Method for proposing mutant sequence information, proposal system, and program

The digitally-driven protein improvement method using deep generative models and evaluation values addresses inefficiencies in current protein improvement techniques by selecting variant sequences efficiently, reducing the need for wet experiments and minimizing local solutions, thereby accelerating biosynthesis.

WO2024236722A1PCT designated stage expired Publication Date: 2024-11-21HITACHI LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2023/018227
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-16
Publication Date
2024-11-21

AI Technical Summary

Technical Problem

Current methods for improving proteins involved in biosynthesis are inefficient and require numerous wet experiments, leading to high research and development costs and a risk of falling into local solutions with limited improvement.

Method used

A digitally-driven protein improvement method that generates and evaluates variant sequence information using deep generative models, calculating likelihood and diversity evaluation values to select promising sequences without prior wet experiments, thereby reducing the risk of local solutions and accelerating protein improvement.

Benefits of technology

This method enables efficient selection and proposal of variant sequence information with desired functions, reducing the need for extensive wet experiments and minimizing the risk of local solutions, thus accelerating the biosynthesis process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023018227_21112024_PF_FP_ABST
    Figure JP2023018227_21112024_PF_FP_ABST
Patent Text Reader

Abstract

Provided is a method for proposing mutant sequence information having a desired function, the method comprising: a generation step for generating at least three sets of mutant sequence information on the basis of biological sequence information of a protein having the desired function; an evaluation value calculation step for calculating, on the basis of the mutant sequence information, a likelihood evaluation value indicating the likelihood that a protein according to the mutant sequence information has the desired function, and a diversity evaluation value that represents the diversity of the mutant sequence information; and a selection step for selecting and proposing at least two sets among the at least three sets of mutant sequence information on the basis of the likelihood evaluation value and the diversity evaluation value.
Need to check novelty before this filing date? Find Prior Art

Description

Method, system and program for proposing variant sequence information

[0001] The present invention relates to a method, a system, and a program for proposing variant sequence information.

[0002] One of the major industrial applications of synthetic biology is the synthesis of useful substances using microorganisms (biosynthesis). It is expected to be used in the synthesis of natural compounds that are difficult to synthesize using conventional chemical synthesis, as well as in the synthesis of platform compounds that serve as the basis for various useful compounds.

[0003] To create microorganisms that produce useful substances through biosynthesis, it is necessary to artificially modify the microorganisms. After modifying the gene sequences and metabolic pathways involved in the production of useful substances to increase or decrease metabolic rate (Design), the designed biological cells are constructed (Build), evaluated in various evaluation systems (Test), and the results are learned (Learn) and fed back into the design again. By repeating this cycle (DBTL cycle), the desired microorganisms are created.

[0004] In this cycle, there are many cases where proteins related to the production of useful substances are desired to be improved to have higher functionality. However, obtaining the desired improved protein is not easy, and requires many wet experiments and prior knowledge, which takes a great deal of time and research and development costs. Therefore, technology to more easily improve proteins related to the production of useful substances is extremely important, as it will lead to the acceleration of the social implementation of biosynthesis.

[0005] In response to this demand, digitally driven protein improvement methods have been developed that incorporate digital technology into previously wet experiment-driven protein improvement methods as a more efficient and effective method for improving proteins (Patent Document 1, Patent Document 2).

[0006] International Publication No. 2020 / 246617 Special Publication No. 2022-548841

[0007] The use of digital technology has made it easy to generate a large number of novel mutant sequence information on a computer that did not exist before. To obtain an improved protein, it is necessary to actually create the mutant sequence information based on the computer-generated mutant sequence information. Therefore, it is important to select which mutant sequence information should be created from among these many mutant sequence information.

[0008] Conventional methods for selecting an arbitrary number of mutant sequence information from a large number of novel mutant sequence information generated on a computer include two opposing methods: one is to calculate the likelihood (likelihood) of having a desired biological function value or some other functional value based on the results of wet experiments or simulations, and use this as an evaluation measure, and the other is to select the sequence information randomly.

[0009] Protein improvement requires repeated cycles of selecting and creating mutant sequence information and evaluating it using an evaluation system. Conventional selection methods based on likelihood, which has some functional value, are strongly influenced by the results of the selected sequences in the early cycles, and there is a concern that they may fall into local solutions that limit improvement. Furthermore, in many cases, data acquisition through prior wet experiments is required, which can be a bottleneck for efficient protein improvement.

[0010] On the other hand, random selection methods can be considered to reduce the possibility of falling into a local optimum by selecting diverse sequences. However, considering that random selection methods will only yield a small number of sequences with the desired biological function value in the vast sequence space, more efficient selection methods are needed.

[0011] The present invention aims to reduce the possibility of falling into a local solution that limits improvement among a large number of mutant sequence information generated on a computer, while maintaining the likelihood of the desired function, and to efficiently select and propose the mutant sequence information that should actually be created without conducting prior wet experiments, etc.

[0012] One example of the method according to the present invention is a method for proposing mutant sequence information having a desired function, comprising: a generation step of generating three or more pieces of mutant sequence information based on biological sequence information of a protein having the desired function; an evaluation value calculation step of calculating, based on the mutant sequence information, a likelihood evaluation value which is an evaluation value representing the likelihood that the protein related to the mutant sequence information has the desired function, and a diversity evaluation value which is an evaluation value representing the diversity of the mutant sequence information; and a selection step of selecting and proposing two or more pieces of mutant sequence information based on the likelihood evaluation value and the diversity evaluation value.

[0013] An example of the proposal system according to the present invention proposes sequence information of a variant having the desired function by executing the above-described method based on biological sequence information of the protein having the desired function.

[0014] An example of a program according to the present invention causes a computer to execute the above-described method.

[0015] According to the present invention, it is possible to select and propose mutant sequence information to be actually produced from among the mutant sequence information generated on a computer, while maintaining the likelihood of the desired function, and reducing the possibility of falling into a local optimum that limits improvement. Furthermore, according to one embodiment of the present invention, it is possible to select and propose mutant sequence information to be actually produced from among the mutant sequence information generated on a computer, while maintaining the likelihood of the desired function, and reducing the possibility of falling into a local optimum that limits improvement, without conducting prior wet experiments or the like.

[0016] FIG. 1 is a diagram showing the configuration of a proposed system according to embodiment 1. FIG. 2 is a schematic diagram of an example of generating mutant sequence information. An example of a score obtained by a likelihood calculation module according to this embodiment. An example of a distance matrix created by a diversity calculation module according to this example. An example of a table representation including cluster information. An example of a method for creating a subset according to example 1. An example of a method for creating a subset according to example 2. An example of a method for creating a subset according to example 3. An example of a creation request input screen for subset creation. An example of a model learning status drawing screen. An example of a subset display screen. An example of a subset detail display screen.

[0017] [Embodiment 1] Hereinafter, embodiment 1 of the present invention will be described with reference to the drawings. The following description and drawings are examples for explaining the present invention, and some omissions and simplifications have been made as appropriate for clarity of explanation. The present invention can also be implemented in various other forms. Furthermore, unless otherwise specified, each component may be singular or plural.

[0018] 1 is a diagram showing the configuration of a system (proposal system) for proposing variant sequence information according to embodiment 1, particularly showing the processing flow and components. The proposal system proposes variant sequence information having a desired function by executing the method described herein based on biological sequence information of a protein having a desired function.

[0019] In input 101 in Fig. 1, protein biosequence information is input. The biosequence information is, for example, but not limited to, information representing an amino acid sequence or a nucleic acid sequence. The format of the biosequence information is, for example, but not limited to, a numeric sequence or a symbol string. An amino acid sequence can be expressed as an array of expressions (for example, alphabets) representing any of 20 types of amino acids. A nucleic acid sequence can be expressed as an array of expressions (for example, alphabets) representing any of four types of bases.

[0020] The biosequence information can include, but is not limited to, biosequence information that can be obtained from known biosequence information databases, such as Uniprot and InterPro. It can also include biosequence information obtained or created in-house by the practitioner. Furthermore, this biosequence information can be subjected to pre-processing such as selection and alignment as necessary.

[0021] The variant sequence generation module 102 in FIG. 1 generates a large number (three or more) of variant sequence information based on input 101 (particularly, based on biosequence information of a protein having a desired function) (generation step). The variant sequence information is information expressed in a format similar to biosequence information. The variant sequence generation module 102 can use a deep generative model (e.g., a model based on a Transformer architecture) and / or a diffusion model. Deep generative models include, for example, long short-term memory (LSTM), recurrent neural networks (RNNs), gated recurrent units (GRUs), generative adversarial networks (GANs), variational autoencoders (VAEs), Flow, or bidirectional encoder representations from transformers (BERT) models. However, the method for generating variant sequence information is not limited to these.

[0022] Based on the numerous mutant sequence information generated by the mutant sequence generation module 102, the likelihood calculation module 103 in FIG. 1 calculates a score (likelihood evaluation value) for each mutant sequence information, representing the likelihood (likelihood) that the protein associated with that mutant sequence information has the desired biological function (first evaluation value calculation step). The definition of the score can be determined appropriately by those skilled in the art. As a specific example, when using biological sequence information for an enzyme with a specific activity, the likelihood that the enzyme associated with the mutant sequence information has a non-zero activity value for that activity can be used as the score. The format of the score can also be determined appropriately by those skilled in the art. For example, the score can be defined so that a higher score indicates a higher likelihood that the mutant sequence information has the desired biological function value, or conversely, the score can be defined so that a lower score indicates a higher likelihood.

[0023] The diversity calculation module 104 in Figure 1 aligns the multiple variant sequence information generated by the variant sequence generation module 102. "Alignment" refers to a process for identifying the correspondence between identical or similar portions between multiple sequences, and alignment makes it possible to understand the correspondence between elements that correspond in content when they are positioned at different positions in different sequences. The specific processing content for alignment can be designed as appropriate by those skilled in the art.

[0024] Furthermore, the diversity calculation module 104 calculates the similarity between each piece of mutant sequence information based on the mutant sequence information, and calculates the distance between each piece of mutant sequence information based on the similarity. Furthermore, a matrix (distance matrix) representing the distance between each piece of mutant sequence information is generated based on the calculated distances (second evaluation value calculation step). The distance matrix can be used as a measure (diversity evaluation value) representing the diversity in the entire mutant sequence information. In Examples 1 to 3 described in detail below, only the distance matrix can be used, or the distance matrix and the results of clustering the mutant sequence information based on the distance matrix can be used.

[0025] 1 selects a specified number (two or more) of mutant sequence information as a subset of the set of mutant sequence information (selection step) based on the likelihood evaluation value calculated by the likelihood calculation module 103 and the diversity evaluation value calculated by the diversity calculation module 104. The number of mutant sequence information to be selected can be specified in advance by the user.

[0026] This selection is performed, for example, by combinatorial optimization. For example, single-objective optimization or multi-objective optimization is performed according to the likelihood evaluation value and the diversity evaluation value, and a subset containing a specified number of mutant sequence information is created from a large number of mutant sequence information. For example, multi-objective optimization can be performed using an objective function that maximizes the likelihood evaluation value and an objective function that maximizes the diversity evaluation value. Note that it is also possible to use a method that does not fall under single-objective optimization or multi-objective optimization.

[0027] The effect verification module 106 in Figure 1 is a module that verifies the effect of the subsets created by the subset creation module 105. For example, it creates a scatter plot based on the information on the numerous mutant sequences created by the mutant sequence generation module 102, and calculates a coverage ratio that indicates how widely the subsets are spatially scattered. A method for calculating the coverage ratio can be designed appropriately by those skilled in the art. Note that the operation of the effect verification module 106 is not limited to this.

[0028] 1 outputs the subsets created by the subset creation module 105. In this way, the proposal system proposes the selected variant sequence information.

[0029] The variant sequence generation module 102, likelihood calculation module 103, diversity calculation module 104, subset creation module 105, and effect verification module 106 can be configured by one or more computers. The computer includes, for example, a calculation means and a storage means. The calculation means includes, for example, a processor, and the storage means includes, for example, a storage medium such as a semiconductor memory device or a magnetic disk device. Some or all of the storage medium may be non-transitory storage media.

[0030] The computer may also include input / output means, which include, for example, input devices such as a keyboard and a mouse, output devices such as a display and a printer, and communication devices such as a network interface.

[0031] The storage means may store a program, which the processor may execute to cause the computer to carry out the methods described herein.

[0032] Hereinafter, more detailed examples of this embodiment will be described.

[0033] [Example 1] Figure 2 shows a schematic diagram of an example in which biosequence information of a protein to be improved, as well as a large number of related biosequence information, are obtained from a known biosequence information database 201, and mutant sequence information 207 is generated using a deep generative model.

[0034] Biosequence information 202 is obtained from a biosequence information database 201. At this time, the biosequence information may include, for example, biosequence information of a protein group belonging to the superfamily of the protein to be improved. Biosequence information of other proteins may also be used as needed.

[0035] The learning unit 203 and mutant sequence generation unit 209 in Figure 2 constitute the mutant sequence generation module 102 in Figure 1. The learning unit 203 includes a preprocessing unit 204. The preprocessing unit 204 selects and discards biosequence information from the biosequence information 202 as needed, and performs preprocessing such as alignment. The preprocessing unit 204 performs a conversion process on the biosequence information 202, for example, so that it is expressed as a one-hot vector used in a language processing program. That is, the biosequence information is converted into an array of one-hot vectors by converting each of the multiple elements (e.g., residues) that make up the sequence into a single one-hot vector.

[0036] In the preprocessing unit 204, the conversion process to one-hot vector representation is just one example, and further amino acid sequence representation grammar functions may be incorporated, and functions for removing representations other than one-hot vector representations and non-existent biological sequence representations may be added.

[0037] Furthermore, the learning unit 203 can refer to physical property information 208 corresponding to the biosequence information 202 as necessary. Examples of the physical property information 208 include, but are not limited to, stability, solubility, and the like.

[0038] Next, the sequence feature encoding unit 205 reads the preprocessed biological sequence information 202 and the preprocessed physical property information 208, and generates a distributed representation having at least two elements as a feature based on these. The sequence feature encoding unit 205 can be configured using the above-mentioned deep generative model or diffusion model.

[0039] The sequence feature decoding unit 206 reads the distributed representation and generates a biosequence representation based on it (i.e., converts it back into biosequence information). Note that the sequence feature decoding unit 206 can also be configured using the above-mentioned deep generative model or diffusion model.

[0040] The learning unit 203 adjusts the parameters of the sequence feature encoding unit 205 and the sequence feature decoding unit 206 so that the input biosequence information 202 and the output biosequence information are the same. Through this series of processes, the learning unit 203 performs machine learning.

[0041] Next, a mutant sequence generation unit 209 generates a large number of mutant sequence information 207 based on the biosequence information of a protein to be improved. At this time, the biosequence information of the protein to be improved may be included in the biosequence information 202. Furthermore, physical property information of the protein to be improved may be referenced as necessary.

[0042] As a specific example of processing, the biological sequence information of the protein to be improved is input to the trained sequence feature encoding unit 205 and converted into a distributed representation, and then the distributed representation is subjected to probabilistic variation. In this way, a large number of varied distributed representations are obtained based on the biological sequence information of a single protein. These are input to the sequence feature decoding unit 206 and converted back to a biological sequence representation, thereby generating a large number of mutant sequence information 207. Note that any random variable can be used in the probabilistic variation. For example, the variation of the distributed representation can be selected to follow a multidimensional Gaussian distribution with a standard deviation of 2σ, but this is not limited to this.

[0043] Fig. 3 shows an example of a score calculated by the likelihood calculation module 103 according to this embodiment. Fig. 3 includes variant sequence information and a score calculated by the likelihood calculation module 103, which correspond to each other one-to-one.

[0044] The score may represent a relative evaluation value between the mutant sequence information, and the score may be calculated using, but is not limited to, the loss function used in the learning unit 203, any other simulation, specific physical property information, etc.

[0045] 4 shows an example of a distance matrix created by the diversity calculation module 104 according to this embodiment. The distance matrix shows, for all pairs of mutant sequence information, distances that can be calculated from the similarities between the mutant sequence information after the diversity calculation module 104 calculates and aligns the mutant sequence information.

[0046] In this way, the proposed system aligns the generated variant sequence information with each other, calculates the similarity, and creates a distance matrix based on the similarity.

[0047] The distance value of each cell may represent a relative evaluation value between the variant sequence information. The similarity between the variant sequence information after alignment may be used to calculate the distance. A specific method for calculating the distance between the variant sequence information can be appropriately defined by a person skilled in the art based on the number, position, type, etc. of different sequence elements (e.g., residues).

[0048] FIG. 5 shows an example of a table representation in which clustering is performed based on a distance matrix by the diversity calculation module 104 according to this embodiment and cluster information is included.

[0049] FIG. 5 includes variant sequence information, scores calculated by the likelihood calculation module 103, and cluster information, which correspond to each other in a 1:1:1 relationship.

[0050] Clustering can be performed by any method, for example, at least one of the following methods: PCA, t-SNE, MDS (Multi-Dimensional Scaling), and K-means, but is not limited to these.

[0051] Next, the subset creation module 105 selects two or more pieces of mutant sequence information based on the mutant sequence information, the score by the likelihood calculation module 103, and the clustering results (e.g., cluster number) by the diversity calculation module 104, and creates a subset.

[0052] 6 shows an example of a technique for creating a subset according to Example 1. According to this example, the subset can be created using integer linear programming.

[0053] In FIG. 6, S represents a set of variant sequence information generated by the variant sequence generation module 102, and s represents each variant sequence information included in S.

[0054] Variable x s is a binary variable that takes 0 or 1, and when a certain variant sequence information s is selected as a subset, x s = 1, and if not selected, x s =0.

[0055] C i means a group of variant sequence information having the i-th cluster number calculated by the diversity calculation module 104. ∪ represents a union, ∩ represents a product, and φ represents an empty set (i.e., each variant sequence information has only one of the cluster numbers).

[0056] score s means the score of the variant sequence information s. 1 is the number of variant sequence information to be selected as a subset, which can be arbitrarily designated by the user, or may be a fixed constant. 2 is the upper limit of the number of variant sequence information groups having the same cluster number to be selected as a subset, and can be arbitrarily designated by the user, but may also be a fixed constant.

[0057] By solving the integer linear programming problem of Equation 1, m 1 A subset is created as a combination of an arbitrary number of optimal mutant sequence information specified as follows: Note that the formula in Figure 6 is an example of a formulation (single-objective optimization method) of an integer linear programming problem with a single-objective function to be solved in this embodiment, and is not limited to this.

[0058] [Embodiment 2] Embodiment 2 will be described below, focusing mainly on the differences from embodiment 1, and the description of the commonalities with embodiment 1 will be simplified or omitted.

[0059] In this embodiment, a subset containing a specified number of variant sequence information is created based on the score calculated by the likelihood calculation module 103 and the distance matrix created by the diversity calculation module 104.

[0060] 7 shows an example of a method for creating a subset according to Example 2. According to this example, the subset can be created using a multi-objective optimization method.

[0061] In FIG. 1 and f 2 Each of these represents a different objective function. ij is the distance between the variant sequence information i and the variant sequence information j, and is an element of the distance matrix created by the diversity calculation module 104. s means the score of the variant sequence information s.

[0062] w 1 and w 2 means the weight of each objective function. The user can specify any value for these in advance. m is the number of variant sequence information to be selected as a subset, and can be specified by the user, but may also be a fixed constant.

[0063] By solving an integer programming problem based on the formula in Figure 7, a subset is created as a combination of an arbitrary number of optimal variant sequence information designated as m. According to the formula in Figure 7, f 1 Due to the contribution of the above, variant sequence information that is far from other variant sequence information is preferentially selected, thereby improving the diversity in the subset.

[0064] The formula in FIG. 7 is an example of a formulation of an integer programming problem using a multi-objective function to be solved in this embodiment, and is not limited to this.

[0065] [Example 3] Example 3 will be described below, focusing mainly on the differences from Examples 1 and 2, and explanations of the commonalities with Examples 1 and 2 will be simplified or omitted.

[0066] 8 shows an example of a method for creating a subset according to Example 3. According to this example, the subset can be created using a multi-objective optimization method.

[0067] In FIG. 1 and f 2 Each of these represents a different objective function.ij is the distance between the variant sequence information i and the variant sequence information j, and is an element of the distance matrix created by the diversity calculation module 104. s means the score of the variant sequence information s. ij is x i and x j It can be thought of as a value representing the logical AND with

[0068] w 1 and w 2 means the weight of each objective function. The user can specify any value for these in advance. m is the number of variant sequence information to be selected as a subset, and can be specified by the user, but may also be a fixed constant.

[0069] By solving the integer programming problem based on the formula in Figure 8, subsets are created as combinations of optimal mutant sequence information of any number specified as m. According to the formula in Figure 8, as in Example 2 (Figure 7), f 1 Due to the contribution of the above, variant sequence information that is far from other variant sequence information is preferentially selected, thereby improving the diversity in the subset.

[0070] The formula in FIG. 8 is an example of a formulation of an integer programming problem with a multi-objective function to be solved in this embodiment, and is not limited to this.

[0071] [Example 4] Example 4 will be described below, focusing mainly on the differences from Examples 1 to 3, and explanations of the commonalities with Examples 1 to 3 will be simplified or omitted.

[0072] FIG. 9 shows an example of a graphic user interface in the proposed system, and shows an example of a creation request input screen for subset creation.

[0073] The graphic user interface includes a subset creation request input screen 901. The proposed system starts processing based on request input fields (e.g., an input field 902 for biosequence information of a certain protein and an input field 903 for biosequence information of proteins belonging to the same superfamily).

[0074] The graphic user interface may include input fields for hyperparameters required for processing (e.g., an input field 904 for the number of variant sequence information to be selected as a subset, an input field 905 for the number of subsets to be created, an input field 906 for the sampling range, an input field 907 for the number of variant sequences to be generated, an input field 908 for the clustering method, an input field 909 for the number of clusters to be created by clustering, and an input field 910 for the phylogenetic tree creation method, but are not limited to these).

[0075] In FIG. 9, when a subset creation button 911 is pressed on the subset creation request input screen 901, the graphic user interface transitions to the screen shown in FIG.

[0076] FIG. 10 shows an example of a model learning state drawing screen 1003 as an example of a graphic user interface in the proposed system.

[0077] The model learning status 1004 shows the transition of the numerical value output by the loss function as an example of a display format. The display format is not limited to this and may be any format.

[0078] Depending on the model learning status 1004, the user can select and operate a Start Subset Creation button 1005, a Continue Learning button 1006, or an Interrupt Processing button 1007. When the Start Subset Creation button 1005 is operated, the proposed system completes learning and starts the process of generating variant sequence information. When the Continue Learning button 1006 is operated, the proposed system continues the learning process. When the Interrupt Processing button 1007 is operated, the proposed system interrupts the learning process and / or the process of generating variant sequence information.

[0079] FIG. 11 is an example of a graphic user interface, showing an example of a subset display screen 1101 after subsets have been created.

[0080] The subset display screen 1101 displays the created subsets. An example of the display format is shown as a scatter plot. In the example of FIG. 11, the scatter plot is two-dimensional, and it is possible to display values ​​obtained by converting the feature quantities generated by the sequence feature encoding unit 205 into two dimensions. Note that the display format is not limited to this, and any format such as a table format, a list format, or a phylogenetic tree display may be used.

[0081] In this way, the proposed system displays the distribution of the generated variant sequence information (e.g., all variant sequence information) and the variant sequence information selected as a subset (display step). Such a display allows the user to easily grasp the diversity within the subset.

[0082] The scatter plot shows the position of each variant sequence information, and the display format may be different for those selected as a subset and those not. For example, different colors or shapes may be used. This improves visibility. In Figure 11, variant sequence information selected as a subset is indicated by a star, and variant sequence information not selected as a subset is indicated by a dot. However, the display format is not limited to these.

[0083] After multiple subsets have been created, when one subset is selected by the user in FIG. 11, the graphic user interface transitions to the screen shown in FIG.

[0084] 12 is an example of a graphic user interface in the proposal system, and shows an example of a subset detail display screen 1202. The subset detail display screen 1202 displays detailed information about a selected subset.

[0085] The detailed information includes, for example, a scatter plot 1203, a phylogenetic tree 1204, information on the frequency of occurrence of mutated residues in the subset 1205, a histogram of the number of mutations in the subset 1206, information on mutant sequences in the subset 1207, likelihood scores of mutant sequences in the subset 1208 (e.g., scores calculated by the likelihood calculation module 103), and a cluster number 1209. The cluster number 1209 may be replaced with the coverage calculated by the effect verification module 106.

[0086] In the phylogenetic tree 1204, the variant sequence information selected as the subset is marked with a black frame, making it distinguishable from the variant sequence information that was not selected. In the frequency of occurrence information 1205, the horizontal axis represents the position of the residue within the sequence, and the vertical axis represents the number of sequences in which that residue is mutated in the subset. In the mutation number histogram 1206, the horizontal axis represents the number of mutations within the sequence, and the vertical axis represents the cumulative number of sequences containing that number of mutations. The intra-subset variant sequence information 1207, intra-subset variant sequence likelihood score 1208, and cluster number 1209 can be the same as those shown in FIG. 5.

[0087] The subset detail display screen 1202 in FIG. 12 is an example, and the display format and display contents are not limited to this.

[0088] DESCRIPTION OF SYMBOLS 101...Input 102...Mutant sequence generation module 103...Likelihood calculation module 104...Diversity calculation module 105...Subset creation module 106...Effect verification module 107...Output 201...Biosequence information database 202...Biosequence information 203...Learning unit 204...Preprocessing unit 205...Sequence feature encoding unit 206...Sequence feature decoding unit 207...Mutant sequence information 208...Physical property information 209...Mutant sequence generation unit 901...Subset creation request input screen 902-910...Input fields 911...Subset creation button 1003...Model learning status drawing screen 1004...Model learning status 1005...Subset creation start button 1006...Learn continuation button 1007...Processing interrupt button 1101...Subset display screen 1202...Subset detail display screen 1203...Scatter plot 1204... Phylogenetic tree 1205... Occurrence frequency information 1206... Mutation number histogram 1207... Mutant sequence information within subset 1208... Mutant sequence likelihood score within subset 1209... Cluster number

Claims

1. A method for proposing mutant sequence information having a desired function, comprising: a generation step of generating three or more mutant sequence information based on biological sequence information of a protein having the desired function; an evaluation value calculation step of calculating, based on the mutant sequence information, a likelihood evaluation value which is an evaluation value representing the likelihood that the protein related to the mutant sequence information has the desired function, and a diversity evaluation value which is an evaluation value representing the diversity of the mutant sequence information; and a selection step of selecting and proposing two or more of the three or more mutant sequence information based on the likelihood evaluation value and the diversity evaluation value.

2. The method according to claim 1, further comprising a display step of displaying a distribution of the three or more variant sequence information and the selected two or more variant sequence information.

3. The method according to claim 1, further comprising the steps of: calculating similarity after aligning the three or more mutant sequence information with each other; and creating a distance matrix based on the similarity; and wherein the selection step uses integer linear programming.

4. The method according to claim 3, further comprising the step of clustering the variant sequence information based on the distance matrix, and the selection step is further performed based on a result of the clustering.

5. The method of claim 1, wherein the selection step uses a single-objective or multi-objective optimization technique.

6. A proposal system which proposes sequence information of a mutant having the desired function by executing the method of claim 1 based on biological sequence information of a protein having the desired function.

7. A program for causing a computer to execute the method according to claim 1.

Citation Information

Patent Citations

  • Method, system, and software for identifying functional bio-molecule

    JP2009277235A

  • Structure-based predictive modeling

    JP2016537699A

  • Machine learning methods and apparatus for evolutionary data-driven design of proteins and other sequence-defined biomolecules

    JP2022548841A