant effect on the antibiotic at different concentrations of the mediator. The results are shown in Figures 18(a) to (c).

[0135] Figures 18(a) to (c) show the relationship between mediator concentration and adjuvant effect, as experimentally determined in this embodiment. In the experiment, the ΔsoxS strain of Escherichia coli was used, and carbenicillin was used as the antibiotic. The mediators for Figures 18(a) to (c) were BB24, PYO, and BDM88855, respectively. BDM88855 is a known drug efflux pump inhibitor and was used as a reference for BB24 and PYO, respectively. The adjuvant effect was calculated based on the following equation (7).

[0136]

number

[0137] However, V(x) is a function defined by the following equation (8).

[0138]

number

[0139] Note that CFU stands for colony-forming unit. Equation (7) is an equation in which the adjuvant effect is set to 100 when the difference between the bactericidal effect of high concentration (100 μg / ml) carbenicillin and the bactericidal effect of low concentration (50 μg / ml) carbenicillin disappears.

[0140] As shown in Figures 18(a) and (b), both BB24 and PYO showed a tendency for the adjuvant effect to increase as the mediator concentration increased. This trend is identical to the known trend observed in BDM88855, shown in Figure 18(c). These results clearly demonstrate that both BB24 and PYO function as drug efflux pump inhibitors for carbenicillin.

[0141] Next, we will describe a mediator discovery device according to this embodiment that identifies drug efflux pump inhibitors based on the above principle.

[0142] Figure 19 is a functional configuration diagram of the mediator search device according to this embodiment. As shown in Figure 19, in addition to the parts 33 to 41 described in the first embodiment, this mediator search device 3 further includes a generation unit 42 and a third identification unit 43.

[0143] The generation unit 42 is a processing unit that generates a predictive model that predicts experimental values ​​of electrical properties from the chemical properties of each mediator. The explanatory variables of the predictive model are the experimental values ​​of the chemical properties of the mediator, and the dependent variable is the experimental value of the electrical properties. The type of predictive model is not particularly limited. In this example, the generation unit 42 generates a regression line using linear regression as the predictive model. The generation unit 42 may also generate a predictive model using nonlinear regression instead of linear regression. Examples of nonlinear regression include k-nearest neighbors, regression trees, random forests, neural networks, and support vector regression.

[0144] The dataset used to generate the predictive model is training data 6 (see Figure 19). As described in the first embodiment, training data 6 is updated until "YES" is determined in the conditional branch of step S12 (see Figure 10). For example, suppose that the number of times training data 6 is updated before "YES" is determined in step S12 is N. In this case, the generation unit 42 uses the experimental values ​​of mediators and electrical characteristics included in the training data 6 updated in the last Nth update as the dataset to generate the predictive model. This makes it possible to obtain a predictive model based on training data 6 that reflects the insights gained with the maximum number of updates.

[0145] As mentioned above, mediators that function as drug efflux pump inhibitors have the characteristic of having a large current value. To capture this characteristic, it is preferable that the electrical characteristics, which are the target variable of the prediction model, are characteristics related to current. In this example, the maximum value of the current flowing through the solution at a predetermined time is adopted as the electrical characteristic, as shown in Figures 15 to 17. In this case, data relating "mediator" and "maximum current value" is adopted as the learning data 6 (see Figure 2), and the generation unit 42 can refer to this learning data 6 to identify the maximum current value for each mediator.

[0146] Furthermore, instead of generating a predictive model for all mediators included in the training data 6, the generation unit 42 may generate a predictive model only for mediators in the training data 6 whose maximum current value is of a predetermined rank or higher. The predetermined rank is not particularly limited and may be, for example, 10th place or higher. This makes it possible to generate a predictive model while considering mediators with large currents that are likely to function as drug efflux pump inhibitors, while reducing the computational cost of generating the predictive model.

[0147] Furthermore, in order to determine whether the mediator was expelled from the cell by cell membrane permeability or by the action of drug efflux pumps, it is preferable to use properties related to the mediator's ability to permeate the cell membrane as explanatory variables in the predictive model. Such properties include the number of rotatable bonds (see Figure 16) and the number of hydrogen bond donors (see Figure 17), as mentioned above.

[0148] Alternatively, instead of properties related to the ability to permeate the cell membrane, the redox potential of the mediator (see Figure 15) may be adopted as an explanatory variable in the predictive model. As shown in Figure 15, the redox potential correlates with the maximum current, and BB24 and PYO, mentioned above, are above the regression line 51. Therefore, even by adopting the redox potential, it is possible to screen multiple mediators and identify drug efflux pump inhibitors.

[0149] The generation unit 42 may generate a predictive model for one of several different chemical properties, such as redox potential, number of rotatable bonds, and number of hydrogen bond donors, as specified by the user, or it may generate a predictive model for each of the multiple chemical properties.

[0150] Chemical properties such as redox potential, number of rotatable bonds, and number of hydrogen bond donors can be obtained from literature or chemical databases. For example, the user may pre-store information relating chemical properties to mediators in the memory unit 31 (see Figure 19), and the generation unit 42 may identify the chemical properties of each mediator based on that information.

[0151] The method for generating the prediction model is not particularly limited. As mentioned above, when generating a regression line as the prediction model, the generation unit 42 can generate the regression line using the least squares method.

[0152] The third identification unit 43 is a processing unit that identifies mediators whose experimental electrical characteristics are greater than the predicted values ​​from the prediction model as drug efflux pump inhibitors. For example, the third identification unit 43 compares the experimental electrical characteristics with the predicted electrical characteristics predicted from the prediction model for each mediator. Then, the third identification unit 43 identifies mediators whose experimental electrical characteristics are greater than the predicted values ​​as drug efflux pump inhibitors.

[0153] The mediators to be searched may be mediators included in the training data 6, or mediators not included in the training data 6. For example, as with the PYO mentioned above, mediators not included in the training data 6 may be included in the mediators to be searched. In this case, for example, the user may pre-store experimental values ​​of the electrical and chemical properties of that mediator in the storage unit 31, and the third identification unit 43 may refer to these experimental values ​​to determine whether the experimental value of the electrical properties is greater than the predicted value.

[0154] Furthermore, as mentioned above, the generation unit 42 may generate prediction models for each of several different types of chemical properties, and the third identification unit 43 may identify mediators whose experimental electrical properties are greater than the predicted values ​​from all prediction models as drug efflux pump inhibitors. This increases the likelihood that the identified mediators will function as drug efflux pump inhibitors compared to using only one prediction model.

[0155] The bacteria that can be targeted by drug efflux pump inhibitors discovered in this way are not particularly limited to pathogenic bacteria. For example, in addition to the aforementioned Escherichia coli, Staphylococcus aureus, Salmonella, Vibrio cholerae, Helicobacter pylori, Mycobacterium tuberculosis, Streptococcus, Pseudomonas aeruginosa, Clostridium botulinum, Bordetella pertussis, Streptococcus pyogenes, Staphylococcus epidermidis, and Campylobacter can also be targeted by drug efflux pump inhibitors.

[0156] Drug efflux pump inhibitors are administered together with antibiotics, but the antibiotics are not particularly limited. For example, in addition to carbenicillin mentioned above, antibiotics such as penicillin, ampyricin, piperacillin, amoxicillin, flomox, erythromycin, kanamycin, gentamicin, vancomycin, cephalosporins, cephamycins, oxacephalosporins, ciprofloxacin, chloramphenicol, tetracycline, novobiocin, and lincomycin can also be administered together with drug efflux pumps.

[0157] Next, an example of the processing performed by the mediator search device 3 according to this embodiment will be described.

[0158] Figure 20 is a flowchart showing an example of the processing performed by the mediator search device 3 according to this embodiment. This flowchart is triggered, for example, by the completion of the processing shown in Figure 10 in the first embodiment.

[0159] First, the generation unit 42 uses the training data 6 as a dataset to generate a predictive model that predicts experimental values ​​of electrical properties from the chemical properties of each mediator included in the training data 6 (step S21). The generation unit 42 generates, for example, a regression line as the predictive model.

[0160] Next, the third identification unit 43 identifies mediators whose experimental electrical characteristics are greater than the predicted values ​​from the prediction model as drug efflux pump inhibitors (step S22).

[0161] This completes the processing performed by the mediator search device 3.

[0162] According to the embodiment described above, a regression line is generated using the training data 6 generated in the first embodiment as a dataset (step S21), and mediators above the regression line are identified as drug efflux pump inhibitors (step S22). Therefore, not only can mediators whose experimental values ​​of electrical characteristics meet the criteria be discovered as in the first embodiment, but mediators that function as drug efflux pump inhibitors can also be discovered. As a result, drug efflux pump inhibitors can be administered to drug-resistant bacteria together with antibiotics to kill those resistant bacteria, thereby reducing the public health threat associated with the spread of drug-resistant bacteria.

[0163] While drug efflux pump inhibitors were used as an example of adjuvants in the above example, the search is not limited to these. For example, adjuvants that enhance the effectiveness of vaccines when administered together could also be searched for.

[0164] <Hardware Configuration> Next, we will describe the hardware configuration of the mediator search device 3.

[0165] Figure 21 is an example of a hardware configuration diagram of the mediator search device 3 according to the first and second embodiments. As shown in Figure 21, the mediator search device 3 includes a storage device 101, memory 102, processor 103, communication interface 104, and media reader 105. These components are interconnected by a bus 106.

[0166] Of these, the storage device 101 is a non-volatile storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores the mediator search program 110 according to this embodiment.

[0167] Alternatively, the mediator search program 110 may be recorded on a computer-readable recording medium 111, and the processor 103 may read the mediator search program 110 via a media reader 105.

[0168] Examples of such recording media 111 include physically portable recording media such as CD-ROMs (Compact Disc - Read Only Memory), DVDs (Digital Versatile Discs), and USB (Universal Serial Bus) memory. Alternatively, semiconductor memory such as flash memory or hard disk drives may also be used as recording media 111. These recording media 111 are not temporary media like carrier waves that do not have a physical form.

[0169] Furthermore, the mediator search program 110 may be stored in a device connected to a public network, the internet, or a LAN. In that case, the processor 103 can read and execute the mediator search program 110.

[0170] On the other hand, memory 102 is hardware that temporarily stores data, such as DRAM (Dynamic Random Access Memory).

[0171] The processor 103 is hardware such as a CPU (Central Processing Unit) and a GPU (Graphical Processing Unit) that controls each part of the mediator search device 3. The processor 103 also works in cooperation with the memory 102 to execute the mediator search program 110.

[0172] In this way, the memory 102 and the processor 103 work together to execute the mediator search program 110, thereby realizing the control unit 30 (see Figures 8 and 19).

[0173] Furthermore, the storage unit 31 (see Figures 8 and 19) is realized by the storage device 101 and the memory 102.

[0174] Furthermore, the communication interface 104 is hardware such as a NIC (Network Interface Card) for connecting the mediator search device 3 to the network. The communication unit 32 (see Figures 8 and 19) is realized through this communication interface 104.

[0175] The media reader 105 is hardware such as a CD drive, DVD drive, and USB interface for reading the recording medium 111. [Explanation of Symbols]

[0176] 1…Mediator search system, 2…Electrochemical measuring device, 3…Mediator search device, 4…Compound database, 5…Network, 6…Training data, 7…Compound list, 10…Well plate, 10a…Top surface, 10b…Bottom surface, 10c, 10w, 10r…Through-hole, 11…Measurement section, 12…Well, 12a…Bottom surface, 13c…Counter electrode, 13r…Reference electrode, 13w…Working electrode, 14…Resin, 15…Wiring, 16c…Conductive pad for counter electrode, 16r…Conductive pad for reference electrode, 16w…Working electrode Conductive pad for use, 20...bacteria, 21, 22...mediator, 30...control unit, 31...storage unit, 32...communication unit, 33...acquisition unit, 34...determination unit, 35...learning unit, 36...first identification unit, 37...estimation unit, 38...list presentation unit, 39...second identification unit, 40...extraction unit, 41...addition unit, 42...generation unit, 43...third identification unit, 101...storage device, 102...memory, 103...processor, 104...communication interface, 105...mediation device, 106...bus, 110...mediator search program, 111...recording medium.

Claims

1. Computers This involves learning training data that associates experimental values ​​of the electrical properties of a solution containing a mediator and cells with the mediator, Based on the results of the learning, estimate the electrical characteristics of each of the multiple mediators not included in the learning data, Adding experimental values ​​of the electrical characteristics of at least some of the mediators whose electrical characteristics have been estimated, along with the mediators themselves, to the training data, The process is repeated until the experimental values ​​of the electrical characteristics meet the criteria, thereby searching for a mediator whose experimental values ​​meet the criteria. Mediator search method.

2. The aforementioned electrical characteristics are the characteristics of the current flowing through the solution. The mediator search method according to claim 1.

3. The aforementioned electrical characteristic is the maximum value of the time variation of the current. The mediator search method according to claim 2.

4. The experimental values ​​of the electrical characteristics are measured values ​​obtained by taking measurements for each of the multiple wells holding the solution for each type of mediator. A mediator search method according to any one of claims 1 to 3.

5. Some of the multiple wells hold the solution containing the mediator of the same type, The experimental value of the electrical characteristics is the average value of the measured values ​​in some of the wells. The mediator search method according to claim 4.

6. The learning process includes generating a function that includes a kernel function indicating the similarity between the mediators included in the learning data and mediators not included in the learning data, and which returns an estimate of the electrical characteristics. The estimation of the electrical properties is performed by estimating the electrical properties of each of the compounds in the database that are not included as mediators in the training data, based on the function generated as a result of the training. The mediator search method according to claim 1.

7. In the aforementioned database, the compound is associated with information indicating whether the compound is commercially available. The computer further performs the task of identifying commercially available compounds from among the multiple compounds stored in the database. The calculation of the aforementioned estimated value is performed for the identified compound. The mediator search method according to claim 6.

8. The aforementioned computer, The compounds are presented in order of their estimated electrical properties, from best to worst. The mediator search method according to claim 6 or claim 7.

9. The aforementioned computer, Further, the compounds whose electrical properties are of a rank higher than a predetermined rank are extracted as some mediators. The addition to the training data is performed by adding the extracted mediators and the experimental values ​​of the electrical characteristics of those mediators to the training data. The mediator search method according to claim 8.

10. The solution comprises the mediator and Escherichia coli composed of the cells. A mediator search method according to any one of claims 1 to 9.

11. To generate a predictive model that predicts the experimental values ​​of the electrical properties from the chemical properties of at least some of the mediators among the plurality of mediators, The method further includes identifying mediators as adjuvants in which the experimental values ​​of the aforementioned electrical characteristics are greater than the predicted values ​​obtained by the prediction model. The mediator search method according to claim 1.

12. The aforementioned electrical characteristic is the maximum value of the current flowing through the solution at a predetermined time. The mediator search method according to claim 11.

13. The aforementioned chemical properties are related to the mediator's ability to penetrate the cell membrane. A mediator search method according to claim 11 or claim 12.

14. The aforementioned chemical property is the number of rotatable bonds or the number of hydrogen bond donors. The mediator search method according to claim 13.

15. In generating the aforementioned prediction model, the prediction model is generated for each of several different types of chemical properties, In identifying the mediator as an adjuvant, the mediator is identified as an adjuvant if the experimental value of the electrical properties is greater than all of the predicted values ​​from the prediction model for each of the multiple chemical properties. A mediator search method according to any one of claims 11 to 14.

16. A mediator consisting of thionine or chlorazole black E.

17. A mediator mixture comprising the mediator and cells according to claim 16.

18. A mediator that functions as an adjuvant for either BB24 (Basic Blue 24) or PYO (Pyocianine).

19. This involves learning training data that associates experimental values ​​of the electrical properties of a solution containing a mediator and cells with the mediator, Based on the results of the learning, estimate the electrical characteristics of each of the multiple mediators not included in the learning data, Adding experimental values ​​of the electrical characteristics of at least some of the mediators whose electrical characteristics have been estimated, along with the mediators themselves, to the training data, A mediator search program that causes a computer to perform a process of searching for a mediator whose experimental values ​​satisfy the criteria by repeating the process until the experimental values ​​of the electrical characteristics satisfy the criteria.

20. A learning unit that learns learning data relating experimental values ​​of the electrical properties of a solution containing a mediator and cells with the mediator, An estimation unit that estimates the electrical characteristics of each of the multiple mediators not included in the training data based on the results of the learning, An additional unit that adds experimental values ​​of the electrical characteristics of at least some of the plurality of mediators whose electrical characteristics have been estimated, and the mediators themselves, to the learning data. It includes a determination unit that determines whether the added experimental values ​​meet the criteria, If the learning unit determines that the experimental value does not meet the criteria, it will perform the learning using the additional learning data. Mediator search device.

Images