Information processing system, information processing method, and program

Abstract

[Problem] To provide, inter alia, an information processing system with which it is possible to accurately estimate binding force to lipid [Solution] According to one aspect of the present invention, provided is an information processing system comprising a processor configured to execute the next steps by reading a program. In an acquisition step, an amino acid sequence for which an inference is to be carried out is acquired. In an inference step, the binding force, to lipid, of a protein having the amino acid sequence for which an inference is to be carried out is inferred on the basis of the amino acid sequence for which an inference is to be carried out, and reference information. The reference information includes a correlation between the amino acid sequence and the binding force, the correlation being constructed on the basis of actually measured values of the binding force.

Description

Information Processing System, Information Processing Method, and Program

[0001] The present invention relates to an information processing system, an information processing method, and a program.

[0002] Patent Document 1 discloses a system that uses a machine learning model to search for the amino acid sequence of a protein.

[0003] Japanese Patent No. 7516368

[0004] In the above system, there is no guarantee that the binding force to lipids can be accurately estimated as a characteristic of proteins.

[0005] In view of the above circumstances, the present invention aims to provide an information processing system and the like that can accurately estimate the binding force to lipids.

[0006] According to one aspect of the present invention, there is provided an information processing system including a processor configured to execute the following steps by reading a program. In an acquisition step, an amino acid sequence to be inferred is acquired. In an inference step, based on the amino acid sequence to be inferred and reference information, the binding force of a protein having the amino acid sequence to be inferred to lipids is inferred. Here, the reference information includes a correlation between an amino acid sequence and a binding force, which is constructed based on the actually measured value of the binding force.

[0007] According to such an aspect, by specifying an amino acid sequence, the binding force of a protein having the amino acid sequence to lipids can be accurately estimated.

[0008] This is a diagram showing the configuration of the information processing system 1. This is a block diagram showing the hardware configuration of the information processing device 10. This is a block diagram showing the hardware configuration of the user terminal 20. This is a block diagram showing the functions realized by the information processing device 10 (control unit 11) and the user terminal 20 (control unit 21). This is a schematic diagram showing an example of the flow of lipid binding strength estimation by the estimation unit 112. This is a graph showing the relationship between the binding strength to phospholipids estimated by the estimation unit 112 and the measured value of said binding strength for multiple proteins (amino acid sequences). This is a graph showing the binding strength to a specific lipid of a protein (nanobody N1) having a randomly mutated amino acid sequence and a protein (nanobody N2) having an amino acid sequence searched by the search unit 113. This is a graph showing the binding strength of four proteins ("Ev44", "Ev53", "Ev56", and "Ev66") that have specific binding properties and were searched by the search unit 113. This is an activity diagram showing an example of the flow of information processing (amino acid sequence search processing) performed by the information processing system 1.

[0009] Embodiments of the present invention will be described below with reference to the drawings. The various features shown in the embodiments below can be combined with each other.

[0010] Incidentally, the program for realizing the software appearing in one embodiment may be provided as a computer-readable non-transitor-readable medium, or it may be provided so that it can be downloaded from an external server, or it may be provided so that the program is launched on an external computer and its functions are realized on a client terminal (so-called cloud computing).

[0011] Furthermore, in various information processing according to one embodiment, an input and an output corresponding to the input can be realized. Here, as long as an output is obtained as a result of the input, the form of the information referenced in such information processing (hereinafter referred to as "reference information") is not limited. The reference information may be, for example, rule-based information such as a database, a lookup table, or a predetermined function (including a decision formula such as a regression equation constructed by a statistical method), or a pre-trained model that has learned the correlation between input and output in advance, or a large-scale language model that can output a desired result by inputting a prompt.

[0012] Furthermore, in one embodiment, "part" may include, for example, hardware resources implemented by a circuit in a broad sense, and the information processing of software that can be specifically realized by these hardware resources. Also, in one embodiment, various types of information are handled, and this information can be represented, for example, by the physical values ​​of signal values ​​representing voltage and current, the high or low values ​​of signal values ​​as a set of binary bits composed of 0s or 1s, or by quantum superposition (so-called qubits), and communication and calculations can be performed on a circuit in a broad sense.

[0013] Furthermore, a circuit in a broad sense is a circuit realized by combining at least an appropriate combination of circuits, circuits, processors, and memory. The processor may be a general-purpose processor or a dedicated circuit. In other words, this includes application-specific integrated circuits (ASICs), programmable logic devices (for example, simple programmable logic devices (SPLDs), complex programmable logic devices (CPLDs), and field programmable gate arrays (FPGAs)), etc.

[0014] 1. Hardware Configuration This section describes the hardware configuration.

[0015] <Information Processing System 1> Figure 1 is a configuration diagram representing Information Processing System 1. Information Processing System 1 comprises a communication line 2, an information processing device 10, and a plurality of user terminals 20. The information processing device 10 and the user terminals 20 are configured to communicate with each other via the communication line 2. The connection between the information processing device 10 and the user terminals 20 may be wired or wireless.

[0016] In one embodiment of the information processing system 1, the information processing system 1 consists of one or more devices or components. These components will be described below.

[0017] <Information Processing Device 10> Figure 2 is a block diagram showing the hardware configuration of the information processing device 10. As shown in Figure 2, the information processing device 10 comprises a control unit 11, a storage unit 12, a communication unit 13, and a communication bus 14. The control unit 11, the storage unit 12, and the communication unit 13 are electrically connected within the information processing device 10 via the communication bus 14.

[0018] <Control Unit 11> The control unit 11 performs processing and control of the overall operation related to the information processing device 10. The control unit 11 is, for example, a Central Processing Unit (CPU). The control unit 11 realizes various functions related to the information processing device 10 by reading predetermined programs stored in the memory unit 12. That is, information processing by software stored in the memory unit 12 is concretely realized by the control unit 11, which is an example of hardware, and can be executed as each functional unit included in the control unit 11. These will be described in more detail in the next section. Note that the control unit 11 is not limited to being a single unit, and the information processing device 10 may have multiple control units 11 for each function. Also, the information processing device 10 may have a configuration that combines these.

[0019] <Storage Unit 12> The storage unit 12 stores various types of information as defined above. This can be implemented, for example, as a storage device such as a solid-state drive (SSD) that stores various programs related to the information processing device 10 executed by the control unit 11, or as a memory such as a random access memory (RAM) that stores temporarily necessary information (arguments, arrays, etc.) related to program calculations. The storage unit 12 stores various programs, variables, etc. related to the information processing device 10 executed by the control unit 11.

[0020] <Communication Unit 13> The communication unit 13 preferably uses wired communication methods such as USB, IEEE 1394, Thunderbolt®, and wired LAN network communication, but may also include wireless LAN network communication, mobile communication such as LTE / 5G, and Bluetooth® communication as needed. In other words, it is more preferable to implement it as a collection of these multiple communication methods. That is, the information processing device 10 may communicate various information from the outside via the communication unit 13 and the network.

[0021] The information processing device 10 may be on-premise or in a cloud-based configuration. In the case of a cloud-based information processing device 10, for example, the above-mentioned functions and processing may be provided in the form of SaaS (Software as a Service) or cloud computing.

[0022] <User Terminal 20> The user terminal 20 is an information processing terminal used by a user who utilizes the information processing functions provided by the information processing device 10. Figure 3 is a block diagram showing the hardware configuration of the user terminal 20. As shown in Figure 3, the user terminal 20 comprises a control unit 21, a storage unit 22, a communication unit 23, an input unit 24, an output unit 25, and a communication bus 26. The control unit 21, storage unit 22, communication unit 23, input unit 24, and output unit 25 are electrically connected within the user terminal 20 via the communication bus 26. The explanation of the control unit 21, storage unit 22, and communication unit 23 is the same as the explanation of each part in the information processing device 10, so it is omitted.

[0023] <Input Unit 24> The input unit 24 receives action inputs made by the user. The action input is transmitted to the control unit 21 via the communication bus 26 as a command signal. The control unit 21 can perform predetermined controls or calculations based on the transmitted command signal as needed. The input unit 24 may be included in the casing of the user terminal 20 or it may be externally attached. For example, the input unit 24 may be implemented as a touch panel integrated with the output unit 25. When the input unit 24 is implemented as a touch panel, the user can input tap actions, swipe actions, etc. to the input unit 24. Instead of a touch panel, the input unit 24 can be a switch button, mouse, trackpad, QWERTY keyboard, etc.

[0024] <Output Unit 25> The output unit 25 displays a graphical user interface (GUI) screen that allows the user to take action. The output unit 25 may be included in the casing of the user terminal 20 or it may be an external component. Specifically, the output unit 25 can be implemented as a display device such as a CRT display, liquid crystal display, organic EL display, or plasma display. It is preferable that these display devices be used in accordance with the type of user terminal 20.

[0025] 2. Functional Configuration This section describes the functional configuration of this embodiment. Information processing by software stored in the memory unit 12 is specifically realized by the control unit 11, which is an example of hardware, and can be executed as each functional unit included in the control unit 11 (the processor provided by the information processing system 1).

[0026] Figure 4 is a block diagram showing the functions realized by the information processing device 10 (control unit 11) and the user terminal 20 (control unit 21).

[0027] As shown in Figure 4A, the information processing device 10 (control unit 11) comprises an acquisition unit 111, an inference unit 112, a search unit 113, and an artificial intelligence unit 120. As shown in Figure 4B, the user terminal 20 (control unit 21) comprises a display unit 211 and an operation acquisition unit 212.

[0028] The acquisition unit 111 is configured to acquire the amino acid sequence to be predicted (the amino acid sequence of a protein whose binding affinity to lipids is predicted). The acquisition unit 111 acquires the amino acid sequence to be predicted by, for example, receiving input of the amino acid sequence from the user terminal 20. Alternatively, the acquisition unit 111 may acquire the amino acid sequence to be predicted from a storage location (network address) of the amino acid sequence data by receiving a specification of the storage location (network address) from the user terminal 20. Furthermore, the acquisition unit 111 may acquire a mutant amino acid sequence created by the search unit 113, described later, as the amino acid sequence to be predicted.

[0029] The number of amino acids in each amino acid sequence acquired by the acquisition unit 111 (the number of amino acids in one sequence) is not particularly limited, and is, for example, 10 or more and 3000 or less.

[0030] <Prediction Unit 112> The prediction unit 112 is configured to predict the lipid binding strength (hereinafter, "lipid binding strength") of a protein having the amino acid sequence to be predicted (hereinafter, the predicted protein) based on the amino acid sequence to be predicted and reference information acquired by the acquisition unit 111.

[0031] Lipid binding strength is typically a value that indicates the strength (activity) of a lipid's binding to a particular lipid. Alternatively, lipid binding strength may also indicate the strength of the binding to a lipid group, which is a grouping of lipids. Lipid groups include, for example, those grouped by chemical structure (simple lipids, complex lipids, derived lipids), those grouped by fatty acid type, and those grouped by the polar head of the lipid molecule. Furthermore, "strength of binding strength" may include statistical values ​​such as maximum, minimum, mean, and median values.

[0032] Examples of proteins to be predicted include antibody mimetics such as nanobodies, monobodies, and aphibodies, antibodies, proteins contained in the lipid-binding domain region of naturally occurring proteins, and artificial proteins designed using computers. The prediction unit 112 is preferably configured to predict the lipid-binding affinity of nanobodies. This makes it possible to design or search for nanobodies with high binding affinity to specific lipids.

[0033] The reference information includes the correlation between amino acid sequences and lipid binding strength, constructed based on measured values ​​of protein binding strength to lipids. The reference information is stored, for example, in the memory unit 12. The reference information is an estimator constructed to output lipid binding strength from an amino acid sequence. The reference information may also include, for example, tables, functions, simple algorithms, etc., that show the correlation between feature quantities (e.g., vector data) obtained by information extraction from or information transformation of amino acid sequences and lipid binding strength. The correlations included in the reference information can be constructed, for example, by statistically analyzing data recording lipid binding strength measured from actual proteins.

[0034] The reference information may include a binding force prediction model that has been pre-trained to take an amino acid sequence as input and output lipid binding force. In this case, the prediction unit 112 inputs the amino acid sequence into the binding force prediction model included in the artificial intelligence unit 120, and causes the binding force prediction model to output the lipid binding force. The binding force prediction model is constructed, for example, by machine learning using records that associate the amino acid sequence of a certain protein with the measured value of the lipid binding force of this protein as training data. In the binding force prediction model, parameters calculated and tuned through learning constitute the correlation of the reference information.

[0035] The reference information may include first reference information and second reference information. The first reference information includes the correlation between the amino acid sequence and the feature vector. The second reference information includes the correlation between the feature vector and the lipid binding strength. In this case, the prediction unit 112 obtains the feature vector based on the amino acid sequence to be predicted and the first reference information, and predicts the lipid binding strength based on the obtained feature vector and the second reference information. This allows for the construction of the first reference information for obtaining the feature vector and the construction of the second reference information for predicting the lipid binding strength from the feature vector to be performed separately, thereby improving the accuracy of lipid binding strength prediction and reducing costs in the construction and updating of the information processing system 1.

[0036] <First Reference Information> The first reference information includes, for example, tables, functions, simple algorithms, etc., for obtaining feature vectors by extracting information from amino acid sequences or by transforming information from amino acid sequences. The "correlation between amino acid sequences and feature vectors" in the first reference information may include not only relationships that directly represent the relationship between the two, but also indirect relationships that mathematically link the two (for example, laws that mechanically transform amino acid sequences into feature vectors, or parameters that define such laws).

[0037] Furthermore, the first reference information may include a deep learning model trained to generate feature vectors representing protein characteristics from amino acid sequences. In this case, the prediction unit 112 inputs the amino acid sequence to be predicted into the deep learning model of the artificial intelligence unit 120 and obtains feature vectors from the deep learning model. This makes it possible to obtain feature vectors of the amino acid sequence to be predicted based on pattern analysis of a large amount of amino acid sequences. As a result, the accuracy of lipid binding strength prediction is improved.

[0038] The "protein features" output by deep learning models include, for example, secondary structure, tertiary structure, structural stability, and hydrophobicity. Deep learning models generate feature vectors that directly or indirectly represent these features from the amino acid sequence.

[0039] The feature vectors obtained from a deep learning model may be the output information of the deep learning model itself, or they may be intermediate information generated as a byproduct when the deep learning model generates the output information. The dimension of the feature vectors may be on the order of, for example, 1000 dimensions.

[0040] As a deep learning model, for example, the so-called protein language model (pLM) can be used. Examples of pLMs include UniRep® and ESM.

[0041] A typical deep learning model uses a large-scale protein amino acid sequence database as training data, and the parameters of the neural network are tuned to predict the next amino acid to be placed in a given amino acid sequence (or the amino acid to be placed in a masked portion of the sequence). These tuned parameters correspond to the "correlation between amino acid sequences and feature vectors" in the first reference information.

[0042] In such a deep learning model, in order to obtain the target output (predicted amino acid sequence), the input amino acid sequence is converted into a high-dimensional feature vector. That is, the deep learning model generates an embedding vector from the amino acid sequence. More specifically, in the deep learning model, for example, using Transformer (encoder), the amino acid sequence is embedded into a feature vector, and further a predicted amino acid sequence is generated from the feature vector.

[0043] <Second reference information> The second reference information includes, for example, a table, a function, a simple algorithm, etc. that define the correlation between the feature vector and the lipid binding force.

[0044] The second reference information may be constructed using machine learning data (teacher data) that combines the feature vector obtained from the amino acid sequence of the learning protein and the measured value of the lipid binding force in the learning protein. Thereby, second reference information with relatively high inference accuracy can be constructed at a relatively low cost.

[0045] The machine learning data is prepared, for example, by the following procedure. That is, for the data acquisition protein prepared for learning, the binding force to the target lipid is measured by an appropriate assay. In addition, the amino acid sequence of the data acquisition protein is converted into a feature vector using the first reference information (for example, a deep learning model), and the feature vector is associated with the measured binding force. Thereby, machine learning data that combines the feature vector as the input data and the binding force as the correct answer data is obtained. Using these machine learning data, when the feature vector is input, the parameters of the regression model (regression formula or algorithm) included in the second reference information are adjusted so that the estimated value of the lipid binding force of the corresponding protein is output. <​​​The second reference information may include a deep learning model that is machine-learned to infer the lipid binding ability from the feature vector. In this case, the inference unit 112 inputs the feature vector into the deep learning model of the artificial intelligence unit 120 and obtains an inferred value of the lipid binding ability from the deep learning model.

[0048] The data (measured values of lipid binding ability) used to construct the second reference information is significantly less than the data (protein amino acid sequence data) used to construct the first reference information. The number of data for constructing the first reference information is, for example, on the order of 100,000 to 10 billion, while the number of data for constructing the second reference information is, for example, on the order of 10 to 100,000.

[0049] FIG. 5 is a schematic diagram showing an example of the flow of inferring the lipid binding ability by the inference unit 112. First, the inference unit 112 inputs the amino acid sequence AC to be inferred into the deep learning model PM1. Next, the inference unit 112 inputs the feature vector CV generated by the deep learning model into the regression model PM2. The inference unit 112 obtains the lipid binding ability LB output by the regression model PM2 as the inference result.

[0050] The inference unit 112 may infer the binding ability of the target protein to each of a plurality of types of lipids or to each of a plurality of lipid groups. Thereby, for the input of one amino acid sequence, the binding ability to each of a plurality of types of lipids can be inferred simultaneously.

[0051] For example, the inference unit 112 converts the amino acid sequence into a feature vector by inputting it into the deep learning model, and inputs this feature vector into a plurality of regression models prepared for each type of lipid to obtain an inferred value of the lipid binding ability for each type of lipid. Also, the inference unit 112 may receive the selection of the lipid for which the binding ability is to be inferred from the user terminal 20 and input the feature vector into the regression model for inferring the binding ability to the selected lipid.

[0052] As described above, the estimation unit 112 estimates lipid binding strength using a common first reference information that is independent of the type of lipid, and a plurality of second reference information prepared for each type of lipid. Each of the plurality of second reference information is constructed based on data of the lipid binding strength of proteins measured for each type of lipid. By making the first reference information common to all types of lipids in this way, a model can be constructed (added) to estimate the binding strength of new lipids for which the binding strength had not been estimated before, by constructing only the second reference information without having to reconstruct the first reference information. Therefore, the construction of estimation models for new lipids can be done relatively easily and at low cost.

[0053] Figure 6 is a graph showing the relationship between the phospholipid binding strength estimated by the estimation unit 112 and the measured values ​​of that binding strength for several proteins (amino acid sequences). The horizontal axis of the four graphs in Figure 6 shows the estimated values ​​of the binding strength to PI(3,5)P2 (phosphatidylinositol 3,5-bisphosphate), PI(3,4)P2 (phosphatidylinositol 3,4-bisphosphate), PI(4,5)P2 (phosphatidylinositol 4,5-bisphosphate), and PS (phosphatidylserine), respectively, and the vertical axis shows the measured values ​​of the binding strength to each phosphate. As shown in each graph of Figure 6, a high correlation is observed between the estimated values ​​of lipid binding strength by the estimation unit 112 and the measured values ​​of lipid binding strength.

[0054] <Search Unit 113> The search unit 113 is configured to search for amino acid sequences with high lipid binding affinity by mutating the amino acid sequence to be predicted while referring to the lipid binding affinity predicted by the prediction unit 112.

[0055] Specifically, the search unit 113 first causes the estimation unit 112 to estimate the lipid binding strength of a mutant amino acid sequence, which is a mutation of a portion of the initial amino acid sequence (the amino acid sequence input by the user terminal 20). The mutation of the amino acid sequence is performed, for example, by randomly replacing several randomly selected amino acids in the amino acid sequence with other amino acids.

[0056] Next, the search unit 113 records the mutated amino acid sequence and the estimated lipid binding strength. The search unit 113 also determines whether to maintain the preceding mutation in the mutated amino acid sequence based on the estimated lipid binding strength. Specifically, the search unit 113 maintains the mutation if the lipid binding strength satisfies predetermined conditions. The predetermined conditions for maintaining the mutation are set, for example, by the difference between the lipid binding strength of the amino acid sequence before the mutation and the lipid binding strength of the amino acid sequence after the mutation. Specifically, the mutation is maintained if the value obtained by subtracting the lipid binding strength of the amino acid sequence after the mutation from the lipid binding strength of the amino acid sequence before the mutation is greater than or equal to a threshold. Such a threshold may be a predetermined fixed value, but it is preferable that it be a value (function) that gradually increases according to the depth of the search (number of mutations). In the initial stages of the search, the threshold may be a negative value. This makes it possible to search for the maximum value of lipid binding strength in the entire search region, rather than the so-called local maximum value of lipid binding strength.

[0057] If the mutation is maintained (i.e., lipid binding affinity is improved or maintained), the search unit 113 introduces another mutation into the mutated amino acid sequence. If the mutation is not maintained (i.e., lipid binding affinity is decreased), the search unit 113 cancels the previous mutation in the mutated amino acid sequence (i.e., returns the substituted amino acid to the amino acid before the mutation) and introduces another mutation into the mutated amino acid sequence. After introducing the new mutation, the search unit 113 again has the prediction unit 112 predict the lipid binding affinity of the mutated amino acid sequence.

[0058] The search unit 113 repeats the process of introducing mutations into the amino acid sequence and the process of estimating lipid binding strength until the search termination conditions are met. Examples of search termination conditions include obtaining an amino acid sequence with a lipid binding strength exceeding a predetermined target value, reaching a predetermined number of searches (number of mutations), or the elapsed of a predetermined time.

[0059] Figure 7 is a graph showing the binding affinity to a specific lipid between a protein with a randomly mutated amino acid sequence (nanobody N1) and a protein with an amino acid sequence searched by the search unit 113 (nanobody N2). The vertical axis of Figure 7 shows the estimated value of the lipid binding affinity by the prediction unit 112, and the horizontal axis shows the distance from known nanobodies (degree of mutation). As shown in Figure 7, some nanobodies N2 searched by the search unit 113 have lipid binding affinity that cannot be reached by the randomly mutated nanobodies N1. Furthermore, the nanobodies N2 searched by the search unit 113 have novel sequences that are far removed from similar sequences of known nanobodies.

[0060] Furthermore, the search unit 113 can search for amino acid sequences that exhibit specific binding affinity, such as binding to a target lipid (e.g., PI(3,5)P2) but not to other lipids (e.g., PS). In other words, the search unit 113 can search for proteins that have high binding affinity to the target lipid and low binding affinity to lipids other than the target lipid. In contrast, with random mutations as described above, it is difficult to obtain amino acid sequences that exhibit such specific binding affinity.

[0061] Figure 8 is a graph showing the binding affinity of four proteins ("Ev44", "Ev53", "Ev56", and "Ev66") that exhibit specific binding affinity and were searched by the search unit 113. The vertical axis of the four graphs in Figure 8 represents the magnitude of lipid binding affinity. As shown in Figure 8, Ev44 has no binding affinity to PS but a high binding affinity to PI(3,4)P2. Ev53 also has no binding affinity to PS but a high binding affinity to PI(3,5)P2. Ev56 has a low binding affinity to PS but a high binding affinity to PI(3,5)P2. Ev66, like Ev53, has no binding affinity to PS but a high binding affinity to PI(3,5)P2. However, Ev66 has lower binding affinity to PI(4,5)P2 and PI(3,4)P2 than Ev53, and exhibits stronger specificity to PI(3,5)P2 than Ev53.

[0062] <Artificial Intelligence Unit 120> The artificial intelligence unit 120 is configured to receive input from each functional unit and return the instructed output. The artificial intelligence used by the information processing device 10 in each functional unit may be common to all units, or it may be prepared individually for each functional unit.

[0063] The artificial intelligence unit 120 may include learning models such as a transformer or a recurrent neural network (RNN), including a generative AI.

[0064] Furthermore, specific machine learning algorithms used to construct learning models include nearest neighbors, naive Bayes, decision trees, support vector machines, and deep learning using neural networks. The artificial intelligence unit 120 can apply the above algorithms as appropriate.

[0065] The artificial intelligence unit 120 may have a trained model constructed by a learning method such as supervised learning, unsupervised learning, or self-supervised learning. In supervised learning, machine learning is performed using training data (machine learning data). Training data consists of pairs of input data and output data (correct answer data) for learning. Furthermore, the language model may not only be one trained for a specific task, but also a general-purpose model that can be used universally for a wide range of tasks. The trained model included in the artificial intelligence unit 120 can undergo additional learning as transfer learning or fine-tuning.

[0066] <Display Unit 211> The display unit 211 of the user terminal 20 is configured to display the screen indicated by the screen data transmitted from the information processing device 10 on the output unit 25.

[0067] <Operation Acquisition Unit 212> The operation acquisition unit 212 of the user terminal 20 is configured to accept operations from the user of the user terminal 20.

[0068] 3. Information Processing Method This section describes the information processing method of the information processing device 10. In this information processing method, each part of the information processing device 10 is executed by the computer as a step.

[0069] Specifically, this information processing method comprises an acquisition step, an inference step, and a search step. In the acquisition step, the amino acid sequence to be inferred is acquired. In the inference step, the lipid binding affinity of a protein having the amino acid sequence to be inferred is inferred based on the amino acid sequence to be inferred and reference information. In the search step, amino acid sequences with high lipid binding affinity are searched for by mutating the amino acid sequence to be inferred, while referring to the lipid binding affinity.

[0070] Figure 9 is an activity diagram showing an example of the flow of information processing (amino acid sequence search processing) performed by the information processing system 1. The information processing will be explained below in accordance with each activity in this activity diagram.

[0071] The amino acid sequence search process begins with the user inputting an amino acid sequence. The user inputs the amino acid sequence on the user terminal 20 (Activity A110). The information processing device 10 acquires the amino acid sequence input on the user terminal 20 as the initial amino acid sequence (Activity A120).

[0072] After obtaining the initial amino acid sequence, the information processing device 10 estimates the lipid binding strength of the initial amino acid sequence (Activity A130). Subsequently, the information processing device 10 creates a mutant amino acid sequence by mutating a part of the initial amino acid sequence (Activity A140). After creating the mutant amino acid sequence, the information processing device 10 estimates the lipid binding strength of the mutant amino acid sequence (Activity A150).

[0073] After estimating the lipid binding affinity of the mutated amino acid sequence, the information processing device 10 determines whether the termination conditions for the amino acid sequence search have been met (Activity A160). If the termination conditions are met, the information processing device 10 outputs the search results (the searched amino acid sequence and its lipid binding affinity) to the user terminal 20 (Activity A170). As a result, the search results are displayed on the user terminal 20 (Activity A180).

[0074] On the other hand, if the termination condition is not met in activity A160, the information processing device 10 determines whether the lipid binding strength of the immediately predicted mutated amino acid sequence has improved compared to the lipid binding strength of the amino acid sequence immediately before the mutation (activity A190). If the lipid binding strength has improved or is maintained (for example, if the lipid binding strength is above a threshold), the information processing device 10 maintains the immediately preceding mutation (activity A200). On the other hand, if the lipid binding strength has decreased (for example, if the lipid binding strength is below a threshold), the information processing device 10 cancels the immediately preceding mutation (activity A210).

[0075] After deciding whether to maintain or cancel the mutation, the information processing device 10, in activity A140, creates a new mutant amino acid sequence by adding another mutation to a part of the amino acid sequence in which the mutation was maintained or canceled. If the mutation is maintained, yet another mutation is added to the mutant amino acid sequence. In activity A150, the information processing device 10 estimates the lipid binding affinity of the newly created mutant amino acid sequence. The information processing device 10 repeats activities A140 and A150, and either activity A200 or activity A210, until the search termination conditions are met.

[0076] 4. Effect The effect of this embodiment can be summarized as follows: By specifying an amino acid sequence, the binding affinity of a protein having that amino acid sequence to lipids can be estimated with high accuracy.

[0077] Although embodiments of the present invention have been described above, the present invention is not limited thereto and can be modified as appropriate without departing from the technical spirit of the invention.

[0078] 5. In the above embodiment, the information processing device 10 performed various storage and control functions, but instead of the information processing device 10, multiple external devices may be used. That is, various information and programs may be stored in a distributed manner across multiple external devices using blockchain technology or the like.

[0079] The embodiments of this model are not limited to the information processing system 1, but may also be an information processing method or a program. The information processing method comprises steps executed by the information processing system 1. The program causes a computer to execute each step of the information processing system 1.

[0080] At least one of the devices included in the information processing system 1 may be located outside the country in which the functions of the information processing system 1 are performed.

[0081] The information processing system 1 may consist only of the information processing device 10. In other words, the information processing system 1 does not necessarily have to include a user terminal 20.

[0082] The information processing system 1 does not necessarily have to include a search unit 113. For example, the information processing system 1 may only have the function of estimating the lipid binding affinity of a specified amino acid sequence.

[0083] The product may be provided in any of the following embodiments.

[0084] (1) An information processing system comprising a processor configured to perform the following steps by reading a program, wherein in the acquisition step, an amino acid sequence to be predicted is acquired, and in the prediction step, the binding force of a protein having the amino acid sequence to be predicted to a lipid is predicted based on the amino acid sequence to be predicted and reference information, wherein the reference information includes a correlation between the amino acid sequence and the binding force, constructed based on measured values ​​of the binding force.

[0085] (2) An information processing system as described in (1) above, wherein the reference information includes first reference information and second reference information, the first reference information includes a correlation between an amino acid sequence and a feature vector, the second reference information includes a correlation between the feature vector and the binding force, and in the estimation step, the system obtains the feature vector based on the amino acid sequence to be estimated and the first reference information, and estimates the binding force based on the feature vector and the second reference information.

[0086] (3) An information processing system described in (2) above, wherein the second reference information is constructed using machine learning data that combines the feature vector obtained from the amino acid sequence of the learning protein and the measured value of the binding force in the learning protein.

[0087] (4) An information processing system according to (2) or (3) above, wherein the first reference information includes a deep learning model trained to generate a feature vector representing the characteristics of a protein from an amino acid sequence, and in the prediction step, the amino acid sequence to be predicted is input to the deep learning model, and the feature vector is obtained from the deep learning model.

[0088] (5) An information processing system according to any one of (1) to (4) above, wherein the estimation step estimates the binding affinity of the protein to each of the multiple types of lipids or each of the multiple lipid groups.

[0089] (6) An information processing method comprising each step performed by the information processing system described in any one of (1) to (5) above.

[0090] (7) A program that causes a computer to perform each step of the information processing system described in any one of (1) to (5) above.

[0091] 1: Information processing system, 2: Communication line, 10: Information processing device, 11: Control unit, 12: Memory unit, 13: Communication unit, 14: Communication bus, 20: User terminal, 21: Control unit, 22: Memory unit, 23: Communication unit, 24: Input unit, 25: Output unit, 26: Communication bus, 111: Acquisition unit, 112: Prediction unit, 113: Search unit, 120: Artificial intelligence unit, 211: Display unit, 212: Operation acquisition unit, PM1: Deep learning model, PM2: Regression model

Claims

1. An information processing system comprising a processor configured to perform the following steps by reading a program, wherein in the acquisition step, an amino acid sequence to be predicted is acquired; and in the prediction step, the binding force of a protein having the amino acid sequence to a lipid is predicted based on the amino acid sequence to be predicted and reference information, wherein the reference information includes a correlation between the amino acid sequence and the binding force, constructed based on measured values ​​of the binding force.

2. An information processing system according to claim 1, wherein the reference information includes first reference information and second reference information, the first reference information includes a correlation between an amino acid sequence and a feature vector, the second reference information includes a correlation between the feature vector and the binding force, and in the estimation step, the system obtains the feature vector based on the amino acid sequence to be estimated and the first reference information, and estimates the binding force based on the feature vector and the second reference information.

3. An information processing system according to claim 2, wherein the second reference information is constructed using machine learning data that combines the feature vector obtained from the amino acid sequence of the learning protein and the measured value of the binding force in the learning protein.

4. An information processing system according to claim 2 or claim 3, wherein the first reference information includes a deep learning model trained to generate a feature vector representing the characteristics of a protein from an amino acid sequence, and the prediction step includes inputting the amino acid sequence to be predicted into the deep learning model and obtaining the feature vector from the deep learning model.

5. An information processing system according to any one of claims 1 to 4, wherein the estimation step estimates the binding affinity of the protein to each of a plurality of types of lipids or each of a plurality of lipid groups.

6. An information processing method comprising each step performed by the information processing system described in any one of claims 1 to 5.

7. A program for causing a computer to perform each step of the information processing system described in any one of claims 1 to 5.

Images