Small molecule generation method and device considering conformational change of protein pocket

Abstract

The invention provides a small molecule generation method and device considering conformational change of a protein pocket, and the method comprises the steps: generating a plurality of candidate small molecules according to a given original protein pocket; obtaining a plurality of protein pocket conformations in binding ligand states based on the plurality of candidate small molecules; determining a target protein pocket from a plurality of binding ligand state protein pocket conformations; and generating a target small molecule according to the target protein pocket. According to the method, the conformational change possibly generated when the protein pockets are combined with the small molecules is considered, so that the corresponding small molecules can be generated for the protein pockets combined with the ligands, and the corresponding small molecules can also be generated for the protein pockets not combined with the ligands; a protein pocket conformation in a better binding ligand state is generated for a protein pocket, and a target small molecule is generated according to the protein pocket conformation, so that the success rate and efficiency of drug development and the quality of a final drug are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of drug design, and in particular to a method and device for generating small molecules taking into account conformational changes of protein pockets. Background Art

[0002] At present, the field of protein-ligand binding (PLB) prediction mainly adopts the structure-based drug design (SBDD) method. Most traditional SBDD models assume that proteins and their binding pockets are static, ignoring the conformational changes that may occur during ligand binding. This static assumption limits the accuracy of SBDD models in predicting flexible protein interactions and conformational changes in binding sites.

[0003] Furthermore, existing SBDD models are usually trained based on known binding pocket-ligand complexes and cannot be directly applied to binding pockets without known ligands. Specifically, when only the APO state structure of the binding pocket is provided, these models cannot generate the corresponding HOLO state binding pocket-ligand pair. This means that existing SBDD models are mainly limited to scenarios with known ligand binding pockets, while for many new drug targets, their ligand information is unknown, and existing SBDD models have limited ability to predict binding pockets for unknown ligands.

[0004] Therefore, how to solve the problem that the existing SBDD model ignores the conformational changes that may occur during ligand binding and cannot generate corresponding binding pocket-ligand pairs for the binding pocket of unknown ligands is an important issue that needs to be urgently solved in the field of drug design. Summary of the invention

[0005] The present invention provides a method and device for generating small molecules taking into account conformational changes of protein pockets, so as to overcome the defects of the existing SBDD model that ignores the conformational changes that may occur during ligand binding and cannot generate corresponding binding pocket-ligand pairs for the binding pocket of unknown ligands, thereby generating a better protein pocket conformation in the ligand-bound state and further generating the target small molecule.

[0006] On the one hand, the present invention provides a method for generating small molecules taking into account changes in protein pocket conformation, comprising: generating multiple candidate small molecules based on a given original protein pocket; wherein the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in an unligand-bound state; based on the multiple candidate small molecules, obtaining multiple protein pocket conformations in a ligand-bound state; determining a target protein pocket from the multiple protein pocket conformations in a ligand-bound state; and generating a target small molecule based on the target protein pocket.

[0007] Further, generating a plurality of candidate small molecules according to the given original protein pocket includes: inputting the backbone of the original protein pocket into a pre-trained molecule generation model to obtain a plurality of output candidate small molecules; wherein, the molecule generation model is trained and optimized based on protein pocket backbone samples and corresponding small molecule samples, and the plurality of candidate small molecules can specifically bind to the original protein pocket.

[0008] Further, obtaining a plurality of protein pocket conformations in the bound ligand state based on the plurality of candidate small molecules includes: performing flexible docking on the plurality of candidate small molecules to generate initial protein-ligand complex conformations; performing relaxation processing on the selected initial protein-ligand complex conformations to obtain a plurality of protein pocket conformations in the bound ligand state.

[0009] Further, determining the target protein pocket from the plurality of protein pocket conformations in the bound ligand state includes: scoring the plurality of protein pocket conformations in the bound ligand state based on a pre-trained confidence model; determining the protein pocket conformation with the highest score as the target protein pocket; wherein, the confidence model is trained and optimized based on positive samples and negative samples.

[0010] Further, training and optimizing the confidence model specifically includes: based on the pre-trained molecule generation model, generating training small molecules according to the collected protein pocket backbone samples; calculating the docking score of each training small molecule and the average docking score of all training small molecules; determining the training small molecules higher than the average docking score as positive samples, and determining the training small molecules lower than the average docking score as negative samples; using the positive samples and negative samples as model inputs and the predicted scores of positive samples and negative samples as model outputs, and iteratively optimizing the confidence model through a preset loss function.

[0011] Further, the process of determining the target protein pocket is a cyclic iterative process, specifically including: in the case where the current round does not reach the set number of rounds, using the target protein pocket determined in the current round as the original protein pocket in the next round and re-executing the steps of determining the target protein pocket; in the case where the current round reaches the set number of rounds, stopping the iteration and using the target protein pocket determined in the current round as the final protein pocket; correspondingly, generating a target small molecule according to the target protein pocket includes: inputting the final protein pocket into a pre-trained target molecule generation model to obtain the output target small molecule; wherein, the target molecule generation model is trained and optimized based on known protein pocket-ligand small molecule complexes.

[0012] Second aspect, the present invention further provides a small molecule generation device considering conformational changes of protein pockets, including: a candidate small molecule generation module for generating a plurality of candidate small molecules according to a given original protein pocket; wherein the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in a ligand-unbound state; a protein pocket conformation acquisition module for obtaining a plurality of protein pocket conformations in a ligand-bound state based on the plurality of candidate small molecules; a target protein pocket determination module for determining a target protein pocket from the plurality of protein pocket conformations in a ligand-bound state; and a target small molecule generation module for generating a target small molecule according to the target protein pocket.

[0013] Third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, it implements the small molecule generation method considering conformational changes of protein pockets as described in any one of the above.

[0014] Fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the small molecule generation method considering conformational changes of protein pockets as described in any one of the above.

[0015] Fifth aspect, the present invention further provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the small molecule generation method considering conformational changes of protein pockets as described in any one of the above.

[0016] The small molecule generation method considering conformational changes of protein pockets provided by the present invention generates a plurality of candidate small molecules according to a given original protein pocket; wherein the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in a ligand-unbound state; obtains a plurality of protein pocket conformations in a ligand-bound state based on the plurality of candidate small molecules; determines a target protein pocket from the plurality of protein pocket conformations in a ligand-bound state; and generates a target small molecule according to the target protein pocket. By considering the conformational changes that may occur when a protein pocket binds to a small molecule, this method can not only generate corresponding small molecules for protein pockets in a ligand-bound state, but also generate corresponding small molecules for protein pockets in a ligand-unbound state; by generating better protein pocket conformations in a ligand-bound state for the protein pocket and generating target small molecules accordingly, it effectively improves the success rate, efficiency of drug development, and the quality of the final drug. Description of the Drawings

[0017] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a schematic flowchart of a small molecule generation method considering protein pocket conformational changes provided by an embodiment of the present invention.

[0019] Figure 2 It is a schematic overall flowchart of a small molecule generation method considering protein pocket conformational changes provided by an embodiment of the present invention.

[0020] Figure 3 It is a schematic structural diagram of a small molecule generation device considering protein pocket conformational changes provided by an embodiment of the present invention.

[0021] Figure 4 It is a schematic physical structure diagram of an electronic device provided by an embodiment of the present invention. Detailed Embodiments

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present invention belong to the scope of protection of the present invention.

[0023] It should be noted that most current SBDD models assume that proteins and their pockets are static, ignoring the conformational changes that may occur when binding small molecules. In addition, these models are usually trained based on known protein pocket-ligand complexes. For protein pockets without known ligands, only inputting the protein pocket conformation in the APO state cannot generate the corresponding protein pocket-ligand pairs in the HOLO state.

[0024] Considering this, the present invention proposes a small molecule generation method considering protein pocket conformational changes. Specifically, Figure 1 shows a schematic flowchart of a small molecule generation method considering protein pocket conformational changes provided by an embodiment of the present invention.

[0025] As Figure 1 shown, the method includes steps S110 - S140. The following will elaborate on steps S110 - S140 and related steps in detail.

[0026] S110. Generate multiple candidate small molecules based on a given original protein pocket, where the original protein pocket is a protein pocket in a ligand-bound state or a ligand-unbound state.

[0027] A protein pocket, also known as a protein-binding pocket, refers to a cavity on the surface or inside of a protein that is suitable for binding ligands (such as small molecules, drugs, or other biomolecules). The shape, location, physicochemical properties, and function of a protein pocket are mainly determined by the surrounding amino acid residues. The kinetic properties of a protein pocket, such as flexibility and motion, are crucial for the specific interactions of the protein. These properties allow the protein pocket to open, close, or adapt, thereby regulating the ligand-binding process and performing specific protein functions.

[0028] It is easy to understand that the three-dimensional structure of the original protein can be obtained from a protein database, and the structure file of the original protein can be downloaded. Then, various software tools can be used to identify the pockets in the original protein structure, that is, the original protein pocket can be obtained.

[0029] For a given original protein pocket, multiple corresponding candidate small molecules can be generated through a pre-trained molecular generation model (such as an improved SBDD model), or multiple candidate small molecules that may have good interactions with the original protein pocket can be found through virtual screening methods (such as molecular docking software AutoDock Vina, GOLD, etc.). Specific limitations are not provided here.

[0030] It is worth mentioning that the original protein pocket given in this embodiment can be not only a protein pocket in a ligand-bound state (HOLO state), but also a protein pocket in a ligand-unbound state (APO state).

[0031] A protein pocket in the HOLO state refers to a protein that has bound its natural ligand (such as a substrate, cofactor, or inhibitor) or other small molecules. In this state, the protein pocket is usually "filled", meaning it has been occupied by one or more molecules.

[0032] A protein pocket in the APO state refers to a protein that has not bound any ligand (such as a substrate, cofactor, or inhibitor) or other small molecules. In this state, the protein pocket may be empty or in a state ready to bind.

[0033] Based on generating multiple candidate small molecules according to the given original protein pocket in step S110, further, step S120 is executed.

[0034] S120. Based on the multiple candidate small molecules, obtain multiple conformations of protein pockets in the ligand-bound state.

[0035] It is easily understandable that, given the original protein pocket and the corresponding multiple candidate small molecules, the protein pocket conformations in multiple ligand-bound states can be determined through a flexible docking step and a relaxation step.

[0036] In the flexible docking step, not only the flexibility of the candidate small molecules is considered, but also the side chains of the original protein pocket are allowed to be adjusted to simulate a more realistic biological binding process. Since conformational changes occur when the protein pocket binds to the ligand, compared with traditional rigid docking, the flexible docking in this embodiment is closer to the actual docking situation.

[0037] In the relaxation step, the initial conformation of the original protein pocket-candidate small molecule complex is optimized to make it more stable and reflect the actual state under physiological conditions. Relaxation helps to eliminate unnatural contacts, release stress, and explore possible low-energy states.

[0038] Specifically, multiple candidate small molecules can be first flexibly docked to obtain the protein pocket-ligand complex conformation, and then the protein pocket-ligand complex conformation can be relaxed to obtain multiple protein pocket conformations in the ligand-bound state.

[0039] Among them, the so-called "protein pocket conformation" refers to the spatial structure and chemical properties of a specific region on the surface or inside of a protein, including but not limited to the geometric shape of the pocket, chemical properties (such as hydrophobicity, polarity, charge distribution, etc.), and dynamic changes (such as side chain adjustment).

[0040] Based on obtaining multiple protein pocket conformations in multiple ligand-bound states in step S120, further, step S130 is executed.

[0041] S130, determine the target protein pocket from multiple protein pocket conformations in multiple ligand-bound states.

[0042] It is easily understandable that the protein pocket conformations in multiple ligand-bound states can be scored separately, and the protein pocket conformation with the highest score can be determined as the target protein pocket.

[0043] Among them, the scoring criteria include but not limited to the structural rationality, energy stability, dynamic behavior, key interactions, binding affinity, and ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) characteristics of the protein pocket conformation, etc. A good protein pocket conformation should be as consistent with the existing experimental data as possible, have a low free energy, be able to remain stable for a certain period of time, and not undergo significant conformational changes.

[0044] The process of scoring the conformational states of protein pockets can be achieved by various methods, such as scoring functions based on physical and chemical principles, machine learning models, and experimental data verification, etc., which are not specifically limited herein.

[0045] On the basis of determining the target protein pocket (conformational state) from multiple protein pocket conformations in the bound ligand state in step S130, further, step S140 is executed.

[0046] S140, generate a target small molecule according to the target protein pocket.

[0047] It is easy to understand that a pre-trained target molecule generation model (such as the SBDD model) can be used to generate the target small molecule. Specifically, the target protein pocket is input into the pre-trained target molecule generation model, and the output target small molecule can be obtained.

[0048] It is worth mentioning that the target protein pocket is a better protein pocket compared to multiple protein conformations. On this premise, using the target protein pocket to generate the target small molecule can effectively improve the success rate, efficiency of drug development, and the quality of the final drug.

[0049] It should be noted that the target molecule generation model applicable in this step is trained and optimized based on known protein pocket-ligand small molecule complexes, and should be different from the molecule generation model applicable in step S110. Specifically, the latter (molecule generation model) removes the limitation that the side chain conformation of the protein pocket remains fixed, which helps to find better and more diverse small molecules.

[0050] In this embodiment, multiple candidate small molecules are generated according to the given original protein pocket; wherein, the original protein pocket is a protein pocket in the bound ligand state or a protein pocket in the unbound ligand state; based on the multiple candidate small molecules, multiple protein pocket conformations in the bound ligand state are obtained; the target protein pocket is determined from the multiple protein pocket conformations in the bound ligand state; and the target small molecule is generated according to the target protein pocket. By considering the conformational changes that may occur when the protein pocket binds to a small molecule, this method can not only generate corresponding small molecules for protein pockets in the bound ligand state, but also generate corresponding small molecules for protein pockets in the unbound ligand state; by generating better protein pocket conformations in the bound ligand state for the protein pocket and generating the target small molecule accordingly, the success rate, efficiency of drug development, and the quality of the final drug are effectively improved.

[0051] Based on the above embodiments, optionally, multiple candidate small molecules are generated according to a given original protein pocket, including: inputting the backbone of the original protein pocket into a pre-trained molecule generation model to obtain multiple output candidate small molecules; wherein, the molecule generation model is trained and optimized based on protein pocket backbone samples and corresponding small molecule samples, and the multiple candidate small molecules can specifically bind to the original protein pocket.

[0052] It is easy to understand that in this embodiment, a molecule generation model is pre-trained. Different from the existing SBDD model, the input of the molecule generation model is only the backbone structure (such as backbone coordinates) of the original protein pocket, which removes the restriction that the side chain structure is completely fixed, rather than the overall structure of the original protein pocket (backbone structure + side chain structure).

[0053] After inputting the known backbone structure of the original protein pocket into the pre-trained molecule generation model, multiple output candidate small molecules can be obtained, and these multiple candidate small molecules can specifically bind to the original protein pocket.

[0054] When training the molecule generation model, the protein pocket backbone sample is used as the model input, the generated training small molecule is used as the model output, and the difference between the generated training small molecule and the corresponding small molecule sample is used as the training loss, and the molecule generation model is iteratively optimized to obtain a trained molecule generation model.

[0055] In this embodiment, by inputting the backbone of the original protein pocket into a pre-trained molecule generation model, multiple output candidate small molecules are obtained, and multiple protein pocket conformations in the ligand-bound state are obtained based on the multiple candidate small molecules. A target protein pocket is determined from the multiple protein pocket conformations in the ligand-bound state to generate a target small molecule according to the target protein pocket. By considering the conformational changes that the protein pocket may undergo when binding to a small molecule, this method can not only generate corresponding small molecules for the protein pocket in the ligand-bound state, but also generate corresponding small molecules for the protein pocket in the unliganded state; by generating better protein pocket conformations in the ligand-bound state for the protein pocket and generating the target small molecule accordingly, the success rate, efficiency of drug development and the quality of the final drug are effectively improved.

[0056] Based on the above embodiments, optionally, multiple protein pocket conformations in the ligand-bound state are obtained based on multiple candidate small molecules, including: performing flexible docking on the multiple candidate small molecules to generate initial protein-ligand complex conformations; performing relaxation processing on the selected initial protein-ligand complex conformations to obtain multiple protein pocket conformations in the ligand-bound state.

[0057] It is easy to understand that multiple candidate small molecules are first subjected to flexible docking. Specifically, during the docking process, the flexibility of the side chains of the protein pocket is allowed to better simulate the interactions under real biological conditions; it is ensured that an appropriate flexibility (such as the number of rotatable bonds) is defined for each candidate small molecule so that the candidate small molecule can adjust its conformation during docking. By performing flexible docking in this way, a series of possible initial protein-ligand complex conformations can be generated.

[0058] Then, all or representative initial protein-ligand complex conformations can be selected for relaxation. The selected initial protein-ligand complex conformations are placed under physiological conditions (such as temperature, pH value, ion concentration, etc.), and molecular dynamics (MD) simulations on a short time scale are performed to help stabilize the initial protein-ligand complex conformations, eliminate unnatural contacts, and explore possible low-energy states. Alternatively, if long-term molecular dynamics simulations are not required, local structure optimization methods such as molecular mechanics optimization (MM), quantum mechanics (QM) optimization, or a method that combines molecular mechanics optimization and quantum mechanics optimization can also be used to improve the energy distribution of the initial protein-ligand complexes.

[0059] After relaxation, multiple protein-ligand complex conformations in the ligand-bound state can be obtained. By extracting the protein pocket conformations among them, multiple protein pocket conformations in the ligand-bound state can be obtained.

[0060] It should be noted that the embodiment of first performing flexible docking and then relaxation is a preferred embodiment. In other embodiments, single flexible docking or relaxation treatment can be performed on the candidate small molecules, and multiple protein pocket conformations in the ligand-bound state can also be obtained, which is not specifically limited here.

[0061] In this embodiment, by performing flexible docking on multiple candidate small molecules, the dynamic behaviors of the protein pocket and the ligand small molecule are considered, making the generated initial protein-ligand complex conformations closer to the actual situation. By performing relaxation treatment on the initial protein-ligand complexes, more stable protein pocket conformations that reflect the actual biological conditions can be obtained. In addition, it is also ensured that the generated protein pocket conformations in the ligand-bound state have sufficient diversity to comprehensively understand all possible changes in the protein pocket.

[0062] Based on the above embodiments, optionally, determining the target protein pocket from multiple protein pocket conformations in the ligand-bound state includes: scoring multiple protein pocket conformations in the ligand-bound state based on a pre-trained confidence model; determining the protein pocket conformation with the highest score as the target protein pocket; wherein the confidence model is trained and optimized based on positive samples and negative samples.

[0063] It is easy to understand that this embodiment requires pre-training a confidence model, including: based on a pre-trained molecular generation model, generating training small molecules according to the collected protein pocket backbone samples (such as protein pocket backbone samples not used when training the molecular generation model); calculating the docking score of each training small molecule and the average docking score of all training small molecules; determining the training small molecules with docking scores higher than the average docking score as positive samples and those with docking scores lower than the average docking score as negative samples; using the positive samples and negative samples as the model input and the predicted scores of the positive samples and the predicted scores of the negative samples as the model output, and iteratively optimizing the confidence model through a preset loss function.

[0064] Specifically, input the collected protein pocket backbone samples into the pre-trained molecular generation model, and the output training small molecules can be obtained. Subsequently, a scoring function of molecular docking software (such as force field scoring function, empirical scoring function, semi-empirical scoring function, knowledge-driven scoring function, physical chemistry scoring function, etc.) can be used to score each training small molecule to obtain the docking score of each training small molecule. On this basis, taking the average of the docking scores of all training small molecules, the average docking score can be obtained.

[0065] Immediately afterwards, according to the average docking score and the docking score of each training small molecule, the training small molecules are divided into positive samples and negative samples. Specifically, if the docking score of a training small molecule is higher than the average docking score, then the training small molecule is regarded as a positive sample; conversely, if the docking score of a training small molecule is lower than the average docking score, then the training small molecule is regarded as a negative sample.

[0066] After constructing the positive samples and negative samples, use the positive samples and negative samples to iteratively optimize the confidence model. During training, use the positive samples and negative samples as the model input and the predicted scores of the positive samples and the predicted scores of the negative samples as the model output, and iteratively optimize the confidence model through the preset loss function described by the following formula (1) Iteratively optimize the confidence model.

[0067] (1).

[0068] In formula (1), represents the positive sample, represents the predicted score of the confidence model for the positive sample, represents the negative sample, represents the predicted score of the confidence model for the negative sample.

[0069] According to the above, the pre-trained confidence model can be obtained.

[0070] During application, multiple protein pocket conformations in the ligand-bound state are respectively input into a pre-trained confidence model for scoring, and scores corresponding to the multiple protein pocket conformations can be obtained. The protein pocket with the highest score is determined as the target protein pocket.

[0071] Further, the target protein pocket is input into a pre-trained target molecule generation model, and the output target small molecule can be obtained. The target molecule generation model here is an existing SBDD model, which is trained and optimized based on known protein pocket-ligand small molecule complexes.

[0072] In this embodiment, by using a pre-trained confidence model to score multiple protein pocket conformations in the ligand-bound state, and determining the protein pocket conformation with the highest score as the target protein pocket, and then generating a target small molecule according to the target protein pocket. This method can generate corresponding small molecules not only for protein pockets in the ligand-bound state but also for protein pockets in the unligand-bound state by considering the conformational changes that may occur when the protein pocket binds to a small molecule; by generating a better protein pocket conformation in the ligand-bound state for the protein pocket and generating a target small molecule accordingly, the success rate, efficiency, and quality of the final drug in drug development are effectively improved.

[0073] Based on the above embodiment, optionally, the process of determining the target protein pocket is a cyclic iterative process, specifically including: when the current round has not reached the set number of rounds, the target protein pocket determined in the current round is used as the original protein pocket in the next round, and the step of determining the target protein pocket is re-executed; when the current round reaches the set number of rounds, the iteration is stopped, and the target protein pocket determined in the current round is used as the final protein pocket.

[0074] It is easy to understand that the process of determining the target protein pocket is cyclic and iterative. The target protein pocket obtained in each round can be used as the original protein pocket in the next round, and the above steps S110 - S130 are executed, thereby realizing the continuous optimization of the target protein pocket and finally obtaining an optimal protein pocket.

[0075] Specifically, this embodiment has a limit on the set number of rounds. At the start of each round, the current round is recorded, and at the end of each round, the current round is compared with the set number of rounds. If the current round does not reach the set number of rounds, it indicates that further optimization is possible. The target protein pocket determined in the current round is used as the original protein pocket in step S110, and steps S110 - S130 are executed to obtain the target protein pocket for another round. This process is repeated until the current round reaches the set number of rounds. In this case, the iteration stops, and the target protein pocket determined in the current round (i.e., the last round) is used as the final and best protein pocket.

[0076] Among them, the set number of rounds can be set according to actual needs and is not specifically limited here.

[0077] Furthermore, the final and best protein pocket is used as the target protein pocket and input into a pre-trained target molecule generation model, and the output target small molecule can be obtained. The target molecule generation model is trained and optimized based on known protein pocket-ligand small molecule complexes, which will not be elaborated here.

[0078] In this embodiment, when the current round does not reach the set number of rounds, the target protein pocket determined in the current round is used as the original protein pocket for the next round, and the steps to determine the target protein pocket are re-executed. When the current round reaches the set number of rounds, the iteration stops, and the target protein pocket determined in the current round is used as the final protein pocket. Then, the target small molecule is generated based on the final protein pocket. By considering the conformational changes that may occur when the protein pocket binds to a small molecule, this method can not only generate corresponding small molecules for the protein pocket in the ligand-bound state but also for the protein pocket in the unligand-bound state. By generating a better conformational state of the protein pocket in the ligand-bound state and generating the target small molecule accordingly, it effectively improves the success rate, efficiency, and quality of the final drug in drug development.

[0079] In addition, in some embodiments, Figure 2 shows the overall flowchart of the small molecule generation method considering protein pocket conformational changes provided by the embodiments of the present invention.

[0080] Such as Figure 2As shown, first, for a given original protein pocket, input it into a pre-trained molecular generation model to generate multiple candidate small molecules. Subsequently, perform flexible docking and / or relaxation processing on the multiple candidate small molecules to obtain multiple protein pocket conformations in the bound ligand state. Further, input the multiple protein pocket conformations in the bound ligand state into a pre-trained confidence model for scoring, and take the protein pocket conformation with the highest score as the current better protein pocket, that is, the target protein pocket.

[0081] According to Figure 2 It can be seen that after obtaining the target protein pocket, it can be used as the input in a pre-trained target molecular generation model (such as the HOLO SBDD model) to generate target small molecules; alternatively, the target protein pocket can also be used as a new original protein pocket to continue to execute the above steps S110 - S130 to determine the target protein pocket in a new round.

[0082] Thereafter, the target protein pocket in each round can be selected to continue iterative cycling or input into the target molecular generation model. The selection basis can be the set number of rounds described in the above embodiments or other rule settings (such as actual situations or requirements), which are not specifically limited herein.

[0083] Corresponding to the small molecule generation method considering protein pocket conformation changes described in the above embodiments, the present invention also proposes a small molecule generation device considering protein pocket conformation changes.

[0084] Specifically, Figure 3 shows a schematic structural diagram of the small molecule generation device considering protein pocket conformation changes provided by the embodiments of the present invention.

[0085] As Figure 3 shown, the device includes: a candidate small molecule generation module 310 for generating multiple candidate small molecules according to a given original protein pocket, where the original protein pocket is a protein pocket in the bound ligand state or the unbound ligand state; a protein pocket conformation acquisition module 320 for obtaining multiple protein pocket conformations in the bound ligand state based on the multiple candidate small molecules; a target protein pocket determination module 330 for determining the target protein pocket from the multiple protein pocket conformations in the bound ligand state; and a target small molecule generation module 340 for generating target small molecules according to the target protein pocket.

[0086] In this embodiment, a candidate small molecule generation module 310 generates a plurality of candidate small molecules according to a given original protein pocket, where the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in an unligand-bound state; a protein pocket conformation acquisition module 320 acquires a plurality of protein pocket conformations in a ligand-bound state based on the plurality of candidate small molecules; a target protein pocket determination module 330 determines a target protein pocket from the plurality of protein pocket conformations in a ligand-bound state; and a target small molecule generation module 340 generates a target small molecule according to the target protein pocket. By considering the conformational changes that may occur when a protein pocket binds to a small molecule, this device can not only generate corresponding small molecules for protein pockets in a ligand-bound state, but also generate corresponding small molecules for protein pockets in an unligand-bound state; by generating better protein pocket conformations in a ligand-bound state for the protein pocket and generating target small molecules accordingly, the success rate, efficiency, and quality of the final drug in drug development are effectively improved.

[0087] It should be noted that the small molecule generation device considering protein pocket conformational changes provided in the embodiments of the present invention can be correspondingly referred to the small molecule generation method considering protein pocket conformational changes described in the above embodiments, and will not be elaborated here.

[0088] Figure 4 An entity structure diagram of an electronic device is exemplified, as Figure 4 shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440. Among them, the processor 410, the communication interface 420, and the memory 430 complete communication with each other through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the small molecule generation method considering protein pocket conformational changes, and the method includes: generating a plurality of candidate small molecules according to a given original protein pocket, where the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in an unligand-bound state; acquiring a plurality of protein pocket conformations in a ligand-bound state based on the plurality of candidate small molecules; determining a target protein pocket from the plurality of protein pocket conformations in a ligand-bound state; and generating a target small molecule according to the target protein pocket.

[0089] In addition, when the logical instructions in the above-mentioned memory 430 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0090] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the small molecule generation method considering the conformational changes of protein pockets provided by the above-mentioned various methods. The method includes: generating a plurality of candidate small molecules according to a given original protein pocket; wherein, the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in an unliganded state; based on the plurality of candidate small molecules, obtaining a plurality of conformational states of the protein pocket in the ligand-bound state; determining a target protein pocket from the plurality of conformational states of the protein pocket in the ligand-bound state; and generating a target small molecule according to the target protein pocket.

[0091] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the small molecule generation method considering the conformational changes of protein pockets provided by the above-mentioned various methods. The method includes: generating a plurality of candidate small molecules according to a given original protein pocket; wherein, the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in an unliganded state; based on the plurality of candidate small molecules, obtaining a plurality of conformational states of the protein pocket in the ligand-bound state; determining a target protein pocket from the plurality of conformational states of the protein pocket in the ligand-bound state; and generating a target small molecule according to the target protein pocket.

[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative work.

[0093] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments or equivalently replace some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating small molecules considering conformational changes of protein pockets, characterized in that, Comprising: Generating a plurality of candidate small molecules based on a given original protein pocket; wherein, the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in an unliganded state; Obtaining a plurality of protein pocket conformations in a ligand-bound state based on the plurality of candidate small molecules; Determining a target protein pocket from the plurality of protein pocket conformations in a ligand-bound state; Generating a target small molecule according to the target protein pocket.

2. The method for generating small molecules considering conformational changes of protein pockets according to claim 1, wherein The generating a plurality of candidate small molecules based on a given original protein pocket includes: Inputting the backbone of the original protein pocket into a pre-trained molecular generation model to obtain the output plurality of candidate small molecules; Wherein, the molecular generation model is trained and optimized based on protein pocket backbone samples and corresponding small molecule samples, and the plurality of candidate small molecules can specifically bind to the original protein pocket.

3. The small molecule generation method considering conformational changes of protein pockets according to claim 1, characterized in that, The obtaining a plurality of protein pocket conformations in a ligand-bound state based on the plurality of candidate small molecules includes: Performing flexible docking on the plurality of candidate small molecules to generate initial protein-ligand complex conformations; Performing relaxation processing on the selected initial protein-ligand complex conformations to obtain a plurality of protein pocket conformations in a ligand-bound state.

4. The method for generating small molecules considering conformational changes of protein pockets according to claim 1, characterized in that The determining a target protein pocket from the plurality of protein pocket conformations in a ligand-bound state includes: Scoring the plurality of protein pocket conformations in a ligand-bound state based on a pre-trained confidence model; Determining the protein pocket conformation with the highest score as the target protein pocket; Wherein, the confidence model is trained and optimized based on positive samples and negative samples.

5. The method for generating small molecules considering conformational changes of protein pockets according to claim 4, wherein Training and optimizing the confidence model specifically includes: Based on a pre-trained molecular generation model, generating training small molecules according to the collected protein pocket backbone samples; Calculating the docking score of each training small molecule and the average docking score of all training small molecules; Determining the training small molecules higher than the average docking score as positive samples and the training small molecules lower than the average docking score as negative samples; Using the positive samples and negative samples as model inputs and the predicted scores of the positive samples and the predicted scores of the negative samples as model outputs, and iteratively optimizing the confidence model through a preset loss function.

6. The method for generating small molecules considering conformational changes of protein pockets according to any one of claims 1-5, characterized in that, The process of determining the target protein pocket is a cyclic iterative process, specifically including: When the current round does not reach the set number of rounds, using the target protein pocket determined in the current round as the original protein pocket in the next round, and re-executing the steps of determining the target protein pocket; When the current round reaches the set number of rounds, stopping the iteration and using the target protein pocket determined in the current round as the final protein pocket; Correspondingly, generating a target small molecule according to the target protein pocket includes: Inputting the final protein pocket into a pre-trained target molecule generation model to obtain the output target small molecule; Wherein, the target molecule generation model is trained and optimized based on known protein pocket-ligand small molecule complexes.

7. A small molecule generation device that takes into account conformational changes in protein pockets, characterized in that, Comprising: A candidate small molecule generation module for generating a plurality of candidate small molecules according to a given original protein pocket, wherein the original protein pocket is a protein pocket in a ligand-bound state or a protein pocket in a ligand-unbound state; A protein pocket conformation acquisition module for acquiring a plurality of protein pocket conformations in a ligand-bound state based on a plurality of candidate small molecules; A target protein pocket determination module for determining a target protein pocket from a plurality of protein pocket conformations in a ligand-bound state; A target small molecule generation module for generating a target small molecule according to the target protein pocket.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the small molecule generation method considering protein pocket conformation changes according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium, having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the small molecule generation method considering protein pocket conformation changes according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the small molecule generation method considering protein pocket conformation changes according to any one of claims 1 to 6.

Images