Non-natural amino acid production methods, devices, apparatuses, media and products

Abstract

The embodiment of the invention discloses a non-natural amino acid generation method, device and equipment, a medium and a product, and the method comprises the following steps: under the condition that a to-be-modified amino acid molecule is obtained, segmenting the to-be-modified amino acid molecule to obtain a molecular skeleton and a modified fragment; inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain an amino acid modified fragment, and obtaining generated amino acid molecules based on the molecular skeleton and the amino acid modified fragment; and scoring the preset target pharmacochemical properties of the generated amino acid molecules to obtain a scoring result, and ending the generation process of the non-natural amino acid molecules according to the scoring result. The technical scheme provided by the embodiment of the invention solves the problems of small generation quantity and low quality of amino acids generated by modifying natural amino acids at present, can automatically generate non-natural amino acids according with target pharmacochemical properties through the pre-trained model, has the advantages of large generation quantity and high generation speed, and provides a basis for development of polypeptide drugs.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of artificial intelligence technology, and in particular, to a method, device, equipment, medium and product for generating unnatural amino acids. Background Art

[0002] Currently, the generation of non-natural amino acids is generally based on traditional drug development methods, which are produced by modifying natural amino acids. Amino acid modification, as an important means of peptide modification, can effectively change the physical and chemical properties of peptide compounds, increase water solubility, prolong the duration of action in the body, change its biodistribution, eliminate immunogenicity, reduce toxic side effects, etc., and thus be used in the development of peptide drugs.

[0003] Traditional amino acid modification methods are unable to provide a large number of novel or diverse molecules to support the screening of lead compounds. Secondly, it is impossible to quickly optimize candidate compounds that meet the target pharmaceutical properties. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, equipment, medium and product for generating non-natural amino acids, which can automatically generate non-natural amino acids that meet the target pharmaceutical properties through a pre-trained amino acid molecule generation model, generate a large number of non-natural amino acids at a fast speed, and provide a basis for the development of polypeptide drugs.

[0005] In a first aspect, an embodiment of the present invention provides a method for producing an unnatural amino acid, the method comprising:

[0006] When the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton and a modified fragment;

[0007] Inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generating amino acid molecules based on the molecular skeleton and the amino acid modification fragments;

[0008] The preset target pharmaceutical properties of the generated amino acid molecules are scored to obtain a scoring result, and the non-natural amino acid molecule generation process is terminated based on the scoring result.

[0009] In a second aspect, an embodiment of the present invention provides a device for producing non-natural amino acids, the device comprising:

[0010] An amino acid molecule segmentation module is used to segment the amino acid molecule to be modified to obtain a molecular skeleton and a modified fragment after obtaining the amino acid molecule to be modified;

[0011] An amino acid molecule generation module is used to input a molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generate amino acid molecules based on the molecular skeleton and the amino acid modification fragments;

[0012] The amino acid molecule scoring module is used to score the preset target pharmaceutical properties of the generated amino acid molecules, obtain the scoring results, and end the non-natural amino acid molecule generation process based on the scoring results.

[0013] In a third aspect, an embodiment of the present invention further provides a computer device, the computer device comprising:

[0014] one or more processors;

[0015] A memory for storing one or more programs;

[0016] When the one or more programs are executed by one or more processors, the one or more processors implement the method for generating non-natural amino acids provided in any embodiment of the present invention.

[0017] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for generating non-natural amino acids as provided in any embodiment of the present invention.

[0018] In a fifth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the method for generating non-natural amino acids as provided in any embodiment of the present invention.

[0019] The embodiments of the above invention have the following advantages or beneficial effects:

[0020] In the embodiment of the present invention, when the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton and a modified fragment; the molecular skeleton is input into a pre-trained amino acid molecule generation model to obtain an amino acid modified fragment, and the generated amino acid molecule is obtained based on the molecular skeleton and the amino acid modified fragment; the preset target pharmaceutical properties of the generated amino acid molecule are scored to obtain a scoring result, and according to the scoring result, the non-natural amino acid molecule generation process is terminated. The technical solution of the embodiment of the present invention solves the problem that the current generation of amino acids through natural amino acid modification has a small number of generated amino acids and low quality. The pre-trained amino acid molecule generation model can be used to automatically generate non-natural amino acids that meet the target pharmaceutical properties, and the number of generated amino acids is large and the generation speed is fast, providing a basis for the development of polypeptide drugs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1is a flow chart of a method for producing unnatural amino acids provided by an embodiment of the present invention;

[0022] Figure 2 is a flow chart of a method for producing unnatural amino acids provided by an embodiment of the present invention;

[0023] Figure 3 is a flow chart of a method for producing unnatural amino acids provided by an embodiment of the present invention;

[0024] Figure 4 is a schematic diagram of a model training process provided by an embodiment of the present invention;

[0025] Figure 5 It is a flowchart of the overall algorithm of a method for generating non-natural amino acids provided by an embodiment of the present invention;

[0026] Figure 6 is a schematic structural diagram of a non-natural amino acid production device provided by an embodiment of the present invention;

[0027] Figure 7 It is a structural schematic diagram of a computer device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0029] Figure 1 A flowchart of a method for generating a non-natural amino acid provided in an embodiment of the present invention is provided. This embodiment can be applied to the scenario of generating non-natural amino acids, especially the scenario of generating non-natural amino acids through a neural network model, so as to provide non-natural amino acids with certain pharmaceutical properties for relevant technical personnel. The method can be performed by a non-natural amino acid generating device, which can be implemented by software and / or hardware and integrated into a computer device with application development function.

[0030] like Figure 1 As shown, the method for producing the non-natural amino acid of this embodiment comprises the following steps:

[0031] S110, when the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton and modified fragments.

[0032] The amino acid molecule to be modified can be a natural amino acid molecule or a non-natural amino acid molecule. A new modification fragment can be added to the backbone structure of the amino acid molecule to be modified to generate a new amino acid molecule.

[0033] The skeleton of an amino acid molecule can be composed of a carbon chain, with some functional groups as modified fragments connected to the skeleton, which is a larger molecule. Cutting the macromolecule can help the amino acid molecule generation model to better analyze its structure and reactivity, so that the introduction of new modified fragments into the generated amino acid molecule can give the molecule new properties and application value.

[0034] To split the amino acid molecules to be modified, several common methods of molecular splitting can be used, such as determining the functional groups in the molecules, such as hydroxyl, carboxyl, amino, etc., based on functional group analysis. Functional groups are often the more active and reactive parts of the molecules, and can be used as the starting point for splitting. For example, the type of chemical bond, the splitting position is selected according to the strength and reactivity of the chemical bond. For example, bonds that are easier to break, such as ester bonds and ether bonds, can be used as splitting sites. And based on chemical reaction conditions, consider the possible cleavage pathways of the molecules under specific reaction conditions. For example, under conditions of high temperature, strong acid, strong base, etc., some chemical bonds are more likely to break.

[0035] By dividing the amino acid molecule to be modified, the molecular skeleton and modified fragments of the amino acid molecule to be modified can be obtained.

[0036] S120, inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generating amino acid molecules based on the molecular skeleton and the amino acid modification fragments.

[0037] Specifically, the molecular skeleton may be represented in the form of SMILES (Simplified Molecular Input Line Entry System). The molecular skeleton is input into the pre-trained amino acid molecule generation model, that is, the SMILES corresponding to the molecular skeleton is input into the pre-trained amino acid molecule generation model.

[0038] The pre-trained amino acid molecule generation model can extract and decode features such as amino acid molecular structure and reactivity of the data input into the model to output a new amino acid modification fragment, and combine the molecular skeleton and the new amino acid modification fragment to generate the amino acid molecule.

[0039] The pre-trained amino acid molecule generation model may be a neural network model, such as a recurrent neural network model (RNN, Recurrent Neural Network).

[0040] S130, scoring the preset target pharmaceutical properties of the generated amino acid molecule to obtain a scoring result, and ending the non-natural amino acid molecule generation process based on the scoring result.

[0041] The pharmacochemical properties of molecules include, but are not limited to, solubility, stability, acid-base properties, and reactivity, which may depend on the modified fragments on the molecular backbone. The pharmacochemical properties of molecules can be evaluated by dimensions such as molecular weight, number of aromatic rings, number of hydrogen donors, and number of hydrogen acceptors.

[0042] Among them, the preset target pharmaceutical property can be a pharmaceutical property that the generated amino acid molecule is expected to have. By analyzing dimensions such as the molecular weight of the generated amino acid molecule, the number of aromatic rings, the number of hydrogen donors and the number of hydrogen acceptors, it can be determined whether the generated amino acid molecule meets the standards of the preset target pharmaceutical property in the corresponding dimension.

[0043] The process of scoring the preset target pharmaceutical properties of the generated amino acid molecules and obtaining the scoring results can be based on the quantitative evaluation methods corresponding to the above-mentioned analysis dimensions to obtain an evaluation result.

[0044] If the evaluation result reaches the corresponding standard, it means that the generated amino acid is an amino acid molecule that meets expectations, and the functionality of the pre-trained amino acid molecule generation model is better.

[0045] If the evaluation result does not meet the corresponding standard, it means that the amino acid molecules generated by the pre-trained amino acid molecule generation model cannot meet the amino acid generation requirements, the model stability needs to be improved, and the model can be further trained and optimized. For example, based on the evaluation results and the policy gradient algorithm of reinforcement learning, the above pre-trained amino acid molecule generation model is updated to obtain a more robust amino acid molecule generation model.

[0046] Furthermore, the qualified generated amino acid molecules can be saved in the result data set, and whether the generated amino acid molecules are natural amino acid molecules can be determined based on the natural amino acid data set. When the generated amino acid molecules are non-natural amino acid molecules, the preset NSAA (Non-standard amino acids, non-standard amino acids) data set classifies the generated amino acid molecules into known non-natural amino acids and new non-natural amino acids.

[0047] The technical solution of this embodiment is to obtain the amino acid molecule to be modified by segmenting the amino acid molecule to be modified to obtain the molecular skeleton and the modified fragment; input the molecular skeleton into the pre-trained amino acid molecule generation model to obtain the amino acid modified fragment, and obtain the generated amino acid molecule based on the molecular skeleton and the amino acid modified fragment; score the preset target pharmaceutical properties of the generated amino acid molecule to obtain the scoring result, and end the non-natural amino acid molecule generation process according to the scoring result. The technical solution of the embodiment of the present invention solves the problem that the current generation of amino acids through natural amino acid modification has a small number of generated amino acids and low quality. The pre-trained amino acid molecule generation model can be used to automatically generate non-natural amino acids that meet the target pharmaceutical properties, with a large number of generated amino acids and a fast generation speed, providing a basis for the development of polypeptide drugs.

[0048] Figure 2 A flowchart of a method for generating a non-natural amino acid provided in an embodiment of the present invention, this embodiment and the method for generating a non-natural amino acid in the above embodiment belong to the same inventive concept, and further describes the process of ending the generation process of the non-natural amino acid molecule according to the scoring result. The method can be performed by a non-natural amino acid generation device, which can be implemented by software and / or hardware, and integrated into a computer device with application development function.

[0049] like Figure 2 As shown, the method for producing the non-natural amino acid of this embodiment comprises the following steps:

[0050] S210, when the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton and modified fragments.

[0051] S220, inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generating amino acid molecules based on the molecular skeleton and the amino acid modification fragments.

[0052] The pre-trained amino acid molecule generation model can be an agent model that has been reinforced with a policy gradient algorithm. An agent model is a model that can make decisions autonomously and take actions to achieve a certain goal.

[0053] In an optional embodiment, before the molecular skeleton is input into the pre-trained amino acid molecule generation model, it is determined whether the molecular skeleton contains both amino groups and carboxyl groups.

[0054] Only molecules that contain both amino and carboxyl groups in their molecular skeletons belong to amino acid molecules, so model reasoning is only performed on molecules that contain both amino and carboxyl groups in their molecular skeletons.

[0055] S230, scoring the preset target pharmaceutical properties of the generated amino acid molecules to obtain a scoring result.

[0056] The preset target pharmaceutical properties may specifically be molecular structural properties. In an optional embodiment, the preset target pharmaceutical properties of the generated amino acid molecules are scored, and at least one of the molecular weight, number of aromatic rings, number of hydrogen donors, and number of hydrogen acceptors of the generated amino acid molecules is scored; wherein each scoring item corresponds to a preset scoring rule.

[0057] The preset scoring rule may be a rule based on the study of molecular structure properties. The preset scoring rule may be in the form of a scoring function. Specifically, the scoring may be performed based on whether the relevant properties of the generated amino acid molecules are within the preset range of the corresponding scoring item properties.

[0058] For example, for the molecular weight scoring item, the generated amino acid molecules within the preset molecular weight range of 100-300 can get a higher score for this item, while the generated amino acid molecules with a molecular weight that is too small, such as less than 100, or too large, such as greater than 300, can get a lower score for this item.

[0059] The scoring result may be the total scoring result of all scoring items. Optionally, the scoring result may also be the total scoring result of weighted calculation of all scoring items. Different weight values ​​may be assigned to different scoring items according to their influence and importance on the target pharmaceutical properties, and the total scoring result may be obtained by weighted calculation according to the weight values.

[0060] S240. When the scoring result meets the preset scoring threshold and the generated amino acid molecule is a non-natural amino acid molecule, the generated amino acid molecule is saved, and the non-natural amino acid molecule generation process is ended.

[0061] The preset scoring threshold can be set according to actual needs, and this embodiment does not limit it. For example, it can be a value between 0 and 1. When the scoring result is greater than the preset scoring threshold, the generated amino acid molecule is a non-natural amino acid molecule, the generated amino acid molecule is saved, and the non-natural amino acid molecule generation process ends.

[0062] It is understandable that, without setting restrictions, the model may repeatedly generate the same or highly similar generated amino acid molecules. To avoid this situation, in an optional embodiment, the generated amino acid molecules are scored for novelty through a diversity filter (DF) to obtain a novelty score result. The novelty score result is summed with the score result of the preset target pharmaceutical property to obtain a total score of the dual property. When the total score of the dual property meets the preset total score threshold, for example, exceeds the preset total score threshold, the generated amino acid molecule is saved and the non-natural amino acid molecule generation process is terminated. The preset total score threshold can be set according to actual needs, for example, 0.3.

[0063] Diversity filter is a technique used in generative models to avoid mode collapse and repeated generation of the same or similar compounds. In this embodiment, only novel generated amino acid molecules are saved to the result data set through diversity filter scoring.

[0064] S250. When the scoring result does not meet the preset scoring threshold, the amino acid molecule generation model is updated, and a non-natural amino acid corresponding to the amino acid molecule to be modified is generated based on the updated amino acid molecule generation model.

[0065] When the scoring result does not meet the preset scoring threshold, the generated amino acid molecule cannot reach the preset scoring threshold and may not have good target pharmaceutical properties. Therefore, it is necessary to update the amino acid molecule generation model, that is, update the model parameters of the pre-trained amino acid molecule generation model.

[0066] For example, the parameters of the Agent model are updated by the policy gradient algorithm, and based on the updated amino acid molecule generation model, i.e., the Agent model, the non-natural amino acid corresponding to the amino acid molecule to be modified is generated, and steps S230 and S240 are performed. When the non-natural amino acid satisfies step S240, the generated amino acid molecule is saved, and the non-natural amino acid molecule generation process is terminated. If not, S250 is continued until S240 is satisfied.

[0067] The technical solution of this embodiment is to obtain the amino acid molecule to be modified by segmenting the amino acid molecule to be modified to obtain the molecular skeleton and the modified fragment; input the molecular skeleton into the pre-trained amino acid molecule generation model to obtain the amino acid modified fragment, and obtain the generated amino acid molecule based on the molecular skeleton and the amino acid modified fragment; score the preset target pharmaceutical properties of the generated amino acid molecule to obtain the scoring result, and when the scoring result meets the preset scoring threshold and the generated amino acid molecule is a non-natural amino acid molecule, save the generated amino acid molecule and end the non-natural amino acid molecule generation process; when the scoring result does not meet the preset scoring threshold, update the amino acid molecule generation model, and generate the non-natural amino acid corresponding to the amino acid molecule to be modified based on the updated amino acid molecule generation model. Score at least one of the scoring items of the molecular weight, the number of aromatic rings, the number of hydrogen donors and the number of hydrogen acceptors of the generated amino acid molecule; wherein each scoring item corresponds to a preset scoring rule. Before inputting the molecular skeleton and the modified fragment into the pre-trained amino acid molecule generation model, it is determined whether the molecular skeleton contains both amino and carboxyl groups. The technical solution of the embodiment of the present invention solves the problem that the current generation of amino acids through modification of natural amino acids has a small number of generated amino acids and low quality. It can automatically generate non-natural amino acids that meet the target pharmaceutical properties through a pre-trained amino acid molecule generation model and preset target pharmaceutical properties for scoring. The non-natural amino acids generated are large in number and fast in speed, providing a basis for the development of polypeptide drugs.

[0068] Figure 3 A flowchart of a method for generating a non-natural amino acid provided in an embodiment of the present invention, this embodiment and the method for generating a non-natural amino acid in the above embodiment belong to the same inventive concept, and further describes the process of training an amino acid molecule generation model. The method can be performed by a non-natural amino acid generation device, which can be implemented by software and / or hardware, and integrated into a computer device with an application development function.

[0069] like Figure 3 As shown, the amino acid molecule generation model training process in the non-natural amino acid generation method of this embodiment includes the following steps:

[0070] S310, performing molecular segmentation on the amino acid molecules in the first preset data set to obtain corresponding molecular skeletons, modified fragments, and original amino acid molecules before segmentation to form a first data set triple.

[0071] For example, the first data set triplet can be represented as {molecular skeleton, modified fragment, original amino acid molecule before segmentation}, and the molecular skeleton, modified fragment and original amino acid molecule before segmentation in the triplet are all represented by SMILES. The complete amino acid molecule, i.e., the original amino acid molecule before segmentation, can be segmented based on a known chemical reaction, for example, the complete amino acid molecule is segmented according to the break position of the chemical bond in the known chemical reaction.

[0072] The first preset data set is a small molecule data set, including amino acid molecules and non-amino acid molecules. The first preset data set can be a ChEMBL data set. The ChEMBL data set is a large, public medicinal chemistry database that contains a wealth of amino acid molecules and is suitable as a data set for preliminary model training.

[0073] S320, using the molecular skeleton in the triplet of the first data set as an input sample, and using the corresponding modified fragments and original amino acid molecules as sample labels to train the initial amino acid molecule generation model to obtain a first amino acid molecule generation model.

[0074] The molecular skeleton in the triplet of the first data set is used as the input sample, and the corresponding modified fragments and original amino acid molecules are used as the expected output of the model, i.e., the sample labels, to determine the initial amino acid molecule generation model, and the initial amino acid molecule generation model is trained through the samples.

[0075] Specifically, the molecular skeleton in the triplet of the first data set may be input into the initial amino acid molecule generation model to obtain a generated modified fragment. Then, the generated modified fragment is combined with the molecular skeleton input into the initial amino acid molecule generation model to obtain the generated amino acid molecule. Furthermore, the generated amino acid molecule, the generated modified fragment, the corresponding modified fragment and the original amino acid molecule may be compared to calculate the learning loss of the model to update the parameters of the initial amino acid molecule generation model.

[0076] In an optional embodiment, the molecular skeleton and modified fragments in the triplet of the first data set can be used as input samples, and the corresponding original amino acid molecules can be used as the expected output of the model, i.e., the sample labels, to train an amino acid molecule generation model to generate generated amino acid molecules.

[0077] refer to Figure 4The initial amino acid molecule generation model can be an RNN model composed of LSTM (Long Short-Term Memory) and Attention (attention mechanism), and the RNN model includes a preset number of encoders and decoders. For example, the ChEMBL dataset is used to train the RNN model. The starting learning rate is set to 0.0001, the minimum learning rate is set to 1e-06, and the Adam optimizer is used to automatically adjust the learning rate. When updating parameters, the Adam optimizer not only considers the current gradient, but also the historical information of the gradient. It can automatically adjust the learning rate during the training process, reduce the learning rate for frequently updated parameters, and increase the learning rate for infrequently updated parameters, thereby accelerating the convergence speed and improving the training effect.

[0078] Finally, torch.nn.NLLLOSS is used as the objective function, and the parameter values ​​of the RNN model are updated according to the objective function and the back propagation algorithm. torch.nn.NLLLoss (Negative Log Likelihood Loss) is used to calculate the negative log likelihood loss and update the parameter values ​​of the RNN model according to the negative log likelihood loss.

[0079] The first amino acid molecule generation model can be understood as an initial model with the function of generating amino acid molecules. In order to improve the model reasoning effect and generate non-natural amino acid molecules, or generate non-natural amino acid molecules with target pharmaceutical properties, the initial model can be further trained.

[0080] S330, performing molecular segmentation on the amino acid molecules in the second preset data set to obtain corresponding molecular skeletons, modified fragments, and original amino acid molecules before segmentation to form a second data set triple.

[0081] The molecules in the second preset data set are non-natural amino acid molecules.

[0082] The amount of data in the second preset data set is smaller than that in the first preset data set, but it only includes non-natural amino acid molecules, and the data is more refined. The second preset data set can be an NSAA (Non-standard amino acids, non-standard amino acids) data set, specifically, an NSAA data set after removing natural amino acids.

[0083] The molecular segmentation and triplet format may refer to the description of step S310 .

[0084] S340. Using the molecular skeleton in the triplet of the second data set as an input sample, and using the corresponding modified fragments and the original amino acid molecules as sample labels to train the first amino acid molecule generation model, to obtain a second amino acid molecule generation model as a pre-trained amino acid molecule generation model.

[0085] The molecular skeleton in the triple of the NSAA data set is used as an input sample, and the corresponding modified fragment and the original amino acid molecule are used as sample labels to train the updated RNN model, i.e., the second amino acid molecule generation model. According to the back propagation algorithm as described in step S310, a smaller initial learning rate, such as 0.00001, is used compared to the training of the first amino acid molecule generation model, and the model parameters, such as the weight value of the model, are further updated to migrate the first amino acid molecule generation model to the NSAA target domain, so as to obtain the second amino acid molecule generation model, i.e., the Prior model.

[0086] In an optional embodiment, before model training, the first data set triples and the description information of the molecular skeleton and modified fragments in the shown first data set triples can also be randomly represented to generate new first data set triples and new second data set triples.

[0087] The randomized representation can be a randomized SMILES format. Specifically, the description information of the molecular skeleton and the modified fragment in the first and second data set triples, i.e., the SMILES string, is randomly transformed to generate different SMILES representations. The random transformation is, for example, a left-right position swap. The amount of data will increase after the randomized representation. The amount of data in the new first data set triple and the new second data set triple is larger than the original first data set triple and the second data set triple. By expanding the data set through randomized representation, the learning space of the model can be expanded and the learning effect of the model can be improved.

[0088] In an optional embodiment, before model training, the amino acid molecules in the first preset data set are filtered according to the amino acid properties of the amino acid molecules in the second preset data set.

[0089] Specifically, the amino acid molecules in the first preset data set that have the same or similar amino acid properties as the amino acid molecules in the second preset data set are retained. Amino acid properties include pharmaceutical properties, molecular structure characteristics, chemical reaction characteristics, etc. Amino acid molecules with similar amino acid properties, for example, higher than a preset similarity threshold. Accordingly, amino acid molecules below the preset similarity threshold are filtered out. Since the amino acid molecules in the second preset data set have better pharmaceutical properties, this embodiment improves the training efficiency of the model by filtering the amino acid molecules based on the amino acid properties, so that the amino acid molecules generated by the model have better pharmaceutical properties.

[0090] The technical solution of this embodiment is to perform molecular segmentation on the amino acid molecules in the first preset data set to obtain the first data set triples consisting of the corresponding molecular skeleton, modified fragments and original amino acid molecules before segmentation; the molecular skeleton in the first data set triples is used as input samples, and the corresponding modified fragments and original amino acid molecules are used as sample labels to train the initial amino acid molecule generation model to obtain the first amino acid molecule generation model; wherein the first preset data set is a small molecule data set, including amino acid molecules and non-amino acid molecules. The amino acid molecules in the second preset data set are molecularly segmented to obtain the corresponding molecular skeleton, modified fragments and the original amino acid molecules before segmentation to form a second data set triple; the molecular skeleton in the second data set triple is used as an input sample, and the corresponding modified fragment and the original amino acid molecule are used as sample labels to train the first amino acid molecule generation model to obtain a second amino acid molecule generation model as a pre-trained amino acid molecule generation model; wherein the molecules in the second preset data set are non-natural amino acid molecules; before model training, the description information of the molecular skeleton and modified fragment in the first data set triple and the first data set triple shown is randomly represented to generate a new first data set triple and a new second data set triple; according to the amino acid properties of the amino acid molecules in the second preset data set, the amino acid molecules in the first preset data set are filtered. The technical solution of the embodiment of the present invention solves the problem that the current generation of amino acids through natural amino acid modification is small in number and low in quality. The amino acid molecule generation model can be trained based on different data sets, the training data is filtered, the random representation is purified, and the data is expanded, and the non-natural amino acids that meet the target pharmacological properties are automatically generated, and the number of generated is large and the generation speed is fast, providing a basis for the development of polypeptide drugs.

[0091] In a specific application example of non-natural amino acid generation, the overall flow chart of the algorithm for non-natural amino acid generation can be referred to Figure 5 The flow chart shown. Figure 5 In the specific process of generating unnatural amino acids, the steps are as follows:

[0092] 1. Construct ChemBL and NSAA datasets.

[0093] Collect as much non-natural amino acid data (NSAA) as possible and download the ChEMBL database (small molecule data set). Filter the ChEMBL data set according to the amino acid properties in the collected non-natural amino acid data set (NSAA).

[0094] 2. Slice the complete molecule based on chemical reactions to generate a {skeleton, modified fragment, original molecule} triple.

[0095] The complete molecule is split based on known chemical reactions to generate a large number of {skeleton, modified fragment, original molecule} triplets.

[0096] 3. Expand the ChemBL dataset and NSAA dataset.

[0097] The ChEMBL and NSAA datasets were augmented with randomized SMILES formulas representing backbone and modified fragments as supervised training sets.

[0098] 4. Build the RNN model.

[0099] Build an RNN model consisting of LSTM and Attention.

[0100] 5. Use the ChemBL dataset to train the RNN model.

[0101] The ChEMBL dataset is used to train the RNN model. The starting learning rate is set to 0.0001, the minimum learning rate is set to 1e-06, and the optimizer is Adam. Finally, torch.nn.NLLLOSS is used as the objective function, and the parameter values ​​of the RNN model are updated according to the objective function and the back propagation algorithm.

[0102] 6. Use the NSAA dataset to train the updated RNN model and obtain the Prior model.

[0103] The updated RNN model is trained using the NSAA dataset as input. The objective function propagation algorithm is continued as described above, with a smaller initial learning rate of 0.00001, and the model weights are further updated to migrate the RNN model to the NSAA target domain to obtain the Prior model.

[0104] 7. Input single amino acids, fragments or {backbone, modified fragment, original molecule} triplets.

[0105] The amino acid molecule to be modified is divided into {skeleton, modified fragment, original molecule} triplets according to the above-mentioned segmentation method. If the skeleton contains both amino and carboxyl groups, proceed to the next step.

[0106] 8. Determine whether the skeleton contains both amino and carboxyl groups. If not, interrupt the program. If yes, go to step 9.

[0107] 9. Set the training steps, use the reinforcement learning agent model to generate molecules, and score them.

[0108] Set the target pharmaceutical properties, such as molecular weight, number of aromatic rings, number of hydrogen donors, number of hydrogen acceptors, etc. The SMILES of each skeleton is subjected to the reinforcement learning agent model to generate amino acids and score the target pharmaceutical properties.

[0109] 10. Determine whether the total score of the medicinal property score + the diversity filter > the target threshold. If so, go to step 11; if not, go to step 12.

[0110] If the total score of the drug-chemical property score and the diversity filter is greater than 0.3, it is included in the result dataset. Otherwise, the Agent model is updated according to the policy gradient algorithm.

[0111] 11. Screen out new unnatural amino acids.

[0112] From the result data set, the natural amino acid data set was excluded, and then classified into known unnatural amino acids and novel unnatural amino acids according to the NSAA data set.

[0113] 12. Update Agent parameters using the policy gradient algorithm and go to step 9 until step 10 is met.

[0114] Figure 6 The present invention provides a schematic diagram of the structure of a non-natural amino acid generation device. This embodiment can be applied to the scenario of non-natural amino acid generation, especially the situation of non-natural amino acid generation through a neural network model, so as to provide clinical reference information for relevant personnel. The non-natural amino acid generation device can be implemented by software and / or hardware, and integrated into a computer terminal device with application development function.

[0115] like Figure 6 As shown, the non-natural amino acid generation device includes: an amino acid molecule segmentation module 410, an amino acid molecule generation module 420 and an amino acid molecule scoring module 430.

[0116] Among them, the amino acid molecule segmentation module 410 is used to segment the amino acid molecule to be modified when the amino acid molecule to be modified is obtained, so as to obtain a molecular skeleton and a modified fragment; the amino acid molecule generation module 420 is used to input the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modified fragments, and to obtain generated amino acid molecules based on the molecular skeleton and the amino acid modified fragments; the amino acid molecule scoring module 430 is used to score the preset target pharmaceutical properties of the generated amino acid molecule, obtain a scoring result, and end the non-natural amino acid molecule generation process according to the scoring result.

[0117] The technical solution of this embodiment is to obtain the amino acid molecule to be modified by segmenting the amino acid molecule to be modified to obtain the molecular skeleton and the modified fragment; input the molecular skeleton into the pre-trained amino acid molecule generation model to obtain the amino acid modified fragment, and obtain the generated amino acid molecule based on the molecular skeleton and the amino acid modified fragment; score the preset target pharmaceutical properties of the generated amino acid molecule to obtain the scoring result, and end the non-natural amino acid molecule generation process according to the scoring result. The technical solution of the embodiment of the present invention solves the problem that the current generation of amino acids through natural amino acid modification has a small number of generated amino acids and low quality. The pre-trained amino acid molecule generation model can be used to automatically generate non-natural amino acids that meet the target pharmaceutical properties, with a large number of generated amino acids and a fast generation speed, providing a basis for the development of polypeptide drugs.

[0118] In an optional embodiment, the amino acid molecule scoring module 430 is specifically used for:

[0119] When the scoring result meets the preset scoring threshold and the generated amino acid molecule is a non-natural amino acid molecule, the generated amino acid molecule is saved and the non-natural amino acid molecule generation process is terminated; when the scoring result does not meet the preset scoring threshold, the amino acid molecule generation model is updated, and the non-natural amino acid corresponding to the amino acid molecule to be modified is generated based on the updated amino acid molecule generation model.

[0120] In an optional embodiment, the amino acid molecule scoring module 430 may also be used for:

[0121] Scoring is performed for at least one of the molecular weight, number of aromatic rings, number of hydrogen donors and number of hydrogen acceptors of the generated amino acid molecule; wherein each scoring item corresponds to a preset scoring rule.

[0122] In an optional embodiment, the above device further includes:

[0123] The judgment module is used to judge whether the molecular skeleton contains both amino groups and carboxyl groups.

[0124] In an optional embodiment, the above device further includes:

[0125] The model training module is used to perform molecular segmentation on the amino acid molecules in the first preset data set to obtain the first data set triples consisting of the corresponding molecular skeleton, modified fragments and original amino acid molecules before segmentation; the molecular skeleton in the first data set triples is used as input samples, and the corresponding modified fragments and original amino acid molecules are used as sample labels to train the initial amino acid molecule generation model to obtain the first amino acid molecule generation model; wherein the first preset data set is a small molecule data set, including amino acid molecules and non-amino acid molecules.

[0126] In an optional implementation, the above model training module is specifically used for:

[0127] The amino acid molecules in the second preset data set are molecularly segmented to obtain a second data set triple consisting of a corresponding molecular skeleton, a modified fragment, and the original amino acid molecule before segmentation; the molecular skeleton in the second data set triple is used as an input sample, and the corresponding modified fragment and the original amino acid molecule are used as sample labels to train the first amino acid molecule generation model to obtain a second amino acid molecule generation model as a pre-trained amino acid molecule generation model; wherein the molecules in the second preset data set are non-natural amino acid molecules.

[0128] In an optional implementation, the above model training module may also be used for:

[0129] The first data set triplet and the description information of the molecular skeleton and the modified fragment in the first data set triplet are randomly represented to generate a new first data set triplet and a new second data set triplet.

[0130] In an optional implementation, the above model training module may also be used for:

[0131] The amino acid molecules in the first preset data set are filtered according to the amino acid properties of the amino acid molecules in the second preset data set.

[0132] The non-natural amino acid production device provided in the embodiment of the present invention can execute the non-natural amino acid production method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0133] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Figure 7 A block diagram of an exemplary computer device 12 suitable for use in implementing embodiments of the present invention is shown. Figure 7 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention. The computer device 12 can be any terminal device with computing capabilities, such as an intelligent controller and server, a mobile phone and other terminal devices.

[0134] like Figure 7 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0135] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0136] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0137] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 7 not shown, usually called a "hard drive"). Although Figure 7 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The system memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0138] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. Program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0139] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. Furthermore, the computer device 12 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with the other modules of the computer device 12 via the bus 18. It should be understood that although Figure 7 Not shown, other hardware and / or software modules may be used in conjunction with computer device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RFID systems, tape drives, and data backup storage systems.

[0140] The processing unit 16 executes various functional applications and data processing by running the program stored in the system memory 28, for example, implementing the method for generating an unnatural amino acid provided in the embodiment of the present invention, the method comprising:

[0141] When the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton and a modified fragment;

[0142] Inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generating amino acid molecules based on the molecular skeleton and the amino acid modification fragments;

[0143] The preset target pharmaceutical properties of the generated amino acid molecules are scored to obtain a scoring result, and the non-natural amino acid molecule generation process is terminated based on the scoring result.

[0144] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for generating an unnatural amino acid as provided in any embodiment of the present invention, the method comprising:

[0145] When the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton and a modified fragment;

[0146] Inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generating amino acid molecules based on the molecular skeleton and the amino acid modification fragments;

[0147] The preset target pharmaceutical properties of the generated amino acid molecules are scored to obtain a scoring result, and the non-natural amino acid molecule generation process is terminated based on the scoring result.

[0148] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0149] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0150] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0151] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, Python, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0152] An embodiment of the present invention further provides a computer program product, including a computer program, which, when executed by a processor, implements the method for generating non-natural amino acids as provided in any embodiment of the present application.

[0153] In the process of implementation, the computer program product can be written in one or more programming languages ​​or a combination thereof to perform the computer program code of the present invention, including object-oriented programming languages, such as Java, Smalltalk, Python, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0154] It should be understood by those skilled in the art that the modules or steps of the present invention described above can be implemented by a general-purpose computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, optionally, they can be implemented by a program code executable by a computer device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0155] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for producing an unnatural amino acid, characterized in that: include: When the amino acid molecule to be modified is obtained, the amino acid molecule to be modified is segmented to obtain a molecular skeleton; Inputting the molecular skeleton into a pre-trained amino acid molecule generation model to obtain amino acid modification fragments, and generating amino acid molecules based on the molecular skeleton and the amino acid modification fragments; The preset target pharmaceutical properties of the generated amino acid molecule are scored to obtain a scoring result, and the non-natural amino acid molecule generation process is terminated according to the scoring result.

2. The method according to claim 1, characterized in that The step of ending the non-natural amino acid molecule generation process according to the scoring result comprises: When the scoring result meets a preset scoring threshold and the generated amino acid molecule is a non-natural amino acid molecule, the generated amino acid molecule is stored, and the non-natural amino acid molecule generation process is terminated; When the scoring result does not meet the preset scoring threshold, the amino acid molecule generation model is updated, and the non-natural amino acid corresponding to the amino acid molecule to be modified is generated based on the updated amino acid molecule generation model.

3. The method according to claim 1, characterized in that The scoring of the preset target pharmaceutical properties of the generated amino acid molecule includes: Scoring is performed on at least one of the molecular weight, number of aromatic rings, number of hydrogen donors and number of hydrogen acceptors of the generated amino acid molecule; wherein each of the scoring items corresponds to a preset scoring rule.

4. The method according to claim 1, characterized in that: Before inputting the molecular skeleton into the pre-trained amino acid molecule generation model, the method further comprises: Determine whether the molecular skeleton contains both amino groups and carboxyl groups.

5. The method according to any one of claims 1 to 4, characterized in that: The process of training the pre-trained amino acid molecule generation model comprises: Molecular segmentation is performed on the amino acid molecules in the first preset data set to obtain corresponding molecular skeletons, modified fragments, and original amino acid molecules before segmentation to form a first data set triple; Using the molecular skeleton in the triplet of the first data set as an input sample, and using the corresponding modified fragment and the original amino acid molecule as sample labels to train the initial amino acid molecule generation model, to obtain a first amino acid molecule generation model; Wherein, the first preset data set is a small molecule data set, including amino acid molecules and non-amino acid molecules.

6. The method according to claim 5, characterized in that The process of training the pre-trained amino acid molecule generation model also includes: Molecular segmentation is performed on the amino acid molecules in the second preset data set to obtain corresponding molecular skeletons, modified fragments, and original amino acid molecules before segmentation to form a second data set triple; Using the molecular skeleton in the triplet of the second data set as an input sample, and using the corresponding modified fragments and the original amino acid molecules as sample labels to train the first amino acid molecule generation model, to obtain a second amino acid molecule generation model as the pre-trained amino acid molecule generation model; Wherein, the molecules in the second preset data set are non-natural amino acid molecules.

7. The method according to claim 6, characterized in that Before model training, it also includes: The first data set triplet and the description information of the molecular skeleton and the modified fragment in the first data set triplet are randomly represented to generate a new first data set triplet and a new second data set triplet.

8. The method according to claim 6, characterized in that Before model training, it also includes: The amino acid molecules in the first preset data set are filtered according to the amino acid properties of the amino acid molecules in the second preset data set.

9. A non-natural amino acid production device, characterized in that include: An amino acid molecule segmentation module is used to segment the amino acid molecule to be modified to obtain a molecular skeleton and a modified fragment when the amino acid molecule to be modified is obtained; An amino acid molecule generation module, used for inputting the molecular skeleton and the modified fragment into a pre-trained amino acid molecule generation model to obtain a generated amino acid molecule; The amino acid molecule scoring module is used to score the preset target pharmaceutical properties of the generated amino acid molecule, obtain a scoring result, and end the non-natural amino acid molecule generation process based on the scoring result.

10. A computer device, characterized in that: The computer device comprises: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for producing an unnatural amino acid as described in any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for producing an unnatural amino acid as claimed in any one of claims 1 to 8 is implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the computer program implements the method for producing an unnatural amino acid according to any one of claims 1 to 8.

Images