Drug discovery assistance device, method for operating drug discovery assistance device, and program for operating drug discovery assistance device

Abstract

The processor: acquires a first prediction value, for each ion channel and for each addition amount by which a candidate substance for a medicine is added, by causing a prediction model to output first prediction values for the degree of inhibition indicating the degree by which the flow of ions in a plurality of ion channels present in iPS cardiomyocytes is inhibited by a candidate substance, on the basis of a first variation waveform that is a variation waveform of the extracellular potential of iPS cardiomyocytes (which are cardiomyocytes derived from human iPS cells) and that is measured when the candidate substance is added to the iPS cardiomyocytes while varying the addition amount; derives, for each ion channel, an index value indicating the dose reaction relationship for the candidate substance, on the basis of the first prediction values for different addition amounts; and presents, to the user, estimation reference information that corresponds to the index value and is referred to in order to estimate the mechanism of action by the candidate substance.

Description

Drug discovery support device, drug discovery support device operation method, and drug discovery support device operation program

[0001] The technology of the present disclosure relates to a drug discovery support device, an operating method for the drug discovery support device, and an operating program for the drug discovery support device.

[0002] As a test to evaluate the safety of drug candidate substances, the International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use (ICH) non-clinical trial guideline, Safety Pharmacology Test S7B, requires a test to evaluate the presence or absence of cardiotoxicity of the candidate substance. Among cardiotoxicities, a fatal arrhythmia called Torsades de Pointes (TdP) is considered the most important toxicity. The cause of fatal arrhythmia is said to be QT prolongation, a phenomenon in which the QT interval on an electrocardiogram is prolonged.

[0003] In conventional tests based on the ICH S7B guideline, the inhibitory effect of candidate substances on the hERG (human Ether-a-go-go Related Gene) channel, which is one of the ion channels expressed in cardiomyocytes, was examined, and candidate substances with a high probability of causing fatal arrhythmias were identified by evaluating the QT interval of electrocardiograms using non-human animals. However, substances with strong inhibitory effects on the hERG channel do not necessarily cause QT prolongation. Therefore, to improve the accuracy of the evaluation, it was necessary to also evaluate the effects on other ion channels, such as Ca channels and Na channels.

[0004] Incidentally, cardiomyocytes derived from human iPS (induced pluripotent stem) cells (hereinafter referred to as iPS cardiomyocytes) possess not only hERG channels but also ion channels expressed in human cardiac tissue, such as Na channels and Ca channels. Therefore, iPS cardiomyocytes have been validated as a tool capable of evaluating the presence or absence of QT prolongation in drug candidate substances. A large-scale, multicenter validation study demonstrated the usefulness of a method for measuring the electrical activity of iPS cardiomyocytes as an electrocardiogram-like waveform to evaluate the presence or absence of QT prolongation. This method was included in a follow-up study of the ICH S7B guideline. Specifically, the electrocardiogram-like waveform is the waveform of fluctuations in the extracellular potential of iPS cardiomyocytes.

[0005] The measurement method of electrical activity of iPS cardiomyocytes using MEA can determine whether QT prolongation occurs. However, when QT prolongation occurs, it is not possible to determine the mechanism of action of the candidate substance, such as which ion channel's ion flow is inhibited or activated by the candidate substance, causing QT prolongation. Therefore, K. H. Jaeger, et al “Identifying Drug Response by Combining Measurements of the Membrane Potential, the Cytosolic Calcium Concentration, and the Extracellular Potential in Microphysiological Systems” Frontiers in Pharmacology 08 February 2021. (hereinafter referred to as Non-Patent Document 1), and Fabien Raphel, et al., "A greedy classifier optimization strategy to assess ion channel blocking activity and pro-arrhythmia in hiPSC-cardiomyocytes," PLOS Computational Biology, September 25, 2020 (hereinafter referred to as Non-Patent Document 2), describe techniques that contribute to the estimation of the mechanism of action of candidate substances.

[0006] In Non-Patent Document 1, the height of the first peak and the conduction velocity of the fluctuation waveform are derived as feature quantities from the fluctuation waveform of the extracellular potential of iPS cardiomyocytes measured with an MEA, and the derived feature quantities are input into a prediction model, which outputs a predicted value of the degree of inhibition of the candidate substance on the Na channel (a numerical value indicating the extent to which the candidate substance inhibits the flow of Na ions through the Na channel). Furthermore, the degree of inhibition of the candidate substance on the hERG channel (a numerical value indicating the extent to which the candidate substance inhibits the flow of K ions through the hERG channel) is obtained by measuring the intracellular potential of iPS cardiomyocytes. Furthermore, the degree of inhibition of the candidate substance on the Ca channel (a numerical value indicating the extent to which the candidate substance inhibits the flow of Ca ions through the Ca channel) is obtained by measuring the calcium concentration.

[0007] Non-Patent Document 2 uses a simulation model to generate a waveform of fluctuations in the intracellular potential of iPS cardiomyocytes, and converts the generated waveform of fluctuations in the intracellular potential of iPS cardiomyocytes into a waveform of fluctuations in the extracellular potential of iPS cardiomyocytes using an MEA. Then, the authors propose an algorithm to predict the degree of inhibition of a candidate substance on hERG channels, Na channels, and Ca channels based on feature quantities derived from the waveform of fluctuations in the extracellular potential of iPS cardiomyocytes.

[0008] In estimating the mechanism of action of a candidate substance, an index value indicating the dose-response relationship of the candidate substance, such as the median inhibitory concentration (IC50), is thought to be very useful. However, the techniques described in Non-Patent Documents 1 and 2 do not derive an index value. Therefore, there is a risk that estimating the mechanism of action of a candidate substance will be time-consuming.

[0009] One embodiment of the technology of the present disclosure provides a drug discovery support device, an operating method for the drug discovery support device, and an operating program for the drug discovery support device, which are capable of easily estimating the mechanism of action of a candidate substance.

[0010] The drug discovery support device disclosed herein includes a processor, and the processor acquires first predicted values ​​for each ion channel and each added amount obtained by outputting from a prediction model first predicted values ​​of the degree of inhibition indicating the degree to which the flow of ions in multiple ion channels present in the iPS cardiomyocytes is inhibited by the candidate substance based on a first fluctuation waveform, which is a fluctuation waveform of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, measured when a candidate pharmaceutical substance is added to the iPS cardiomyocytes while changing the added amount, and derives an index value indicating the dose-response relationship of the candidate substance for each ion channel based on the first predicted values ​​for each added amount, and presents to the user estimated reference information corresponding to the index value, which is referenced to estimate the mechanism of action of the candidate substance.

[0011] The processor preferably searches for a dose-response curve that fits the first predicted value for each added dose, and derives the index value from the searched dose-response curve.

[0012] The first predicted value is not only a predicted value of the degree of inhibition, but also a predicted value of the degree of activity indicating the degree to which the flow of ions in the ion channel has been activated by the candidate substance, and there are two types of dose-response curves: a first dose-response curve when the flow of ions in the ion channel is inhibited by the candidate substance, and a second dose-response curve when the flow of ions in the ion channel is activated by the candidate substance, and it is preferable that the processor derives an index value from either the first dose-response curve or the second dose-response curve, whichever is more suitable for the first predicted value for each added amount.

[0013] The processor preferably presents the discovered dose-response curve to the user as estimated reference information.

[0014] The curve used to search for the dose-response curve is preferably a logistic curve.

[0015] The first predicted value is preferably obtained by inputting a feature amount derived from the first fluctuation waveform into a prediction model.

[0016] Preferably, the processor acquires the first fluctuation waveform, derives features from the first fluctuation waveform, and inputs the features into the prediction model, thereby outputting the first predicted value from the prediction model.

[0017] It is preferable that the processor performs noise reduction processing on the first fluctuation waveform before deriving the feature amount.

[0018] Preferably, the first fluctuation waveform has a periodicity corresponding to the pulsation of iPS cardiomyocytes, and the processor performs an averaging process of a plurality of periodic portions of the first fluctuation waveform as noise reduction processing.

[0019] It is preferable that the first fluctuation waveform is measured using multiple electrodes for one iPS cardiomyocyte, and that the processor selects one of the multiple first fluctuation waveforms measured by the multiple electrodes as the first fluctuation waveform from which the feature is derived, in accordance with predetermined conditions.

[0020] The feature amount preferably includes the conduction velocity of the first fluctuation waveform.

[0021] The feature amount is preferably normalized by a reference feature amount derived from a reference first fluctuation waveform measured when the candidate substance is not added to the iPS cardiomyocytes.

[0022] When there are experimental values ​​of index values ​​that indicate the dose-response relationship of the candidate substance, it is preferable that the first predicted value be obtained by inputting the experimental values ​​in addition to the feature quantities into the prediction model.

[0023] The prediction model is preferably a model that has been trained in two ways, one where experimental values ​​are input and one where they are not input.

[0024] It is preferable that the first predicted value be obtained by inputting into the prediction model, in addition to the feature, a second predicted value of an index value indicating the dose-response relationship of the candidate substance, which is derived based on structural information of the candidate substance.

[0025] The prediction model is preferably a model trained using training data including simulation data.

[0026] It is preferable that the simulation data be generated using a first simulation model that reproduces a second fluctuation waveform, which is the fluctuation waveform of the intracellular potential of iPS cardiomyocytes, and a second simulation model that converts the second fluctuation waveform into the first fluctuation waveform.

[0027] The method of operating the drug discovery support device disclosed herein includes obtaining a first predicted value for each ion channel and each added amount obtained by outputting a first predicted value of the degree of inhibition indicating the degree to which the flow of ions in multiple ion channels present in the iPS cardiomyocytes is inhibited by the candidate substance from a prediction model based on a first fluctuation waveform, which is a fluctuation waveform of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, measured when a candidate pharmaceutical substance is added to the iPS cardiomyocytes while changing the added amount; deriving an index value indicating the dose-response relationship of the candidate substance for each ion channel based on the first predicted value for each added amount; and presenting to a user estimated reference information corresponding to the index value, which is referenced to estimate the mechanism of action of the candidate substance.

[0028] The operating program for the drug discovery support device disclosed herein causes a computer to execute processes including: acquiring first predicted values ​​for each ion channel and each added amount obtained by outputting, from a prediction model, first predicted values ​​of the degree of inhibition indicating the degree to which the flow of ions in multiple ion channels present in iPS cardiomyocytes is inhibited by the candidate substance, based on a first fluctuation waveform that is a fluctuation waveform of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, measured when a candidate pharmaceutical substance is added to iPS cardiomyocytes while changing the added amount; deriving, for each ion channel, an index value indicating the dose-response relationship of the candidate substance based on the first predicted values ​​for each added amount; and presenting to a user estimated reference information corresponding to the index value, which is referenced to estimate the mechanism of action of the candidate substance.

[0029] According to the technology of the present disclosure, it is possible to provide a drug discovery support device, an operating method for a drug discovery support device, and an operating program for a drug discovery support device that are capable of easily estimating the mechanism of action of a candidate substance.

[0030] 1 is a diagram illustrating a drug discovery support server, a user terminal, and a well plate. FIG. 2 is a table showing the amount of drug candidate substance added to each well. FIG. 3 is a block diagram illustrating computers constituting the drug discovery support server and a user terminal. FIG. 4 is a block diagram illustrating a processing unit of a CPU of the drug discovery support server. FIG. 5 is a diagram illustrating processing by a preprocessing unit. FIG. 6 is a diagram illustrating noise reduction processing. FIG. 7 is a diagram illustrating selection processing. FIG. 8 is a diagram illustrating selection processing. FIG. 9 is a diagram illustrating processing by a feature derivation unit. FIG. 10 is a diagram illustrating normalization of features by reference features. FIG. 11 is a diagram illustrating feature information. FIG. 12 is a diagram illustrating a group of prediction models. FIG. 13 is a diagram illustrating processing by a prediction unit that inputs features to each prediction model and outputs a first predicted value from each prediction model. FIG. 14 is a diagram illustrating prediction result information. FIG. 15 is a diagram illustrating processing in the learning phase of a prediction model for a hERG channel. FIG. 16 is a diagram illustrating a group of dose-response curves. FIG. 17 is a graph illustrating a first dose-response curve. FIG. 18 is a graph illustrating a second dose-response curve. FIG. 19 is a flowchart illustrating the processing procedure of a search unit. FIG. 19 is a diagram illustrating derivation of a first predicted value for each ion channel and for each added amount. 1 is a diagram showing a state in which the squared error of the searched second dose-response curve is smaller than the squared error of the searched first dose-response curve, and the half effective concentration is derived from the searched second dose-response curve. FIG. 2 is a block diagram showing the processing unit of the CPU of the user terminal. FIG. 3 is a diagram showing a waveform input screen. FIG. 4 is a diagram showing an evaluation result display screen. FIG. 5 is a flowchart showing the processing procedure of the drug discovery support server. FIG. 6 is a diagram showing a second embodiment in which experimental values ​​of index values ​​indicating the dose-response relationship of a candidate substance are input into the prediction model in addition to feature quantities. FIG. 7 is a diagram showing processing in the learning phase of the prediction model for hERG channels of the second embodiment. FIG. 8 is a diagram showing a third embodiment in which second predicted values ​​of index values ​​indicating the dose-response relationship of a candidate substance, derived based on structural information of the candidate substance, are input into the prediction model in addition to feature quantities. FIG. 9 is a diagram showing a fourth embodiment in which simulation data is included in the learning data of the prediction model. FIG. 10 is a table showing another example of allocation of candidate substances and the amounts of candidate substances added to each well.

[0031] First Embodiment As shown in FIG. 1 as an example, a drug discovery support server 10 is connected to a user terminal 11 via a network 12. The drug discovery support server 10 is an example of a "drug discovery support device" according to the technology of the present disclosure. The user terminal 11 is installed, for example, at a pharmaceutical company developing drugs or an organization contracted by a pharmaceutical company to develop drugs, i.e., a contract research organization (CRO). The user terminal 11 is operated by a user U involved in drug development at the pharmaceutical company or contract research organization. The network 12 is, for example, a wide area network (WAN) such as the Internet or a public communication network. Note that while only one user terminal 11 is connected to the drug discovery support server 10 in FIG. 1 , multiple user terminals 11 from multiple pharmaceutical companies or contract research organizations are actually connected to the drug discovery support server 10.

[0032] A well plate 13 is connected to the user terminal 11. The well plate 13 has a plurality of wells 14. The wells 14 are arranged at equal intervals in the vertical and horizontal directions. FIG. 1 illustrates a well plate 13 having 8 x 6 = 48 wells 14. The number of wells 14 is not limited to the illustrated 48, and may be 24, 96, or the like.

[0033] The well 14 is a cylindrical depression with an open top. The well 14 is filled with a culture medium, in which iPS cardiomyocytes 15, which are cardiomyocytes derived from human iPS cells, are cultured. The iPS cardiomyocytes 15 are sheet-shaped and grow to the point where they can beat spontaneously. To culture the iPS cardiomyocytes 15 under a constant environment, the well plate 13 is placed in a thermostatic bath (not shown).

[0034] As an example, as shown in Table 25 of FIG. 2 , of the 48 wells 14, wells 14 numbered 1 to 40 are doped with candidate pharmaceutical substances. More specifically, wells 14 numbered 1 to 10 are doped with candidate pharmaceutical substances at an amount AA1, and wells 11 to 20 are doped with candidate pharmaceutical substances at an amount AA2. Wells 14 numbered 21 to 30 are doped with candidate pharmaceutical substances at an amount AA3, and wells 14 numbered 31 to 40 are doped with candidate pharmaceutical substances at an amount AA4. The magnitude relationship between the amounts AA1 to AA4 is AA1<AA2<AA3<AA4. In contrast, wells 14 numbered 41 to 48 are doped with candidate pharmaceutical substances. In this way, candidate pharmaceutical substances are doped to iPS cardiomyocytes 15 while the amount AA is changed. Note that the amount AA is not limited to the four types shown. For example, it may be two types or ten types.

[0035] Returning to FIG. 1 , a planar microelectrode array 16 is provided on the bottom surface of the well 14. The planar microelectrode array 16 is composed of a plurality of microelectrodes 17 arranged in a square. FIG. 1 illustrates a planar microelectrode array 16 having 8 x 8 = 64 microelectrodes 17. As with the well 14, the microelectrodes 17 are numbered. The microelectrodes 17 are an example of an "electrode" according to the technology of the present disclosure. Note that the number of microelectrodes 17 is not limited to the illustrated 64, and may be 16, 100, or the like.

[0036] The microelectrodes 17 are exposed on the bottom surface of the well 14 and come into contact with the iPS cardiomyocytes 15. The microelectrodes 17 are connected to a potential measurement circuit (not shown). The potential measurement circuit measures the electrical signal from each of the multiple microelectrodes 17, i.e., the extracellular potential of the iPS cardiomyocytes 15. Therefore, the extracellular potential is measured for each of the multiple microelectrodes 17, 64 microelectrodes 17 in this example. This method of measuring the extracellular potential using a planar microelectrode array 16 is much simpler and less expensive than a method known as manual patch clamp, in which electrodes are inserted directly into the iPS cardiomyocytes 15 to measure the extracellular potential.

[0037] The measurement results of the extracellular potential by the potential measurement circuit are input into the user terminal 11. The user terminal 11 generates a first variation waveform 18, which is a variation waveform that shows the temporal change in the measurement results of the extracellular potential. Here, the measurement of the extracellular potential is performed for a predetermined period, such as 10 minutes. Therefore, the first variation waveform 18 shows the temporal change in the measurement results of the extracellular potential for the predetermined period. The number of the well 14 and the number of the microelectrode 17 from which the extracellular potential was measured are stored in association with the first variation waveform 18.

[0038] The first fluctuation waveform 18 contains several pulses resulting from the pulsation of the iPS cardiomyocytes 15. Therefore, the first fluctuation waveform 18 has a periodicity corresponding to the pulsation of the iPS cardiomyocytes 15. In other words, the first fluctuation waveform 18 is a collection of multiple periodic portions 19 separated by one interspike interval (ISI).

[0039] The user terminal 11 transmits an evaluation request 20 to the drug discovery support server 10. The evaluation request 20 is a request for the drug discovery support server 10 to evaluate the cardiotoxicity of a drug candidate substance. The evaluation request 20 includes a plurality of first fluctuation waveforms 18. The number of first fluctuation waveforms 18 included in the evaluation request 20 depends on the number of wells 14 and microelectrodes 17. In this example, there are 48 wells 14 and 64 microelectrodes 17, so the evaluation request 20 includes 48 x 64 = 3072 first fluctuation waveforms 18. Although not shown in the figure, the evaluation request 20 also includes a terminal ID (identification data) for uniquely identifying the user terminal 11 that transmitted the evaluation request 20.

[0040] When receiving the evaluation request 20, the drug discovery support server 10 derives estimated reference information 21. The drug discovery support server 10 distributes the estimated reference information 21 to the user terminal 11 that sent the evaluation request 20. When receiving the estimated reference information 21, the user terminal 11 makes the estimated reference information 21 available for viewing by the user U.

[0041] The estimated reference information 21 is information that is referenced by the user U to estimate the mechanism of action of a candidate substance. The mechanism of action of a candidate substance means, for example, when QT prolongation occurs in iPS cardiomyocytes 15, which of multiple ion channels present in iPS cardiomyocytes 15 has its ion flow inhibited or activated by the candidate substance, causing QT prolongation.

[0042] The multiple ion channels present in iPS cardiomyocytes 15 are the hERG channel through which K ions pass, the Na channel through which Na ions pass, and the Ca channel through which Ca ions pass. These ion channels play a very important role in the pulsation of iPS cardiomyocytes 15. For this reason, it is thought that inhibition or activation of the flow of each ion in these ion channels results in the occurrence of QT prolongation, which is a cause of fatal arrhythmia.

[0043] 3, the computers that make up the drug discovery support server 10 and the user terminal 11 basically have the same configuration, and include a storage 30, a memory 31, a CPU (Central Processing Unit) 32, a communication unit 33, a display 34, and an input device 35. These are interconnected via a bus line 36.

[0044] The storage 30 is a hard disk drive built into the computer that constitutes the drug discovery support server 10 and the user terminal 11, or connected via a cable or network. Alternatively, the storage 30 is a disk array with multiple hard disk drives connected in series. The storage 30 stores control programs such as an operating system, various application programs (hereinafter referred to as APs (Application Programs)), and various data associated with these programs. Note that a solid state drive may be used instead of a hard disk drive.

[0045] The memory 31 is a work memory for the CPU 32 to execute processing. The CPU 32 loads programs stored in the storage 30 into the memory 31 and executes processing in accordance with the programs. In this way, the CPU 32 comprehensively controls each part of the computer. The CPU 32 is an example of a "processor" according to the technology of the present disclosure. The memory 31 may be built into the CPU 32.

[0046] The communication unit 33 is a network interface that controls the transmission of various information via the network 12, etc. The display 34 displays various screens. The various screens are equipped with operation functions using a GUI (Graphical User Interface). The computers that make up the drug discovery support server 10 and the user terminal 11 accept input of operation instructions from an input device 35 via the various screens. The input device 35 is a keyboard, a mouse, a touch panel, a microphone for voice input, etc.

[0047] In the following explanation, the parts of the computer that make up the drug discovery support server 10 (storage 30 and CPU 32) are distinguished by adding the suffix "A" to their symbols, and the parts of the computer that make up the user terminal 11 (storage 30, CPU 32, display 34, and input device 35) are distinguished by adding the suffix "B" to their symbols.

[0048] 4, an operating program 40 is stored in the storage 30A of the drug discovery support server 10. The operating program 40 is an AP for causing a computer to function as the drug discovery support server 10. In other words, the operating program 40 is an example of an "operating program for a drug discovery support device" according to the technology of the present disclosure. The storage 30 also stores a group of prediction models 41, a group of dose-response curves 42, and the like.

[0049] When the operating program 40 is started, the CPU 32A of the computer constituting the drug discovery support server 10 works in cooperation with the memory 31, etc. to function as a request receiving unit 45, a read / write (hereinafter abbreviated as RW (Read Write)) control unit 46, a preprocessing unit 47, a feature derivation unit 48, a prediction unit 49, a search unit 50, and a screen distribution control unit 51.

[0050] The request receiving unit 45 receives various requests from the user terminal 11, including the evaluation request 20. When the evaluation request 20 is received, the request receiving unit 45 outputs the first fluctuation waveform 18 included in the evaluation request 20 to the RW control unit 46. Although not shown in the figure, the request receiving unit 45 also outputs the terminal ID of the user terminal 11 included in the evaluation request 20 to the screen distribution control unit 51.

[0051] The RW control unit 46 controls the storage of various data in the storage 30A and the reading of various data from the storage 30A. In particular, the RW control unit 46 controls the storage of the first fluctuation waveform 18 in the storage 30A and the reading of the first fluctuation waveform 18 from the storage 30A. The RW control unit 46 outputs the read first fluctuation waveform 18 to the preprocessing unit 47. The RW control unit 46 also reads the prediction model group 41 from the storage 30A and outputs the read prediction model group 41 to the prediction unit 49. The RW control unit 46 also reads the dose-response curve group 42 from the storage 30A and outputs the read dose-response curve group 42 to the search unit 50.

[0052] The preprocessing unit 47 performs preprocessing on the first fluctuation waveform 18. The preprocessing unit 47 outputs the preprocessed first fluctuation waveform 18 (hereinafter referred to as first fluctuation waveform 18AT) to the feature derivation unit 48.

[0053] The feature derivation unit 48 derives multiple types of feature quantities 65 from the first fluctuation waveform 18AT using, for example, a machine learning model that outputs feature quantities 65 (see FIG. 9 ) when the first fluctuation waveform 18AT is input. The feature derivation unit 48 generates feature quantity information 55 that summarizes the multiple types of derived feature quantities 65 and outputs the feature quantity information 55 to the prediction unit 49.

[0054] The prediction unit 49 causes the prediction model group 41 to predict the degree of action of the candidate substance on each ion channel of the iPS cardiomyocytes 15 based on the feature amount 65 of the feature amount information 55. The prediction unit 49 generates prediction result information 56 summarizing the prediction results of the degree of action of the candidate substance on each ion channel of the iPS cardiomyocytes 15, and outputs the prediction result information 56 to the search unit 50.

[0055] The search unit 50 searches for a dose-response curve that matches the prediction result information 56 based on the group of dose-response curves 42. The search unit 50 derives an index value indicating the dose-response relationship of the candidate substance from the searched dose-response curve. The search unit 50 generates estimated reference information 21 according to the index value and outputs the estimated reference information 21 to the screen distribution control unit 51.

[0056] The screen delivery control unit 51 controls the delivery of various screens to the user terminal 11. Specifically, the screen delivery control unit 51 delivers and outputs various screens to the user terminal 11 that has sent the various requests in the form of screen data for web delivery created using a markup language such as XML (Extensible Markup Language). At this time, the screen delivery control unit 51 identifies the user terminal 11 that has sent the various requests based on the terminal ID from the request receiving unit 45. Note that instead of XML, other data description languages ​​such as JSON (Javascript (registered trademark) Object Notation) may be used.

[0057] The various screens include a waveform input screen 90 (see FIG. 23) for inputting the first fluctuation waveform 18, and an evaluation result display screen 95 (see FIG. 24) for presenting the estimated reference information 21 to the user U. In addition to these processing units 45 to 51, the CPU 32A also includes an instruction receiving unit that receives various operation instructions from the input device 35.

[0058] 5, the pre-processing unit 47 performs a noise reduction process 60 and a selection process 61 as pre-processing on the first fluctuation waveform 18. The noise reduction process 60 is, for example, a process shown in FIG. 6, and the selection process 61 is, for example, a process shown in FIGS. 7 and 8.

[0059] As shown in Fig. 6, the preprocessing unit 47 performs the noise reduction process 60 in the following manner. First, the preprocessing unit 47 extracts multiple periodic portions 19 from a first fluctuation waveform 18 measured at a certain microelectrode 17 in a certain well 14. Then, from the multiple periodic portions 19, a set number of periodic portions 19 that satisfy a preset selection criterion are selected (step ST10). The selection criterion may be, for example, that the pulsation interval ISI is within a set range, that the maximum amplitude is equal to or less than a set value, and / or that the S / N ratio is equal to or less than a set value. The set number is, for example, 30.

[0060] The preprocessing unit 47 averages the selected number of periodic portions 19 to obtain the periodic portion 19AV (step ST11). This completes the noise reduction process 60. In other words, the noise reduction process 60 is an averaging process of multiple periodic portions 19 of the first fluctuation waveform 18. The preprocessing unit 47 repeats steps ST10 and ST11 until there are no more first fluctuation waveforms 18 that have not been subjected to the noise reduction process 60. In other words, the preprocessing unit 47 performs the noise reduction process 60 on all first fluctuation waveforms 18. Therefore, the number of periodic portions 19AV generated is equal to the number of first fluctuation waveforms 18. FIG. 6 shows an example in which the noise reduction process 60 is performed on the first fluctuation waveform 18 measured at the No. 1 microelectrode 17 in the No. 1 well 14. Note that the noise reduction process 60 may include smoothing using a low-pass filter on the first fluctuation waveform 18 in addition to or instead of the averaging process.

[0061] As shown in Figures 7 and 8, the preprocessing unit 47 selects one periodic portion 19AV_E from multiple periodic portions 19AV generated from the first fluctuation waveform 18 measured at each microelectrode 17 in a certain well 14. The preprocessing unit 47 selects the periodic portion 19AV_E according to preset conditions (step ST20). The periodic portion 19AV_E is used to derive a feature 65 in the feature derivation unit 48. In other words, the periodic portion 19AV_E is an example of a "first fluctuation waveform from which a feature is derived" according to the technology of the present disclosure. Thus, the selection process 61 is a process for selecting a so-called golden channel.

[0062] Here, selecting the periodic portion 19AV_E according to preset conditions refers to, for example, selection according to the method disclosed in International Patent No. 2022 / 176310. The method disclosed in International Patent No. 2022 / 176310 is a method of selecting one waveform (periodic portion 19AV_E in this example) from multiple selection target waveforms by repeating clustering analysis between multiple selection target waveforms (multiple periodic portions 19AV in this example) and multiple teacher waveforms labeled as ideal waveforms or not. Note that conditions may be set based on feature amounts obtained from an ideal waveform, and a periodic portion 19AV_E that satisfies the conditions may be selected.

[0063] The pre-processing unit 47 repeats the process of step ST20 until there are no wells 14 that have not been subjected to the selection process 61. In other words, the pre-processing unit 47 performs the selection process 61 on all wells 14. Therefore, periodic portions 19AV_E are selected for the number of wells 14. The pre-processing unit 47 outputs the periodic portions 19AV_E for the number of wells 14 as the first fluctuation waveform 18AT to the feature deriving unit 48. Note that FIG. 7 shows an example in which the periodic portion 19AV associated with the No. 7 microelectrode 17 is selected as the periodic portion 19AV_E from the multiple periodic portions 19AV associated with the microelectrodes 17 of the No. 1 well 14. Also, FIG. 8 shows an example in which the periodic portion 19AV associated with the No. 7 microelectrode 17 is selected as the periodic portion 19AV_E from the multiple periodic portions 19AV associated with the microelectrodes 17 of the No. 48 well 14. 8 shows an example in which the periodic portion 19AV associated with the microelectrode 17 No. 45 is selected as the periodic portion 19AV_E. According to FIG. 2, the well 14 No. 48 is a well 14 to which no candidate substance is added. Therefore, the periodic portion 19AV_E selected in FIG. 8 is an example of the "reference first fluctuation waveform" according to the technology of the present disclosure.

[0064] 9, the feature deriving unit 48 derives the following features 65 from the periodic portion 19AV_E: Field Potential Duration (FPD) cF, Depolarization Amplitude (DA), Repolarization Center (RC), Repolarization Width (RW), Repolarization Amplitude (RA), Area Under Curve of the repolarization wave (AUCr), and Conduction Velocity (CV).

[0065] FPDcF is the value obtained by dividing the time interval FPD (field potential duration) between depolarization wave P1 and repolarization wave P2, which are pulses caused by the pulsation of iPS cardiomyocytes 15, by the 1 / 3 power of the pulsation interval ISI. That is, FPDcF=FPD / (ISI). １／３ where DA is the amplitude of the depolarization wave P1. RC is the time interval from the depolarization wave P1 to the rising edge of the repolarization wave P2 (repolarization center). RW is the time interval of the repolarization wave P2 (repolarization width). RA is the amplitude of the repolarization wave P2. AUCr is the area under the curve of the repolarization wave P2. CV is the conduction velocity of the first fluctuation waveform 18. FPDcF (FPD), DA, RC, RW, RA, and AUCr are feature quantities listed in Non-Patent Document 2. CV is a feature quantity listed in Non-Patent Document 1. CV is calculated based on the placement position of the microelectrode 17.

[0066] As an example, as shown in Figure 10, the feature derivation unit 48 standardizes the feature 65 derived from the periodic portions 19AV_E of wells 14 Nos. 1 to 40 to which the candidate substance was added by dividing it by a reference feature 65S derived from the periodic portions 19AV_E of wells 14 Nos. 41 to 48 to which the candidate substance was not added, to obtain a feature 65AS. As an example, as shown in Figure 11, the feature information 55 is a collection of the feature 65AS of wells 14 Nos. 1 to 40. The reference feature 65S is a representative value, such as the average or median, of the feature 65 derived from the periodic portions 19AV_E of wells 14 Nos. 41 to 48.

[0067] As an example, as shown in FIG. 12 , the prediction model group 41 is composed of a hERG channel prediction model 70, a Na channel prediction model 71, and a Ca channel prediction model 72. The hERG channel prediction model 70 predicts the degree of inhibition (hereinafter, sometimes referred to as "inhibition degree DI"; see FIG. 17 ) indicating the degree of inhibition of K ion flow through the hERG channel by a candidate substance, or the activity (hereinafter, sometimes referred to as "activity degree DAC"; see FIG. 18 ) indicating the degree of activation. The Na channel prediction model 71 predicts the degree of inhibition or activity indicating the degree of activation of Na ion flow through the Na channel by a candidate substance. The Ca channel prediction model 72 predicts the degree of inhibition or activity indicating the degree of activation of Ca ion flow through the Ca channel by a candidate substance. Each of these prediction models 70-72 is a machine learning model constructed using a machine learning technique, such as a neural network. Furthermore, each of the prediction models 70 to 72 is a machine learning model that supports multi-label regression using a technique such as regressor chain, and therefore each of the prediction models 70 to 72 is a model that takes into account the correlation between each ion channel.

[0068] 13 , the prediction unit 49 inputs the feature quantity 65AS into each prediction model 70-72, causing each prediction model 70-72 to output first predicted values ​​75A, 75B, and 75C. The first predicted value 75A output from the hERG channel prediction model 70 is a predicted value of the degree of inhibition indicating the degree to which the candidate substance inhibits or activates the flow of K ions through the hERG channel. The first predicted value 75B output from the Na channel prediction model 71 is a predicted value of the degree of inhibition indicating the degree to which the candidate substance inhibits or activates the flow of Na ions through the Na channel. The first predicted value 75C output from the Ca channel prediction model 72 is a predicted value of the degree of inhibition indicating the degree to which the candidate substance inhibits or activates the flow of Ca ions through the Ca channel. These first predicted values ​​75A to 75C take a value greater than 0 and less than 1 when it is predicted that the flow of ions is inhibited by the candidate substance, and take a value greater than 1 when it is predicted that the flow of ions is activated by the candidate substance. When it is predicted that the flow of ions is neither inhibited nor activated by the candidate substance, the first predicted values ​​75A to 75C are 1. In the following, unless there is a particular need to distinguish between them, the first predicted values ​​75A to 75C will be simply referred to as first predicted values ​​75.

[0069] 14, the prediction result information 56 is a collection of first predicted values ​​75A to 75C for each of wells 14 No. 1 to 40. In other words, the prediction result information 56 is a collection of first predicted values ​​75A to 75C for each ion channel and each added amount.

[0070] The prediction unit 49 also uses a prediction model other than the prediction models 70 to 72 to predict whether or not addition of a candidate substance will result in QT prolongation in the iPS cardiomyocytes 15. The prediction result information 56 also includes the prediction result of whether or not QT prolongation will occur in the iPS cardiomyocytes 15. The prediction model that predicts whether or not QT prolongation will occur in the iPS cardiomyocytes 15 receives the feature quantity 65AS as input, as with the prediction models 70 to 72. Alternatively, the first prediction values ​​75A to 75C are input.

[0071] As an example, as shown in FIG. 15 , the hERG channel prediction model 70 is trained using training data 78. The training data 78 is a set of training features 65ASL and correct answer data 75ACA. The training features 65ASL are features 65AS of candidate substances that have been actually tested in the past. The correct answer data 75ACA are the results of actual measurements of the inhibition or activity of the candidate substance that provided the training features 65ASL in tests that have actually been conducted in the past.

[0072] The training feature 65ASL is input to the hERG channel prediction model 70. As a result, the hERG channel prediction model 70 outputs a first training predicted value 75AL. The first training predicted value 75AL is compared with the correct answer data 75ACA, and a loss calculation is performed on the hERG channel prediction model 70 using a loss function based on the comparison result. Then, internal parameters such as the filter coefficients of the hERG channel prediction model 70 are updated according to the result of the loss calculation, and the hERG channel prediction model 70 is updated according to the updated setting.

[0073] The above-described series of processes, including input of the training features 65ASL to the hERG channel prediction model 70, output of the first training predicted value 75AL from the hERG channel prediction model 70, loss calculation, update setting, and update of the hERG channel prediction model 70, are repeatedly performed while the training data 78 is changed. The above-described series of processes is terminated when the prediction accuracy of the first training predicted value 75AL for the correct answer data 75ACA reaches a preset level. The hERG channel prediction model 70, whose prediction accuracy has thus reached the preset level, is stored in storage 30A. Training may be terminated after a predetermined number of repetitions of the above-described series of processes, regardless of the prediction accuracy. Training of the hERG channel prediction model 70 may also be continued after storage in storage 30A.

[0074] The processing in the learning phase of hERG channel prediction model 70 has been described above, but the processing in the learning phase of Na channel prediction model 71 and Ca channel prediction model 72 is basically the same except for the content of learning data 78. For this reason, illustrations and descriptions of Na channel prediction model 71 and Ca channel prediction model 72 will be omitted.

[0075] 16, the group of dose-response curves 42 is composed of a first dose-response curve 80 and a second dose-response curve 81. The first dose-response curve 80 is a dose-response curve obtained when the flow of ions through the ion channel is inhibited by a candidate substance, whereas the second dose-response curve 81 is a dose-response curve obtained when the flow of ions through the ion channel is activated by a candidate substance.

[0076] As an example, as shown in FIG. 17, a first dose-response curve 80 is a logistic curve expressed by the following formula (1).

[0077]

[0078] Here, DI is the degree of inhibition, AA is the amount of the candidate substance added, and IC50 is the half maximal inhibitory concentration. In a two-dimensional space with the degree of inhibition DI on the vertical axis and the amount AA on the horizontal axis, the first dose-response curve 80 depicts an inverted S-shaped curve in which the degree of inhibition DI gradually decreases from 1 to 0 as the amount AA increases.

[0079] As an example, as shown in FIG. 18, a second dose-response curve 81 is a logistic curve expressed by the following formula (2).

[0080]

[0081] Here, DAC is activity, AA is the amount of the candidate substance added as in the case of the first dose-response curve 80, EC50 (Median Effective Concentration) is the median effective concentration, and M is the value of the asymptote. In a two-dimensional space with activity DAC on the vertical axis and added amount AA on the horizontal axis, the second dose-response curve 81 describes an S-shaped curve in which activity DAC gradually increases from 1 toward M as added amount AA increases.

[0082] The searching unit 50 performs processing according to the procedure shown in FIG. 19 as an example. First, the searching unit 50 derives first predicted values ​​75A-75C for each addition amount AA (step ST30). More specifically, as shown in FIG. 20 as an example, the searching unit 50 determines the representative value of the first predicted values ​​75A-75C for wells 14 Nos. 1-10 as first predicted values ​​75A1-75C1 for addition amount AA1, and the representative value of the first predicted values ​​75A-75C for wells 14 Nos. 11-20 as first predicted values ​​75A2-75C2 for addition amount AA2. Similarly, the searching unit 50 determines the representative value of the first predicted values ​​75A-75C for wells 14 Nos. 21-30 as first predicted values ​​75A3-75C3 for addition amount AA3, and the representative value of the first predicted values ​​75A-75C for wells 14 Nos. 21-30 as first predicted values ​​75A4-75C4 for addition amount AA4. The representative value of the first predicted values ​​75A to 75C for wells 14 31 to 40 is set as first predicted values ​​75A4 to 75C4 for the addition amount AA4. This completes the first predicted values ​​75 for each ion channel and each addition amount. The representative value may be, for example, the average value or the median value.

[0083] 19 , the search unit 50 searches for a first dose-response curve 80 that fits the first predicted values ​​75A to 75C for each dose of a certain ion channel (step ST31). Specifically, the search unit 50 changes the IC50 of the first dose-response curve 80, and each time the IC50 is changed, the search unit 50 calculates the squared error between the first dose-response curve 80 and the first predicted values ​​75A to 75C for each dose. The first dose-response curve 80 having the IC50 that minimizes the squared error is then determined to be the first dose-response curve 80 that fits the first predicted values ​​75A to 75C for each dose.

[0084] The searching unit 50 also searches for a second dose-response curve 81 that fits the first predicted values ​​75A to 75C for each dose of a certain ion channel (step ST32). Specifically, the searching unit 50 calculates the squared error between the second dose-response curve 81 and the first predicted values ​​75A to 75C for each dose while changing the M and EC50 of the second dose-response curve 81. The second dose-response curve 81 having the M and EC50 that minimize the squared error is then determined to be the second dose-response curve 81 that fits the first predicted values ​​75A to 75C for each dose.

[0085] Next, the search unit 50 compares the magnitude of the squared error of the first dose-response curve 80 searched for in step ST31 with the magnitude of the squared error of the second dose-response curve 81 searched for in step ST32 (step ST33). If the squared error of the first dose-response curve 80 searched for is smaller than the squared error of the second dose-response curve 81 searched for (YES in step ST33), i.e., if the first dose-response curve 80 is a better fit to the first predicted values ​​75A to 75C for each added dose than the second dose-response curve 81, the search unit 50 derives the half maximal inhibitory concentration (IC50) from the searched first dose-response curve 80 (step ST34). This half maximal inhibitory concentration (IC50) derived from the first dose-response curve 80 is an example of an "index value" according to the technology of the present disclosure.

[0086] On the other hand, if the squared error of the searched second dose-response curve 81 is smaller than the squared error of the searched first dose-response curve 80 (NO in step ST33), i.e., if the second dose-response curve 81 is a better fit to the first predicted values ​​75A-75C for each added dose than the first dose-response curve 80, the search unit 50 derives the half effective concentration (EC50) from the searched second dose-response curve 81 (step ST35). This half effective concentration (EC50) derived from the second dose-response curve 81 is also an example of an "index value" according to the technology of the present disclosure. The search unit 50 repeats the processes of steps ST31 to ST35 until it has not derived the half inhibitory concentrations (IC50) or half effective concentrations (EC50) for all ion channels (NO in step ST36). In other words, the search unit 50 derives the half inhibitory concentrations (IC50) or half effective concentrations (EC50) for all ion channels. The search unit 50 outputs the first predicted value 75, the searched first dose-response curve 80 or the searched second dose-response curve 81, and the derived half inhibitory concentration IC50 or half effective concentration EC50 to the screen delivery control unit 51 as estimated reference information 21. The search unit 50 also outputs the prediction result of whether or not QT prolongation will occur in the iPS cardiomyocytes 15 upon addition of the candidate substance to the screen delivery control unit 51 as estimated reference information 21. Note that the estimated reference information 21 may be only the half inhibitory concentration IC50 or half effective concentration EC50.

[0087] 21 shows an example in which the Na channel is the target. Fig. 21 illustrates a case in which the square error of the second dose-response curve 81, indicated by the solid line, is smaller than the square error of the first dose-response curve 80, indicated by the dashed line. The example also illustrates a case in which the half effective concentration (EC50) is derived from the second dose-response curve 81.

[0088] As an example, as shown in Figure 22, an evaluation AP 85 is stored in the storage 30B of the user terminal 11. The evaluation AP 85 is installed in the user terminal 11 by the user U. The evaluation AP 85 is an AP for causing the drug discovery support server 10 to evaluate whether or not a candidate substance has cardiotoxicity. When the evaluation AP 85 is started, the CPU 32B of the user terminal 11 functions as a browser control unit 87 in cooperation with the memory 31, etc. The browser control unit 87 controls the operation of a web browser dedicated to the evaluation AP 85.

[0089] The browser control unit 87 reproduces various screens based on various screen data from the drug discovery support server 10 and displays the reproduced various screens on the display 34B. The browser control unit 87 also accepts various operation instructions input by the user U from the input device 35B via the various screens. The browser control unit 87 transmits various requests, including the evaluation request 20, to the drug discovery support server 10 in response to the operation instructions.

[0090] When the evaluation AP 85 is activated, a waveform input screen 90, as shown in FIG. 23 for example, is displayed on the display 34B under the control of the browser control unit 87. The waveform input screen 90 has an input box 91 for the first fluctuation waveform 18. A file of the first fluctuation waveform 18 can be dropped into the input box 91. After dropping the file of the desired first fluctuation waveform 18 into the input box 91, the user U selects the evaluation button 92. When the evaluation button 92 is selected, the browser control unit 87 generates an evaluation request 20 including the first fluctuation waveform 18 input into the input box 91 and transmits the generated evaluation request 20 to the drug discovery support server 10.

[0091] Furthermore, when the drug discovery support server 10 evaluates the cardiotoxicity of a candidate substance, an evaluation result display screen 95, as shown in FIG. 24 , is displayed on the display 34B under the control of the browser control unit 87. The evaluation result display screen 95 displays the estimated reference information 21. That is, the evaluation result display screen 95 displays a graph of the first dose-response curve 80 or the second dose-response curve 81 with a plot of the first predicted value 75, as well as the half maximal inhibitory concentration (IC50) or half maximal effective concentration (EC50) for each ion channel. The evaluation result display screen 95 also displays the prediction result of whether or not the candidate substance will induce QT prolongation in iPS cardiomyocytes 15. In this way, the estimated reference information 21 is presented to the user U in the form of screen data distribution.

[0092] A save button 96 and an OK button 97 are provided at the bottom of the evaluation result display screen 95. When the save button 96 is selected, the display contents of the evaluation result display screen 95, including the estimated reference information 21, are stored in the storage 30B of the user terminal 11. When the OK button 97 is selected, the display of the evaluation result display screen 95 is erased.

[0093] Next, the operation of the above configuration will be described with reference to the flowchart shown in Fig. 25 as an example. When the operating program 40 is started in the drug discovery support server 10, the CPU 32A of the drug discovery support server 10 functions as a request receiving unit 45, an RW control unit 46, a preprocessing unit 47, a feature derivation unit 48, a prediction unit 49, a search unit 50, and a screen distribution control unit 51, as shown in Fig. 4. When the evaluation AP 85 is started in the user terminal 11, the CPU 32B of the user terminal 11 functions as a browser control unit 87, as shown in Fig. 22.

[0094] 23 is displayed on the display 34B of the user terminal 11 under the control of the browser control unit 87. When the user U inputs a file of the desired first fluctuation waveform 18 into the input box 91 on the waveform input screen 90 and selects the evaluation button 92, an evaluation request 20 is transmitted from the browser control unit 87 to the drug discovery support server 10.

[0095] In the drug discovery support server 10, the request receiving unit 45 receives the evaluation request 20 (YES in step ST100). The first fluctuation waveform 18 included in the evaluation request 20 is output from the request receiving unit 45 to the RW control unit 46 and stored in the storage 30A under the control of the RW control unit 46 (step ST110). In addition, the terminal ID of the user terminal 11 included in the evaluation request 20 is output from the request receiving unit 45 to the screen distribution control unit 51.

[0096] The first fluctuation waveform 18 is read from the storage 30A by the RW control unit 46 (step ST120). The first fluctuation waveform 18 is output from the RW control unit 46 to the pre-processing unit 47.

[0097] 5 to 8, the pre-processing unit 47 performs pre-processing such as noise reduction processing 60 and selection processing 61 on the first fluctuation waveform 18 (step ST130). As a result, the first fluctuation waveform 18AT, which is composed of the periodic portion 19AV_E for each well 14, is output from the pre-processing unit 47 to the feature derivation unit 48.

[0098] 9 and 10 , the feature deriving unit 48 derives the feature 65AS from the periodic portion 19AV_E (step ST140). Then, the feature information 55 shown in FIG. 11 , which is a collection of the feature 65AS for each well 14, is output from the feature deriving unit 48 to the prediction unit 49.

[0099] 13, in the prediction unit 49, the feature quantity 65AS is input to each of the prediction models 70 to 72, and first predicted values ​​75A to 75C are output from each of the prediction models 70 to 72 (step ST150). Then, prediction result information 56 shown in FIG. 14, which is a collection of the first predicted values ​​75A to 75C for each well 14, is output from the prediction unit 49 to the search unit 50.

[0100] 19 to 21, the search unit 50 searches for a first dose-response curve 80 or a second dose-response curve 81 that fits the first predicted value 75 for each added dose. Then, a half maximal inhibitory concentration (IC50) or half maximal effective concentration (EC50) is derived as an index value from the searched first dose-response curve 80 or second dose-response curve 81 (step ST160). Estimated reference information 21, including the searched first dose-response curve 80 or second dose-response curve 81, the derived half maximal inhibitory concentration (IC50) or half maximal effective concentration (EC50), and the like, is output from the search unit 50 to the screen distribution control unit 51.

[0101] The screen delivery control unit 51 generates screen data for the evaluation result display screen 95 shown in Fig. 24 based on the estimated reference information 21. Under the control of the screen delivery control unit 51, the screen data for the evaluation result display screen 95 is delivered to the user terminal 11 that sent the evaluation request 20 (step ST170).

[0102] In the user terminal 11, under the control of the browser control unit 87, the screen data of the evaluation result display screen 95 is reproduced, and the reproduced evaluation result display screen 95 is displayed on the display 34B. In this way, the estimated reference information 21 is presented to the user U.

[0103] As described above, the prediction unit 49 of the drug discovery support server 10 obtains first predicted values ​​75 for each of multiple ion channels present in iPS cardiomyocytes 15, which are cardiomyocytes derived from human iPS cells, and for each added amount of a drug candidate substance. The first predicted values ​​75 are output from the prediction models 70-72 based on the first fluctuation waveform 18. The first fluctuation waveform 18 is a fluctuation waveform of the extracellular potential of the iPS cardiomyocytes 15, and is measured when the candidate substance is added to the iPS cardiomyocytes 15 while changing the added amount. The first predicted value 75 is a predicted value of the degree of inhibition indicating the degree to which the candidate substance inhibits the flow of ions through multiple ion channels.

[0104] The search unit 50 derives, for each ion channel, a 50% inhibitory concentration (IC50), which is an index value indicating the dose-response relationship of the candidate substance, based on the first predicted value 75 for each added amount. The screen delivery control unit 51 delivers screen data of the evaluation result display screen 95, which includes the estimated reference information 21, to the user terminal 11, thereby presenting the estimated reference information 21 to the user U. The estimated reference information 21 is information corresponding to the 50% inhibitory concentration (IC50), which is an index value, and is referenced to estimate the mechanism of action of the candidate substance.

[0105] By viewing the estimated reference information 21 through the evaluation result display screen 95, the user U can grasp at a glance the extent to which the candidate substance inhibits the flow of ions through each ion channel. Therefore, for example, if QT prolongation occurs in iPS cardiomyocytes 15, it becomes possible to easily estimate the mechanism of action by which the candidate substance caused QT prolongation, such as which ion channel's ion flow was inhibited by the candidate substance, resulting in QT prolongation. Conversely, if QT prolongation does not occur in iPS cardiomyocytes 15, it becomes possible to easily estimate the mechanism of action by which the candidate substance did not cause QT prolongation. If it becomes possible to easily estimate the mechanism of action by a candidate substance, it becomes easy to obtain guidelines for the structure of candidate substances that do not cause QT prolongation, which will greatly facilitate the development of pharmaceuticals.

[0106] The searching unit 50 searches for a first dose-response curve 80 or a second dose-response curve 81 that fits the first predicted value 75 for each added amount, and derives the half inhibitory concentration IC50 or half effective concentration EC50 from the searched first dose-response curve 80 or second dose-response curve 81. Therefore, it is possible to derive a more reliable half inhibitory concentration IC50 or half effective concentration EC50.

[0107] The first predicted value 75 is not only a predicted value of the degree of inhibition, but also a predicted value of the degree of activation indicating the degree to which the candidate substance activates ion flow through the ion channel. There are two types of dose-response curves: a first dose-response curve 80, which indicates the case where the candidate substance inhibits ion flow through the ion channel, and a second dose-response curve 81, which indicates the case where the candidate substance activates ion flow through the ion channel. The search unit 50 derives the half inhibitory concentration (IC50) or half effective concentration (EC50) from either the first dose-response curve 80 or the second dose-response curve 81, whichever is more suitable for the first predicted value 75 for each dose. This allows the mechanism of action of the candidate substance to be estimated by taking into account not only the inhibitory effect of the candidate substance on ion flow through the ion channel but also its activating effect. Furthermore, a more reliable half inhibitory concentration (IC50) or half effective concentration (EC50) can be derived.

[0108] The screen distribution control unit 51 presents the searched first dose-response curve 80 or the searched second dose-response curve 81 to the user U as the estimated reference information 21. This allows the user U to immediately understand which ion channel's ion flow is inhibited or activated by the candidate substance.

[0109] The curve used to search for the dose-response curve is a logistic curve. Logistic curves are commonly used to search for dose-response curves. Therefore, following the conventional method, it is possible to easily search for a dose-response curve that fits the first predicted value 75 for each added amount. Note that the curve used to search for the dose-response curve may also be a probit curve or the like.

[0110] The first predicted value 75 is obtained by inputting the feature quantity 65AS derived from the first fluctuation waveform 18 into the prediction models 70 to 72. This makes it possible to easily improve the prediction accuracy of the first predicted value 75. Note that the first fluctuation waveform 18 itself may be input to the prediction models 70 to 72 instead of or in addition to the feature quantity 65AS.

[0111] The request receiving unit 45 receives the evaluation request 20 to acquire the first fluctuation waveform 18. The feature derivation unit 48 derives the feature 65AS from the first fluctuation waveform 18. The prediction unit 49 inputs the feature 65AS into the prediction models 70-72, causing the prediction models 70-72 to output first predicted values ​​75A-75C. In this way, the drug discovery support server 10 single-handedly handles the process of acquiring the first fluctuation waveform 18, the process of deriving the feature 65AS from the first fluctuation waveform 18, and the process of outputting the first predicted values ​​75A-75C using the prediction models 70-72. Therefore, the processing load on the CPU 32B of the user terminal 11 can be reduced compared to when the CPU 32B of the user terminal 11 is responsible for, for example, the process of deriving the feature 65AS from the first fluctuation waveform 18 and the process of outputting the first predicted values ​​75A-75C using the prediction models 70-72.

[0112] Of course, if the CPU 32B of the user terminal 11 has a relatively high processing capacity, the CPU 32B of the user terminal 11 may be responsible for the process of deriving the feature quantity 65AS and the process of outputting the first predicted values ​​75A to 75C, etc. Furthermore, a CPU of a device other than the drug discovery support server 10 and the user terminal 11 may be responsible for the process of deriving the feature quantity 65AS and the process of outputting the first predicted values ​​75A to 75C, etc.

[0113] Prior to deriving the feature quantity 65AS, the preprocessing unit 47 performs noise reduction processing 60 on the first fluctuation waveform 18. This makes it possible to derive a more reliable feature quantity 65AS, and as a result, it is possible to improve the prediction accuracy of the first predicted value 75.

[0114] The first fluctuation waveform 18 has a periodicity corresponding to the pulsation of the iPS cardiomyocytes 15. The pre-processing unit 47 performs noise reduction processing 60 by averaging a plurality of periodic portions 19 of the first fluctuation waveform 18. This makes it possible to effectively reduce noise on the first fluctuation waveform 18.

[0115] The first fluctuation waveform 18 is measured using multiple microelectrodes 17 for one iPS cardiomyocyte 15. The preprocessing unit 47 selects one of the multiple first fluctuation waveforms 18 (periodic portions 19AV) measured using the multiple microelectrodes 17 according to preset conditions as the first fluctuation waveform 18 (periodic portions 19AV_E) from which the feature value 65AS is derived. This allows for the derivation of a more reliable feature value 65AS, thereby improving the prediction accuracy of the first predicted value 75.

[0116] The feature quantity 65AS includes the conduction velocity CV of the first fluctuation waveform 18. According to Non-Patent Document 1, the conduction velocity CV of the first fluctuation waveform 18 contributes to improving the prediction accuracy of the first predicted value 75, particularly the first predicted value 75B related to the Na channel. This further improves the prediction accuracy of the first predicted value 75.

[0117] The feature 65AS is normalized by the reference feature 65S derived from the reference first fluctuation waveform (such as the periodic portion 19AV_E shown in FIG. 8 ) measured when no candidate substance was added to the iPS cardiomyocytes 15. This makes it possible to eliminate individual differences among the iPS cardiomyocytes 15.

[0118] Second Embodiment Conventionally, a hERG test to evaluate the inhibitory effect of a candidate substance on the hERG channel has generally been conducted as a safety pharmacology test S7B. For this reason, some candidate substances have already undergone hERG testing prior to evaluation by the drug discovery support server 10, and experimental values ​​of the half maximal inhibitory concentration (IC50) or half maximal effective concentration (EC50) for the hERG channel (hereinafter referred to as "experimental IC50 (EC50) value") 110 have been measured. Therefore, in the second embodiment, as shown in FIG. 26 as an example, if an experimental IC50 (EC50) value 110 from the hERG test is available, the experimental IC50 (EC50) value 110 is input to each prediction model 70-72 in addition to the feature 65AS, and first prediction values ​​75A-75C are output from each prediction model 70-72.

[0119] In this case, as shown in FIG. 27 as an example, the training data 112 for the hERG channel prediction model 70 includes the training feature 65ASL and the correct answer data 75ACA of the first embodiment, as well as the training experimental value 110L. The training data 112 may or may not include the training experimental value 110L. Therefore, the hERG channel prediction model 70 is trained in two ways: with and without the training experimental value 110L. As in the case of FIG. 15 , the processing in the training phase for the Na channel prediction model 71 and the Ca channel prediction model 72 is essentially the same, except for the content of the training data 112. Therefore, illustration and description of the Na channel prediction model 71 and the Ca channel prediction model 72 are omitted.

[0120] As described above, in the second embodiment, when there is an IC50 (EC50) experimental value 110, which is an experimental value of an index value indicating the dose-response relationship of a candidate substance, the first predicted value 75 is obtained by inputting the IC50 (EC50) experimental value 110 in addition to the feature quantity 65AS into the prediction models 70 to 72. Since the IC50 (EC50) experimental value 110 provides an additional clue for prediction, the prediction accuracy of the first predicted value 75 can be further improved.

[0121] Furthermore, the prediction models 70 to 72 are models that have been trained in two ways: when the IC50 (EC50) experimental value 110 is input and when it is not input, and therefore can handle both cases where the IC50 (EC50) experimental value 110 is present and where it is not present.

[0122] 26 shows an example of an IC50 (EC50) experimental value 110 from a hERG test, but is not limited to this. If there are experimental values ​​related to Na channels and / or Ca channels, they may be input into the prediction models 70 to 72.

[0123] 28 as an example, in the third embodiment, the prediction unit 49 derives second predicted values ​​116A, 116B, and 116C of index values ​​indicating the dose-response relationship of a candidate substance based on the candidate substance structural information 115. Then, in addition to the feature quantity 65AS, the derived second predicted values ​​116A to 116C are input to the prediction models 70 to 72, and the prediction models 70 to 72 output first predicted values ​​75A to 75C.

[0124] The structural information 115 is a character string representing the chemical structure of a candidate substance using SMILES (Simplified Molecular Input Line Entry System) notation. The structural information 115 is input to an index value prediction model 117. The index value prediction model 117 is a machine learning model such as Light Gradient Boosting Machine (LightGBM), which derives descriptors from the structural information 115 using an RDKit and outputs second predicted values ​​116A to 116C in response to the input of the descriptors. The second predicted value 116A is a predicted value of the half maximal inhibitory concentration (IC50) or half maximal effective concentration (EC50) for the hERG channel. The second predicted value 116B is a predicted value of the half maximal inhibitory concentration (IC50) or half maximal effective concentration (EC50) for the Na channel. The second predicted value 116C is a predicted value of the half maximal inhibitory concentration IC50 or half maximal effective concentration EC50 for the Ca channel.

[0125] As described above, in the third embodiment, the first predicted values ​​75A-75C are obtained by inputting, into the prediction models 70-72, the second predicted values ​​116A-116C, which are index values ​​indicating the dose-response relationship of the candidate substance, derived based on the structural information 115 of the candidate substance, in addition to the feature quantity 65AS. The second predicted values ​​116A-116C provide additional clues for prediction, further improving the prediction accuracy of the first predicted value 75. Furthermore, when the experimental IC50 (EC50) value 110 of the second embodiment is not available, the second predicted values ​​116A-116C can be input into the prediction models 70-72 instead of the experimental IC50 (EC50) value 110.

[0126] The structural information 115 is not limited to a character string representing the chemical structure of the candidate substance using the SMILES notation shown as an example. It may also be a Molecular Design Limited (MOL) file or a Structure-Data File (SDF) representing the chemical structure of the candidate substance. In any case, a description method that can uniquely determine a three-dimensional structure such as an isomer is preferable, and a description method that can represent three-dimensional coordinate information of a molecule is even more preferable.

[0127] The index value prediction model 117 may be a graph neural network that derives a graph structure of the candidate substance from the structural information 115 and makes a prediction based on the graph structure. Alternatively, the index value prediction model 117 may make a prediction by docking simulation based on the structural information 115 and the three-dimensional structure of each ion channel.

[0128] The second embodiment and the third embodiment may be combined. That is, in addition to the feature 65AS, the IC50 (EC50) experimental value 110 and second predicted values ​​116B and 116C derived based on the structural information 115 of the candidate substance may be input to the prediction models 70 to 72, causing the prediction models 70 to 72 to output first predicted values ​​75A to 75C.

[0129] [Fourth Embodiment] In the first embodiment, the training data 78 is generated from data of tests that have actually been conducted in the past, but this is not limiting. For example, as shown in Fig. 29, simulation data may be included in the training data 78.

[0130] 29 , the first simulation model 120 uses the degree of inhibition or activity of ion flow in each ion channel as a parameter to reproduce a second fluctuation waveform 121SM according to the set degree of inhibition or activity. The second fluctuation waveform 121SM is a fluctuation waveform of the intracellular potential of an iPS cardiomyocyte 15. The first simulation model 120 outputs the second fluctuation waveform 121SM to the second simulation model 122.

[0131] The second simulation model 122 converts the second fluctuation waveform 121SM into a first fluctuation waveform 18SM. The training feature 65ASL and the supervised data 75ACA derived based on the first fluctuation waveform 18SM are used as training data 78.

[0132] The first simulation model 120 may be the ORd (O'Hara-Rudy dynamic) model published in 2011, the ToR (Tomek-Rodriguez)-ORd-dynCl (dynamic intracellular chloride) model published in 2020 which improves the behavior of Na channels when Na ions are inhibited, or the Paci model published in 2019 which models the behavior of iPS cardiomyocytes 15. The second simulation model 122 may be a bidomain model or an EMI (extracellular-membrane-intracellular) model.

[0133] As described above, in the fourth embodiment, the prediction models 70 to 72 are models trained using the training data 78 including the feature quantity 65AS based on the first fluctuation waveform 18SM, which is simulation data. Therefore, even if the amount of data from actual tests conducted in the past is relatively small, the amount of training data 78 can be increased. Consequently, the prediction accuracy of the first predicted values ​​75A to 75C by the prediction models 70 to 72 can be improved.

[0134] The first fluctuation waveform 18SM is generated using a first simulation model 120 that reproduces a second fluctuation waveform 121SM, which is a fluctuation waveform of the intracellular potential of the iPS cardiomyocyte 15, and a second simulation model 122 that converts the second fluctuation waveform 121SM into the first fluctuation waveform 18SM. This makes it possible to generate a first fluctuation waveform 18SM that is more suitable for the training data 78.

[0135] The allocation of candidate substances and the amounts AA of candidate substances to be added to each well 14 is not limited to the example shown in Table 25 in Figure 2. For example, the allocation shown in Table 130 in Figure 30 may be used. Table 130 shows an example in which a different candidate substance is added every five wells 14, such as candidate substance Z1 to wells 14 Nos. 1 to 5, candidate substance Z2 to wells 14 Nos. 6 to 10, and so on. Table 130 also shows an example in which wells 14 Nos. 1, 6, and so on are added with candidate substance AA1, wells 14 Nos. 2, 7, and so on are added with candidate substance AA2, wells 14 Nos. 3, 8, and so on are added with candidate substance AA3, wells 14 Nos. 4, 9, and so on are added with candidate substance AA4, and wells 14 Nos. 5, 10, and so on are not added with candidate substance.

[0136] The index value is not limited to the exemplified median inhibitory concentration (IC50) and median effective concentration (EC50), but may be, for example, median lethal concentration (LC50) or the like.

[0137] The drug discovery support server 10 may be installed in a pharmaceutical company or a pharmaceutical development contract organization, or may be installed in a data center independent of the pharmaceutical company or the pharmaceutical development contract organization.

[0138] Instead of delivering screen data of the evaluation result display screen 95 including the estimated reference information 21 to the user terminal 11, the estimated reference information 21 itself may be delivered to the user terminal 11. In this case, the user terminal 11 generates the evaluation result display screen 95 based on the estimated reference information 21 under the control of the browser control unit 87.

[0139] The method of presenting the estimated reference information 21 to the user U is not limited to the example of delivering screen data. The estimated reference information 21 may be presented to the user U by printing it on a paper medium, or by attaching it to an email and sending it to the user terminal 11.

[0140] The hardware configuration of the computer constituting the drug discovery support server 10 according to the technology of the present disclosure can be modified in various ways. For example, the drug discovery support server 10 can be configured with multiple computers separated as hardware in order to improve processing power and reliability. For example, the functions of the request reception unit 45, RW control unit 46, and preprocessing unit 47, and the functions of the feature derivation unit 48, prediction unit 49, search unit 50, and screen distribution control unit 51 can be distributed and performed by two computers. In this case, the drug discovery support server 10 is configured by two computers. Some or all of the functions of the drug discovery support server 10 may be performed by the user terminal 11.

[0141] In this way, the hardware configuration of the computer of the drug discovery support server 10 can be changed as appropriate depending on the required performance such as processing power, safety, reliability, etc. Furthermore, not only the hardware but also APs such as the operating program 40 can be duplicated or stored in multiple storages in order to ensure safety and reliability.

[0142] In each of the above embodiments, the hardware structure of the processing units that execute various processes, such as the request receiving unit 45, the RW control unit 46, the preprocessing unit 47, the feature derivation unit 48, the prediction unit 49, the search unit 50, the screen delivery control unit 51, and the browser control unit 87, can be any of the various processors listed below. As described above, the various processors include the CPUs 32A and 32B, which are general-purpose processors that execute software (operating program 40 and evaluation AP 85) and function as various processing units, as well as programmable logic devices (PLDs) that are processors whose circuit configuration can be changed after manufacture, such as FPGAs (Field Programmable Gate Arrays), and dedicated electrical circuits that are processors having a circuit configuration designed specifically for executing specific processing, such as ASICs (Application Specific Integrated Circuits).

[0143] A single processing unit may be configured with one of these various processors, or may be configured with a combination of two or more processors of the same or different types (e.g., a combination of multiple FPGAs and / or a combination of a CPU and an FPGA).Furthermore, multiple processing units may be configured with a single processor.

[0144] Examples of configuring multiple processing units with a single processor include, first, a form in which one processor is configured with a combination of one or more CPUs and software, as typified by computers such as client and server, and this processor functions as multiple processing units. Second, a form in which a processor is used to realize the functions of the entire system including multiple processing units with a single IC (Integrated Circuit) chip, as typified by systems on chips (SoCs). In this way, various processing units are configured using one or more of the above-mentioned various processors as a hardware structure.

[0145] Furthermore, more specifically, the hardware structure of these various processors can be an electric circuit (circuitry) that combines circuit elements such as semiconductor elements.

[0146] From the above description, the technology described in the following supplementary paragraphs can be understood.

[0147] [Supplementary Item 1] A drug discovery support device comprising: a processor, wherein the processor acquires, for each ion channel and for each added amount, first predicted values ​​of inhibition indicating the degree to which ion flow in multiple ion channels present in the iPS cardiomyocytes is inhibited by the candidate substance, based on first fluctuation waveforms that are fluctuation waveforms of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, and the first predicted values ​​are output from a prediction model based on first fluctuation waveforms measured when a drug candidate substance is added to the iPS cardiomyocytes at varying addition amounts; derives, for each ion channel, an index value indicating a dose-response relationship of the candidate substance based on the first predicted values ​​for each added amount; and presents to a user estimated reference information corresponding to the index value, the estimated reference information being referenced for inferring a mechanism of action of the candidate substance. [Supplementary Item 2] The drug discovery support device according to Supplementary Item 1, wherein the processor searches for a dose-response curve that fits the first predicted value for each added amount, and derives the index value from the found dose-response curve. [Supplementary Item 3] The drug discovery support device according to Supplementary Item 2, wherein the first predicted value is not only a predicted value of the degree of inhibition but also a predicted value of activity indicating the degree to which ion flow in the ion channel has been activated by the candidate substance, and the dose-response curves are of two types: a first dose-response curve when ion flow in the ion channel is inhibited by the candidate substance, and a second dose-response curve when ion flow in the ion channel is activated by the candidate substance, and the processor derives the index value from one of the first dose-response curve and the second dose-response curve that better matches the first predicted value for each added amount. [Supplementary Item 4] The drug discovery support device according to Supplementary Item 2 or Supplementary Item 3, wherein the processor presents the found dose-response curve to the user as the estimated reference information. [Supplementary Item 5] The drug discovery support device according to any one of Supplementary Item 2 to Supplementary Item 4, wherein the curve used to find the dose-response curve is a logistic curve. [Supplementary Item 6] The drug discovery support device according to any one of Supplementary Items 1 to 5, wherein the first predicted value is obtained by inputting a feature derived from the first fluctuation waveform into the prediction model.[Supplementary Item 7] The drug discovery support device according to Supplementary Item 6, wherein the processor acquires the first fluctuation waveform, derives the feature value from the first fluctuation waveform, and inputs the feature value into the prediction model to output the first predicted value from the prediction model. [Supplementary Item 8] The drug discovery support device according to Supplementary Item 7, wherein the processor performs noise reduction processing on the first fluctuation waveform prior to deriving the feature value. [Supplementary Item 9] The drug discovery support device according to Supplementary Item 8, wherein the first fluctuation waveform has periodicity according to pulsation of the iPS cardiomyocytes, and the processor performs averaging processing of multiple periodic portions of the first fluctuation waveform as the noise reduction processing. [Supplementary Item 10] The drug discovery support device according to any one of Supplementary Items 7 to 9, wherein the first fluctuation waveform is measured using multiple electrodes for one iPS cardiomyocyte, and the processor selects one of the multiple first fluctuation waveforms measured by the multiple electrodes as the first fluctuation waveform from which the feature value is to be derived in accordance with preset conditions. [Supplementary Item 11] The drug discovery support device according to any one of Supplementary Items 6 to 10, wherein the feature amount includes the conduction velocity of the first fluctuation waveform. [Supplementary Item 12] The drug discovery support device according to any one of Supplementary Items 6 to 11, wherein the feature amount is normalized by a reference feature amount derived from a reference first fluctuation waveform measured when the candidate substance is not added to the iPS cardiomyocytes. [Supplementary Item 13] The drug discovery support device according to any one of Supplementary Items 6 to 12, wherein, when an experimental value of the index value indicating a dose-response relationship of the candidate substance is available, the first predicted value is obtained by inputting the experimental value in addition to the feature amount into the prediction model. [Supplementary Item 14] The drug discovery support device according to Supplementary Item 13, wherein the prediction model is a model trained in two ways, one where the experimental value is input and one where it is not input. [Supplementary Item 15] The drug discovery support device according to any one of Supplementary Items 6 to 14, wherein the first predicted value is obtained by inputting, into the prediction model, a second predicted value of the index value that indicates a dose-response relationship of the candidate substance, derived based on structural information of the candidate substance, in addition to the feature amount.[Supplementary Item 16] The drug discovery support device according to any one of Supplementary Items 1 to 15, wherein the prediction model is a model trained using training data including simulation data. [Supplementary Item 17] The drug discovery support device according to Supplementary Item 16, wherein the simulation data is generated using a first simulation model that reproduces a second fluctuation waveform that is a fluctuation waveform of an intracellular potential of the iPS cardiomyocyte, and a second simulation model that converts the second fluctuation waveform into the first fluctuation waveform.

[0148] The technology of the present disclosure can be appropriately combined with the various embodiments and / or various modified examples described above. Furthermore, it is not limited to the above embodiments, and various configurations can be adopted without departing from the spirit of the present disclosure. Furthermore, the technology of the present disclosure extends not only to programs, but also to storage media that non-temporarily store programs, and computer program products that include programs.

[0149] The above-described description and illustrations are a detailed explanation of the parts related to the technology of the present disclosure and are merely an example of the technology of the present disclosure. For example, the above description of the configuration, functions, actions, and effects is an explanation of an example of the configuration, functions, actions, and effects of the parts related to the technology of the present disclosure. Therefore, it goes without saying that unnecessary parts may be deleted, new elements may be added, or replacements may be made to the above-described description and illustrations within the scope of the gist of the technology of the present disclosure. Furthermore, to avoid confusion and facilitate understanding of the parts related to the technology of the present disclosure, the above-described description and illustrations omit explanations of common technical knowledge that do not require particular explanation to enable the implementation of the technology of the present disclosure.

[0150] In this specification, "A and / or B" is synonymous with "at least one of A and B." In other words, "A and / or B" means that it may be only A, only B, or a combination of A and B. Furthermore, in this specification, the same concept as "A and / or B" is also applied when three or more things are expressed by connecting them with "and / or."

[0151] All publications, patent applications, and technical standards mentioned in this specification are herein incorporated by reference to the same extent as if each individual publication, patent application, or technical standard was specifically and individually indicated to be incorporated by reference.

Claims

1. A drug discovery support device comprising a processor, wherein the processor obtains, for each ion channel and for each added amount, a first predicted value obtained by outputting from a prediction model a first predicted value of inhibition degree indicating the degree to which the flow of ions in multiple ion channels present in the iPS cardiomyocytes is inhibited by the candidate substance based on a first fluctuation waveform, which is a fluctuation waveform of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, measured when a candidate drug substance is added to the iPS cardiomyocytes while changing the amount of addition; deriving an index value indicating the dose-response relationship of the candidate substance for each ion channel based on the first predicted value for each added amount; and presenting to a user estimated reference information corresponding to the index value, which is referenced to infer a mechanism of action of the candidate substance.

2. The drug discovery support device according to claim 1, wherein the processor searches for a dose-response curve that fits the first predicted value for each added amount, and derives the index value from the dose-response curve found.

3. The drug discovery support device of claim 2, wherein the first predicted value is, in addition to being a predicted value of the degree of inhibition, also a predicted value of the degree of activity indicating the degree to which the flow of ions in the ion channel has been activated by the candidate substance, and the dose-response curves are of two types: a first dose-response curve when the flow of ions in the ion channel is inhibited by the candidate substance, and a second dose-response curve when the flow of ions in the ion channel is activated by the candidate substance, and the processor derives the index value from either the first dose-response curve or the second dose-response curve, whichever is more suitable to the first predicted value for each added amount.

4. The drug discovery support device according to claim 2, wherein the processor presents the discovered dose-response curve to the user as the estimated reference information.

5. The drug discovery support device according to claim 2, wherein the curve used to search for the dose-response curve is a logistic curve.

6. A drug discovery support device as described in claim 1, wherein the first predicted value is obtained by inputting a feature derived from the first fluctuation waveform into the prediction model.

7. The drug discovery support device of claim 6, wherein the processor acquires the first fluctuation waveform, derives the feature from the first fluctuation waveform, and inputs the feature into the prediction model, thereby outputting the first predicted value from the prediction model.

8. The drug discovery support device according to claim 7, wherein the processor performs noise reduction processing on the first fluctuation waveform prior to deriving the feature amount.

9. A drug discovery support device as described in claim 8, wherein the first fluctuation waveform has a periodicity corresponding to the pulsation of the iPS cardiomyocytes, and the processor performs an averaging process of multiple periodic portions of the first fluctuation waveform as the noise reduction process.

10. The drug discovery support device described in claim 7, wherein the measurement of the first fluctuation waveform is performed using multiple electrodes on one of the iPS cardiomyocytes, and the processor selects one of the multiple first fluctuation waveforms measured by the multiple electrodes in accordance with preset conditions as the first fluctuation waveform from which the feature is derived.

11. A drug discovery support device as described in claim 6, wherein the feature amount includes a conduction velocity of the first fluctuation waveform.

12. A drug discovery support device as described in claim 6, wherein the feature is normalized by a reference feature derived from a reference first fluctuation waveform measured when the candidate substance is not added to the iPS cardiomyocytes.

13. A drug discovery support device as described in claim 6, wherein, when there is an experimental value of the index value indicating the dose-response relationship of the candidate substance, the first predicted value is obtained by inputting the experimental value in addition to the feature into the prediction model.

14. The drug discovery support device according to claim 13, wherein the prediction model is a model that has been trained in two ways, that is, when the experimental values ​​are input and when they are not input.

15. A drug discovery support device as described in claim 6, wherein the first predicted value is obtained by inputting into the prediction model a second predicted value of the index value indicating the dose-response relationship of the candidate substance, derived based on structural information of the candidate substance, in addition to the feature amount.

16. The drug discovery support device according to claim 1, wherein the predictive model is a model trained using training data including simulation data.

17. A drug discovery support device as described in claim 16, wherein the simulation data is generated using a first simulation model that reproduces a second fluctuation waveform, which is the fluctuation waveform of the intracellular potential of the iPS cardiomyocyte, and a second simulation model that converts the second fluctuation waveform into the first fluctuation waveform.

18. A method for operating a drug discovery support device comprising: obtaining a first predicted value for each ion channel and for each added amount obtained by outputting a first predicted value of inhibition degree indicating the degree to which the flow of ions in multiple ion channels present in the iPS cardiomyocytes is inhibited by the candidate substance from a prediction model based on a first fluctuation waveform that is a fluctuation waveform of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, when a candidate drug substance is added to the iPS cardiomyocytes while changing the added amount; deriving an index value indicating the dose-response relationship of the candidate substance for each ion channel based on the first predicted value for each added amount; and presenting to a user estimated reference information corresponding to the index value, which is referenced to infer the mechanism of action of the candidate substance.

19. An operating program for a drug discovery support device that causes a computer to execute processes including: obtaining a first predicted value for each ion channel and for each added amount obtained by outputting a first predicted value of the degree of inhibition indicating the degree to which the flow of ions in multiple ion channels present in the iPS cardiomyocytes is inhibited by the candidate substance from a prediction model based on a first fluctuation waveform that is a fluctuation waveform of the extracellular potential of iPS cardiomyocytes, which are cardiomyocytes derived from human iPS cells, when a candidate drug substance is added to the iPS cardiomyocytes while changing the amount of addition; deriving an index value indicating the dose-response relationship of the candidate substance for each ion channel based on the first predicted value for each added amount; and presenting to a user estimated reference information corresponding to the index value, which is referenced to infer the mechanism of action of the candidate substance.

Images