Optimize capsid monomer sequences using machine learning

Abstract

A method, computer system, and apparatus for optimizing capsids for predefined characteristics are disclosed, including a computer program encoded on a computer storage medium. The system maintains candidate monomer data representing a population of one or more candidate capsid monomers. At each iteration in a series of iterations, the system updates (i) the candidate monomer data and (ii) the current values ​​of the model parameters of a first machine learning model. After performing the multiple iterations, the system selects one or more target capsid monomers from the population of candidate capsid monomers and outputs data specifying the selected target capsid monomers.

Description

Priority Claim

[0001] This application claims priority to U.S. Provisional Application No. 63 / 595,610, filed November 2, 2023, and European Application No. 23315432.7, filed November 21, 2023, the contents of which are hereby incorporated herein by reference. Background Technology

[0002] This manual relates to the use of machine learning to optimize viral capsid monomer sequences.

[0003] A virus can refer to a submicroscopic infectious pathogen that can replicate within the living cells of an organism, and the capsid can refer to the protein shell of a virus, which encapsulates the virus's genetic material.

[0004] The capsid may comprise an assembly of repeating protein structural units called “capsid monomers” or simply “monomers”. The capsid may comprise any suitable number of monomers, such as 20 monomers, 50 monomers, 100 monomers, etc. Capsid monomers may comprise an amino acid sequence (chain), wherein each amino acid is selected from a possible set of amino acids, such as the standard set of 20 α-amino acids, for example, glycine, alanine, valine, etc. The amino acid sequence of the capsid monomer may have any suitable length (i.e., where the “length” of the amino acid sequence refers to the number of positions occupied by amino acids in the amino acid sequence), for example, lengths of 10, 20, 30, etc.

[0005] Adeno-associated virus (AAV) is a small virus that can infect humans but elicits a mild immune response; it is currently unclear whether it causes disease. AAV possesses characteristics that make it a suitable candidate for constructing viral vectors for gene therapy, such as its apparent non-pathogenicity, its ability to infect non-dividing cells, and its ability to stably integrate into specific sites within the host cell genome. The AAV capsid can comprise a mixture of capsid monomers VP1, VP2, and VP3, totaling 60 monomers, which assemble together to form a capsid with icosahedral symmetry.

[0006] A machine learning model is a computational model that learns patterns and relationships in data and then uses that knowledge to make predictions or decisions about new data. A neural network is a machine learning model that uses one or more non-linear units to predict outputs from received inputs. The output of each hidden layer serves as the input to the next layer in the network (i.e., the next hidden layer or output layer). Each layer of the network generates its output from the received inputs based on the current values ​​of its corresponding set of parameters. Summary of the Invention

[0007] This disclosure describes methods, computer systems, and apparatus for optimizing a garment for predefined characteristics, including computer programs encoded on a computer storage medium. Predefined characteristics can refer to any suitable characteristic of the garment, such as the manufacturability of the garment.

[0008] In this specification, the "manufacturability" of the capsid characterizes the efficiency of capsid production. Specifically, capsid manufacturability can characterize the rate at which a plasmid encoding the capsid can be transfected into one or more production cells. More specifically, capsid manufacturability can be based at least in part on the number of capsid instances generated when a plasmid encoding the capsid is transfected into one or more cells. Capsid manufacturability indirectly characterizes various structural and biochemical properties of the capsid that affect its production, such as the ease with which capsid monomers assemble into the three-dimensional structure of the capsid, the stability of the capsid, etc.

[0009] In this specification, "embedding" can refer to an ordered set of values ​​(e.g., vectors, matrices, or other numerical tensors).

[0010] In one aspect, this disclosure provides a method for optimizing capsid monomer sequences. This method can be implemented by a system comprising one or more computers. The system maintains candidate monomer data characterizing a population of one or more candidate capsid monomers. For each candidate capsid monomer in the population, the candidate monomer data includes: (i) data defining the corresponding amino acid sequence of the corresponding candidate capsid monomer, and (ii) data specifying a corresponding fitness score characterizing a predefined property of the capsid containing the corresponding candidate capsid monomer. The system initializes the population of candidate capsid monomers using (i) one or more initial capsid monomer sequences and (ii) the corresponding fitness score for each corresponding initial capsid monomer sequence. The system initializes the current values ​​of the model parameters of a first machine learning model. At each iteration in a series of iterations, the system updates (i) the candidate monomer data and (ii) the current values ​​of these model parameters of the first machine learning model. To update (i) the candidate capsid data and (ii) the current values ​​of the model parameters of the first machine learning model, the system: uses the first machine learning model to generate data representing a set of query capsids based on the current values ​​of the model parameters and the corresponding amino acid sequences of one or more candidate capsids in the population; processes the corresponding data representing each of these query capsids using a second machine learning model to generate a corresponding output specifying a corresponding baseline fitness score for that query capsid; updates the current values ​​of the model parameters of the first machine learning model using the baseline fitness scores of these query capsids; adds one or more of these query capsids to the population of candidate capsids; and updates the candidate capsid data by including, for each query capsid in the set of query capsids: (i) data defining the corresponding amino acid sequence of the corresponding query capsid, and (ii) data specifying the corresponding baseline fitness score of the corresponding query capsid. After performing multiple iterations, the system selects one or more target capsids from the population of candidate capsids and outputs data specifying the selected target capsids.

[0011] In some implementations, in order to select one or more target capsid monomers, the system selects the target capsid monomer with the highest fitness score from the population of candidate capsid monomers.

[0012] In some implementations, this predefined characteristic characterizes the manufacturability of the capsid. For example, the manufacturability of the capsid can be defined based on the number of capsid instances generated when a plasmid encoding the capsid is transfected into one or more cells. The number of capsid instances generated when a plasmid encoding the capsid is transfected into one or more cells can be normalized by the number of wild-type sequences generated.

[0013] In some implementations, this predefined characteristic characterizes the ability of a virus containing the capsid to avoid being neutralized.

[0014] In some implementations, this predefined characteristic characterizes the immunoreactivity of the capsid.

[0015] In some implementations, the predefined characteristics of the capsid characterize the predictive ability of the virus containing the capsid to penetrate target tissue.

[0016] In some implementations, the predicted fitness score characterizes the packing capacity of the cap.

[0017] In some implementations, the predicted fitness score characterizes the predictive ability of a virus containing the capsid to integrate into the host genome.

[0018] In some implementations, the capsid corresponds to the capsid of adeno-associated virus (AAV).

[0019] In some embodiments, the system further physically generates one or more viruses, each containing a protein having one of the selected target capsid monomers. The system can then be further applied to a subject to achieve a therapeutic effect.

[0020] In some embodiments, the second machine learning model includes: an embedded neural network configured to process inputs defining an amino acid sequence of the corresponding query capsid monomer to generate a corresponding embedding of the amino acid sequence; and a fitness prediction machine learning model configured to process the corresponding embedding of the amino acid sequence to generate an output characterizing a corresponding predicted fitness score of the corresponding query capsid monomer. The second machine learning model may further include an error prediction machine learning model configured to process inputs representing the amino acid sequence of the corresponding query capsid monomer to generate a corresponding fitness error, wherein the corresponding fitness error is an estimate of the error of the corresponding predicted fitness score generated by the fitness prediction machine learning model. The corresponding baseline fitness score can be calculated by combining (i) the corresponding predicted fitness score generated by the fitness prediction machine learning model and (ii) the corresponding fitness error generated by the error prediction machine learning model.

[0021] In some implementations, the fitness prediction machine learning model has a different model architecture than the error prediction machine learning model.

[0022] In some implementations, the embedded neural network has been trained on a training capsid set using self-supervised learning. For example, the self-supervised learning may include training the embedded neural network on the training capsid set to perform a token demasking task. In another example, the self-supervised learning may include training the embedded neural network on the training capsid set to perform a next token prediction task.

[0023] In some implementations, the fitness prediction machine learning model has been trained on a set of training examples, wherein each training example corresponds to a corresponding training captive and includes: (i) the embedding of the training captive, and (ii) the target fitness score of the training captive.

[0024] In some implementations, generating data representing the query capsid monomer set includes: generating data representing a set of mutant capsid monomers that mutate from one or more candidate capsid monomers among these candidate capsid monomers; initializing the query capsid monomer set by selecting from the mutant capsid set; and updating the query capsid monomer set using the first machine learning model.

[0025] In some implementations, the first machine learning model is configured to process a first input representing the input capsid monomer to predict the adjusted fitness score of the input capsid monomer.

[0026] In some embodiments, the first input further characterizes a mutation of the input capsid monomer relative to one or more of its mutated wild-type capsid monomers. In some embodiments, the adjusted fitness score measures a combination of: (i) a predicted fitness characterizing a predefined property of the input capsid monomer, and (ii) a mutation distance between the input capsid monomer and the corresponding wild-type capsid monomer.

[0027] In some implementations, the first machine learning model includes: an encoder network configured to process data of corresponding input capsid fragments adjacent to each mutation of the input capsid from the corresponding wild-type capsid to generate corresponding feature vectors; a pooling network layer configured to combine the corresponding feature vectors generated from these input capsid fragments into a combined feature vector; and a decoder network configured to process the combined feature vector to generate a predicted score for the input capsid.

[0028] In some implementations, updating the query capsid set using the first machine learning model includes, at each iteration in a plurality of iterations, performing the following operations: predicting an adjusted fitness score of a subset of query capsids using the first machine learning model, selecting one or more seed capsids from the subset of query capsids based on the predicted adjusted fitness score, generating data representing an additional set of mutant capsids mutated from the selected seed capsids, and updating the query capsid set by including one or more mutant capsids from the additional set of mutant capsids.

[0029] This disclosure also provides a system comprising one or more computers and one or more storage devices storing instructions that, when executed by one or more computers, cause the one or more computers to perform the optimization method described above.

[0030] This disclosure also provides one or more computer storage media that store instructions that, when executed by one or more computers, cause the one or more computers to perform the optimization method described above.

[0031] The subject matter described in this disclosure may be implemented in particular embodiments to achieve one or more advantages.

[0032] Capsid monomer engineering is widely used in virology and gene therapy to modify the structure and function of viral capsid proteins. The capsid is the outer protein shell of a virus. In the context of gene therapy, viral capsids (such as those derived from adeno-associated virus (AAV)) are often used as vectors to deliver therapeutic genes into target cells.

[0033] Gene therapy holds the potential to transform healthcare, for example by enabling the treatment of diseases such as hemophilia, leukemia, and melanoma with single-dose therapy using therapeutic viral vectors (e.g., AAV vectors). However, gene therapy treatments are typically expensive, costing $1 million or more to treat a single patient in some cases. This high cost is a major obstacle to fully realizing the potentially transformative effects of gene therapy. A significant factor contributing to the cost of gene therapy is the manufacturing cost of the viral vector capsid.

[0034] To optimize capsid properties (e.g., manufacturability), one approach is to perform high-throughput screening of a large number of synthesized candidate capsid mutants to identify those with the desired properties. However, this approach can present significant challenges. For example, the AAV VP1 capsid monomer sequence consists of 735 amino acids. Screening all possible mutants along the entire 735-amino acid length is impractical. For instance, even considering only substitution mutations, 20...735 The search space expands significantly when insertions and deletions are included. Even when focusing on a small region of length 14-aa, the number of possible sequences still reaches 20. 14 One, that is, about 10 18 The practical cloning limit will be limited to <10 unique candidates that can be generated at the plasmid level. 8 When packaging the viral library, this number is further reduced to a fraction of that number (<10). 7 Therefore, over 99.999999999% of potential candidates were lost.

[0035] To aid in the engineering of viral capsid mutants, machine learning methods can be applied. For example, instead of using high-throughput screening, predictive models trained on large-scale AAV mutants can be used to predict fitness scores of predefined properties characterizing new sequences of AAV mutants. However, this approach also presents significant challenges.

[0036] First, the accuracy and robustness of a machine learning model can be limited by the quantity and quality of the training data available to train that model. In many cases, generating capsid fitness training data for training a fitness prediction system involves conducting laboratory experiments, such as measuring the number of capsid instances generated by transfecting cells with plasmids encoding these capsids. These laboratory experiments can be costly and time-consuming, and may produce noisy (e.g., inaccurate or inconsistent) data.

[0037] Furthermore, the quality of the capsid (e.g., manufacturability) can arise from complex interactions of biology, chemistry, and physics, thus requiring machine learning models to perform sophisticated implicit reasoning to predict capsid quality. The reasoning ability of a machine learning model can depend on the number of its parameters; for example, increasing the number of parameters can improve its reasoning ability. Some machine learning models performing complex prediction tasks may include a large number of parameters, such as millions or billions. However, increasing the number of parameters in a machine learning model can increase the likelihood of "overfitting" during training, especially when some residual variations (noise) in the training data are captured as variations in the underlying structure of the training data. The risk of overfitting can be exacerbated when the training data is quantitatively limited and noisy, as is the case with the capsid fitness training data described above.

[0038] Furthermore, as mentioned above, the sequence space of possible viral capsid mutants is large (e.g., for the AAV VP1 monomer, there are 20). 735(Number of possible sequences). Attempting to exhaustively sample this space would require an impractical amount of computational resources, even with computationally efficient prediction models.

[0039] The capsid optimization system described in this specification implements various innovations to address these problems.

[0040] This system implements a high-quality discriminative machine learning model to predict the corresponding fitness of a capsid, including the candidate capsid, for each candidate capsid monomer. In some implementations, the discriminative machine learning model includes (i) an embedded neural network and (ii) a fitness prediction model, the embedded neural network being configured to generate embeddings of the input amino acid sequence, and the fitness prediction model being configured to process the embeddings to predict the fitness score of the sequence. The embedded neural network can be pre-trained using self-supervised learning techniques on a large amount of protein data for which labels (e.g., capsid fitness scores) are unavailable to generate efficient embeddings of the capsid monomer sequence. Training the embedded neural network to perform a self-supervised task enables the embedded neural network to learn to generate capsid monomer embeddings that compactly encode the biological / chemical / mechanical properties of the capsid monomers, which can be used by the fitness prediction model to generate accurate fitness scores. Furthermore, training the embedded neural network on a self-supervised task does not depend on capsid fitness training data and is therefore not affected by limitations in the quantity and quality of available capsid fitness training data.

[0041] Furthermore, instead of using exhaustive sampling to explore the sequence space, the capsid optimization system employs model-guided search; that is, it uses a generative or inferential machine learning model to generate or guide the generation of candidate capsid monomer sequences in each iteration of multiple iterations. A discriminative machine learning model is used to evaluate the generated candidate capsid monomer sequences in each iteration, and the evaluation data is used to train the generative or inferential machine learning model in each iteration. In particular, in some implementations, the system can use generative or inferential machine learning to perform proximal exploration of the sequence space, i.e., considering both the fitness of the inferred candidate capsid monomer sequences and the mutational distance between the candidate capsid monomer and its corresponding wild-type capsid monomer. By using proximal exploration, the system can efficiently explore the sequence space with fewer queries, thereby identifying high-quality capsid monomer sequences.

[0042] Details of one or more embodiments of the subject matter described in this disclosure are set forth in the accompanying drawings and the description below. Other features, aspects, and advantages of this subject matter will become apparent from the specification, drawings, and claims. Attached Figure Description

[0043] Figure 1 An example environment for optimizing capsid monomer sequences is shown.

[0044] Figure 2 An example shell optimization system is shown.

[0045] Figure 3 An example discriminant model for benchmarking capsid fitness is shown.

[0046] Figure 4 This is a flowchart of an example process for optimizing the coat for predefined characteristics.

[0047] Figure 5 This is a flowchart of an example process for generating a query capsid monomer sequence.

[0048] Figure 6 This paper presents a comparison between capsid quality achieved through the capsid optimization system described in this specification and capsid quality achieved through random mutation of candidate capsid monomer sequences.

[0049] Figure 7 This is a block diagram of an example computer system.

[0050] In each of the accompanying drawings, the same reference numerals and names indicate the same elements. Detailed Implementation

[0051] Figure 1 An example environment 100 for designing a target capsid monomer is shown, which is optimized for predefined characteristics using a capsid optimization system 200.

[0052] Initial capsid monomer data 102 defines an initial set of capsid monomers, wherein each capsid monomer is represented by a corresponding amino acid sequence. Initial capsid monomer data 102 further includes data specifying the fitness scores of the corresponding initial set of capsid monomers. The fitness scores of capsid monomers can characterize predefined properties of the corresponding capsid.

[0053] The initial capsid monomer data 102 can be generated in any of a variety of possible ways. For example, some or all of the capsid monomers in the initial capsid monomers defined by the initial capsid monomer data 102 can be mutations of the “original” capsid monomers (e.g., VP1, VP2, or VP3 capsid monomers of AAV). More specifically, each initial capsid monomer can be generated by modifying the identity of the corresponding amino acid at one or more positions in the amino acid sequence of the original capsid monomer. The positions in the amino acid sequence of the original capsid monomer can be selected in any suitable manner for mutation, for example, by random selection or selection according to predefined rules. The identity of the new amino acid replacing the position in the amino acid sequence of the original capsid monomer can be selected in any suitable manner, for example, by random selection from a probability distribution of possible sets of amino acids. The corresponding amino acid sequence of each capsid monomer in the initial capsid monomer data can differ from the amino acid sequence of the original capsid monomer at any suitable number of positions, for example, 1 position, 3 positions, or 10 positions. The fitness score of the initial capsid monomers can be measured experimentally or predicted by a prediction system. In some implementations, the initial capsid monomer can be selected from capsid monomers based on the measured or predicted fitness score of the capsid monomer.

[0054] The capsid optimization system 200 is configured to generate target capsid monomer sequences 106 optimized for predefined properties.

[0055] In some implementations, system 200 generates a target capsid monomer sequence 106 optimized for the manufacturability of the corresponding capsid. Capsid manufacturability characterizes the rate at which a capsid can be generated when a plasmid encoding the capsid is transfected into one or more cells. (More specifically, capsid manufacturability may be based at least in part on the number of capsid instances generated when a plasmid encoding the capsid is transfected into one or more cells). That is, capsid manufacturability characterizes the efficiency of capsid generation. Capsid manufacturability indirectly characterizes various structural and biochemical properties of the capsid that affect its manufacture, such as the ease with which capsid monomers assemble into the three-dimensional structure of the capsid, the stability of the capsid, etc.

[0056] In some implementations, system 200 may generate a target capsid monomer sequence 106 optimized for the ability of a virus containing a corresponding capsid to avoid neutralization in an organism. Virus neutralization refers to, for example, the process of weakening the ability of a virus to infect cells by binding an antibody to an epitope on the surface of the virus.

[0057] In some implementations, system 200 may generate a target capsid monomer sequence 106 optimized for the immunoreactivity of the corresponding capsid. Capsid immunoreactivity refers to a measure of the immune response elicited by the capsid.

[0058] In some implementations, system 200 may generate a target capsid monomer sequence 106 optimized for the ability of a virus containing the capsid to penetrate target tissues in an organism. Target tissues may be, for example, tissues corresponding to specific organs, such as liver tissue, brain tissue, eye tissue, etc.

[0059] In some implementations, system 200 may generate a target capsid monomer sequence 106 optimized for the packing capacity of the corresponding capsid. The packing capacity of the capsid is a measure of the amount of genetic material (e.g., deoxyribonucleic acid (DNA) or ribonucleic acid (RNA)) that the capsid can hold.

[0060] In some implementations, system 200 may generate a target capsid monomer sequence 106 optimized for the ability of a virus containing a corresponding capsid to integrate into the host genome in a site-specific manner.

[0061] Then, for each target capsid monomer 106, a virus comprising a capsid corresponding to the target capsid monomer 106 can be manufactured 108 (i.e., physically generated) using appropriate manufacturing techniques. In some embodiments, the capsid is expressed in cells. For example, a transfer plasmid carrying the target gene (GOI) can be co-transfected into packaging cells (e.g., HEK293T) with a Rep-cap plasmid (including the capsid sequence) and an auxiliary plasmid encoding adenovirus genes (e.g., E4, E2A, and VA) that mediate AAV replication. After a short period of culture, viral particles can be harvested from cell lysates or supernatants and concentrated, for example, by PEG precipitation. The viral particles can be further purified and concentrated, for example, by cesium chloride (CsCl) gradient ultracentrifugation. The generated virus can be used in any of a variety of applications; for example, the generated virus can be used as a therapeutic agent 110 applied to subject 112 to achieve a therapeutic effect on the subject. For example, the generated virus can be used as a viral vector for gene therapy, as described above. The generated virus can be applied to any suitable subjects, such as mice, cats, dogs, pigs, or humans, to achieve any appropriate therapeutic effect, such as treating diseases like hemophilia, leukemia, melanoma, etc.

[0062] Figure 2 The operation of the shell optimization system 200 is illustrated. System 200 is an example of a system implemented as a computer program on one or more computers at one or more locations, in which the systems, components and techniques described below can be implemented.

[0063] System 200 maintains candidate monomer data 210 characterizing a population of one or more candidate capsid monomers. Specifically, for each candidate capsid monomer in the population, candidate monomer data 210 includes: (i) data defining the corresponding amino acid sequence of the corresponding candidate capsid monomer, and (ii) data specifying a corresponding fitness score characterizing a predefined property of the corresponding capsid containing the corresponding candidate capsid monomer.

[0064] System 200 uses initial capsid data 102 to initialize a population of candidate capsids, which defines an initial set of capsids and specifies the corresponding fitness score for that initial set of capsids.

[0065] System 200 uses a first machine learning model 220 to generate data representing a set of query capsid monomers 230. System 200 uses a second machine learning model 250 to process data representing each of these query capsid monomers to generate a corresponding output specifying a corresponding baseline fitness score 260 for the respective query capsid monomer. Training engine 270 is configured to update the current values ​​of the model parameters of the first machine learning model using the baseline fitness scores of the query capsid monomers. System 200 can update candidate capsid monomer data 210 by adding one or more of the query capsid monomers 230 to a population of candidate capsid monomers. The above process can be repeated for multiple iterations until a stopping condition is met, for example, until a predefined number of iterations have been performed, or until the fitness scores of one or more candidate capsid monomers have reached a threshold fitness score. In some implementations, system 200 can perform the exploration in two different ways: (1) a broader search, and (2) a deeper search. For a broader search, system 200 can perform a broad mutational exploration starting from the initial sequence in an outer search loop. For deeper searches, System 200 can perform more mutation iterations within its internal search loop, but starts with a more limited set of seed sequences. This dual approach enhances the system's adaptability, enabling it to expand its search scope or focus more deeply on specific sequence optimizations.

[0066] After performing multiple iterations, system 200 can (e.g., based on corresponding fitness scores) select one or more target capsids from the population of candidate capsids. System 200 can output data specifying the selected capsid.

[0067] In some implementations, the first machine learning model 220 may be a generative model configured to generate a query capsid sequence from candidate capsid sequences, such as a variational autoencoder (VAE).

[0068] In other embodiments, the first machine learning model 220 may be an inference model configured to infer characteristics of an input capsid monomer sequence. The system 200 may use the inferred capsid monomer characteristics to guide the generation of a query capsid monomer sequence. Specifically, the inference model 220 may infer an adjusted fitness score for the input capsid monomer. In a particular example, the adjusted fitness score measures a combination of: (i) a predicted fitness characteristic of a predefined feature of the input capsid monomer, and (ii) the mutation distance between the input capsid monomer and its corresponding wild-type capsid monomer from which the input capsid monomer mutates.

[0069] The inference model 220 can employ any suitable architecture. In a particular model, the inference model 220 includes: (i) an encoder network configured to process data of corresponding input capsid fragments adjacent to each mutation of an input capsid from a corresponding wild-type capsid to generate corresponding feature vectors; (ii) a pooling network layer configured to combine the corresponding feature vectors generated from the input capsid fragments into a combined feature vector; and (iii) a decoder network configured to process the combined feature vector to generate a predicted score for the input capsid. Examples of inference model 220 can be found in the following literature: Ren et al., “ProximalExploration for Model-guided Protein Sequence Design”, Proceedings of the 39th International Conference on Machine Learning, in Proceedings of MachineLearning Research 162:18520-18536, 2022; and Sinai et al., “Generative AAV capsid diversification by latent interpolation”, bioRxiv, doi: 10.1101 / 2021.04.16.440236, April 2021.

[0070] Reference Figure 5 Describe an example procedure for generating a query capsid sequence.

[0071] Figure 3An example of a second machine learning model 250 as a fitness prediction model is shown. Fitness prediction model 250 is configured to process data defining the amino acid sequence of a query capsid monomer 230 to generate a baseline fitness score 260 for the query capsid monomer 230. The baseline fitness score 260 characterizes the predicted quality of the capsid corresponding to the query capsid monomer 230. For example, as referenced above... Figure 1 As described, the baseline fitness score 260 can characterize the predictive manufacturability of the capsid, the predictive ability of a virus containing the capsid to avoid neutralization, the predictive immune reactivity of the capsid, the predictive ability of a virus containing the capsid to penetrate target tissues, the predictive packaging ability of the capsid, or the predictive ability of a virus containing the capsid to integrate into the host genome.

[0072] The fitness prediction model 250 includes an embedded neural network 251, a fitness prediction machine learning model 253, and an error prediction machine learning model 254, which will be described in more detail below.

[0073] Embedded neural network 251 is configured to process data defining the amino acid sequence of query capsid monomer 230 to generate embedding 252 of query capsid monomer 230.

[0074] The embedded neural network 251 can be configured to process any suitable representation of the amino acid sequence of the query capsid monomer 230. For example, the embedded neural network 251 can process the representation of the amino acid sequence of the query capsid monomer 230 into an amino acid embedding sequence. The amino acid embedding at each position in the amino acid embedding sequence can represent the identity of the corresponding amino group in the amino acid sequence of the capsid monomer. The amino acid embedding representing the amino acid can be, for example, a one-hot embedding, or any other suitable predefined embedding.

[0075] The embedded neural network 251 may have any suitable neural network architecture that enables it to perform the functions described herein. In particular, the embedded neural network 251 may include any suitable number (e.g., 5 layers, 10 layers, 50 layers, etc.) of any suitable type of neural network layers (e.g., fully connected layers, convolutional layers, recurrent layers, attention layers, etc.) connected in any suitable configuration (e.g., as a linear sequence of layers).

[0076] In some embodiments, the embedded neural network 251 may be a recurrent neural network configured to sequentially process each amino acid embedding in the amino acid sequence representing the query capsid monomer 230.

[0077] The system can train the embedded neural network 251 on a set of training examples, where each training example corresponds to a specific protein and defines the amino acid sequence of that protein. Specifically, the embedded neural network 251 can be trained on the set of training examples using self-supervised learning techniques (i.e., training techniques that do not rely on labeling training examples with fitness scores). For example, self-supervised learning may include: (i) training the embedded neural network 251 on a set of training capsid individuals to perform a token demasking task, and / or (ii) training the embedded neural network on a set of training capsid individuals to perform a next token prediction task.

[0078] Training an embedded neural network 251 using self-supervised learning techniques, for example to perform tasks such as token demasking, next token prediction, or both, can enable the embedded neural network 251 to learn to generate embeddings that densely encode rich information characterizing proteins (e.g., capsid monomers).

[0079] The fitness prediction machine learning model 253 is configured to process the embedding 252 of the query garment singleton 230 to generate a baseline fitness score 260 that characterizes the predicted quality of the garment corresponding to the query garment singleton 230.

[0080] The fitness prediction machine learning model 253 can have any suitable machine learning model architecture that enables it to perform the functions described herein. For example, the fitness prediction machine learning model can be a parametric machine learning model, such as a neural network model, a random forest model, or a support vector regression model. As another example, the fitness prediction machine learning model can be a non-parametric model, such as a k-nearest neighbor model.

[0081] The system can train a fitness prediction machine learning model on a training example set using supervised learning. Each training example corresponds to a specific capsid monomer, and the following are defined: (i) the embedding of the amino acid sequence of the capsid monomer generated using an embedded neural network; and (ii) the target fitness score associated with the capsid monomer. The target fitness score of the capsid monomer represents the prediction target of the fitness prediction system, i.e., the "true" fitness that the fitness prediction system should generate for the capsid monomer.

[0082] The fitness prediction machine learning model 253 can be trained using a target fitness score generated in any suitable manner (such as next-generation sequencing reads from a physical experiment). For example, the fitness score of a capsid monomer can describe the manufacturability of a given capsid. To generate a target fitness score, a plasmid corresponding to a given capsid can be transfected into one or more cells. Experimental measurements from next-generation sequencing assays can determine the number of instances of a given capsid or wild-type capsid that can be generated from one or more cells. The fitness score can be defined as the ratio of the number of instances of a given capsid generated to the number of instances of wild-type capsids.

[0083] Error prediction machine learning model 254 is configured to process the representation of the amino acid sequence of the query capsid monomer 230 to generate fitness error 255. Fitness error 255 is an estimate of the error in fitness score 256 generated by fitness prediction machine learning model 253.

[0084] Fitness prediction model 250 updates the fitness score 256 generated by fitness prediction machine learning model 208 by combining (i) the fitness score 256 and (ii) the fitness error 255. Fitness prediction model 250 can combine the fitness score 256 and the fitness error 255, for example, by summing the fitness score 256 and the fitness error 255. Fitness prediction model 250 can then output the updated fitness score 256 as the baseline fitness score for the query garment entity 230.

[0085] Figure 4 This is a flowchart illustrating an example process 400 for optimizing a coat for predefined characteristics. For convenience, process 400 will be described as being executed by a system consisting of one or more computers located at one or more locations. For example, a coat optimization system appropriately programmed according to this disclosure (e.g., Figure 2 The shell optimization system 200) can execute process 400.

[0086] The system maintains candidate monomer data characterizing a population of one or more candidate capsid monomers. For each candidate capsid monomer in the population, the candidate monomer data includes: (i) data defining the corresponding amino acid sequence of the corresponding candidate capsid monomer, and (ii) data specifying a corresponding fitness score characterizing a predefined property of the capsid containing the corresponding candidate capsid monomer.

[0087] At 410, the system initializes a population of candidate capsids using (i) one or more initial capsid sequence and (ii) the corresponding fitness score for each initial capsid sequence. At 420, the system initializes the current values ​​of the model parameters of the first machine learning model.

[0088] The system performs operations 430 to 460 for multiple iterations to update (i) the candidate individual data and (ii) the current values ​​of the model parameters of the first machine learning model.

[0089] At 430, the system uses a first machine learning model to generate data representing the set of queried capsid monomers based on the current values ​​of the model parameters and the corresponding amino acid sequences of one or more candidate capsid monomers in the population.

[0090] At 440, the system uses a second machine learning model to process the corresponding data representing each of the query capsid monomers to generate the corresponding output for the corresponding baseline fitness score of the specified query capsid monomer.

[0091] At position 450, the system uses the baseline fitness score of the query capsid monomer to update the current values ​​of the model parameters of the first machine learning model.

[0092] At 460, the system adds one or more of the query capsid monomers to the population of candidate capsid monomers and updates the candidate monomer data by including the following for each query capsid monomer in the query capsid monomer set: (i) data defining the corresponding amino acid sequence of the corresponding query capsid monomer, and (ii) data specifying the corresponding baseline fitness score of the corresponding query capsid monomer.

[0093] After performing multiple iterations, at position 470, the system selects one or more target capsid monomers from the population of candidate capsid monomers.

[0094] At 480, the system outputs data for the selected target capsid monomer.

[0095] Figure 5 This is a flowchart illustrating an example of the process 430 used to generate a query capsid monomer sequence.

[0096] At 431, the system generates data representing a set of mutant capsid monomers that mutate from one or more candidate capsid monomers among the candidate capsid monomers.

[0097] At position 432, the system initializes the query capsid set by selecting from the mutant capsid set. The system can use any suitable strategy to select mutant capsids, for example, selecting those with properties that satisfy predefined conditions.

[0098] The system uses a first machine learning model to update the query capsid individual set, for example by performing operations 433 to 436 over multiple iterations.

[0099] At 433, the system uses a first machine learning model to predict the adjusted fitness score of a subset of query capsids. The adjusted fitness score of a query capsid can measure a combination of: (i) the predicted fitness of a predefined characteristic of the input capsid, and (ii) the mutation distance between the input capsid and its corresponding wild-type capsid.

[0100] At position 434, the system selects one or more seed capsids from a subset of query capsids based on the predicted adjusted fitness score. For example, the system can select the query capsid with the highest adjusted fitness score as the seed capsid.

[0101] At position 435, the system generates data representing a set of additional mutant capsid monomers that mutate from the selected seed capsid monomer.

[0102] At 436, the system updates the query capsid monomer set by including one or more mutant capsid monomers from that additional set.

[0103] The process 433 to 436 can be repeated for multiple iterations until a stopping condition is met, such as until a predefined number of iterations have been performed, or until a predefined number of query capsid monomer sequences have been generated. Multiple iterations of 433 to 436 represent a proximal exploration process, in which the query capsid monomer is explored in the direction of the higher adjusted fitness score of the lower-order mutants of the capsid monomer sequence. The adjusted fitness score takes into account both the fitness of the query capsid monomer and the mutation distance, which has proven to be an effective and efficient strategy for exploring the capsid monomer space.

[0104] Figure 6 A graphical comparison is provided between capsid quality achieved through the capsid optimization system 200 and capsid quality achieved through random mutations of candidate capsid monomer sequences. Specifically, Figure 6 The fitness scores of capillary manufacturability were compared between randomly generated capillary monomers and capillary monomers optimized using System 200, and it was shown that System 200 statistically significantly improved the quality of optimized capillary monomers.

[0105] Figure 7This is a block diagram of an example computer system 700 that can be used to perform the operations described above. System 700 includes a processor 710, a memory 720, a storage device 730, and an input / output device 740. Each of components 710, 720, 730, and 740 can be interconnected, for example, using a system bus 750. Processor 710 is capable of processing instructions for execution within system 700. In one embodiment, processor 710 is a single-threaded processor. In another embodiment, processor 710 is a multi-threaded processor. Processor 710 is capable of processing instructions stored in memory 720 or on storage device 730.

[0106] The memory 720 stores information within the system 700. In one embodiment, the memory 720 is a computer-readable medium. In one embodiment, the memory 720 is a volatile memory cell. In another embodiment, the memory 720 is a non-volatile memory cell.

[0107] Storage device 730 provides high-capacity storage for system 700. In one embodiment, storage device 730 is a computer-readable medium. In various other embodiments, storage device 730 may include, for example, a hard disk drive, an optical disk drive, a storage device shared by multiple computing devices over a network (e.g., a cloud storage device), or some other high-capacity storage device.

[0108] Input / output device 740 provides input / output operations for system 700. In one embodiment, input / output device 740 may include one or more network interface devices (e.g., Ethernet cards), serial communication devices (e.g., RS-232 ports), and / or wireless interface devices (e.g., 502.11 cards). In another embodiment, input / output device may include a driver device configured to receive data and send output data to other input / output devices (e.g., keyboards, printers, and display device 760). However, other embodiments may also be used, such as mobile computing devices, mobile communication devices, set-top box television client devices, etc.

[0109] Despite Figure 7 Example processing systems have been described in this disclosure, but implementations of the subjects and functional operations described herein may be carried out in other types of digital electronic circuits, or in computer software, firmware, or hardware (including the structures disclosed herein and their structural equivalents), or a combination thereof.

[0110] This disclosure uses the term "configuration" when referring to system and computer program components. For a system consisting of one or more computers to be configured to perform specific operations or actions, this means that the system has software, firmware, hardware, or combinations thereof installed thereon that cause the system to perform those operations or actions when run. For one or more computer programs to be configured to perform specific operations or actions, this means that the one or more programs include instructions that, when executed by a data processing device, cause that device to perform those operations or actions. Embodiments of the subject matter and functional operations described in this disclosure may be implemented in digital electronic circuits, tangibly embodied computer software or firmware, computer hardware (including the structures disclosed in this disclosure and their structural equivalents), or combinations thereof. Embodiments of the subject matter described in this disclosure may be implemented as one or more computer programs (i.e., one or more computer program instruction modules encoded on a tangible non-transitory storage medium) for execution by a data processing device or for controlling the operation of a data processing device. The computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or combinations thereof. Alternatively or additionally, program instructions may be encoded on artificially generated propagation signals (e.g., machine-generated electrical, optical, or electromagnetic signals) that are generated to encode information for transmission to a suitable receiver device for execution by a data processing device.

[0111] The term "data processing apparatus" refers to data processing hardware and encompasses all kinds of devices, apparatuses, and machines for processing data, including, for example, programmable processors, computers, or multiple processors or computers. The apparatus may also be or further include special-purpose logic circuitry, such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits). In addition to hardware, the apparatus may optionally include code that creates an execution environment for computer programs, such as code constituting processor firmware, protocol stacks, database management systems, operating systems, or combinations thereof.

[0112] Computer programs (which may also be referred to or described as programs, software, software applications, applications, modules, software modules, scripts, or code) can be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages; and they can be deployed in any form, including as standalone programs or as modules, components, subroutines, or other units suitable for use in a computing environment. A program may, but does not necessarily, correspond to a file in a file system. A program may be stored in a portion of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or portions of code). A computer program can be deployed to execute on a single computer, or on multiple computers located in one location or distributed across multiple locations and interconnected via a data communication network.

[0113] In this disclosure, the term "database" is used broadly to refer to any collection of data: the data does not need to be structured in any particular way or at all, and the data can be stored on storage devices in one or more locations. Therefore, for example, an index database may include multiple collections of data, each of which may be organized and accessed in different ways.

[0114] Similarly, in this disclosure, the term "engine" is used broadly to refer to a software-based system, subsystem, or process programmed to perform one or more specific functions. Typically, an engine will be implemented as one or more software modules or components installed on one or more computers located at one or more sites. In some cases, one or more computers will be dedicated to a particular engine; in others, multiple engines may be installed on or run on the same or more computers.

[0115] The processes and logical flows described in this disclosure can be executed by one or more programmable computers, which execute one or more computer programs to perform functions by manipulating input data and generating output. These processes and logical flows can also be executed by special-purpose logic circuitry (e.g., FPGA or ASIC), or by a combination of special-purpose logic circuitry and one or more programmable computers.

[0116] A computer suitable for executing computer programs can be based on a general-purpose microprocessor or a special-purpose microprocessor or both, or on any other type of central processing unit (CPU). Typically, the CPU receives instructions and data from read-only memory or random access memory or both. The basic components of a computer are the CPU for executing instructions and one or more memory devices for storing instructions and data. The CPU and memory may be supplemented by or incorporated into special-purpose logic circuitry. Typically, a computer will also include one or more mass storage devices (e.g., disks, magneto-optical disks, or optical disks) for storing data, or operatively coupled to receive data from or transfer data to, or both. However, a computer does not necessarily need to have such a device. Furthermore, a computer can be embedded in another device, such as a mobile phone, personal digital assistant (PDA), mobile audio or video player, game console, GPS receiver, or portable storage device (e.g., a Universal Serial Bus (USB) flash drive).

[0117] Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media and memory devices, including, for example, semiconductor memory devices such as EPROM, EEPROM and flash memory devices; magnetic disks such as internal hard disks or removable disks; magneto-optical disks; and CD ROMs and DVD-ROMs.

[0118] To provide interaction with the user, embodiments of the subject matter described in this disclosure can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including auditory, voice, or tactile input. Additionally, the computer can interact with the user by sending documents to and receiving documents from the device used by the user; for example, by sending a webpage to a web browser on the user's device in response to a request received from a web browser. Furthermore, the computer can interact with the user by sending text messages or other forms of messages to a personal device (e.g., a smartphone running a messaging application) and receiving response messages returned by the user.

[0119] The data processing apparatus used to implement machine learning models may also include, for example, dedicated hardware accelerator units for handling the computationally intensive portions common in machine learning training or production (i.e., inference) workloads.

[0120] Machine learning models can be implemented and deployed using machine learning frameworks such as PyTorch or TensorFlow.

[0121] Embodiments of the subjects described in this disclosure can be implemented in a computing system that includes back-end components (e.g., as a data server), middleware components (e.g., an application server), front-end components (e.g., a client computer having a graphical user interface, web browser, or application through which a user can interact with embodiments of the subjects described in this disclosure), or any combination of one or more such back-end components, middleware, or front-end components. Components of the system can be interconnected via any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks (LANs) and wide area networks (WANs) (e.g., the Internet).

[0122] A computing system may include clients and servers. Clients and servers are typically geographically separated and usually interact via a communication network. The client-server relationship is established by computer programs running on respective computers and having a client-server relationship with each other. In some embodiments, the server transmits data (e.g., HTML pages) to a user device acting as a client, for example, to display data to a user interacting with that device and to receive user input from that user. Data generated at the user device (e.g., the result of user interaction) can be received from that device at the server.

[0123] While this disclosure contains numerous details of specific implementations, these details should not be construed as limiting the scope of any invention or the scope of what may be claimed, but rather as a description of features that may be specific to a particular embodiment of the invention. Certain features described in this disclosure, in the context of a single embodiment, may also be implemented in combination in a single embodiment. Conversely, different features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments. Furthermore, although features may be described above as functioning in a certain combination and even initially claimed in this way, in some cases, one or more features from the claimed combination may be removed from that combination, and the claimed combination may involve sub-combinations or variations thereof.

[0124] Similarly, although operations are depicted in a specific order in the accompanying drawings and are also recited in a specific order in the claims, this should not be construed as requiring such operations to be performed in the specific order shown or sequentially, or requiring the performance of all illustrated operations to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Furthermore, the separation of various system modules and components in the embodiments described above should not be construed as requiring such separation in all embodiments, and it should be understood that the described program components and systems can generally be integrated together in a single software product or packaged into multiple software products.

Claims

1. A computer-implemented method, comprising: Maintain candidate monomer data for a population characterizing one or more candidate capsid monomers, wherein, for each candidate capsid monomer in the population, the candidate monomer data includes: (i) data defining the corresponding amino acid sequence of the corresponding candidate capsid monomer, and (ii) data specifying a corresponding fitness score characterizing a predefined property of the capsid containing the corresponding candidate capsid monomer; The population of the candidate capsid is initialized using (i) one or more initial capsid monomer sequences and (ii) the corresponding fitness score for each of the respective initial capsid monomer sequences; Initialize the current values ​​of the model parameters for the first machine learning model; At each iteration in the multiple iterations, update (i) the current values ​​of the candidate individual data and (ii) the model parameters of the first machine learning model. This update includes: Using this first machine learning model, data representing the query capsid set is generated based on the current values ​​of these model parameters and the corresponding amino acid sequences of one or more candidate capsid monomers from the population. A second machine learning model is used to process the corresponding data representing each of these query capsids to generate a corresponding output with a corresponding baseline fitness score for that query capsid. Use the baseline fitness scores of these query capsid units to update the current values ​​of these model parameters of the first machine learning model; Add one or more of these query capsid monomers to the population of candidate capsid monomers, and update the candidate monomer data by including the following for each query capsid monomer in the set of query capsid monomers: (i) data defining the corresponding amino acid sequence of the corresponding query capsid monomer, and (ii) data specifying the corresponding baseline fitness score of the corresponding query capsid monomer; After performing these multiple iterations, one or more target capsid monomers are selected from the population of candidate capsid monomers; and Outputs data for the selected target capsid monomer.

2. The method as described in claim 1, wherein, Selecting one or more target capsid monomers involves choosing the target capsid monomer with the highest fitness score from the population of candidate capsid monomers.

3. The method as claimed in claim 1 or claim 2, wherein, This predefined property characterizes the manufacturability of the cap.

4. The method of claim 3, wherein, The manufacturability of the capsid is defined based on the number of capsid instances generated when a plasmid encoding the capsid is transfected into one or more cells.

5. The method of claim 4, wherein, The number of capsid instances generated when the plasmid encoding the capsid is transfected into one or more cells is normalized by the number of wild-type sequences generated.

6. The method as described in any of the preceding claims, wherein, This predefined characteristic characterizes the ability of a virus containing this capsid to avoid being neutralized.

7. The method as described in any of the preceding claims, wherein, This predefined property characterizes the immunoreactivity of the capsid.

8. The method as described in any of the preceding claims, wherein, The predefined characteristics of the capsid characterize the predictive ability of the virus containing the capsid to penetrate target tissues.

9. The method as described in any of the preceding claims, wherein, The predicted fitness score characterizes the packing capacity of the cap.

10. The method as claimed in any of the preceding claims, wherein, The predicted fitness score characterizes the ability of a virus containing the capsid to predict integration into the host genome.

11. The method as claimed in any of the preceding claims, wherein, This capsid corresponds to the capsid of adeno-associated virus (AAV).

12. The method of any of the preceding claims, further comprising: One or more viruses are physically generated, each containing a protein having one of the selected target capsid monomers.

13. The method of claim 12, further comprising: The generated virus is applied to the subject to achieve a therapeutic effect.

14. The method as claimed in any of the preceding claims, wherein, The second machine learning model includes: An embedded neural network configured to process inputs defining the amino acid sequence of the corresponding query capsid monomer to generate a corresponding embedding of that amino acid sequence; and A fitness prediction machine learning model is configured to process the corresponding embedding of the amino acid sequence to generate an output representing the corresponding predicted fitness score of the corresponding query capsid monomer.

15. The method of claim 14, wherein, The second machine learning model further includes an error prediction machine learning model configured to process inputs representing the amino acid sequence of the corresponding query capsid monomer to generate a corresponding fitness error, wherein the corresponding fitness error is an estimate of the error of the corresponding predicted fitness score generated by the fitness prediction machine learning model.

16. The method of claim 15, wherein, The corresponding baseline fitness score is calculated by combining the following: (i) the corresponding predicted fitness score generated by the fitness prediction machine learning model, and (ii) the corresponding fitness error generated by the error prediction machine learning model.

17. The method of claim 15 or claim 16, wherein, The fitness prediction machine learning model has a different model architecture than the error prediction machine learning model.

18. The method according to any one of claims 14 to 17, wherein, The embedded neural network has been trained on a training capsid monomer set using self-supervised learning.

19. The method of claim 18, wherein, The self-supervised learning involves training the embedded neural network on the training capsid set to perform a token demasking task.

20. The method of claim 18 or claim 19, wherein, The self-supervised learning involves training the embedded neural network on the training capsid monomer set to perform the next token prediction task.

21. The method according to any one of claims 14 to 20, wherein, The fitness prediction machine learning model has been trained on a set of training examples, where each training example corresponds to a corresponding training captive and includes: (i) the embedding of the training captive, and (ii) the target fitness score of the training captive.

22. The method as described in any of the preceding claims, wherein, The data generated to represent the collection of shell entities for this query includes: Generate data representing a set of mutant capsid monomers that mutate from one or more of these candidate capsid monomers; The query capsid monomer set is initialized by selecting from the mutant capsid set; and Update the query shell monomer set using the first machine learning model.

23. The method of claim 22, wherein, The first machine learning model is configured to process a first input representing the input capsid monomer to predict the adjusted fitness score of that input capsid monomer.

24. The method of claim 23, wherein, The first input further characterizes the mutation of the input capsid monomer relative to one or more of its corresponding wild-type capsid monomers.

25. The method of claim 24, wherein, The adjusted fitness score measures a combination of the following: (i) the predicted fitness of a predefined property of the input capsid monomer, and (ii) the mutation distance between the input capsid monomer and the corresponding wild-type capsid monomer.

26. The method of claim 25, wherein, The first machine learning model includes: An encoder network is configured to process data of a corresponding input capsid fragment adjacent to each mutation of the input capsid from the corresponding wild-type capsid to generate a corresponding feature vector. A pooling network layer configured to combine the corresponding feature vectors generated from these input capsid monomer fragments into a combined feature vector; and A decoder network is configured to process the combined feature vector to generate a predicted score for the input capsid unit.

27. The method of claim 25 or claim 26, wherein, Updating the query shell monomer set using the first machine learning model includes: In each iteration of a series of iterations, perform the following: The first machine learning model is used to predict the adjusted fitness score of a subset of query capsids; One or more seed capsids are selected from a subset of the queried capsids based on the predicted adjusted fitness scores. Generate data representing an additional set of mutant capsid monomers derived from the selected seed capsid monomers; and The queried capsid set is updated by including one or more mutant capsids from the additional mutant capsid set.

28. A system comprising: One or more computers; as well as One or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform the operations claimed in any one of claims 1 to 27.

29. One or more non-transitory computer storage media storing instructions that, when executed by one or more computers, cause the one or more computers to perform the operations claimed in any one of claims 1 to 27.

Images