KR20230040261A - System and method for optimizing general purpose biological network for drug response prediction using meta-reinforcement learning agent - Google Patents

Abstract

Disclosed is a cancer treatment candidate drug determining method which comprises the following steps of: allowing a simulation device to generate a plurality of specific perturbation networks by reflecting mutation information of a first cancer cell line in each of a plurality of drug response networks for a plurality of drugs; and selecting a plurality of candidate drugs among the plurality of drugs based on a plurality of death probabilities of the first cancer cell line outputted by the plurality of specific perturbation networks.

Description

System and method for optimizing general purpose biological network for drug response prediction using meta-reinforcement learning agent}

The present invention defines a bio-signaling network for predicting the survival and death of cells due to external stimuli to cells, and implements a method of optimizing parameters included in the bio-signaling network using a computing device. it's about

Cells can either live or die. Various proteins included in cells can contribute to cell survival or death while influencing each other. A set of proteins included in a cell can affect the expression levels of other proteins according to their respective expression levels. A meaningful network representing the association between such a set of proteins can be constructed, and it can be referred to as a biosignal transduction network or a biosignal transduction network.

The bio-signaling network may be composed of nodes and links connecting the nodes. Each node may mean a specific protein present in the cell. A weight may be assigned to each link. This weight is the expression level of the first protein representing the first node connected to the first end of both ends of the link corresponding to the weight, and the expression level of the second protein representing the second node connected to the second end of the both ends. It can indicate the degree or strength of the impact on

Various biological signal transduction networks can be defined for one cell. Among them, one specific biosignal transduction network may be specifically related to the expression and death of a specific cancer cell, and the other biosignal transduction network may be particularly related to the expression and death of other cancer cells.

When the cell is normal, a specific biosignal transduction network defined for the cell may be referred to as a nominal biosignal transduction network. The state of the cell may be determined by a combination of values of nodes of the nominal biosignaling network. The value of each node may be determined by a state transition equation that determines the time-dynamics of the nominal bio-signaling network. The state transition equation may depend on the weight of each link.

When a mutation occurs in the cell and the normal cell is changed into a cancer cell, a certain node in the nominal bio-signaling network may have a different type of time-dynamics without following the state transition equation. For example, a node where a mutation has occurred may always have only a specific value over time. Such a mutated biosignal transduction network may be referred to as a cancer cell biosignal transduction network. The cancer cell bio-signaling network may have a feature that prevents the cancer cells from dying over time. At this time, when a specific drug is administered to the cancer cells, the specific drug can affect the expression level of a specific node of the cancer cell bio-signaling network, and the cancer cells are killed by a chain reaction induced therefrom. You may. At this time, it can be said that the specific drug perturbs the specific node. Also, the specific drug may perturb one node or a plurality of nodes. Finding a drug that kills the cancer cells can give a good effect to cancer treatment.

Although it is possible to find the optimal drug through an experiment in which various drugs are sequentially administered to actual cancer cells, this method requires a lot of cost and time, and in the meantime, the condition of the cancer patient may deteriorate, and ultimately, the cancer patient treatment may fail. Therefore, if a simulation using a computing device can be used to quickly find a drug suitable for cancer patients, it can be of great help in the treatment of cancer patients.

Several methods have been disclosed as such a simulation method.

Some of these previously disclosed technologies use the biosignal transfer network described above. At this time, reliability of the simulation result may be determined by a weight assigned to each link of the bio-signal transmission network. Therefore, it is important to find the optimal weight. These optimal weights can be determined by the experience and consideration of the bio-signaling network of the researcher who devises the simulation method, but it can be expected that there may be limitations.

Therefore, the present invention intends to provide a technique using machine learning to determine the optimal weight.

Prior art related to biosignal transduction network and target drug determination method using the same are Korean Patent Application Nos. etc. are presented.

In the present invention, it is intended to provide a technique for determining weights associated with links in a cancer cell bio-signaling network that can be commonly applied to various types of cancer and various patients regardless of the type of cancer and the location of a mutation.

In the present invention, it is intended to provide a technique for optimizing the parameters of a modeled biological signal transduction network through machine learning so as to give biological meaning to its internal structure.

The present invention intends to provide a technique for learning an agent (weight determination agent) that plays a role in determining weights assigned to the links in a bio-signal transfer network composed of nodes and links. In addition, it is intended to provide a technology for selecting a drug suitable for treatment of a new cancer patient by using the learned agent.

According to one aspect of the present invention, in a drug responsive network composed of nodes and links, an agent serving to determine weights assigned to the links may be provided. When the weights assigned to the links of the drug responsiveness network are determined to be appropriate values, the bio-signaling network can output a more accurate cell death probability.

The agent may include a learnable network such as a machine learning network or a neural network, and may include a plurality of layers.

For learning of the agent, a set of cancer cell lines for learning and a plurality of drugs (drug combinations) may be used as learning data. Information on one set of cancer cell lines (N cancer cell lines) and one drug for learning can be used for one learning step.

When the drug is administered to the set of cancer cell lines for learning in an in vitro experiment, the death rate of the set of cancer cell lines for learning can be observed and determined. The mortality rates may be presented as a vector Z consisting of N scalar values, for example.

In addition, a set of specific perturbation networks may be generated by applying mutation information of the set of cancer cell lines for learning to the drug responsiveness network. In addition, a set of death probabilities that can be obtained from the set of specific perturbation networks can be calculated. The mortality rates may be presented as, for example, a vector Y consisting of N scalar values.

The reward calculation unit provided according to one aspect of the present invention may calculate a reward, which is a value to be input to the agent. The reward calculator may calculate the reward using a distance between the vector Y and the vector Z.

The agent may receive the rewards and weights assigned to links of the drug responsiveness network as input data. The agent may output information obtained by updating values of weights to be assigned to links of the drug responsiveness network in a next learning step based on the above input data.

In the present specification, the term 'learning step' means updating the weights of the drug responsiveness network. In contrast, in order for the agent to be learned once, the learning step needs to be executed a plurality of times.

A set of a plurality of continuously executed learning steps may be referred to as a learning episode. For all learning steps within one episode, the number of drugs used as learning data may be limited to one. Only after the episode has been changed can the study drug be changed.

Values of the weights of the drug responsiveness network may be updated once for each learning step. In addition, when the episode is executed once, the agent may be learned once. Each time the episode is repeated, the learning amount of the agent increases.

The sufficiently learned agent can be used to select a drug for killing a new cancer cell line.

According to one aspect of the present invention, a computing device obtains a death probability of the cancer cell line predicted by a specific perturbation network generated by applying mutation information of the cancer cell line to a drug responsiveness network that responds to a specific drug, and the A first step of obtaining an apoptosis rate of the cancer cell line obtained by performing an in-vitro experiment in which the specific drug is administered to the cancer cell line, and calculating a reward value reflecting the difference between the apoptosis probability and the apoptosis rate; a second step of calculating new weights for the links of the drug responsiveness network by inputting the reward values into an agent; and a third step of updating the weights of the links of the drug responsiveness network with the new weights; a program including instructions for determining the weights of the links of the drug responsiveness network by executing a learning step including, A computer-readable non-volatile recording medium may be provided.

At this time, the program further includes instructions for causing the computing device to learn the agent once based on the plurality of rewards obtained in the process of executing the learning step a plurality of times and the plurality of new weights. can include

At this time, in the first step, the cancer cell predicted by the pth specific perturbation network generated by applying the mutation information of the pth cancer cell line [p] among the prepared N cancer cell lines to the drug responsiveness network Obtaining the death probability [y _p ] of the strain [p], and performing an in-vitro experiment in which the specific drug is administered to the cancer cell line [p], the death rate [z _p ] of the cancer cell line [p] The acquiring step may be performed for each of the N cancer cell lines (p=1, 2, 3, ..., N). And the first step is a value inversely proportional to the distance between the vector Y consisting of the death probabilities obtained for the N cancer cell lines and the vector Z consisting of the death rates obtained for the N cancer cell lines. Calculating the reward value based on 1 value; may include.

At this time, in the first step, the cancer cell predicted by the pth specific perturbation network generated by applying the mutation information of the pth cancer cell line [p] among the prepared N cancer cell lines to the drug responsiveness network Obtaining the death probability [y _p ] of the strain [p], and performing an in-vitro experiment in which the specific drug is administered to the cancer cell line [p], the death rate [z _p ] of the cancer cell line [p] The acquiring step may be performed for each of the N cancer cell lines (p=1, 2, 3, ..., N). And the first step is a value inversely proportional to the distance between the vector Y consisting of the death probabilities obtained for the N cancer cell lines and the vector Z consisting of the death rates obtained for the N cancer cell lines. Calculating a value of 1; and calculating the reward value based on a difference between the first value and a previously prepared second value. In this case, the second value may be a value inversely proportional to the distance between the past vector Y and the past vector Z constructed in the past learning step that has already been completed immediately before calculating the first value.

According to another aspect of the present invention, a computing device including a processing unit and a storage unit may be provided. The processing unit is configured to execute an episode, which is a process of learning an agent to determine weights of links of a drug responsiveness network that respond to a specific drug. The processing unit is adapted to execute the step of executing the one episode step of executing a predetermined learning step a plurality of times. In the learning step, the death probability of the cancer cell line predicted by the specific perturbation network generated by applying the mutation information of the cancer cell line to the drug responsiveness network is obtained, and the specific drug is administered to the cancer cell line In-vitro A first step of obtaining an apoptosis rate of the cancer cell line obtained by performing an experiment, and calculating a reward value in which the difference between the apoptosis probability and the apoptosis rate is reflected; a second step of calculating new weights for links of the drug responsiveness network by inputting the reward value to the agent; and a third step of updating the weights of the links of the drug responsiveness network with the new weights.

At this time, in the first step, the cancer cell predicted by the pth specific perturbation network generated by applying the mutation information of the pth cancer cell line [p] among the prepared N cancer cell lines to the drug responsiveness network Obtaining the death probability [y _p ] of the strain [p], and performing an in-vitro experiment in which the specific drug is administered to the cancer cell line [p], the death rate [z _p ] of the cancer cell line [p] The acquiring step may be performed for each of the N cancer cell lines (p=1, 2, 3, ..., N). The first step is a first step, which is a value inversely proportional to the distance between the vector Y consisting of the death probabilities obtained for the N cancer cell lines and the vector Z consisting of the death rates obtained for the N cancer cell lines. Calculating the reward value based on the value; may include.

At this time, the first step is a value inversely proportional to the distance between the vector Y consisting of the death probabilities obtained for the N cancer cell lines and the vector Z consisting of the death rates obtained for the N cancer cell lines. calculating a first value; and calculating the reward value based on a difference between the first value and a previously prepared second value. In this case, the second value may be a value inversely proportional to the distance between the past vector Y and the past vector Z constructed in the past learning step that has already been completed immediately before calculating the first value.

According to one aspect of the present invention, a method for generating a bio-signal transfer network including executing a predetermined learning step by a computing device may be provided. In the learning step, the computing device acquires the death probability of the cancer cell line predicted by the specific perturbation network generated by applying mutation information of the cancer cell line to the drug responsiveness network that responds to a specific drug, and A first step of obtaining a death rate of the cancer cell line obtained by performing an in-vitro experiment in which the specific drug is administered, and calculating a reward value in which the difference between the death probability and the death rate is reflected; a second step of calculating, by the computing device, new weights for links of the drug responsiveness network by inputting the reward value to an agent; and a third step of updating, by a computing device, weights of links of the drug responsiveness network to the new weights.

At this time, the learning step may be repeatedly executed. And in the first step, the cancer cell line predicted by the p-th specific perturbation network generated by applying the mutation information of the p-th cancer cell line [p] among the prepared N cancer cell lines to the drug responsiveness network Obtain the death probability [y _p ] of [p] and obtain the death rate [ _z p ] of the cancer cell line [p] obtained by performing an in-vitro experiment in which the specific drug is administered to the cancer cell line [p] The step of doing may be performed for each of the N cancer cell lines (p = 1, 2, 3, ..., N). At this time, the first step is a value inversely proportional to the distance between the vector Y consisting of the death probabilities obtained for the N cancer cell lines and the vector Z consisting of the death rates obtained for the N cancer cell lines. calculating a first value; and calculating the reward value based on a difference between the first value and a previously prepared second value. In this case, the second value may be a value inversely proportional to the distance between the past vector Y and the past vector Z constructed in the past learning step that has already been completed immediately before calculating the first value.

According to one aspect of the present invention, generating a plurality of specific perturbation networks by reflecting mutation information of a first cancer cell line to a plurality of drug responsiveness networks for a plurality of drugs, respectively, by the simulation apparatus 810; selecting, by the simulation device, a plurality of candidate drugs from among the plurality of drugs based on a plurality of death probabilities of the first cancer cell line output from the plurality of specific perturbation networks; providing information on the determined plurality of candidate drugs to a drug response screening device; executing, by the drug response screening device, an in-vitro experiment in which the plurality of candidate drugs are administered to a plurality of wells in which the first cancer cell line is stored; capturing and analyzing, by the drug response screening device, images of the first cancer cell line in the plurality of wells using a cell image capturing device; and outputting, by the drug reaction screening device, in vitro test results for at least some of the plurality of candidate drugs based on the analyzed results; a method for determining candidate drugs for cancer treatment may be provided.

At this time, prior to the generating step, the computing device 710 performs a process (=episode) of determining the weight of the kth drug responsiveness network that responds to the kth drug among the plurality of drug responsiveness networks. can include more. In addition, the step of performing the process may be configured to use an agent whose learning has been completed by reinforcement learning. At this time, the step of performing the process may include, by the computing device, mutation information of N (= p _k ) cell lines having information on reactivity by an in vitro experiment using the kth drug among the plurality of drugs. obtaining them; generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug; repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent; selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network.

In this case, the agent may be configured to determine the weights of the links of the kth drug responsiveness network in the next learning step based on the reward and the weights of the links of the kth drug responsiveness network in the current learning step.

At this time, in the process of determining the reward, the computing device, in the current learning step among the plurality of learning steps, a vector Y consisting of N death probabilities output by the N specific perturbation networks, and the preparing a vector Z composed of values related to the death rate of the N first cancer cell lines observed by an in vitro experiment in which a k-th drug is administered to the first cancer cell line; calculating, by the computing device, a first value that is inversely proportional to a distance between the vector Y and the vector Z; and calculating, by the computing device, the reward based on a difference between the first value and the second value. The second value may be a value inversely proportional to a distance between the vector Y and the vector Z prepared in a learning step immediately before the current learning step.

In this case, prior to the step of determining the weight of the kth drug responsiveness network (=episode), the computing device 710 may further include training the agent. In the step of learning the agent, the process of learning the agent (=episode) may be repeatedly performed for G different drugs. At this time, in the learning process of the agent performed for the g-th drug, the computing device 710 provides mutation information of p _g cell lines having information on reactivity by an in vitro experiment using the g-th drug. obtaining them; generating, _by the computing device, p g specific perturbation networks by applying the p _g pieces of the mutation information to the p th drug responsiveness network that responds to the p th drug; repeating, by the computing device, a learning step of updating weights of links of the g-th drug responsiveness network a plurality of times using the agent; and learning, by the computing device, the agent using the reward values and the weights obtained in the process of repeating the learning step a plurality of times.

In this case, the agent may be configured to determine the weights of the links of the gth drug responsiveness network in the next learning step based on the reward and the weights of the links of the gth drug responsiveness network in the current learning step.

According to another aspect of the present invention, performing, by a computing device 710, a process (=episode) of determining a weight of a kth drug responsiveness network that responds to a kth drug among the plurality of drug responsiveness networks; generating, by the simulation device 810, a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line in each of the plurality of drug response networks for the plurality of drugs; and selecting, by the simulation device, a plurality of candidate drugs from among the plurality of drugs based on a plurality of death probabilities of the first cancer cell line output by the plurality of specific perturbation networks. \,Cancer treatment candidate drug determination method can be provided. At this time, the step of performing the process is to use an agent whose learning has been completed by reinforcement learning. In the step of performing the process, the computing device retrieves mutation information of N (= p _k ) cell lines having information on reactivity by an in vitro experiment using the kth drug among the plurality of drugs. obtaining; generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug; repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent; selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network.

According to another aspect of the present invention, the simulation device 810; drug response screening device 600; And a computing device 710; cancer treatment candidate drug determination system comprising a; can be provided. The simulation device generates a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line in each of a plurality of drug reactivity networks for a plurality of drugs, and outputs the plurality of specific perturbation networks. Based on the plurality of death probabilities of the first cancer cell line, a plurality of candidate drugs are selected from among the plurality of drugs, and information on the determined plurality of candidate drugs is provided to the drug response screening device. And, the drug response screening device executes an in-vitro experiment in which the plurality of candidate drugs are administered to a plurality of wells in which the first cancer cell line is stored, and the drug response screening device takes a cell image. An image of the first cancer cell line in the plurality of wells is captured and analyzed using a device, and the drug response screening device performs an in vitro experiment on at least some of the plurality of candidate drugs based on the analysis result. It is supposed to output the result.

At this time, the computing device determines the weight of the kth drug responsiveness network that responds to the kth drug among the plurality of drug responsiveness networks before the simulation device generates the plurality of specific perturbation networks. (=episode). Further, the process of performing the process may be configured to use an agent whose learning has been completed by reinforcement learning. In the process of performing the process, the computing device retrieves mutation information of N (= p _k ) cell lines having information on reactivity by an in vitro experiment using the kth drug among the plurality of drugs. obtaining; generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug; repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent; selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network.

In this case, the agent may be configured to determine the weights of the links of the kth drug responsiveness network in the next learning step based on the reward and the weights of the links of the kth drug responsiveness network in the current learning step.

At this time, in the process of determining the reward, the computing device, in the current learning step among the plurality of learning steps, a vector Y consisting of N death probabilities output by the N specific perturbation networks, and the preparing a vector Z consisting of values related to the death rate of the N first cancer cell lines observed by an in vitro experiment in which k-th drug is administered to the first cancer cell line; calculating, by the computing device, a first value that is inversely proportional to the distance between the vector Y and the vector Z; and calculating, by the computing device, the reward based on a difference between the first value and the second value. In this case, the second value may be a value inversely proportional to a distance between the vector Y and the vector Z prepared in a learning step immediately before the current learning step.

In this case, before performing the process (=episode) of determining the weights of the kth drug responsiveness network, the computing device 710 may be configured to train the agent. In addition, in the process of learning the agent, the process of learning the agent (=episode) may be repeatedly performed for G different drugs. At this time, in the learning process of the agent performed for the g-th drug, the computing device 710 provides mutation information of p _g cell lines having information on reactivity by an in vitro experiment using the g-th drug. obtaining them; generating, _by the computing device, p g specific perturbation networks by applying the p _g pieces of the mutation information to the p th drug responsiveness network that responds to the p th drug; repeating, by the computing device, a learning step of updating weights of links of the g-th drug responsiveness network a plurality of times using the agent; and learning, by the computing device, the agent using the reward values and the weights obtained in the process of repeating the learning step a plurality of times.

In this case, the agent may be configured to determine the weights of the links of the gth drug responsiveness network in the next learning step based on the reward and the weights of the links of the gth drug responsiveness network in the current learning step.

Simulation device 810 according to another aspect of the present invention; drug response screening device 600; And a computing device 710; cancer treatment candidate drug determination system comprising a; can be provided. At this time, the computing device is configured to perform a process (=episode) of determining a weight of a kth drug responsiveness network that responds to a kth drug among the plurality of drug responsiveness networks. The simulation device generates a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line in each of a plurality of drug reactivity networks for a plurality of drugs, and the plurality of specific perturbation networks A plurality of candidate drugs are selected from among the plurality of drugs based on the plurality of death probabilities of the first cancer cell line output, and the step of performing the process includes an agent for which learning has been completed by reinforcement learning. is meant to be used. In the step of performing the process, the computing device retrieves mutation information of N (= p _k ) cell lines having information on reactivity by an in vitro experiment using the kth drug among the plurality of drugs. obtaining; generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug; repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent; selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network.

According to the existing technology, the weight of the cancer cell bio-signaling network, which is defined to find the optimal drug to be administered to a specific cancer patient, considers the type of cancer of the cancer patient and the position of a mutation specifically occurring in the cancer patient. it was decided Therefore, the cancer cell bio-signaling network for each cancer patient had to be individually defined.

However, according to the present invention, it is possible to determine weights associated with links of a cancer cell bio-signaling network that can be commonly applied to various types of cancer and various patients, regardless of the type of cancer and the position of the mutation.

According to the present invention, it is possible to optimize the parameters of the modeled biological signal transduction network through machine learning so as to give biological meaning to its internal structure. Therefore, it is possible not only to select the optimal drug for cancer treatment using machine learning, but also to generate data suitable for interpreting the biological significance of the parameters determined through machine learning.

According to the present invention, in a bio-signaling network composed of nodes and links, it is intended to provide a technique for learning an agent that plays a role in determining weights assigned to the links. In addition, it is possible to provide a technique for selecting a drug suitable for treatment of a new cancer patient by using the agent for which learning has been completed.

Figure 1a explains the concept of a biological signal transduction network.
1B is for explaining the concept of a drug responsiveness network, which is a concept used in the present invention.
1c shows a drug responsiveness network in which cell mutation information is reflected.
2A illustrates a method of defining and generating a plurality of different specific perturbation networks from a specific drug responsiveness network according to an embodiment of the present invention.
Fig. 2b illustrates in another way how to generate the specific perturbation network of Fig. 2a.
3 illustrates a method of determining weights assigned to links of a specific drug responsiveness network according to an embodiment of the present invention.
Figure 4 shows the predicted value of death probability y _p calculated for cancer cell line [p] and cancer cells when drug [k] is administered to cancer cell line [p] for the execution of one learning step according to an embodiment of the present invention It is a block diagram explaining the function of the reward calculation unit that calculates the reward value using the observed value z _p of the actual mortality rate of week [p].
5 is a block diagram illustrating a process of updating weights assigned to links of the nominal network using calculated rewards.
6 is a flowchart illustrating a method of updating weights assigned to links of a drug responsiveness network related to a specific drug through one learning step provided by an embodiment of the present invention.
FIG. 7 shows a method of determining the weights of the drug responsiveness network to optimal values using the method of updating the weights of the drug responsiveness network described in FIG. 6 .
8 illustrates a concept of determining a plurality of different drug responsiveness networks from a given nominal network.
9 illustrates a process of finding a drug suitable for patient [x] using a plurality of different drug reactivity networks determined according to an embodiment of the present invention.
FIG. 10 shows a process of determining the optimal drug for patient [k] using K specific perturbation networks [x] [k] prepared as shown in FIG. 9 .
11 illustrates a configuration of a computing device executing a method of completing a drug responsiveness network by determining weights of the drug responsiveness network according to an embodiment of the present invention.
12 illustrates a configuration of a computing device executing a simulation method for determining an optimal drug effective for killing a specific cancer cell line according to an embodiment of the present invention.
13 shows the structure of a cancer treatment candidate drug determination system provided according to an embodiment of the present invention.
FIG. 14 is a framework for explaining a method for learning the completed agent presented in FIG. 3 .
15 shows the configuration of a system for obtaining and providing a death rate Z by administering a specific drug to cancer cell lines according to an embodiment of the present invention.
16A shows an example of a business model using the present invention.
16B shows another example of a business model using the present invention.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be implemented in various other forms. Terms used in this specification are intended to aid understanding of the embodiments, and are not intended to limit the scope of the present invention. Also, the singular forms used herein include the plural forms unless the phrases clearly dictate the contrary.

Figure 1a explains the concept of a biological signal transduction network.

In this specification, the bio-signal transfer network may also be referred to as a bio-signal transfer network.

Reference number 500 conceptually suggests the structure of a specific biological signal transduction network of normal cells. The reference number 500 may be referred to as a 'nominal network'.

In one embodiment, the bio-signal transfer network may be composed of a plurality of nodes and a plurality of links connecting them. At this time, each node represents the activity of the protein in the cell. Each node may be modeled to have a binary value or a real value. Each link represents the effect of the activity of the first node at the start point of the link on the activity of the second node at the end point (arrow or square) of the link. A link whose end point is indicated by an arrow indicates that the activity of the first node has a positive effect on the activity of the second node, and a link whose end point is indicated by a square indicates that the activity of the first node indicates that it has a negative effect on the activity of A weight is assigned to each link, which may represent the strength of the positive or negative influence. The structure of the bio-signaling network may be configured using knowledge revealed by research in the field of existing biomolecules.

The modeling method may be selected from among a plurality of methods. Depending on different modeling methods, the number of types of links and expression methods may be slightly different.

In this specification, when a mutation exists in a specific node of the nominal network, the network in which the mutation exists may be referred to as a specific network. That is, the specific network may mean a nominal network in which cell mutation information is reflected.

Reference number 510 denotes a 'first specific network' representing a node corresponding to a mutation existing in the first cancer cell line, that is, cancer cell line [1], in the nominal network, that is, a 'specific network'. [1]'. Nodes with mutations are marked in black.

Reference number 520 denotes a 'second specific network' representing a node corresponding to a mutation present in a second cancer cell line, that is, cancer cell line [2], in the nominal network, that is, a 'specific network'. [2]'. Nodes with mutations are marked in black.

The cancer cell line [k] may be replaced with the concept and term of cancer cell [k].

In this way, when a node corresponding to a mutation present in cancer cell line [k] exists in the nominal network, it may be referred to as a 'specific network [k]'.

Reference number 521 indicates that when a specific drug is administered to the cancer cell line [2], a target node whose expression level is affected by the specific drug exists in the specific network [2], expressing this. It represents the cipic perturbation network [2]'. Nodes with mutations are shown in black, and the two target nodes are shown in gray.

1B is for explaining the concept of a drug responsiveness network, which is a concept used in the present invention.

A plurality of different drug responsiveness networks can be defined from the nominal network 500 . Each of the drug responsiveness networks may be regarded as a subnetwork composed of some of the spheres of the nominal network 500 .

1B shows a first drug responsiveness network 500[1] composed of nodes of node numbers 3, 5, 6, and 7 and a second drug responsiveness network 500 composed of nodes of node numbers 1, 2, and 3. [2]) is presented.

Although FIG. 1B shows only two drug responsiveness networks defined from the nominal network 500, it can be easily understood that more drug responsiveness networks can be defined. For example, a kth drug reactivity network 500[k] not shown in FIG. 1B may be further defined.

In the kth drug responsiveness network 500[k], state transition equations for determining the state value of each node at each time point may have already been defined. Techniques for such state transition equations are exemplified in, for example, Korean Patent Registration Nos. KR 10-2029297 and KR 10-1975424.

In this case, at least some of the coefficients included in the state transition equations may be determined by a weight assigned to each link of the kth drug responsiveness network 500[k]. The weights should be selected to optimal values. This is an important problem to be solved in the present invention, which determines the value of the optimal weight to be assigned to each link of the kth drug responsiveness network (500[k]), and is solved by a specific embodiment of the present invention described below. Means may be provided.

Different drug responsive networks are sub-networks with different sub-structures of the nominal network. Therefore, even if there is a link common to two different drug responsiveness networks, the weight assigned to the link may have different values depending on the two drug responsiveness networks.

In one embodiment of the present invention, the process of determining weights assigned to links existing in each drug responsiveness network among a plurality of drug responsiveness networks may be independently performed for each drug responsiveness network.

1c shows a drug responsiveness network in which cell mutation information is reflected.

When a mutation exists in node 7 of the nominal network 500, the mutation information is applied to the first drug responsiveness network 500[1] shown in FIG. 1B [7][1] ] can be defined.

Alternatively, in the same manner as above, when a mutation exists in node 6 of the nominal network 500, a drug to which the mutation information is applied to the first drug reactivity network 500 [1] shown in FIG. 1B. Reactive networks [6][1] can be defined.

In this way, a drug responsiveness network in which cell mutation information is reflected may be referred to as a specific perturbation network.

Although FIG. 1C illustrates two specific perturbation networks defined from the first drug responsiveness network 500[1], the fact that a larger number of specific perturbation networks can be defined using other mutation information is can be easily understood.

The above-described FIGS. 1A, 1B, and 1C may collectively be referred to as FIG. 1 .

<Determination process of weights of specific drug responsiveness network>

2A illustrates a method of defining and generating a plurality of different specific perturbation networks from a specific drug responsiveness network according to an embodiment of the present invention.

On the left side of FIG. 2A, the kth drug responsiveness network 500[k], which is a drug responsiveness network for drug [k], is presented. If mutation information of p _k different cell lines is reflected in the kth drug responsiveness network 500[k], p _k different specific perturbation networks can be defined. For example, if the p-th mutation information among the mutation information of a total of p _k different cell lines prepared in advance is reflected in the k-th drug reactivity network 500 [k], a specific perturbation network [p] [k] is generated. can

The specific perturbation network [p][k] can output the predicted value y[p][k] of the death probability of the cancer cell line [p] when the drug [k] is administered to the cancer cell line [p].

The p _k different cell lines may be selected from among the P cell lines of the population (P>p _k ). The mutation information of the p _k different cell lines may be selected from the mutation information of the P cell lines.

At this time, information on reactivity to the drug [k] may not exist for all of the P cell lines. For example, an experiment in which the drug [k] was administered to some of the P cell lines was performed, but an experiment in which the drug [k] was administered to the other cell lines may not be performed. . That is, the information on the reactivity to the drug [k] may exist only for some of the P cell lines.

Among the p _k different cell lines used to generate specific perturbation networks from the k th drug reactivity network 500 [k], information on reactivity to drug [k] among the P cell lines is It may be composed of some existing cell lines.

The number of specific perturbation networks each obtained from different drug responsiveness networks may be different. For example, the number of specific perturbation networks obtained from the first drug responsiveness network for drug [1] is p ₁ , and the number of specific perturbed networks obtained from the second drug responsiveness network for drug [2] is If the number is p ₂ , p ₁ may be different from p ₂ .

In the specific perturbation network [p][k], state transition equations for determining the state value of each node over time may be already defined.

For example, the state transition equations for the specific perturbation network [p][k] may be basically the same as the state transition equations of the kth drug responsiveness network 500[k]. However, for example, only one or a plurality of state transition equations for determining the states of nodes corresponding to positions of transitions existing in the specific perturbation network [p][k] may be modified.

Fig. 2b illustrates in another way how to generate the specific perturbation network of Fig. 2a.

A specific network [p] may be generated by applying the information (MN[p]) on mutation-producing nodes of the cancer cell line [p] to the nominal network 500 .

The specific perturbation network [p][k] may be generated by applying the information (PT[k]) on the perturbation target node of the drug [k] to the generated specific network [p].

At this time, when the drug [k] is administered to the cancer cell line [p], the specific perturbation network [p][k] can output the predicted value y [p] [k] of the death probability of the cancer cell line [p]. .

The above-described FIGS. 2A and 2B may collectively be referred to as FIG. 2 .

3 illustrates a method of determining weights assigned to links of a specific drug responsiveness network according to an embodiment of the present invention.

3 shows a framework for determining the weights of links of the k-th drug responsiveness network 500 [k] related to drug [k]. As the framework, a set of specific perturbation networks [p][k] generated from the kth drug reactivity network 500[k] described in FIG. 2 may be used (p = 1, 2, 3, ... p _k ). In addition, the framework may use the reward calculation unit 30 and the agent 20 together.

In this specification, agent 20 may also be referred to as a weight determination agent.

The agent 20 may be an information processing module including a network including a neural network. The agent 20 includes a learnable network such as a machine learning network and a neural network, and may include a plurality of layers. The neural network can be trained by reinforcement learning. The above (20) used in FIG. 3 may have already been learned. A specific method of learning the above (20) will be described later in this specification.

From a set of specific perturbation networks [p][k] presented in FIG. 3, prediction values y[p][k] of a set of cancer cell lines [p] can be output (p=1, 2, 3, ... p _k ).

Hereinafter, the y[p][k] output for drug [k] can be simply expressed as y _p , and the index p _k can be expressed by replacing the index N.

Now, a prediction vector Y={y ₁ , y ₂ , y ₃ , ..., y _N } may be generated using the p _k predicted death probability values y _p , that is, the N death probability predicted values y _p .

In addition, when the drug [k] is administered to the cancer cell line [p], the observation on the actual death rate of the cancer cell line [p] can be prepared through an in vitro experiment. The results of the in vitro experiment may be obtained from existing public data. Therefore, when drug [k] is administered to cancer cell line [p], N observation values z _p regarding the actual death rate of cancer cell line [p] may also be prepared from p=1 to p=N (N=p _k ). there is. An observation value vector Z={z ₁ , z ₂ , z ₃ , ..., z _N } can be created using N observation values z _p .

Figure 4 shows the predicted value of death probability y _p calculated for cancer cell line [p] and cancer cells when drug [k] is administered to cancer cell line [p] for the execution of one learning step according to an embodiment of the present invention It is a block diagram explaining the function of the reward calculator that calculates the reward value using the observed value z _p of the actual mortality rate of week [p] (p=1, 2, 3. ..., N).

The reward calculator 30 may calculate the reward value only when both the predicted value y _p and the observed value z _p for all values from p=1 to p=N are input.

In the present specification, the 'predicted value' may be referred to as a 'simulation predicted value', and the 'observed value' may be referred to as an 'in vitro observed value'.

The error calculation unit 31 generates a predicted value vector Y consisting of the predicted values y _p for all values from p = 1 to p = N, and the observed values for all values from p = 1 to p = N The distance between the observation value vectors Z composed of z _p can be obtained and regarded as the prediction error Err(i) of the specific perturbation network. Here, i is an index representing the i-th learning step (learning iteration). Also, the error calculator 31 may output a first value h/Err(i) that is inversely proportional to the prediction error Err(i).

The first value h/Err(i) is stored in the past error storage unit 32 and can be used later. That is, for example, the first value h/Err(i) stored in the past error storage unit 32 may be used in relation to the i+1 th learning step (learning iteration) executed after the storage is made. can

Similarly, the second value h/Err(i-1), which is inversely proportional to the prediction error Err(i-1) obtained in the i-1th learning step, is already stored in the past error storage unit 32, There may be.

The reward calculator 33 may calculate the reward value based on a difference between the first value h/Err(i) and the second value h/Err(i−1).

A specific method of calculating the reward value is as follows.

The reward input to the agent 20 at the i+1th learning step can be calculated as follows.

First of all, Err(0) ~ Err(i-1) of the errors obtained from the 0th learning step to the i-1th learning step are stored.

At this time, values in inverse proportion to the errors obtained from the 0th learning step to the i-1th learning step h/Err(0), h/Err(1), h/Err(2), ..., h/Err Among (i-1), the maximum value can be selected.

Assuming that the maximum value is h/Err(j), the d(i) value can be calculated as in Equation 1 below (j = 0, 1, 2, ..., or i-1)

[Equation 1] d(i) = h/Err(i) - h/Err(j),

Here, Err(i) is the error value obtained in the ith learning step.

Here, if d(i) is a negative number, the reward may be determined as 0 (zero), and if it is a positive number, the reward may be determined as d(i).

5 is a block diagram illustrating a process of updating weights assigned to links of the nominal network using calculated rewards.

During one learning step process, the reward calculation unit 30 outputs the reward value once. The outputted reward value is input to the agent 20. The agent 20 outputs an action based on the reward value. The action means a set of weights assigned to links of the kth drug responsiveness network 500[k] in the next learning step. Weights assigned to each link of the kth drug responsiveness network 500[k] may be updated by applying the output action.

6 is a flowchart illustrating a method of updating weights assigned to links of a drug responsiveness network related to a specific drug through one learning step provided by an embodiment of the present invention.

The flowchart shown in FIG. 6 can be described with reference to FIGS. 2 to 5 together.

The method according to the flowchart of FIG. 6 may be executed by a computing device having a processing unit and a storage unit. The method may include executing, by the computing device, a predetermined learning step.

In this specification, the learning step may be referred to as learning iteration.

The weight of the kth drug responsiveness network 500[k] may be updated once by the learning step once.

At this time, the learning step may include the following steps (S10), (S20), (S30), (S40), (S50), and (S60).

Steps S10, S20, S30, S40, S50, and S60 may be performed in the i-th learning step, and may be repeatedly executed for each different learning step. .

In step S10, the computing device applies the mutation information of N cancer cell lines to the kth drug reactivity network 500[k] prepared for drug [k], so that N specific perturbation networks [p] [k] can be generated (p=1, 2, 3, ... N(=p _k )).

In step S20, the computing device generates a vector Y={y ₁ , y ₂ , y ₃ , .. ., y _N } and a vector Z={z ₁ , z ₂ , z ₃ , ..., z consisting of values related to the death rate of N cancer cell lines observed by in vitro experiments in which drug [k] is administered. _N } can be prepared.

In step S30, the computing device may calculate a first value (h/Err(i)) that is inversely proportional to the distance (dist{Y, Z}) between vector Y and Z.

In step S40, the computing device may calculate a reward value based on a difference between the first value and a predetermined second value.

At this time, the second value is a value (h/Err(i-1)) inversely proportional to the distance between the vector Y and the vector Z prepared in the i-1 th learning step immediately before the i th learning step. can

In step S50, the computing device inputs the reward value to the agent 20 so that the agent 20 calculates new weights for the links of the kth drug responsiveness network 500[k]. can

In step S60, the computing device may update the kth drug responsiveness network 500[k] with the calculated new weights.

The computing device may be configured to repeatedly execute the learning step for a given kth drug responsiveness network 500[k].

Each time the learning step is repeated, the weights of the links of the kth drug responsiveness network 500[k] may be updated once. That is, whenever the learning iteration is performed once, the weights of the links of the kth drug responsiveness network 500[k] may be updated once.

FIG. 7 shows a method of determining the weights of the drug responsiveness network to optimal values using the method of updating the weights of the drug responsiveness network described in FIG. 6 .

A process in which the agent 20 receives an input once and outputs the input may be referred to as a single learning step.

For the k-th drug reactivity network 500[k] defined for a given drug [k], the learning step described in FIG. 6 may be repeated and executed U times. At this time, in the process of executing the u-th learning step [u], the reward calculator 30 may output a reward [u], and the agent 20 may output a weight [u].

That is, by repeating the learning step U times, a total of U rewards can be generated. At this time, the best reward value may be selected from among the total number of U rewards. If a larger reward value is better, the largest reward value can be selected. The reward value selected in this way is an optimal reward value.

At this time, as the learning step is repeated, the reward value may not necessarily change to a better value. That is, as the learning step is repeated, the reward value may increase and then decrease again, or may decrease and then increase again.

Then, the weight calculated in the learning step in which the optimal reward value is generated may be determined as the optimal weight.

The determined optimal weight may be finally determined as a weight of links of the kth drug responsiveness network 500[k].

<Confirmation of K different drug response networks>

8 illustrates a concept of determining a plurality of different drug responsiveness networks from a given nominal network.

The information described in FIGS. 2 to 7 described above may be applied to a specific drug [k]. The techniques described in FIGS. 2 to 7 may be independently applied to each of a plurality of drugs. That is, the weights of K different drug reactivity networks 500{k} defined for K different drugs can be determined by applying the techniques described in FIGS. 2 to 7 .

That is, the structures of K different drug responsiveness networks 500{k} can be easily determined from one nominal network. However, the value of the weight assigned to each link of the K different drug responsiveness networks 500{k} may be determined by applying the technique according to one embodiment of the present invention described in FIGS. 2 to 7 .

<The process of selecting drugs suitable for the treatment of specific cancer patients>

9 illustrates a process of finding a drug suitable for patient [x] using a plurality of different drug reactivity networks determined according to an embodiment of the present invention.

Now, we can assume a situation in which cancer treatment is needed for patient [x], which is a specific cancer patient. It is also assumed that mutation information of cell line [x], which is a cancer cell line of patient [x], can be acquired. And assume that a drug can be selected from a total of K drugs for the treatment of patient [x]. These assumptions are fully feasible at the current level of technology.

And it is assumed that the completed drug reactivity networks for the K drugs are already prepared by the techniques of FIGS. 2 to 8 described above.

As shown in FIG. 9, by applying the mutation information of the cell line [x] to each of the K drug response networks 500 [k], a total of K specific perturbation networks [x] [k] can be generated (k = 1, 2, 3, ..., K).

FIG. 10 shows a process of determining the optimal drug for patient [k] using K specific perturbation networks [x] [k] prepared as shown in FIG. 9 .

As shown in FIG. 10, the K specific perturbation networks [x][k] predict the death probability of the cell line [x] when the drug [k] is administered to the cell line [x]. Simulation predictions can be output.

Accordingly, a drug or drugs corresponding to the most desirable values among the K simulation predicted values may be proposed as a therapeutic agent for the patient [x]. The proposed treatment may be adopted by a physician or drug developer.

11 illustrates a configuration of a computing device executing a method of completing a drug responsiveness network by determining weights of the drug responsiveness network according to an embodiment of the present invention.

The computing device 710 may include an I/O interface unit 711 , a memory 712 , and a CPU 713 .

The memory 712 may store first information, which is information of drug responsiveness networks whose weights have not been determined. The first information may include information 7121 about state transition rules of the drug responsiveness networks.

And, in the memory 712, a drug responsiveness network selection command code (simply, first code) configured to select one drug responsive network for which a weight in the network is to be determined among the drug responsive networks whose weights have not been determined. 7122 may be stored.

And, in the memory 712, a specific perturbation network generation command code (simply, a second specific perturbation network) for generating N different specific perturbation networks by applying mutation information of N different cancer cell lines to the selected drug responsiveness network. Code) 7123 may be stored.

And, in the memory 712, a command code (simply, third code) 7124 for updating the weights of the selected drug responsiveness network at each learning step by using the method described in FIG. 3 . ) may be stored.

And, in the memory 712, an instruction code for determining the weight of the selected drug responsiveness network (simply, fourth weight set) for determining an optimal weight set among a plurality of weight sets output by the agent 20 for each of a plurality of learning steps. code) 7125 may be stored.

In addition, the memory 712 may store second information 7125, which is information 7125 about the selected drug responsiveness network whose weight is determined. The second information may include information about state transition rules of the drug responsiveness networks and determined weight values.

The CPU 713 can read and use the first information 7121.

Also, the CPU 713 may read and execute the first to fourth codes 7122 to 7125.

In addition, the CPU 713 may store, in the memory 712, information on the plurality of drug responsiveness networks, the weights of which are determined, using the weight information generated by the fourth code 7125.

Then, the CPU 713 executes the first code to perform a drug responsiveness network selection process of selecting one drug responsiveness network for which a weight in the network is to be determined among the drug responsiveness networks whose weights have not been determined. can run Thus, for example, the kth drug responsive network 500[k] of FIG. 2A can be prepared.

The CPU 713 executes the second code to generate N different specific perturbation networks by applying mutation information of N different cancer cell lines to the selected drug responsiveness network. A specific perturbation network You can run the creation process. Thus, for example, the p-th drug reactivity network 500 [p] [k] of FIG. 2A can be prepared (p=1, 2, 3, ....., p _k (=N)).

In addition, the CPU 713 executes the third code to execute a weight update process of the selected drug responsiveness network, which updates the weight of the selected drug responsiveness network at each learning step using the method described in FIG. 3 . . This process can be realized, for example, in the manner described in FIG. 3 .

In addition, the CPU 713 executes the fourth code to determine an optimal weight set among a plurality of weight sets output by the agent 20 for each of a plurality of learning steps, the weight of the selected drug responsiveness network. The decision process can be executed. This processor can be realized, for example, by using the results of a plurality of learning steps performed in the execution process of episode [k] shown in FIG. 7 .

The CPU 713 uses the I/O interface 711 to provide information about the selected drug-responsive network whose weight is determined to another computing device, or to perform a subsequent process of the computing device 710. It can be provided as information for execution.

12 illustrates a configuration of a computing device executing a simulation method for determining an optimal drug effective for killing a specific cancer cell line according to an embodiment of the present invention.

The computing device 810 may include an I/O interface unit 811 , a memory 812 , and a CPU 813 .

The computing device 810 may receive information about K drug candidates and mutation information of the first cancer cell line through the I/O interface 811 . The information on the K drug candidates may be information capable of specifying K drugs. The first cancer cell line may be obtained from the body of a first patient, which is a specific patient.

The memory 812 may store third information 8121, which is information of drug responsiveness networks whose weights have been determined. The drug responsiveness networks for which the weight is determined may include K drug responsiveness networks generated from the K drugs. The third information 8121 may be the same as the second information 7126 stored in the memory 712 of FIG. 11 . The third information may include drug reactivity networks related to a number of drugs greater than the K number.

And, in the memory 812, a command code (simply, fifth code) for generating K specific perturbation networks by applying the mutation information of the first cancer cell line to each of the K drug response networks of which weights are determined. 8122 may be stored.

In addition, the memory 812 may store a command code (simply, a sixth code) 8123 for calculating death probabilities obtained from each of the K specific perturbation networks. At this time, the death probability obtained from the k th specific perturbation network among the K specific perturbation networks is determined by the fact that the first cancer cell line will die when the drug [k] is administered to the first cancer cell line. It may be a simulated value representing a probability.

And in the memory 812, M drugs corresponding to the M death probabilities selected from the K death probabilities obtained from the K specific perturbation networks are determined and included in the optimal drug candidate group, and the optimal drug candidate group is output A command code (simply, seventh code) 8124 may be stored (M<=K). The output of the optimal drug candidate group may be executed through the I/O interface unit 811 .

In a preferred embodiment, a drug corresponding to the highest death probability among the K death probabilities may be included in the optimal drug candidate group.

The CPU 813 may read and use the third information 8121, which is information 7125 about the selected drug responsive network for which the link weight is determined. The third information may include, for example, information about the drug responsiveness network 500 [k] shown in FIG. 9 (k=1, 2, 3, ..., K).

Further, the CPU 813 may read and execute the fifth to seventh codes 8122 to 8124.

And the CPU 813 executes the fifth code to generate K specific perturbation networks by applying the mutation information of the first cancer cell line to each of the K drug response networks of which weights are determined. You can run a specific perturbation network generation process. Thus, for example, information on the specific perturbation network 500[x][k] shown in FIG. 9 may be included (k=1, 2, 3, ..., K, and x are the first 1 Index representing cancer cell lines).

Further, the CPU 813 may execute a process of calculating death probabilities for each specific perturbation network, which calculates death probabilities obtained from each of the K specific perturbation networks by executing the sixth code. Thus, for example, when the drug [k] shown in FIG. 10 is administered to the first cancer cell line, it is possible to obtain a simulated value of the death probability of the first cancer cell line (the first cancer cell line corresponds to cell line [x]). ).

Then, the CPU 813 executes the sixth code to determine M drugs corresponding to the selected M death probabilities among the K death probabilities obtained from the K specific perturbation networks, and include them in the optimal drug candidate group. and an optimal drug candidate group determination and output process of outputting the optimal drug candidate group may be executed.

The CPU 813 uses the I/O interface 811 to provide information on the determined optimal drug candidate group to another computing device, or information for execution of a subsequent process by the computing device 810. can be provided as

The computing device 710 of FIG. 11 and the computing device 810 of FIG. 12 may be provided independently or may be provided as one integrated device.

13 shows the structure of a cancer treatment candidate drug determination system provided according to an embodiment of the present invention.

The cancer treatment candidate drug determination system may include a drug response screening device 600 and a simulation device 80 (computing device 810). The cancer treatment candidate drug determination system may further include the computing device 710 of FIG. 12 .

The drug reaction screening device 600 includes a computing device 610, a drug bank 620, a drug combination device 630, a micro pipette 640, a well-matrix dish 650, and a cell image capturing device 660. can include

The computing device 610 may include an I/O interface unit 611 , a memory 612 , and a CPU 613 . The memory 612 may store a drug combination command code 6121, a real-time cell image analysis command code 6122, and an optimal drug presentation command code 6123. The CPU 613 reads the drug combination command code 6121, the real-time cell image analysis command code 6122, and the optimal drug presentation command code 6123, and performs a drug combination process 6131 corresponding to them, respectively, and a real-time cell image An analysis process 6132 and an optimal drug presentation process 6133 can be executed.

The simulation device 80 of FIG. 13 may be the computing device 810 of FIG. 12 or an integrated device combining the computing device 710 of FIG. 11 and the computing device 810 of FIG. 12 .

As described in FIG. 11 , the simulation device 80 receives first cancer cell line mutation information and K drug candidate group information, and provides information on M selected drugs to the drug response screening device 600. can

The I/O interface unit 611 may deliver information about the M selected drugs to the CPU 613 . The drug combination instruction process 6131 executed in the CPU 613 extracts the M selected drugs from the drug bank 620 using the information on the M selected drugs, and prepares a well-matrix dish 650. A command to inject may be transmitted to the drug combination device 630 and the micropipette 640. The command may be delivered through the I/O interface unit 611.

The drug bank 620 may be a drug storage in which a plurality of drugs including at least the M number of selected drugs are prepared.

Alternatively, the drug bank 620 may be a drug storage in which a plurality of drugs including at least the K drug candidate groups are prepared.

The drug combination device 630 may be a mechanical device capable of extracting a plurality of drugs stored in the drug bank 620 and supplying the extracted drugs to the micro pipette 640 .

When any one of the M selected drugs is a single drug, the first drug, the drug combination device 630 extracts the first drug from the drug bank 620 and provides it to the micropipette 640 can do.

If any one of the M selected drugs is a combination drug of a first drug and a second drug, the drug combination device 630 extracts the first drug and the second drug from the drug bank 620 Thus, combination drugs may be provided to the micropipette 640 .

The well-matrix dish 650 may be a dish in which a plurality of wells are formed.

The drug response screening device 600 may be configured to inject and store the culture medium of the first cancer cell line into M wells of the well-matrix dish 650 .

The micro pipette 640 may inject the drug or drug combination provided from the drug combination device 630 into one of a plurality of wells formed in the well-matrix dish 650 .

The M selected drugs may be respectively injected into M wells in which culture solutions of the first cancer cell line are stored.

The survival rate and death rate of the first cancer cell line stored in the M wells may be determined according to the administered drug.

The real-time cell image analysis command process 6132 captures an image of the first cancer cell line in the M wells to the cell image capturing device 660 through the I/O interface 611 and outputs the resulting image in real time. The video analysis command process 6132 may be instructed to reply.

The optimal drug presentation command process 6133 is a value related to the degree of cell growth and the degree of cell death in each of the M wells based on the images transmitted by the cell image photographing device 660, that is, , cell death rate, cell growth rate, and information such as the area of the cell region in the well can be generated. In addition, based on the generated information, information on drugs that are determined to be effective in killing cancer cells among the M selected drugs as a result of in vitro experiments can be output using a display screen, a speaker, a printer, or the like. Alternatively, the drug response screening device 600 may output in vitro test results of each of the M selected drugs.

In one embodiment, all of the K drug candidates may be drugs approved for administration to actual patients. For example, the K drug candidates or drugs included in the drug bank may all be FDA-approved drugs that can be directly prescribed to patients. In this case, information output from the drug response screening device 600 may be regarded as a final drug candidate for treating cancer patients of the first cancer cell line. At this time, the in vitro test results of the M selected drugs output by the interface unit 611 can be treated as useful information to a doctor who diagnoses and treats patients.

In another embodiment, all of the K drug candidates may be a single drug or a combination drug that is effective for cancer among candidate substances to be developed into drugs. That is, the drugs included in the K drug candidate groups or drug bank may be composed of new drug candidates under development. Furthermore, the drugs included in the K drug candidate group or drug bank may not yet be approved by the FDA. In this case, information output from the drug response screening device 600 may not be used as information for treating cancer patients of the first cancer cell line. However, the in vitro test results of M selected drugs output by the interface unit 611 can be treated as useful information for new drug developers.

In this way, by changing the composition constituting the drug bank included in the drug response screening device provided according to one embodiment of the present invention, a technology that is directly used in the field of patient treatment is provided, or can be directly used in the field of new drug development technology can be provided.

At least one of the computing device 710 of FIG. 11 , the computing device 810 of FIG. 12 , and the computing device 610 of FIG. 13 may be provided as one integrated device.

Hereinafter, FIGS. 14A and 14B to be described below may be collectively referred to as FIG. 14 .

FIG. 14 is a framework for explaining a method for learning the completed agent presented in FIG. 3 .

The agent 20 may be responsible for determining weights assigned to links of the specific perturbation network composed of nodes and links. When the weights assigned to the links of the specific perturbation network are determined to be appropriate values, the specific perturbation network can more accurately output the death probability of the cell line.

Although the structure of the agent 20 may be designed in advance, the input/output characteristics of the agent 20 or the values of parameters given to the inside of the agent 20 for the operation of the agent 20 may vary from a predetermined initial value. It should be updated to an optimal value. For this update, the agent 20 must be trained.

For learning of the agent 20, some or all of a plurality of cancer cell lines for learning and a plurality of drugs (drug combinations) may be used as learning data.

A process in which the agent 20 receives an input once and outputs the input may be referred to as a single learning step.

Information on a set of cancer cell lines for learning and one drug may be used for the one learning step. Using the in vitro experiment device 90, when the one drug is administered to each of the one set of cancer cell lines for learning in an in vitro experiment, the death rate of the one set of cancer cell lines for learning can be observed and determined. The mortality rates may be presented as a vector Z consisting of N scalar values, for example.

In addition, a set of specific perturbation networks may be generated by perturbing a set of specific networks each modeling the set of cancer cell lines for learning. At this time, the node perturbed in each specific network is a node corresponding to a protein on which the selected drug acts.

In addition, a set of death probabilities that can be obtained from the set of specific perturbation networks can be calculated. The death rates may be presented as a vector Y consisting of N scalar values, for example.

The agent 20 may receive the reward and/or the weights of the links allocated to the drug responsiveness network as input data.

The agent 20 may output information obtained by updating values of weights to be assigned to links of the drug responsiveness network in a next learning step based on data input to the agent 20 .

A set of a plurality of continuously executed learning steps may be referred to as a learning episode.

In one embodiment, for all learning steps within one episode, the number of drugs used as learning data may be limited to one. Only after the episode has been changed can the study drug be changed. However, the configurations of the first set of cancer cell lines for learning used in the first learning step in one episode and the second set of cancer cell lines for learning used in the second learning step in the one episode may be different.

Values of the weights of the nominal network may be updated once for each learning step. The agent 20 may be learned for each episode consisting of a plurality of learning steps. Each time the episode is repeated, the learning amount of the agent 20 increases.

In one embodiment of the present invention, a total of K episodes may be executed to train the agent 20 . In this specification, an 'episode' refers to a unit in which the agent 20 is trained once. That is, if an episode is executed a total of K times, the agent 20 is learned K times. In one embodiment of the present invention, one episode is associated with only one drug.

One episode can be executed by executing the learning step mentioned in FIG. 7 a plurality of times. Hereinafter, this will be described in detail.

FIG. 14A shows the framework of the kth episode among K episodes for training the unfinished agent 20 .

The structure shown in FIG. 14A is the same as the structure shown in FIG. 3 . However, while the agent 20 of FIG. 3 has completed learning a total of K times, the agent presented in FIG. 14a is different in that a total of K times of learning has not yet been completed.

14B illustrates a process of completing agent learning by executing a plurality of episodes according to an embodiment of the present invention.

When the execution of one episode ends, the agent 20 may be trained once.

The kth episode may include steps as follows (k=1, 2, 3, ..., K).

First, p _k number of cell lines for which information on reactivity to drug [k] is prepared through an in-vitro experiment may be selected, and mutation information may be prepared for the p _k number of cell lines.

Second, p _k specific perturbation networks may be generated by applying mutation information of the prepared p _k cell lines to the drug reactivity network for drug [k].

Third, the framework of FIG. 14a can be constructed using the generated p _k specific perturbation networks.

Fourth, the learning step may be executed U _k times using the framework of FIG. 14A of FIG. 14A configured above.

When the k th episode is completed, using the U _k set of link weights output by the agent 20 in the process of executing the learning step U _k times and the U _k number of rewards input to the agent 20, The agent 20 may be trained once.

In one embodiment, if k1 and k2 are different, p _k1 and p _k2 may be different from each other, and U _k1 and U _k2 may be different from each other.

Since the agent 20 is learned through a process of determining weights of K different drug responsiveness networks, it may not be used only to determine weights of drug responsiveness networks for a specific drug.

<Method of obtaining death rate Z by administering a specific drug to cancer cell lines>

15 shows the configuration of a system for obtaining and providing a death rate Z by administering a specific drug to cancer cell lines according to an embodiment of the present invention.

The biological signal transmission network generation system 100 may include a computing device 50 , a cell line experiment device 60 , and a data server 70 .

The cell line experiment device 60 may include a cell line container 61 , a drug administration device 62 , and a cell line state observation device 63 .

For example, cancer cell lines may be provided separately in the plurality of wells provided in the cell line container 61 .

The drug administration device 62 may administer a selected specific drug to the cancer cell lines provided in the cell line container 61 .

The cell line state monitoring device 63 may observe and output the death rate of the cancer cell lines after the specific drug is administered.

The cell line experiment device 60 may be configured to provide the observed death rates to the computing device 50 .

The data server 70 may provide the computing device 70 with information about a drug responsiveness network related to a specific drug. The information on the drug responsive network may include a configuration of an interconnection structure of nodes and links of a subnetwork portion that responds to the drug among the nominal networks of the cancer cell lines. In addition, when the specific drug is administered to the cancer cell lines, the data server 70 may provide information about nodes affected by the specific drug to the computing device 70 .

The computing device 50 may include a processing unit 51 , a storage unit 52 , and a user interface 53 .

The user interface 53 may receive information indicating the specific drug and information indicating the cancer cell lines from a user.

The computing device 50 transmits the input information indicating the cancer cell lines and the information indicating the specific drug to the cell line experiment device 60, and after administering the specific drug to the cancer cell lines, the cancer cell lines die. Observation values for the ratio may be requested from the cell line experiment device 60 . The observed death rate obtained from the cell line experiment apparatus 60 may be stored in the storage unit 52 of the computing device 50 .

The processing unit 51 is configured to execute a step of performing a plurality of learning steps for determining the weight of the drug responsiveness network of the specific drug. This example is illustrated in Figures 3-7.

<Operation principle of the agent>

The operating principle of the agent 20 will be described below.

In the current learning step, the current weights and node characteristic values, which are the weights assigned to the links of the network 500, 520, or 521, are input to the graph neural network, and the reward value obtained for the current weight is connected to the RNN (Recurrent It is used as an input for a neural network). The RNN can output an action by combining the input values (weight, reward) in the current learning step and the hidden state containing information in the previous learning step. The action may be an update weight used in the next learning step. The update weight is a weight assigned to links of the network 500, 520, or 521 in the next learning step.

The agent 20 may include three parts: an input layer, a submodule layer, and a main layer.

The input layer may include a graph module for embedding a graph and a message module for communication between agents. The graph module has two graph neural networks having the same structure. One (G) is a module for a global state estimator and a main layer, and the other ( ^GC ) is a module for a context estimator. The message module (G ^m ) may be used in all modules of the sub-module layer and the main layer. All modules in this layer can have a graph neural network structure. The graph module may receive node features (eight centrality measures) and link features (edge betweenness centrality) at each learning step and output nodes, links, and global features. The message module receives the 0 (zero) vector input at the beginning of each episode, and can recursively receive the previous calculation value in the subsequent learning step. Based on each link, the output of the three modules can be reconstructed by concatenating the node features of the source, the target node, the link feature, and the global state, and can be used as an input state for the next module. there is. The state of the i-th link reconstructed in each of the three modules (G, G ^C , and G ^m ) is expressed as Equation 1.

[Formula 1]

L _i =(n _{i_s} , n _{i_t} , l _i , g),

L _i ^C = (n _{i_s} ^C , n _{i_t} ^C , l _i ^C , g ^C ),

L _i ^m =(N _{i_s} ^m , N _{i_t} ^m , L _i ^m , g ^m )

Submodules and modules of the main layer can receive the reconfigured state as input.

The submodule layer may include the context estimator module and the global state estimator module. The two modules are one-layer LSTMs and share weights for individual coordinate inputs, but can maintain independent hidden states. The context estimator is a module that estimates information about the environment. L ^C , L ^m and the reward value of the previous learning step are input into the LSTM, followed by two dense layers to which elu activation is applied. information can be output. The global state estimator is a module for learning the inter-agent communication protocol, and can receive inputs of L, L ^m , and the action of the current step and output the state of the next step (the L of the next step).

In the main layer, L, L ^m , the reward value of the previous learning step, and the outputs of the two submodules are input to the 2-layer LSTM, and an action can be output through one dense layer to which Elu activation is applied.

<Business model using the present invention>

16A shows an example of a business model using the present invention.

The technology provider may provide the computing device 710 with the agent 20 that has completed learning with the technology shown in FIG. 14 . In addition, the computing device 710 may output structure and parameter (weight) information of the plurality of drug responsive networks whose weights are determined. This information can be passed on to technology consumers. A task of learning the agent may also be executed in the computing device 710 .

The technology consumer may input structure and parameter (weight) information of the plurality of drug responsive networks for which the weight is determined into the computing device 810, and input mutation information of the first cancer cell line and K drug candidate group information (Fig. 13). The computing device 810 may provide M pieces of selected drug information to the drug response screening device 600 . The drug reaction screening device 600 may output in vitro test results of M selected drugs required by medical personnel or new drug developers.

16B shows another example of a business model using the present invention.

The technology provider may provide the agent 20 that has completed learning with the technology shown in FIG. 14 to the technology consumer.

A technology consumer may install the provided agent 20 in the computing device 710 . The computing device 710 may output structure and parameter (weight) information of the plurality of drug responsive networks whose weights are determined. The computing device 810 may input structure and parameter (weight) information of the plurality of drug responsiveness networks whose weights are determined, and also input mutation information of the first cancer cell line and K drug candidate information (see FIG. 13 ). . The computing device 810 may provide M pieces of selected drug information to the drug response screening device 600 . The drug reaction screening device 600 may output in vitro test results of M selected drugs required by medical personnel or new drug developers.

In FIGS. 16A and 16B , the in vitro test results of the M selected drugs can be used by doctors. In this case, the K drugs corresponding to the K drug candidate group information input to the computing device 810 are FDA-approved drugs, and may be a set of marketable drugs that can be immediately utilized by a doctor.

Alternatively, the in vitro test results of the M selected drugs can be utilized by new drug developers. The K drugs corresponding to the K drug candidate group information input to the computing device 810 may be a set of substances having pharmacological effects prior to drug development.

Using the above-described embodiments of the present invention, those belonging to the technical field of the present invention will be able to easily implement various changes and modifications without departing from the essential characteristics of the present invention. The content of each claim of the claims may be combined with other claims without reference relationship within the scope understandable through this specification.

<Sasa>

This invention is the result of the following research project.

* Assignment identification number: SRFC-IT1802-02

*Research project title: (G01210368) Universal cancer cell simulator (year 2021)

*Account number: G01210368

*Account manager: Cho Kwang-hyeon

* Host organization: Korea Advanced Institute of Science and Technology

* Research management institution: Samsung Electronics Co., Ltd.

*Researcher number: 10103248

*Research Period: 2021-07-01 ~ 2021-11-30

*Contribution rate: 100

Claims (14)

generating, by the simulation device 810, a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line in each of the plurality of drug response networks for the plurality of drugs;
selecting, by the simulation device, a plurality of candidate drugs from among the plurality of drugs based on a plurality of death probabilities of the first cancer cell line output from the plurality of specific perturbation networks;
providing information on the determined plurality of candidate drugs to a drug response screening device;
executing, by the drug response screening device, an in-vitro experiment in which the plurality of candidate drugs are administered to a plurality of wells in which the first cancer cell line is stored;
capturing and analyzing, by the drug response screening device, images of the first cancer cell line in the plurality of wells using a cell image capturing device; and
outputting, by the drug response screening device, in vitro test results for at least some of the plurality of candidate drugs based on the analyzed results;
including,
Methods for determining drug candidates for cancer treatment. According to claim 1,
Prior to the generating step, further comprising, by the computing device 710, performing a process (=episode) of determining a weight of a k-th drug responsiveness network that responds to a k-th drug among the plurality of drug responsiveness networks. and
The step of performing the process is to use an agent whose learning has been completed by reinforcement learning,
To carry out the above process,
obtaining, by the computing device, mutation information of N (=p _k ) cell lines having information on reactivity by an in vitro experiment using a kth drug among the plurality of drugs;
generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug;
repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent;
selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and
determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network;
including,
Methods for determining drug candidates for cancer treatment. According to claim 2,
The agent is configured to determine weights of links of the kth drug responsiveness network in a next learning step based on the reward and weights of links of the kth drug responsiveness network in a current learning step.
Methods for determining drug candidates for cancer treatment. According to claim 3,
The process of determining the reward,
The computing device, in the current learning step among the plurality of learning steps, a vector Y consisting of N death probabilities output by the N specific perturbation networks, and the k th drug to the first cancer cell line preparing a vector Z composed of values related to the death rate of the N first cancer cell lines observed by in vitro experiments;
calculating, by the computing device, a first value that is inversely proportional to a distance between the vector Y and the vector Z; and
calculating, by the computing device, the reward based on a difference between the first value and the second value;
Including,
The second value is a value inversely proportional to the distance between the vector Y and the vector Z prepared in the learning step immediately before the current learning step,
Methods for determining drug candidates for cancer treatment. According to claim 2,
Prior to performing the process (=episode) of determining the weights of the kth drug responsiveness network, the computing device 710 further comprising the step of training the agent,
In the step of learning the agent, the process of learning the agent (=episode) is repeatedly performed for G different drugs,
The learning process of the agent performed for the g-th drug,
obtaining, by the computing device 710, mutation information of p _g cell lines having information on reactivity by an in vitro experiment using the g-th drug;
generating, _by the computing device, p g specific perturbation networks by applying the p _g pieces of the mutation information to the p th drug responsiveness network that responds to the p th drug;
repeating, by the computing device, a learning step of updating weights of links of the g-th drug responsiveness network a plurality of times using the agent; and
learning, by the computing device, the agent using the reward values and the weights acquired in a process of repeating the learning step a plurality of times;
including,
Methods for determining drug candidates for cancer treatment. According to claim 5,
wherein the agent is configured to determine weights of links of the gth drug responsiveness network in a next learning step based on the reward and weights of links of the gth drug responsiveness network in a current learning step;
Methods for determining drug candidates for cancer treatment. performing, by the computing device 710, a process (=episode) of determining a weight of a kth drug responsiveness network that responds to a kth drug among the plurality of drug responsiveness networks;
generating, by the simulation device 810, a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line in each of the plurality of drug response networks for the plurality of drugs; and
selecting, by the simulation device, a plurality of candidate drugs from among the plurality of drugs based on a plurality of death probabilities of the first cancer cell line output from the plurality of specific perturbation networks;
Including,
The step of performing the process is to use an agent whose learning has been completed by reinforcement learning,
To carry out the above process,
obtaining, by the computing device, mutation information of N (=p _k ) cell lines having information on reactivity by an in vitro experiment using a kth drug among the plurality of drugs;
generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug;
repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent;
selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and
determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network;
including,
Methods for determining drug candidates for cancer treatment. simulation device 810; and
Drug response screening device (600)
Including,
The simulation device,
Generate a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line in each of the plurality of drug response networks for the plurality of drugs;
Selecting a plurality of candidate drugs from among the plurality of drugs based on the plurality of death probabilities of the first cancer cell line output by the plurality of specific perturbation networks, and
To provide information on the determined plurality of candidate drugs to a drug response screening device,
The drug response screening device,
Executing an in-vitro experiment in which the plurality of drug candidates are administered to a plurality of wells in which the first cancer cell line is stored;
The drug response screening device captures and analyzes images of the first cancer cell line in the plurality of wells using a cell image capturing device, and
wherein the drug reaction screening device outputs in vitro test results for at least some of the plurality of candidate drugs based on the analysis results;
Cancer treatment candidate drug decision system. According to claim 8,
Computing device 710; further comprising,
The computing device, before the simulation device generates the plurality of specific perturbation networks, a process of determining a weight of a kth drug responsiveness network that responds to a kth drug among the plurality of drug responsiveness networks (= episode) is to be performed,
The process of performing the above process is to use an agent whose learning has been completed by reinforcement learning,
The process of performing the above process is,
obtaining, by the computing device, mutation information of N (=p _k ) cell lines having information on reactivity by an in vitro experiment using a kth drug among the plurality of drugs;
generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug;
repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent;
selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and
determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network;
including,
Cancer treatment candidate drug decision system. According to claim 9,
The agent is configured to determine weights of links of the kth drug responsiveness network in a next learning step based on the reward and weights of links of the kth drug responsiveness network in a current learning step.
Cancer treatment candidate drug decision system. According to claim 10,
The process of determining the reward,
The computing device, in the current learning step among the plurality of learning steps, a vector Y consisting of N death probabilities output by the N specific perturbation networks, and the k th drug to the first cancer cell line preparing a vector Z composed of values related to the death rate of the N first cancer cell lines observed by in vitro experiments;
calculating, by the computing device, a first value that is inversely proportional to the distance between the vector Y and the vector Z; and
calculating, by the computing device, the reward based on a difference between the first value and the second value;
Including,
The second value is a value inversely proportional to the distance between the vector Y and the vector Z prepared in the learning step immediately before the current learning step,
Cancer treatment candidate drug decision system. According to claim 9,
Prior to performing the process (=episode) of determining the weights of the kth drug responsiveness network, the computing device 710 is configured to train the agent,
In the process of learning the agent, the process of learning the agent (=episode) is repeatedly performed for G different drugs,
The learning process of the agent performed for the g-th drug,
obtaining, by the computing device 710, mutation information of p _g cell lines having information on reactivity by an in vitro experiment using the g-th drug;
generating, _by the computing device, p g specific perturbation networks by applying the p _g pieces of the mutation information to the p th drug responsiveness network that responds to the p th drug;
repeating, by the computing device, a learning step of updating weights of links of the g-th drug responsiveness network a plurality of times using the agent; and
learning, by the computing device, the agent using the reward values and the weights acquired in a process of repeating the learning step a plurality of times;
including,
Cancer treatment candidate drug decision system. According to claim 12,
wherein the agent is configured to determine weights of links of the gth drug responsiveness network in a next learning step based on the reward and weights of links of the gth drug responsiveness network in a current learning step;
Cancer treatment candidate drug decision system. simulation device 810;
drug response screening device 600; and
computing device 710;
Including,
The computing device is configured to perform a process (=episode) of determining weights of a kth drug responsiveness network that responds to a kth drug among the plurality of drug responsiveness networks;
The simulation device,
Generate a plurality of specific perturbation networks by reflecting the mutation information of the first cancer cell line, respectively, on a plurality of drug response networks for a plurality of drugs, and
A plurality of candidate drugs are selected from among the plurality of drugs based on a plurality of death probabilities of the first cancer cell line output by the plurality of specific perturbation networks,
The step of performing the process is to use an agent whose learning has been completed by reinforcement learning,
To carry out the above process,
obtaining, by the computing device, mutation information of N (=p _k ) cell lines having information on reactivity by an in vitro experiment using a kth drug among the plurality of drugs;
generating N specific perturbation networks by applying, by the computing device, the N pieces of mutation information to the kth drug responsiveness network that responds to the kth drug;
repeating, by the computing device, a learning step of updating weights of links of the kth drug responsiveness network a plurality of times using the agent;
selecting, by the computing device, a learning step having the largest value of a reward provided to the agent among the plurality of learning steps; and
determining, by the computing device, link weights output by the agent in the selected learning step as weights of links of the kth drug responsiveness network;
including,
Cancer treatment candidate drug decision system.

Images