Molecular prediction model training method and protein pocket processing method

Abstract

The present invention relates to the technical field of molecular prediction, and in particular to a molecular prediction model training method and a protein pocket processing method. The molecular prediction model training method comprises the following steps: acquiring a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each application scenario; and sequentially using each training subset to train a molecular prediction model. Thus, a trained molecular prediction model can be applied to multiple scenarios.

Description

Training method of molecular prediction model and processing method of protein pocket

[0001] Cross-reference to related applications

[0002] The present application claims priority to the Chinese patent application No. 202411312390.6, filed on September 20, 2024, and entitled "Training method of molecular prediction model and processing method of protein pocket", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0003] The present application relates to the technical field of molecular prediction, in particular to a training method of molecular prediction model and a processing method of protein pocket. BACKGROUND

[0004] Protein pockets are depressions or pits on the surface of protein molecules, usually formed by the arrangement of amino acid residues. These pockets play a key role in the three-dimensional structure of proteins, especially when proteins interact with other molecules such as substrates, ligands, or other proteins. In drug design, medicinal chemists will design molecules that fit the pockets based on their expert experience.

[0005] Since the discovery of suitable molecules for a given condition is a very important step in the early stages of drug design, many research institutions and pharmaceutical companies have invested a lot of resources in this area. In the presence of pockets, research institutions and pharmaceutical companies strive to find molecules that fit specific pockets, which requires a lot of research and experiments to discover suitable molecular structures.

[0006] In the field of molecule generation, deep neural networks are widely used, mainly because they can learn complex data representations and make efficient predictions and generation. However, existing molecular generation models still face some challenges: (1) Data diversity, i.e., the data characteristics in different scenarios can be very different. For example, the pocket NCI (Non-Covalent Interactions) prediction model requires data on pockets and NCI site information, while the pocket-based molecule generation model requires data on pockets and small molecules corresponding to the pockets. Existing models cannot adapt to different scenarios of data input. (2) Task diversity: not only is the data diverse, but the task types can also differ greatly. Some tasks may focus on the prediction of NCI sites on the pocket, while other tasks may focus more on the generation of molecules. Therefore, deep neural network models need to have enough flexibility to adapt to these different task requirements. SUMMARY

[0007] Therefore, the application provides a training method of a molecular prediction model and a processing method of a protein pocket to solve the problem that the existing deep neural network model cannot be applied to various scenes.

[0008] In a first aspect, the application provides a training method of a molecular prediction model, which comprises the following steps: obtaining a training data set, wherein the training data set comprises multiple application scenarios and a training subset corresponding to each application scenario; and training the molecular prediction model using each training subset in sequence.

[0009] The training method of the molecular prediction model provided in the embodiments of the application can make the trained molecular prediction model applicable to various scenes by obtaining a training data set, wherein the training data set comprises multiple application scenarios and a training subset corresponding to each application scenario; and training the molecular prediction model using each training subset in sequence.

[0010] In an optional implementation, the training data set comprises a first application scenario, a second application scenario and a third application scenario; the first training subset corresponding to the first application scenario comprises a pocket data set, an anchor condition set, a small molecule structure data set and a small molecule attribute information set; the second training subset corresponding to the second application scenario comprises the pocket data set and the anchor condition set; and the third training subset corresponding to the third application scenario comprises the pocket data set, the small molecule structure data set and the small molecule attribute information set.

[0011] Thus, the trained molecular prediction model can be applied to the reasoning scenario in which the user has a pocket PDB (Protein Data Bank) file, is clear about the information of the NCI and the anchor point on the pocket and is clear about the expected molecular attribute, hopes to generate a 3D molecule suitable for the pocket; can be applied to the reasoning scenario in which the user only has a pocket PDB file and an expected molecular attribute, hopes to generate a 3D molecule suitable for the pocket; and can be applied to the reasoning scenario in which the user only has a pocket PDB file and hopes to generate anchor point information.

[0012] In some optional implementations, the training of the molecular prediction model using each training subset in sequence comprises: training the molecular prediction model using the first training subset; training the molecular prediction model using the second training subset after the training of the molecular prediction model using the first training subset; and training the molecular prediction model using the third training subset after the training of the molecular prediction model using the second training subset.

[0013] Thus, the generalization ability of the model can be enhanced, that is, by using different training subsets, the model can learn more diverse data features, which helps to improve the prediction ability of the model on unseen data. Each training subset may represent different aspects or different statistical distributions of the data. Such diverse training can reduce the risk of overfitting.

[0014] In some optional embodiments, training the molecular prediction model using the first training subset comprises: inputting the pocket dataset and the anchor condition set into an encoder of the molecular prediction model to obtain a first output vector; inputting the small molecule attribute information set into an attribute controller of the molecular prediction model to obtain attribute representation; and inputting the first output vector, the attribute representation, and the small molecule structure dataset into a decoder of the molecular prediction model to train the molecular prediction model; and / or training the molecular prediction model using the second training subset comprises: inputting the pocket dataset and the anchor condition set into the encoder of the molecular prediction model to train the molecular prediction model; and / or training the molecular prediction model using the third training subset comprises: inputting the pocket dataset into the encoder of the molecular prediction model to obtain a second output vector; inputting the small molecule attribute information set into the attribute controller of the molecular prediction model to obtain attribute representation; and inputting the second output vector, the attribute representation, and the small molecule structure dataset into the decoder of the molecular prediction model to train the molecular prediction model.

[0015] Thus, the model can be adapted to multiple scenarios, achieving the goal of one model with one set of parameters being able to handle multiple scenario problems.

[0016] In some optional embodiments, the pocket dataset comprises a sequence of atom types of the pocket, a sequence of atomic coordinates of the pocket, and a sequence of atom-level features of the pocket, and inputting the pocket dataset and the anchor condition set into the encoder of the molecular prediction model to obtain the first output vector comprises: embedding the sequence of atom types of the pocket to obtain atom type representation with a first dimension of feature; embedding the sequence of atomic coordinates of the pocket to obtain atomic coordinate representation with a second dimension of feature; concatenating the atom type representation with the first dimension of feature and the atomic coordinate representation with the second dimension of feature to obtain atom type / coordinate representation with a third dimension of feature; embedding the sequence of atom-level features of the pocket to obtain atom-level feature representation with a fourth dimension of feature; embedding the anchor condition set to obtain NCI / anchor representation with a fifth dimension of feature; and concatenating the atom-level feature representation with the fourth dimension of feature and the NCI / anchor representation with the fifth dimension of feature to obtain other pocket representation with the third dimension of feature; aligning and adding the atom type / coordinate representation with the third dimension of feature and the other pocket representation with the third dimension of feature to obtain pocket representation with the third dimension of feature, and inputting the pocket representation with the third dimension of feature into the encoder to obtain the first output vector.

[0017] The first output vector obtained in this way can be used to train the molecular prediction model in the first application scenario and the second application scenario, and the trained model has better performance.

[0018] In some optional embodiments, the pocket dataset includes an atomic type sequence of the pocket, an atomic coordinate sequence of the pocket, and an atomic-level feature sequence of the pocket, the inputting the pocket dataset into the encoder of the molecular prediction model to obtain the second output vector includes: embedding the atomic type sequence of the pocket to obtain an atomic type representation with a fifth feature dimension; embedding the atomic coordinate sequence of the pocket to obtain an atomic coordinate representation with a sixth feature dimension; concatenating the atomic type representation with the fifth feature dimension and the atomic coordinate representation with the sixth feature dimension to obtain an atomic type / coordinate representation with a feature dimension H; embedding the atomic-level feature sequence of the pocket to obtain an atomic-level feature representation with a seventh feature dimension, and performing feature projection on the atomic-level feature representation with the seventh feature dimension to obtain a pocket other representation with the feature dimension H; and adding the atomic type / coordinate representation with the feature dimension H and the pocket other representation with the feature dimension H in position to obtain a pocket representation with the feature dimension H, and inputting the pocket representation with the feature dimension H into the encoder to obtain the second output vector.

[0019] The second output vector obtained in this way can be used to train the molecular prediction model in the third application scenario, and the trained model has better performance.

[0020] In some optional embodiments, the small molecule structure dataset comprises a FSMILES type sequence of the molecule and a FSMILES coordinate sequence of the molecule; inputting the first output vector, the property representation, and the small molecule structure dataset into the decoder of the molecular prediction model comprises: embeddingly encoding the FSMILES type sequence of the molecule to obtain a FSMILES type representation with an eighth dimension of feature dimension; embeddingly encoding the FSMILES coordinate sequence of the molecule to obtain a FSMILES coordinate representation with a ninth dimension of feature dimension; concatenating the FSMILES type representation with the eighth dimension of feature dimension and the FSMILES coordinate representation with the ninth dimension of feature dimension to obtain a FSMILES type / coordinate representation with an H dimension of feature dimension; concatenating the FSMILES type / coordinate representation with the H dimension of feature dimension and the property representation in the dimension of sequence to obtain a ligand representation; inputting the first output vector and the ligand label into the decoder of the molecular prediction model; and / or inputting the second output vector, the property representation, and the small molecule structure dataset into the decoder of the molecular prediction model comprises: embeddingly encoding the FSMILES type sequence of the molecule to obtain a FSMILES type representation with a tenth dimension of feature dimension; embeddingly encoding the FSMILES coordinate sequence of the molecule to obtain a FSMILES coordinate representation with an eleventh dimension of feature dimension; concatenating the FSMILES type representation with the tenth dimension of feature dimension and the FSMILES coordinate representation with the eleventh dimension of feature dimension to obtain a FSMILES type / coordinate representation with an H dimension of feature dimension; concatenating the FSMILES type / coordinate representation with the H dimension of feature dimension and the property representation in the dimension of sequence to obtain a ligand representation; inputting the second output vector and the ligand label into the decoder of the molecular prediction model.

[0021] Thus, the molecular prediction model can be trained in the first application scenario and the third application scenario.

[0022] In some optional embodiments, before inputting the first output vector, the property representation, and the small molecule structure dataset into the decoder of the molecular prediction model, or before training the molecular prediction model after inputting the first output vector, the property representation, and the small molecule structure dataset into the decoder of the molecular prediction model, the method further comprises: obtaining a small molecule; dividing the small molecule into a plurality of fragments according to a preset cutting rule; calculating the ring information of each atom in each fragment, and determining the element type of each atom in each fragment and whether each atom has aromaticity to obtain a FSMILES sequence of the small molecule; and obtaining the small molecule structure dataset according to a plurality of FSMILES sequences of a plurality of small molecules.

[0023] Therefore, the ring system in which each atom in the small molecule is located can be determined, so that the quality of the generated 3D molecule is improved to a certain extent, the model learning difficulty is reduced, and the molecule generation capability is improved.

[0024] In some optional embodiments, dividing the small molecule into multiple fragments according to the preset cutting rule comprises: selecting all single bonds in the small molecule that meet the conditions of not belonging to a part of a ring, not being connected to a hydrogen atom, and being connected to a ring at at least one end; and dividing the small molecule into multiple fragments at the selected single bonds.

[0025] This is because the molecule is composed of multiple fragments generated by the molecule prediction model, and is more interpretable, so that the molecule generation capability of the molecule prediction model can be improved.

[0026] In some optional embodiments, the training data set comprises a force field accuracy training data set and an experimental accuracy training data set; wherein the force field accuracy training data set and the experimental accuracy training data set comprise multiple application scenarios and a training subset of force field accuracy corresponding to each application scenario; and the experimental accuracy training data set comprises multiple application scenarios and a training subset of experimental accuracy corresponding to each application scenario.

[0027] That is, a multi-stage training method is adopted, in stage 1, the comprehensive model shown in FIG. 7 is trained by using a force field accuracy training data set of a million scale; in stage 2, the comprehensive model obtained in stage 1 is fine-tuned by using an experimental accuracy training data set of 10,000 scale. Therefore, the problem that a deep learning model needs a large amount of training data can be solved.

[0028] In a second aspect, the present application further provides a protein pocket processing method, comprising the following steps: obtaining to-be-matched protein pocket data; inputting the to-be-matched protein pocket data into a molecule prediction model trained by using the method of the first aspect or any one of the corresponding embodiments thereof, to obtain small molecule structure information and / or anchor point information.

[0029] The protein pocket processing method provided by the embodiments of the present application can be applied to various scenes by obtaining to-be-matched protein pocket data, inputting the to-be-matched protein pocket data into a molecule prediction model, and obtaining small molecule structure information and / or anchor point information.

[0030] In some optional embodiments, the inputting the protein pocket data to be matched into the molecular prediction model trained by the method of the first aspect or any of the corresponding embodiments thereof, the obtaining the small molecule structure information comprises: inputting the protein pocket data to be matched into an encoder of the molecular prediction model to obtain a third output vector; obtaining attribute information of a small molecule to be matched; inputting the attribute information of the small molecule to be matched into an attribute controller of the molecular prediction model to obtain a third attribute representation; and inputting the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information.

[0031] In this way, the molecular prediction model can be applied to a reasoning scenario in which a user only has a pocket PDB file and expects a molecule attribute, and hopes to generate a 3D molecule suitable for the pocket.

[0032] In some optional embodiments, the inputting the protein pocket data to be matched into the molecular prediction model trained by the method of the first aspect or any of the corresponding embodiments thereof, the obtaining the small molecule structure information comprises: obtaining anchor point information corresponding to the protein pocket data to be matched; inputting the protein pocket data to be matched and the anchor point information corresponding to the protein pocket data to be matched into an encoder of the molecular prediction model to obtain the third output vector; obtaining attribute information of a small molecule to be matched; inputting the attribute information of the small molecule to be matched into an attribute controller of the molecular prediction model to obtain the third attribute representation; and inputting the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information.

[0033] In this way, the molecular prediction model can be applied to a reasoning scenario in which a user has a pocket PDB file, is clear about information of NCIs and anchor points on the pocket, and is clear about an expected molecule attribute, and hopes to generate a 3D molecule suitable for the pocket.

[0034] In some optional embodiments, the inputting the protein pocket data to be matched into the molecular prediction model trained by the method of the first aspect or any of the corresponding embodiments thereof, the obtaining the anchor point information comprises: inputting the protein pocket data to be matched into an encoder of the molecular prediction model to obtain the anchor point information.

[0035] In this way, the molecular prediction model can be applied to a reasoning scenario in which a user only has a pocket PDB file and hopes to generate anchor point information.

[0036] In a third aspect, the embodiments of the present application further provide a device for training a molecular prediction model, the device comprising a first obtaining module and a training module; the first obtaining module is configured to obtain a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each of the application scenarios; and the training module is configured to train the molecular prediction model using each of the training subsets in sequence.

[0037] In a fourth aspect, the embodiments of the present application further provide a device for processing a protein pocket, the device comprising a second obtaining module and a prediction module; the second obtaining module is configured to obtain protein pocket data to be matched; and the prediction module is configured to input the protein pocket data to be matched into the molecular prediction model trained by the method of the first aspect or any of the corresponding embodiments thereof, to obtain small molecule structure information and / or anchor point information.

[0038] In a fifth aspect, the present application provides a computer device, comprising a memory and a processor, which are communicatively connected with each other, and the memory stores computer instructions; the processor executes the computer instructions, thereby performing the training method of the molecular prediction model of the above aspect or any of the corresponding embodiments thereof, and / or the processing method of the protein pocket of the above aspect or any of the corresponding embodiments thereof.

[0039] In a sixth aspect, the present application provides a computer readable storage medium, which stores computer instructions for causing a computer to execute the training method of the molecular prediction model of the above aspect or any of the corresponding embodiments thereof, and / or the processing method of the protein pocket of the above aspect or any of the corresponding embodiments thereof.

[0040] In a seventh aspect, the present application provides a computer program product comprising computer instructions for causing a computer to execute the training method of the molecular prediction model of the above aspect or any of the corresponding embodiments thereof, and / or the processing method of the protein pocket of the above aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0041] In order to more clearly illustrate the specific embodiments of the present application or the technical solutions in the prior art, the drawings needed in the specific embodiments or prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0042] FIG. 1 is a flowchart of the training method of the molecular prediction model according to the embodiments of the present application;

[0043] FIG. 2 is a flow chart of a training method of another molecular prediction model according to an embodiment of the present application;

[0044] FIG. 3 is a flow chart of an example of a training method of a molecular prediction model according to an embodiment of the present application;

[0045] FIG. 4 is a flow chart of a training method of yet another molecular prediction model according to an embodiment of the present application;

[0046] FIG. 5 is a schematic diagram of FSMILES encoding according to an embodiment of the present application;

[0047] FIG. 6 is a flow chart of a training method of still another molecular prediction model according to an embodiment of the present application;

[0048] FIG. 7 is a schematic diagram of a two-stage training process according to an embodiment of the present application;

[0049] FIG. 8 is a flow chart of a processing method of a protein pocket according to an embodiment of the present application;

[0050] FIG. 9 is a structural block diagram of a training device of a molecular prediction model according to an embodiment of the present application;

[0051] FIG. 10 is a flow chart of a processing device of a protein pocket according to an embodiment of the present application;

[0052] FIG. 11 is a schematic diagram of a hardware structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will be combined with the accompanying drawings to make a clear and complete description of the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0054] According to an embodiment of the present application, a training method of a molecular prediction model is provided. It should be noted that the steps shown in the flow chart can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flow chart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0055] In the present embodiment, a training method of a molecular prediction model is provided, which can be used in a computer device. FIG. 1 is a flow chart of a training method of a molecular prediction model according to an embodiment of the present application, as shown in FIG. 1, the flow includes the following steps:

[0056] Step S101: obtaining a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each application scenario.

[0057] In an optional embodiment, the training data set comprises a first application scenario, a second application scenario and a third application scenario. The first training subset corresponding to the first application scenario comprises a pocket data set, an anchor point condition set, a small molecule structure data set and a small molecule attribute information set; the second training subset corresponding to the second application scenario comprises a pocket data set and an anchor point condition set; and the third training subset corresponding to the third application scenario comprises a pocket data set, a small molecule structure data set and a small molecule attribute information set.

[0058] Specifically, the first application scenario can be understood as a reasoning scenario in which a user has a pocket PDB file, is clear about the information of NCIs and anchor points on the pocket, and is clear about the expected molecular attributes, and hopes to generate a 3D molecule suitable for the pocket.

[0059] The second application scenario can be understood as a reasoning scenario in which a user only has a pocket PDB file and expected molecular attributes, first predicts NCIs / anchor points through a model, and then generates a 3D molecule suitable for the pocket in the manner of the first scenario.

[0060] The third application scenario can be understood as a reasoning scenario in which a user only has a pocket PDB file and expected molecular attributes, and hopes to generate a 3D molecule suitable for the pocket.

[0061] Step S102: sequentially training a molecular prediction model by using each training subset.

[0062] The training method of the molecular prediction model provided in the embodiment can make the trained molecular prediction model applicable to multiple scenarios.

[0063] In the embodiment, a training method of a molecular prediction model is provided, which can be used for a computer device. FIG. 2 is a flowchart of another training method of a molecular prediction model according to the embodiment of the present application, and FIG. 3 is a flowchart of an example of the training method of the molecular prediction model according to the embodiment of the present application. As shown in FIGS. 2 and 3, the flowchart comprises the following steps:

[0064] Step S201: obtaining a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each application scenario. For details, refer to step S101 of the embodiment shown in FIG. 1, which will not be described here again.

[0065] Step S202: training the molecular prediction model by using the first training subset.

[0066] Specifically, the molecular prediction model comprises an encoder, a decoder and an attribute controller. The encoder is configured to capture features of the pocket; the decoder is configured to output a topological structure and a corresponding 3D conformation of the molecule; and the attribute controller is configured to control a desired attribute of the generated molecule. Thus, the attribute of the small molecule can be controlled.

[0067] As described above, the first training subset corresponding to the first application scenario comprises a pocket dataset, an anchor condition set, a small molecule structure dataset and a small molecule attribute information set.

[0068] In an optional embodiment, training the molecular prediction model by using the first training subset comprises steps S2011-S2013.

[0069] Step S2021: inputting the pocket dataset and the anchor condition set into the encoder of the molecular prediction model to obtain a first output vector.

[0070] In an optional embodiment, the pocket dataset comprises a sequence of atom types of the pocket, a sequence of atomic coordinates of the pocket, and a sequence of atomic-level features of the pocket. That is, based on the input PDB file of the pocket, after specific processing, three sequences can be obtained: the sequence of atom types of the pocket, the sequence of atomic coordinates of the pocket, and the sequence of atomic-level features of the pocket. If the number of atoms on the pocket is less than 500, the above three sequences are completed to a length of 500 by filling 0; if the number of atoms on the pocket exceeds 500, the above three sequences are truncated to a length of 500 by taking the first 500.

[0071] Specifically, the sequence of atom types of the pocket is read in sequence according to the order of heavy atoms in the PDB file and stored in the form of a sequence. The sequence of atomic coordinates of the pocket is read in sequence according to the order of heavy atoms in the PDB file and stored in the form of a sequence. The sequence of atomic-level features of the pocket is read in sequence according to the order of heavy atoms in the PDB file and stored in the form of a sequence.

[0072] In an optional embodiment, after special processing of the NCI / anchor information on the pocket, a sequence of NCI / anchors of the pocket can be obtained. The anchor condition set comprises a plurality of sequences of NCI / anchors of the pocket.

[0073] Specifically, the NCI / anchor sequence of the pocket is obtained according to the rule that the position corresponding to the heavy atom of both NCI and anchor is 3, the position corresponding to the heavy atom of only NCI is 1, the position corresponding to the heavy atom of only anchor is 2, and the remaining positions are 0, based on the input NCI and anchor information, according to the order of heavy atoms in the PDB file.

[0074] Specifically, the pocket dataset and the anchor condition set are input into the encoder of the molecular prediction model to obtain the first output vector, including: embedding coding the atom type sequence of the pocket to obtain an atom type representation with a feature dimension of a first dimension (for example, 0.25H); embedding coding the atom coordinate sequence of the pocket to obtain an atom coordinate representation with a feature dimension of a second dimension (for example, 0.75H); concatenating the atom type representation with the feature dimension of the first dimension and the atom coordinate representation with the feature dimension of the second dimension to obtain an atom type / coordinate representation with a feature dimension of H; embedding coding the feature sequence at the atom level of the pocket to obtain an atom level feature representation with a feature dimension of a third dimension (for example, 0.75H); embedding coding the anchor condition set to obtain an NCI / anchor representation with a feature dimension of a fourth dimension (for example, 0.25H); concatenating the atom level feature representation with the feature dimension of the third dimension and the NCI / anchor representation with the feature dimension of the fourth dimension to obtain other pocket representations with a feature dimension of H; aligning and adding the atom type / coordinate representation with the feature dimension of H and the other pocket representations with the feature dimension of H to obtain a pocket representation with a feature dimension of H, and inputting the pocket representation with the feature dimension of H into the encoder to obtain the first output vector.

[0075] wherein H represents the dimension of the feature, and assuming that H is 512, 0.25H is 128.

[0076] In an optional implementation, the encoder adopts a multi-layer multi-head self-attention mechanism similar to the Transformer to reprocess the input vector to obtain the output vector of the encoder. At this time, since the input contains the NCI / anchor information of the pocket, it is not necessary to predict the NCI / anchor again.

[0077] Step S2022: inputting the small molecule attribute information set into the attribute controller of the molecular prediction model to obtain an attribute representation.

[0078] Step S2023: inputting the first output vector, the attribute representation, and the small molecule structure dataset into the decoder of the molecular prediction model to train the molecular prediction model.

[0079] In an optional implementation, the small molecule structure dataset includes a FSMILES type sequence of the molecule and a FSMILES coordinate sequence of the molecule.

[0080] Specifically, the FSMILES type sequence of the molecule is obtained by splitting the small molecule according to the rules of FSMILES; and the FSMILES coordinate sequence of the molecule is obtained by splitting the small molecule according to the rules of FSMILES, and the 3D space coordinates (x, y, z) corresponding to each FSMILES code are stored in the form of a sequence.

[0081] Specifically, inputting the first output vector, the attribute representation and the small molecule structure dataset into the decoder of the molecular prediction model includes: embedding the FSMILES type sequence of the molecule to obtain the FSMILES type representation with the eighth dimension (for example, 0.25H) of the feature dimension; embedding the FSMILES coordinate sequence of the molecule to obtain the FSMILES coordinate representation with the ninth dimension (for example, 0.75H) of the feature dimension; splicing the FSMILES type representation with the eighth dimension of the feature dimension and the FSMILES coordinate representation with the ninth dimension of the feature dimension to obtain the FSMILES type / coordinate representation with the feature dimension of H; splicing the FSMILES type / coordinate representation with the feature dimension of H and the attribute representation in the dimension of the sequence to obtain the ligand representation; and inputting the first output vector and the ligand label into the decoder of the molecular prediction model.

[0082] Step S203: After training the molecular prediction model by using the first training subset, the molecular prediction model is trained by using the second training subset.

[0083] As described above, the second training subset corresponding to the second application scenario includes the pocket dataset and the anchor condition set. In an optional implementation, training the molecular prediction model by using the second training subset includes: inputting the pocket dataset and the anchor condition set into the encoder of the molecular prediction model to train the molecular prediction model.

[0084] Step S204: After training the molecular prediction model by using the second training subset, the molecular prediction model is trained by using the third training subset.

[0085] As described above, the third training subset corresponding to the third application scenario includes the pocket dataset, the small molecule structure dataset and the small molecule attribute information set.

[0086] In an optional implementation, training the molecular prediction model by using the third training subset includes steps S2041-S2043.

[0087] Step S2041: inputting the pocket dataset into the encoder of the molecular prediction model to obtain the second output vector.

[0088] In an alternative embodiment, the pocket dataset includes a sequence of atom types of the pocket; a sequence of atom coordinates of the pocket; a sequence of atom-level features of the pocket. That is, based on the input pocket PDB file, after specific processing, three sequences can be obtained: a sequence of atom types of the pocket; a sequence of atom coordinates of the pocket; a sequence of atom-level features of the pocket. If the number of atoms on the pocket is less than 500, the above three sequences are completed to a length of 500 by filling 0; if the number of atoms on the pocket exceeds 500, the above three sequences are truncated to a length of 500 by taking the first 500.

[0089] Specifically, inputting the pocket dataset into the encoder of the molecular prediction model to obtain the second output vector includes: embedding the sequence of atom types of the pocket to obtain an atom type representation with a feature dimension of a fifth dimension (e.g., 0.25H); embedding the sequence of atom coordinates of the pocket to obtain an atom coordinate representation with a feature dimension of a sixth dimension (e.g., 0.75H); concatenating the atom type representation with the feature dimension of the fifth dimension and the atom coordinate representation with the feature dimension of the sixth dimension to obtain an atom type / coordinate representation with a feature dimension of H; embedding the sequence of atom-level features of the pocket to obtain an atom-level feature representation with a feature dimension of a seventh dimension (e.g., 0.75H), and performing feature projection on the atom-level feature representation with the feature dimension of the seventh dimension to obtain a pocket other representation with a feature dimension of H; aligning and adding the atom type / coordinate representation with the feature dimension of H and the pocket other representation with the feature dimension of H to obtain a pocket representation with a feature dimension of H, and inputting the pocket representation with the feature dimension of H into the encoder to obtain the second output vector.

[0090] As described above, the encoder adopts a multi-layer multi-head self-attention mechanism similar to the Transformer to reprocess the input vector to obtain the output vector of the encoder. At this time, based on the output of the encoder, it can be predicted which site on the current input pocket sequence is the NCI site and which site is the anchor site.

[0091] Step S2042: inputting the small molecule attribute information set into the attribute controller of the molecular prediction model to obtain an attribute representation.

[0092] The input of the attribute controller is the attribute that the 3D molecule to be generated should have, such as activity, QED (Quantitative Estimate of Drug-likeness), SAS (Synthetic Accessibility score), etc.

[0093] The rule for activity discretization is to divide the small molecule activity into 4 grades: (1) activity less than 100 nm: high activity; (2) activity greater than 100 nm but less than 1000 nm: medium activity; (3) activity greater than 1000 nm: low activity; (4) no measured activity value: unknown activity.

[0094] The rule for QED discretization is to divide the small molecule QED into 2 grades: (1) QED greater than 0.3, attribute value is 1; (2) QED less than or equal to 0.3, attribute value is 0.

[0095] The rule for SAS discretization is to divide the small molecule SAS into 2 grades: (1) SAS greater than or equal to 5, attribute value is 0; (2) SAS less than 5, attribute value is 1.

[0096] The discretized attribute values are extracted as attribute representations in the form of a multi-layer perception, and then used as the initial condition for decoding by the decoder.

[0097] Step S2043: inputting the second output vector, the attribute representation, and the small molecule structure data set into the decoder of the molecular prediction model to train the molecular prediction model.

[0098] In an optional implementation, the small molecule structure data set includes an FSMILES type sequence of the molecule and an FSMILES coordinate sequence of the molecule. If the FSMILES Token of the small molecule is less than 100, the above two sequences are completed to a length of 100 in the form of 0 filling; if the FSMILES Token of the small molecule exceeds 100, the above two sequences are truncated to a length of 100 by taking the first 100.

[0099] Specifically, inputting the second output vector, the attribute representation, and the small molecule structure data set into the decoder of the molecular prediction model includes: embeddingly encoding the FSMILES type sequence of the molecule to obtain an FSMILES type representation with a feature dimension of a tenth dimension (for example, 0.25H); embeddingly encoding the FSMILES coordinate sequence of the molecule to obtain an FSMILES coordinate representation with a feature dimension of an eleventh dimension (for example, 0.75H); concatenating the FSMILES type representation with the feature dimension of the tenth dimension and the FSMILES coordinate representation with the feature dimension of the eleventh dimension to obtain an FSMILES type / coordinate representation with a feature dimension of H; concatenating the FSMILES type / coordinate representation with the feature dimension of H and the attribute representation in the dimension of the sequence to obtain a ligand representation; and inputting the second output vector and the ligand label into the decoder of the molecular prediction model.

[0100] In an alternative embodiment, the decoder employs a multi-layer multi-head masked self-attention mechanism and a cross-modal attention mechanism to rework the ligand representation, obtaining the output of the decoder.

[0101] The embodiment of the present application proposes an integrated model comprising an encoder, a decoder, and an attribute controller, which can effectively guide the generation of molecular attributes; and proposes a multi-scene training method, in which three training tasks are alternately performed, so that the molecular prediction model trained in this way can adaptively generate molecules according to different scenes.

[0102] In the embodiment, a training method of a molecular prediction model is provided, which can be used for a computer device. FIG. 4 is a flowchart of another training method of a molecular prediction model according to the embodiment of the present application. As shown in FIG. 4, the flowchart comprises the following steps:

[0103] Step S401: determining a small molecule structure data set corresponding to a first application scenario.

[0104] In an alternative embodiment, the step of determining a small molecule structure data set corresponding to a first application scenario comprises steps S4011-S4014.

[0105] Step S4011: obtaining a small molecule corresponding to the first application scenario.

[0106] Step S4012: dividing the small molecule corresponding to the first application scenario into multiple fragments according to a preset cutting rule.

[0107] Step S4013: calculating the ring information of each atom in each fragment, and determining the element type of each atom in each fragment and whether each atom has aromaticity, to obtain an FSMILES sequence of the small molecule corresponding to the first application scenario.

[0108] Step S4014: obtaining a small molecule structure data set corresponding to the first application scenario from multiple FSMILES sequences of multiple small molecules corresponding to the first application scenario.

[0109] For example, as shown in FIG. 5, the FSMILES sequence of the small molecule can be obtained through FSMILES encoding, which comprises the following steps.

[0110] a. The small molecule fragment cutting process in FSMILES needs to select single bonds that meet specific criteria, i.e., all single bonds that meet the conditions of not belonging to a ring, not connected to a hydrogen atom, and at least one end connected to a ring, and then the molecule is broken at these single bonds. This cutting process helps to divide the small molecule into multiple fragments.

[0111] b. After the small molecules are divided into multiple fragments according to the cutting rule, the ring information of each atom in the fragment is calculated, that is, firstly, the number of rings in which the atom is located is calculated, then the size of each ring is calculated and sorted in ascending order according to the ring size, and finally the ring number is spliced with the symbol "+" to obtain the ring information of each atom.

[0112] c. The element type of each atom in the fragment is case sensitive according to whether it has aromaticity. The element type of each atom is spliced with the symbol "_" in the ring information calculated in step b, so as to obtain the FSMILES code of each atom. For example, as shown in FIG. 5, 'c_5+6' means that the aromatic carbon atom is located at the connection point of the 5-membered ring and the 6-membered ring. 'C_0' means that the aliphatic carbon atom is not on the ring.

[0113] According to the FSMILES code rule, any small molecule can be represented in the form of FSMILES sequence.

[0114] The embodiment proposes a set of molecular coding rules named FSMILES, which focuses on describing the ring system in which each atom in the small molecule is located, so as to improve the quality of the generated 3D molecule to a certain extent, reduce the model learning difficulty, and improve the molecule generation capability.

[0115] Step S402: Determine a small molecule structure data set corresponding to the third application scenario.

[0116] In an optional embodiment, determining the small molecule structure data set corresponding to the third application scenario includes steps S4021-S4024.

[0117] Step S4021: Obtain a small molecule corresponding to the third application scenario.

[0118] Step S4022: Divide the small molecule corresponding to the third application scenario into multiple fragments according to a preset cutting rule.

[0119] Step S4023: Calculate the ring information of each atom in each fragment, and determine the element type of each atom in each fragment and whether it has aromaticity, to obtain the FSMILES sequence of the small molecule corresponding to the third application scenario.

[0120] Step S4024: Obtain a small molecule structure data set corresponding to the third application scenario according to multiple FSMILES sequences of multiple small molecules corresponding to the third application scenario.

[0121] Step S403: Obtain a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each application scenario.

[0122] In an optional implementation, the training data set comprises a first application scenario, a second application scenario, and a third application scenario. The first training subset corresponding to the first application scenario comprises a pocket data set, an anchor condition set, a small molecule structure data set, and a small molecule attribute information set. The second training subset corresponding to the second application scenario comprises a pocket data set and an anchor condition set. The third training subset corresponding to the third application scenario comprises a pocket data set, a small molecule structure data set, and a small molecule attribute information set.

[0123] Step S404: Train the molecular prediction model using the first training subset.

[0124] As described above, the first training subset corresponding to the first application scenario comprises a pocket data set, an anchor condition set, a small molecule structure data set, and a small molecule attribute information set. In an optional implementation, training the molecular prediction model using the first training subset comprises steps S4041-S4043.

[0125] Step S4041: Input the pocket data set and the anchor condition set into the encoder of the molecular prediction model to obtain a first output vector. For details, please refer to step S2021 of the embodiment shown in FIG. 2, which will not be described here again.

[0126] Step S4042: Input the small molecule attribute information set into the attribute controller of the molecular prediction model to obtain an attribute representation.

[0127] Step S4043: Input the first output vector, the attribute representation, and the small molecule structure data set corresponding to the first application scenario into the decoder of the molecular prediction model to train the molecular prediction model. For details, please refer to step S2023 of the embodiment shown in FIG. 2, which will not be described here again.

[0128] Step S405: After training the molecular prediction model using the first training subset, train the molecular prediction model using the second training subset.

[0129] As described above, the second training subset corresponding to the second application scenario comprises a pocket data set and an anchor condition set. In an optional implementation, training the molecular prediction model using the second training subset comprises inputting the pocket data set and the anchor condition set into the encoder of the molecular prediction model to train the molecular prediction model.

[0130] Step S406: After training the molecular prediction model using the second training subset, train the molecular prediction model using the third training subset.

[0131] As described above, the third training subset corresponding to the third application scenario includes the pocket dataset, the small molecule structure dataset, and the small molecule attribute information set. In an optional embodiment, training the molecular prediction model using the third training subset includes steps S4061-S4063.

[0132] Step S4061: inputting the pocket dataset into the encoder of the molecular prediction model to obtain a second output vector. For details, refer to step S2031 of the embodiment shown in FIG. 2, which will not be described here again.

[0133] Step S4062: inputting the small molecule attribute information set into the attribute controller of the molecular prediction model to obtain an attribute representation.

[0134] Step S4063: inputting the second output vector, the attribute representation, and the small molecule structure dataset corresponding to the third application scenario into the decoder of the molecular prediction model to train the molecular prediction model. For details, refer to step S2033 of the embodiment shown in FIG. 2, which will not be described here again.

[0135] Deep learning models require a large amount of training data, and the experimental data that can be collected through public channels is limited. Based on this, a training method of a molecular prediction model is provided in this embodiment, which can be used in a computer device. FIG. 6 is a flowchart of another training method of a molecular prediction model according to an embodiment of the present application, and FIG. 7 is a schematic diagram of a two-stage training process according to an embodiment of the present application. As shown in FIGS. 6 and 7, the process includes the following steps:

[0136] Step S601: obtaining a force field accuracy training subset, wherein the force field accuracy training dataset and the experimental accuracy training dataset include a plurality of application scenarios and a force field accuracy training subset corresponding to each application scenario.

[0137] In an optional embodiment, the force field accuracy training dataset is obtained by the following method: the present application collects millions of small molecules with activity records from the ChemBL database, and then constructs data through existing computational chemistry methods such as molecular docking, and calculates the attribute values of the small molecules such as activity, QED, and SAS. That is, the force field accuracy training dataset is obtained by computational chemistry methods.

[0138] In an optional embodiment, the experimental accuracy training dataset is obtained by the following method: more than 10,000 pocket-small molecule experimental accuracy data are collected from the public PDBind dataset, and the attribute data of the corresponding small molecules are also collected.

[0139] Step S602: training the molecular prediction model using each force field accuracy training subset in turn.

[0140] Step S603: obtaining an experimental precision training subset, wherein the force field precision training subset corresponding to each application scenario; the experimental precision training data set comprises a plurality of application scenarios and the experimental precision training subset corresponding to each application scenario.

[0141] Step S604: training the molecular prediction model by using each force field precision training subset in sequence.

[0142] That is, the embodiment of the present application adopts a multi-stage training method, in stage 1, the present application adopts a million-scale force field precision training data set to train the comprehensive model as shown in FIG. 7; in stage 2, the present application adopts a 10,000-scale experimental precision training data set to fine-tune the comprehensive model obtained in stage 1.

[0143] Therefore, the embodiment of the present application adopts a deep neural network scheme based on supervised training, constructs a model based on an attention mechanism, and learns by using a training data set collected by computational chemistry and the like, wherein each small molecule has but is not limited to activity, QED, SAS, molecular weight and the like. In order to be able to construct a model capable of generating 3D molecules under a given pocket and capable of artificially controlling the generated molecular properties, and to adapt to the demand of multi-scene 3D molecule generation, after a lot of analysis and experiments, we propose a comprehensive model including an encoder, a decoder and an attribute controller.

[0144] In order to better depict small molecules, the present application proposes a set of molecular encoding rules named FSMILES, which focuses on depicting the ring system in which each atom in the small molecule is located, thereby improving the quality of the generated 3D molecules to a certain extent.

[0145] In order to overcome the problem of less high-quality experimental precision data, the embodiment of the present application adopts a multi-stage training method, in stage 1, the present application adopts a million-scale force field precision training data set to train the comprehensive model as shown in FIG. 7; in stage 2, the present application adopts a 10,000-scale experimental precision training data set to fine-tune the comprehensive model obtained in stage 1.

[0146] In the present embodiment, a protein pocket processing method is provided, which can be used in a computer device. FIG. 8 is a flowchart of the protein pocket processing method according to the embodiment of the present application, as shown in FIG. 8, the flowchart comprises the following steps:

[0147] Step S801: obtaining protein pocket data to be matched.

[0148] Step S802: inputting the protein pocket data to be matched into a molecular prediction model to obtain small molecule structure information and / or anchor point information.

[0149] In an alternative embodiment, inputting the to-be-matched protein pocket data into the molecular prediction model to obtain the small molecule structure information comprises the following steps: inputting the to-be-matched protein pocket data into an encoder of the molecular prediction model to obtain a third output vector; obtaining to-be-matched small molecule attribute information; inputting the to-be-matched small molecule attribute information into an attribute controller of the molecular prediction model to obtain a third attribute representation; and inputting the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information. That is, the decoder outputs the Token type of the next FSMILES, and the bond length, bond angle, and dihedral angle corresponding to the Token in a self-recurrent manner according to the two inputs. The decoder repeats the process until the molecule is generated.

[0150] In another alternative embodiment, inputting the to-be-matched protein pocket data into the molecular prediction model to obtain the small molecule structure information comprises the following steps: obtaining anchor point information corresponding to the to-be-matched pocket data; inputting the to-be-matched pocket data and the anchor point information corresponding to the to-be-matched pocket data into an encoder of the molecular prediction model to obtain the third output vector; obtaining to-be-matched small molecule attribute information; inputting the to-be-matched small molecule attribute information into an attribute controller of the molecular prediction model to obtain the third attribute representation; and inputting the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information. That is, the decoder outputs the Token type of the next FSMILES, and the bond length, bond angle, and dihedral angle corresponding to the Token in a self-recurrent manner according to the two inputs. The decoder repeats the process until the molecule is generated.

[0151] In yet another alternative embodiment, inputting the to-be-matched protein pocket data into the molecular prediction model to obtain the anchor point information comprises the following steps: inputting the to-be-matched protein pocket data into an encoder of the molecular prediction model to obtain the anchor point information.

[0152] The present embodiment also provides a training device of a molecular prediction model, which is used to implement the above-mentioned embodiments and preferred embodiments, and details thereof have been described above. As used below, the term “module” can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, implementation of hardware, or a combination of software and hardware, is also possible and contemplated.

[0153] The present embodiment provides a training device of a molecular prediction model, as shown in FIG. 9, which comprises:

[0154] The acquisition module 901 is configured to acquire a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each application scenario.

[0155] The training module 902 is configured to train the molecular prediction model by using each training subset in sequence.

[0156] In some optional embodiments, the training data set comprises a first application scenario, a second application scenario and a third application scenario. The first training subset corresponding to the first application scenario comprises a pocket data set, an anchor condition set, a small molecule structure data set and a small molecule attribute information set; the second training subset corresponding to the second application scenario comprises the pocket data set and the anchor condition set; and the third training subset corresponding to the third application scenario comprises the pocket data set, the small molecule structure data set and the small molecule attribute information set.

[0157] In some optional embodiments, the training module 902 comprises a first training unit, a second training unit and a third training unit. The first training unit is configured to train the molecular prediction model by using the first training subset; the second training unit is configured to train the molecular prediction model by using the second training subset after training the molecular prediction model by using the first training subset; and the third training unit is configured to train the molecular prediction model by using the third training subset after training the molecular prediction model by using the second training subset.

[0158] In some optional embodiments, the first training unit comprises a first encoder subunit, a first attribute controller subunit and a first decoder subunit. The first encoder subunit is configured to input the pocket data set and the anchor condition set into an encoder of the molecular prediction model to obtain a first output vector; the first attribute controller subunit is configured to input the small molecule attribute information set into an attribute controller of the molecular prediction model to obtain an attribute representation; and the first decoder subunit is configured to input the first output vector, the attribute representation and the small molecule structure data set into a decoder of the molecular prediction model to train the molecular prediction model.

[0159] In some optional embodiments, the second training unit is specifically configured to input the pocket data set and the anchor condition set into the encoder of the molecular prediction model to train the molecular prediction model.

[0160] In some optional embodiments, the third training unit comprises a second encoder subunit, a second attribute controller subunit, and a second decoder subunit. The second encoder subunit is configured to input the pocket dataset into an encoder of the molecular prediction model to obtain a second output vector; the second attribute controller subunit is configured to input the small molecule attribute information set into an attribute controller of the molecular prediction model to obtain attribute representation; and the second decoder subunit is configured to input the second output vector, the attribute representation, and the small molecule structure dataset into a decoder of the molecular prediction model to train the molecular prediction model.

[0161] In some optional embodiments, the first encoder subunit is specifically configured to: perform embedded coding on a sequence of atom types of the pocket to obtain atom type representation with a first dimension of feature; perform embedded coding on a sequence of atom coordinates of the pocket to obtain atom coordinate representation with a second dimension of feature; concatenate the atom type representation with the first dimension of feature and the atom coordinate representation with the second dimension of feature to obtain atom type / coordinate representation with a third dimension of feature; perform embedded coding on a sequence of atom-level features of the pocket to obtain atom-level feature representation with a fourth dimension of feature; perform embedded coding on the anchor condition set to obtain NCI / anchor representation with a fifth dimension of feature; concatenate the atom-level feature representation with the fourth dimension of feature and the NCI / anchor representation with the fifth dimension of feature to obtain other pocket representation with the third dimension of feature; and perform position addition on the atom type / coordinate representation with the third dimension of feature and the other pocket representation with the third dimension of feature to obtain pocket representation with the third dimension of feature, and input the pocket representation with the third dimension of feature into the encoder to obtain the first output vector.

[0162] In some optional embodiments, the second encoder subunit is specifically configured to: perform embedded coding on a sequence of atom types of the pocket to obtain atom type representation with a sixth dimension of feature; perform embedded coding on a sequence of atom coordinates of the pocket to obtain atom coordinate representation with a seventh dimension of feature; concatenate the atom type representation with the sixth dimension of feature and the atom coordinate representation with the seventh dimension of feature to obtain atom type / coordinate representation with the third dimension of feature; perform embedded coding on a sequence of atom-level features of the pocket to obtain atom-level feature representation with an eighth dimension of feature, and perform feature projection on the atom-level feature representation with the eighth dimension of feature to obtain other pocket representation with the third dimension of feature; perform position addition on the atom type / coordinate representation with the third dimension of feature and the other pocket representation with the third dimension of feature to obtain pocket representation with the third dimension of feature, and input the pocket representation with the third dimension of feature into the encoder to obtain the second output vector.

[0163] In some optional embodiments, the first decoder subunit is specifically configured to: perform embedded coding on the FSMILES type sequence of the molecule to obtain a FSMILES type representation with an eighth feature dimension; perform embedded coding on the FSMILES coordinate sequence of the molecule to obtain a FSMILES coordinate representation with a ninth feature dimension; concatenate the FSMILES type representation with the eighth feature dimension and the FSMILES coordinate representation with the ninth feature dimension to obtain a FSMILES type / coordinate representation with a feature dimension of H; concatenate the FSMILES type / coordinate representation with the feature dimension of H and the attribute representation in the sequence dimension to obtain a ligand representation; and input the first output vector and the ligand label into the decoder of the molecular prediction model.

[0164] In some optional embodiments, the second decoder subunit is specifically configured to: perform embedded coding on the FSMILES type sequence of the molecule to obtain a FSMILES type representation with a tenth feature dimension; perform embedded coding on the FSMILES coordinate sequence of the molecule to obtain a FSMILES coordinate representation with an eleventh feature dimension; concatenate the FSMILES type representation with the tenth feature dimension and the FSMILES coordinate representation with the eleventh feature dimension to obtain a FSMILES type / coordinate representation with a feature dimension of H; concatenate the FSMILES type / coordinate representation with the feature dimension of H and the attribute representation in the sequence dimension to obtain a ligand representation; and input the second output vector and the ligand label into the decoder of the molecular prediction model.

[0165] In some optional embodiments, the training device of the molecular prediction model further comprises a small molecule prediction module. Before the first output vector, the attribute representation, and the small molecule structure data set are input into the decoder of the molecular prediction model, or before the molecular prediction model is trained after the first output vector, the attribute representation, and the small molecule structure data set are input into the decoder of the molecular prediction model, the small molecule prediction module is configured to: obtain a small molecule; divide the small molecule into a plurality of fragments according to a preset cutting rule; calculate the ring information of each atom in each fragment, and determine the element type of each atom in each fragment and whether each atom has aromaticity, to obtain a FSMILES sequence of the small molecule; and obtain a small molecule structure data set according to a plurality of FSMILES sequences of a plurality of small molecules.

[0166] In some optional embodiments, the small molecule prediction module comprises a small molecule obtaining unit, a fragment dividing unit, and a processing unit. The small molecule obtaining unit is configured to obtain small molecules. The fragment dividing unit is configured to divide the small molecules into a plurality of fragments according to a preset cutting rule. The processing unit is configured to: calculate ring information of each atom in each fragment, and determine an element type of each atom in each fragment and whether the atom has aromaticity, to obtain FSMILES sequences of the small molecules; and obtain a small molecule structure dataset according to the FSMILES sequences of the small molecules.

[0167] In some optional embodiments, the fragment dividing unit is specifically configured to: select all single bonds in the small molecule that meet the conditions of not belonging to a part of a ring, not being connected to a hydrogen atom, and being connected to a ring at at least one end; and divide the small molecule into a plurality of fragments at the selected single bonds.

[0168] In some optional embodiments, the training dataset comprises a force field accuracy training dataset and an experimental accuracy training dataset. The force field accuracy training dataset and the experimental accuracy training dataset comprise a plurality of application scenarios and a training subset of force field accuracy corresponding to each application scenario. The experimental accuracy training dataset comprises a plurality of application scenarios and a training subset of experimental accuracy corresponding to each application scenario.

[0169] In the present embodiment, a protein pocket processing apparatus is also provided, which is used to implement the above-described embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

[0170] The present embodiment provides a protein pocket processing apparatus, as shown in FIG. 10, comprising:

[0171] A second obtaining module 1001 is configured to obtain to-be-matched protein pocket data.

[0172] A prediction module 1002 is configured to input the to-be-matched protein pocket data into a molecular prediction model trained by the above-described method, to obtain small molecule structure information and / or anchor point information.

[0173] In some optional embodiments, the prediction module 1002 comprises a first prediction unit, where the first prediction unit is specifically configured to: input the to-be-matched protein pocket data into an encoder of the molecular prediction model to obtain a third output vector; obtain to-be-matched small molecule attribute information; input the to-be-matched small molecule attribute information into an attribute controller of the molecular prediction model to obtain a third attribute representation; and input the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information.

[0174] In some optional embodiments, the prediction module 1002 comprises a second prediction unit, where the second prediction unit is specifically configured to: obtain anchor point information corresponding to the to-be-matched pocket data; input the to-be-matched pocket data and the anchor point information corresponding to the to-be-matched pocket data into an encoder of the molecular prediction model to obtain the third output vector; obtain to-be-matched small molecule attribute information; input the to-be-matched small molecule attribute information into an attribute controller of the molecular prediction model to obtain the third attribute representation; and input the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information.

[0175] In some optional embodiments, the prediction module 1002 comprises a third prediction unit, where the third prediction unit is specifically configured to: input the to-be-matched protein pocket data into an encoder of the molecular prediction model to obtain the anchor point information.

[0176] Further function descriptions of the above-mentioned various modules and units are the same as those of the above-mentioned corresponding embodiments, and will not be described here again.

[0177] The training device of the molecular prediction model and / or the processing device of the protein pocket in this embodiment are presented in the form of functional units, where the units refer to ASIC (Application Specific Integrated Circuit, Application Specific Integrated Circuit) circuits, processors and memories executing one or more software or fixed programs, and / or other devices that can provide the above-mentioned functions.

[0178] The embodiment of the present application also provides a computer device having the above-mentioned training device of the molecular prediction model shown in FIG. 9 and / or the processing device of the protein pocket shown in FIG. 10.

[0179] Referring to FIG. 11, FIG. 11 is a structural diagram of a computer device according to an optional embodiment of the present application. As shown in FIG. 11, the computer device includes one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. The various components communicate with each other using different buses, and can be mounted on a common main board or mounted in other manners as needed. The processor can process instructions executed in the computer device, including instructions stored in the memory or on the memory to display graphical information of a GUI on an external input / output device, such as a display device coupled to the interface. In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory banks, if needed. Also, multiple computer devices can be connected, each providing part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). One processor 10 is taken as an example in FIG. 11.

[0180] The processor 10 can be a central processor, a network processor, or a combination thereof. The processor 10 can further include a hardware chip. The hardware chip can be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device can be a complex programmable logic device, a field programmable logic device, a general array logic, or any combination thereof.

[0181] The memory 20 stores instructions executable by the at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiments.

[0182] The memory 20 can include a program storage area and a data storage area. The program storage area can store an operating system and application programs required by at least one function; the data storage area can store data created according to the use of the computer device, and the like. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some optional embodiments, the memory 20 can optionally include a memory remotely arranged with respect to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0183] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state disk; and the memory 20 can further include a combination of the above kinds of memories.

[0184] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 can be connected by a bus or other means, and are connected by a bus in FIG. 11 as an example.

[0185] The input device 30 can receive inputted digital or character information, and generate key signal input related to user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), a tactile feedback device (e.g., a vibration motor), etc. The display device includes but is not limited to a liquid crystal display, a light emitting diode, a display, and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0186] The embodiments of the present application also provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium downloaded through a network and stored in a local storage medium, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk or a solid state disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that the computer, processor, microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, processor or hardware, the method shown in the above embodiments is implemented.

[0187] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can call or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc., and accordingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0188] While embodiments of the application have been described in connection with the preferred embodiments of the various figures, those of ordinary skill in the art will appreciate that various modifications and changes can be made without departing from the spirit and scope of the application, and that such modifications and changes fall within the scope of the appended claims.

Claims

1. A training method for a molecular prediction model, characterized in that, The method comprises: obtaining a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each of the application scenarios; training the molecular prediction model using each of the training subsets in turn; the training data set comprises: a first application scenario, wherein a first training subset corresponding to the first application scenario comprises a pocket data set, an anchor condition set, a small molecule structure data set, and a small molecule attribute information set; a second application scenario, wherein a second training subset corresponding to the second application scenario comprises the pocket data set and the anchor condition set; a third application scenario, wherein a third training subset corresponding to the third application scenario comprises the pocket data set, the small molecule structure data set, and the small molecule attribute information set; the training of the molecular prediction model using each of the training subsets in turn comprises: training the molecular prediction model using the first training subset; after training the molecular prediction model using the first training subset, training the molecular prediction model using the second training subset; after training the molecular prediction model using the second training subset, training the molecular prediction model using the third training subset.

2. The method of claim 1, wherein, the training of the molecular prediction model using the first training subset comprises: inputting the pocket data set and the anchor condition set into an encoder of the molecular prediction model to obtain a first output vector; inputting the small molecule attribute information set into an attribute controller of the molecular prediction model to obtain an attribute representation; inputting the first output vector, the attribute representation, and the small molecule structure data set into a decoder of the molecular prediction model to train the molecular prediction model; and / or, the training of the molecular prediction model using the second training subset comprises: training the molecular prediction model by inputting the pocket data set and the anchor condition set into an encoder of the molecular prediction model; and / or, the training of the molecular prediction model using the third training subset comprises: inputting the pocket data set into an encoder of the molecular prediction model to obtain a second output vector; inputting the small molecule attribute information set into an attribute controller of the molecular prediction model to obtain an attribute representation; inputting the second output vector, the attribute representation, and the small molecule structure data set into a decoder of the molecular prediction model to train the molecular prediction model.

3. The method of claim 2, wherein, the pocket data set comprises a sequence of atom types of a pocket, a sequence of atomic coordinates of the pocket, and a sequence of features at the atomic level of the pocket, and the inputting of the pocket data set and the anchor condition set into the encoder of the molecular prediction model to obtain the first output vector comprises: embeddingly encode the atomic type sequence of the pocket to obtain an atomic type representation with a feature dimension of a first dimension; embeddingly encode the atomic coordinate sequence of the pocket to obtain an atomic coordinate representation with a feature dimension of a second dimension; concatenate the atomic type representation with the feature dimension of the first dimension and the atomic coordinate representation with the feature dimension of the second dimension to obtain an atomic type / coordinate representation with a feature dimension of H; embeddingly encode the atomic-level feature sequence of the pocket to obtain an atomic-level feature representation with a feature dimension of a third dimension; embeddingly encode the anchor point condition set to obtain an NCI / anchor point representation with a feature dimension of a fourth dimension; concatenate the atomic-level feature representation with the feature dimension of the third dimension and the NCI / anchor point representation with the feature dimension of the fourth dimension to obtain other pocket representation with a feature dimension of H; add the atomic type / coordinate representation with the feature dimension of H and the other pocket representation with the feature dimension of H in position to obtain a pocket representation with a feature dimension of H, and input the pocket representation with the feature dimension of H into the encoder to obtain the first output vector.

4. The method of claim 2, wherein, The pocket data set includes an atomic type sequence of a pocket, an atomic coordinate sequence of the pocket, and an atomic-level feature sequence of the pocket, and the inputting of the pocket data set into the encoder of the molecular prediction model to obtain a second output vector includes: embeddingly encode the atomic type sequence of the pocket to obtain an atomic type representation with a feature dimension of a fifth dimension; embeddingly encode the atomic coordinate sequence of the pocket to obtain an atomic coordinate representation with a feature dimension of a sixth dimension; concatenate the atomic type representation with the feature dimension of the fifth dimension and the atomic coordinate representation with the feature dimension of the sixth dimension to obtain an atomic type / coordinate representation with a feature dimension of H; embeddingly encode the atomic-level feature sequence of the pocket to obtain an atomic-level feature representation with a feature dimension of a seventh dimension, and perform feature projection on the atomic-level feature representation with the feature dimension of the seventh dimension to obtain other pocket representation with a feature dimension of H; add the atomic type / coordinate representation with the feature dimension of H and the other pocket representation with the feature dimension of H in position to obtain a pocket representation with a feature dimension of H, and input the pocket representation with the feature dimension of H into the encoder to obtain the second output vector.

5. The method of claim 2, wherein, The small molecule structure data set includes a FSMILES type sequence of a molecule and a FSMILES coordinate sequence of the molecule; and the inputting of the first output vector, the attribute representation, and the small molecule structure data set into the decoder of the molecular prediction model includes: embeddingly encoding the FSMILES type sequence of the molecule to obtain a FSMILES type representation with a feature dimension of an eighth dimension; embeddingly encoding the FSMILES coordinate sequence of the molecule to obtain a FSMILES coordinate representation with a feature dimension of a ninth dimension; concatenating the FSMILES type representation with the feature dimension of the eighth dimension and the FSMILES coordinate representation with the feature dimension of the ninth dimension to obtain a FSMILES type / coordinate representation with a feature dimension of H; concatenating the FSMILES type / coordinate representation with the feature dimension of H and the attribute representation in the dimension of the sequence to obtain a ligand representation; inputting the first output vector and the ligand label into a decoder of the molecule prediction model; and / or; embeddingly encoding the FSMILES type sequence of the molecule to obtain a FSMILES type representation with a feature dimension of an eighth dimension; embeddingly encoding the FSMILES coordinate sequence of the molecule to obtain a FSMILES coordinate representation with a feature dimension of a ninth dimension; concatenating the FSMILES type representation with the feature dimension of the eighth dimension and the FSMILES coordinate representation with the feature dimension of the ninth dimension to obtain a FSMILES type / coordinate representation with a feature dimension of H; concatenating the FSMILES type / coordinate representation with the feature dimension of H and the attribute representation in the dimension of the sequence to obtain a ligand representation; inputting the second output vector and the ligand label into a decoder of the molecule prediction model; before inputting the first output vector, the attribute representation and the small molecule structure data set into the decoder of the molecule prediction model, or before training the molecule prediction model after inputting the first output vector, the attribute representation and the small molecule structure data set into the decoder of the molecule prediction model, further comprising: obtaining a small molecule; dividing the small molecule into a plurality of fragments according to a preset cutting rule; calculating the ring information of each atom in each fragment, and determining the element type of each atom in each fragment and whether it has aromaticity, to obtain the FSMILES sequence of the small molecule; obtaining the small molecule structure data set according to a plurality of FSMILES sequences of a plurality of small molecules; the dividing the small molecule into a plurality of fragments according to a preset cutting rule comprises: selecting all single bonds in the small molecule that meet the conditions of not belonging to a part of a ring, not connected to a hydrogen atom, and at least one end connected to a ring; dividing the small molecule into a plurality of fragments at the selected single bonds. ​ 6. The method of claim 1, wherein, The training data set comprises a force field accuracy training data set and an experimental accuracy training data set; wherein the force field accuracy training data set and the experimental accuracy training data set comprise a plurality of application scenarios and a training subset of force field accuracy corresponding to each of the application scenarios; and the experimental accuracy training data set comprises a plurality of application scenarios and a training subset of experimental accuracy corresponding to each of the application scenarios.

7. A method of processing a protein pocket, characterized by, The method comprises: obtaining protein pocket data to be matched; inputting the protein pocket data to be matched into a molecular prediction model trained by the method of any one of claims 1-6 to obtain small molecule structure information and / or anchor point information.

8. The method of claim 7, wherein, The inputting of the protein pocket data to be matched into the molecular prediction model trained by the method of any one of claims 1-6 to obtain small molecule structure information comprises: inputting the protein pocket data to be matched into an encoder of the molecular prediction model to obtain a third output vector; obtaining small molecule attribute information to be matched; inputting the small molecule attribute information to be matched into an attribute controller of the molecular prediction model to obtain a third attribute representation; inputting the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information.

9. The method of claim 7, wherein, The inputting of the protein pocket data to be matched into the molecular prediction model trained by the method of any one of claims 1-6 to obtain small molecule structure information comprises: obtaining anchor point information corresponding to the pocket data to be matched; inputting the pocket data to be matched and the anchor point information corresponding to the pocket data to be matched into an encoder of the molecular prediction model to obtain a third output vector; obtaining small molecule attribute information to be matched; inputting the small molecule attribute information to be matched into an attribute controller of the molecular prediction model to obtain a third attribute representation; inputting the third output vector and the third attribute representation into a decoder of the attribute controller of the molecular prediction model to obtain the small molecule structure information.

10. The method of claim 7, wherein, The inputting of the protein pocket data to be matched into the molecular prediction model trained by the method of any one of claims 1-6 to obtain anchor point information comprises: inputting the protein pocket data to be matched into an encoder of the molecular prediction model to obtain the anchor point information. 11.A device for training a molecular prediction model, comprising: The device comprises: a first obtaining module configured to obtain a training data set, wherein the training data set comprises a plurality of application scenarios and a training subset corresponding to each of the application scenarios; a training module configured to train the molecular prediction model using each of the training subsets in turn; wherein the training data set comprises: a first application scenario, and a first training subset corresponding to the first application scenario, wherein the first training subset comprises a pocket data set, an anchor point condition set, a small molecule structure data set, and a small molecule attribute information set; a second application scenario, and a second training subset corresponding to the second application scenario, wherein the second training subset comprises the pocket data set and the anchor point condition set; A third application scenario, a third training subset corresponding to the third application scenario includes the pocket dataset, the small molecule structure dataset, and the small molecule attribute information set; The training module is specifically configured to train the molecular prediction model by using the first training subset; train the molecular prediction model by using the second training subset after training the molecular prediction model by using the first training subset; and train the molecular prediction model by using the third training subset after training the molecular prediction model by using the second training subset.

12. A device for processing a protein pocket, characterized in that The device comprises: A first obtaining module configured to obtain protein pocket data to be matched; A prediction module configured to input the protein pocket data to be matched into a molecular prediction model trained by the method of any one of claims 1-6 to obtain small molecule structure information and / or anchor point information.

13. A computer device, comprising: A memory and a processor, which are in communication connection with each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the training method of the molecular prediction model according to any one of claims 1-6 and / or the processing method of the protein pocket according to any one of claims 7-10. The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the training method of the molecular prediction model according to any one of claims 1-6 and / or the processing method of the protein pocket according to any one of claims 7-10.

14. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the training method of the molecular prediction model according to any one of claims 1-6 and / or the processing method of the protein pocket according to any one of claims 7-10.

15. A computer program product, characterised in that, The computer readable storage medium stores computer instructions, and the computer instructions are used to make the computer execute the training method of the molecular prediction model according to any one of claims 1-6 and / or the processing method of the protein pocket according to any one of claims 7-10.

Images