Model training method and apparatus, and molecular structure information recommendation method and apparatus
By constructing three-dimensional molecular map information of protein-degraded chimera and training prediction models, the problem of low drug design efficiency of protein-degraded chimera in the prior art is solved, and accurate prediction of molecular fragments and effective drug design are achieved.
Patent Information
- Application Number
- PCT/CN2024/076376
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-15
- Filing Date
- 2024-02-06
- Publication Date
- 2024-11-21
AI Technical Summary
The existing methods of protein-degrading chimera drug design are inefficient, and the new fragment structures explored are difficult to form effective drug proteins with the original protein-degrading chimera, and the characteristics of the molecules cannot be fully characterized, resulting in poor drug molecular structure design.
Using model training method, three-dimensional molecular graph information is constructed by obtaining the data set of protein-degraded chimera, inputting a prediction model to predict molecular fragments with drug functions, and training the model based on the prediction results to improve the accuracy and efficiency of molecular fragment prediction.
The comprehensive characterization of the three-dimensional molecular structural characteristics of protein-degraded chimera is achieved, which improves the accuracy and efficiency of molecular fragment prediction and enhances the effectiveness of drug design.
Smart Images

Figure CN2024076376_21112024_PF_FP_ABST
Abstract
Description
Model training method and device and molecular structure information recommendation method and device Technical Field
[0001] The present disclosure relates to the fields of artificial intelligence and bioengineering, and in particular to a method and device for model training, a method and device for recommending molecular structure information, an electronic device, and a storage medium. Background Art
[0002] At present, by constructing protein degradation-targeted chimeras with bifunctional fragments, pathogenic proteins in patients can be effectively eliminated. Therefore, the development of protein degradation-targeted chimeras with pharmacological functions is becoming a new exploration direction for the treatment of various diseases.
[0003] For the design of protein degradation chimeras, current drug design methods often use quantitative structure-activity relationship modeling to screen for unknown chemical structures with pharmacological activity. However, the new fragment structures discovered in this way often cannot form effective drug proteins when combined with the original protein degradation-targeting chimera, and this exploration method is often inefficient.
[0004] Summary of the Invention
[0005] The present disclosure provides a method for model training and a method and device for recommending molecular structure information to partially solve the above-mentioned problems existing in the prior art. The present disclosure adopts the following technical solutions.
[0006] In a first aspect, the present disclosure provides a method for model training, comprising: obtaining a dataset of protein degradation targeting chimeras; constructing three-dimensional molecular graph information of a specified protein degradation targeting chimera based on the dataset of the protein degradation targeting chimera; inputting the three-dimensional molecular graph information of the specified protein degradation targeting chimera into a prediction model to be trained, obtaining fragment information of molecular fragments that are combined with the specified protein degradation targeting chimera to have a preset drug function, and using the fragment information of the combined molecular fragments as target fragment information; training the prediction model based on the deviation between multiple predicted target fragment information and multiple fragment information of label molecular fragments corresponding to the specified protein degradation targeting chimeras.
[0007] Optionally, constructing the three-dimensional molecular graph information of the specified protein degradation targeting chimera based on the data set of the protein degradation targeting chimera includes: selecting the data of the specified protein degradation targeting chimera from the data set; determining the characterization data of the specified protein degradation targeting chimera based on the data of the specified protein degradation targeting chimera, wherein the characterization data includes: data for characterizing the type of atoms in the specified protein degradation targeting chimera, the coordinates of each atom in the specified protein degradation targeting chimera in a specified coordinate system, the atomic mass vector of each atom in the specified protein degradation targeting chimera, the nuclear charge number vector of each atom in the specified protein degradation targeting chimera, the molecular mass of the specified protein degradation targeting chimera, The number of atoms in the designated protein degradation targeting chimera, the number of valence electrons of the designated protein degradation targeting chimera, the logP value of the designated protein degradation targeting chimera, the PK value of the designated protein degradation targeting chimera, the molecular structure characteristic data of the designated protein degradation targeting chimera, the pharmaceutical characteristic data of the designated protein degradation targeting chimera, the matrix data used to characterize the molecular structure characteristics of the designated protein degradation targeting chimera, the coordinates of the anchor atoms selected in the designated protein degradation targeting chimera, and the distances between the anchor atoms selected in the designated protein degradation targeting chimera; based on the characterization data of the designated protein degradation targeting chimera, constructing the three-dimensional molecular graph information of the designated protein degradation targeting chimera.
[0008] Optionally, the prediction model includes an encoder and a decoder; obtaining the fragment information of the molecular fragment that is combined with the specified protein degradation targeting chimera to have a preset drug function, and using the fragment information of the combined molecular fragment as the target fragment information includes: inputting the three-dimensional molecular graph information of the specified protein degradation targeting chimera into the encoder, so that the encoder determines the embedding vector of the atomic distance within the specified protein degradation targeting chimera, and determines the three-dimensional molecular graph feature of the specified protein degradation targeting chimera based on the embedding vector; inputting the three-dimensional molecular graph feature into the decoder to obtain the fragment information of the molecular fragment that is combined with the specified protein degradation targeting chimera to have the preset drug function.
[0009] Optionally, according to the embedding vector, determining the three-dimensional molecular graph features of the specified protein degradation targeting chimera according to the embedding vector includes: determining the attention weight for the specified protein degradation targeting chimera through the encoder; determining the invariant features and isovariant features for the specified protein degradation targeting chimera according to the attention weight and the embedding vector through the encoder; and determining the three-dimensional molecular graph features of the specified protein degradation targeting chimera according to the invariant features and the isovariant features through the encoder.
[0010] Optionally, the invariant feature characterizes a molecular feature of the designated protein degradation targeting chimera that does not change with the molecular structure, and the invariant feature is determined by the charge number of each atom in the designated protein degradation targeting chimera.
[0011] Optionally, the isovariant feature characterizes the molecular structural features of the designated protein degradation targeting chimera, and the isovariant feature is determined by the three-dimensional coordinates of each atom in the designated protein degradation targeting chimera.
[0012] Optionally, obtaining the fragment information of the molecular fragments that are combined with the specified protein degradation targeting chimera to have a preset drug function, and using the fragment information of the combined molecular fragments as the target fragment information includes: inputting the three-dimensional molecular graph information of the specified protein degradation targeting chimera into the prediction model to be trained, so that the prediction model predicts the fragment information of multiple molecular fragments that are combined with the specified protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information, as each candidate fragment information; inputting each candidate fragment information into a preset reinforcement learning model, so that the reinforcement learning model can select the target fragment information from the each candidate fragment information.
[0013] Optionally, the inputting of each candidate fragment information into a preset reinforcement learning model to select the target fragment information from the each candidate fragment information through the reinforcement learning model includes: for each candidate fragment information, inputting the candidate fragment information into the reinforcement learning model so that the reinforcement learning model determines the score of the candidate fragment information based on the logP value of the molecular structure and the PK value of the molecular structure; and selecting the target fragment information from the each candidate fragment information according to the score of each candidate fragment information.
[0014] Optionally, the deviation includes a size difference between two fragments in each fragment pair and a weight difference between two fragments in each fragment pair in a plurality of fragment pairs, and a fragment pair includes a fragment corresponding to the predicted target fragment information and a corresponding tag molecule fragment.
[0015] In a second aspect, the present disclosure provides a method for recommending molecular structure information, comprising: obtaining three-dimensional molecular graph information of an original protein degradation targeting chimera; inputting the three-dimensional molecular graph information of the original protein degradation targeting chimera into a pre-trained prediction model, so that the pre-trained prediction model predicts fragment information of a molecular fragment that is combined with the original protein degradation targeting chimera to form a preset drug function based on the three-dimensional molecular graph information of the original protein degradation targeting chimera, and the pre-trained prediction model is trained by the model training method of the first aspect above; based on the predicted fragment information of the molecular fragment, the fragment information of the molecular fragment is sent to the user.
[0016] Optionally, the method further comprises: storing the three-dimensional molecular graph information of the original protein degradation targeting chimera in correspondence with the predicted fragment information.
[0017] In a third aspect, the present disclosure provides a device for model training, comprising: an acquisition module for acquiring a data set of a protein degradation targeting chimera; a construction module for constructing three-dimensional molecular graph information of a specified protein degradation targeting chimera based on the data set of the protein degradation targeting chimera; a prediction module for inputting the three-dimensional molecular graph information of the specified protein degradation targeting chimera into a prediction model to be trained, obtaining fragment information of a molecular fragment that is combined with the specified protein degradation targeting chimera to have a preset drug function, and using the fragment information of the combined molecular fragment as target fragment information; a training module for training the prediction model based on the deviation between multiple predicted target fragment information and multiple fragment information of label molecular fragments corresponding to the specified protein degradation targeting chimera.
[0018] In a fourth aspect, the present disclosure provides a device for recommending molecular structure information, comprising: an acquisition module for acquiring three-dimensional molecular graph information of an original protein degradation targeting chimera; a prediction module for inputting the three-dimensional molecular graph information of the original protein degradation targeting chimera into a pre-trained prediction model, so that the pre-trained prediction model predicts fragment information of a molecular fragment that is combined with the original protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information of the original protein degradation targeting chimera, and the prediction model is trained by the model training method of the first aspect above; a recommendation module for recommending molecular structure information to a user based on the predicted fragment information of the molecular fragment.
[0019] In a fifth aspect, the present disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for model training in the first aspect.
[0020] In a sixth aspect, the present disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the method for recommending molecular structure information according to the second aspect.
[0021] In the seventh aspect, the present disclosure provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the model training method of the first aspect when executing the program.
[0022] In an eighth aspect, the present disclosure provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for recommending molecular structure information according to the second aspect is implemented.
[0023] At least one of the above-mentioned technical solutions adopted in the present disclosure can achieve the following beneficial effects: It can be seen from the above-mentioned method that the present application can construct three-dimensional molecular graph information of a specified protein degradation targeting chimera through the acquired data set. This three-dimensional molecular graph information can fully characterize the various characteristics of the molecular structure of the specified protein degradation targeting chimera. Then, after the three-dimensional molecular graph information is input into the prediction model, the prediction model will predict the fragment information of the molecular fragments that are combined with the specified protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information, as the target fragment information, and then train the prediction model based on the deviation between the predicted target fragment information and the fragment information of the label molecular fragment corresponding to the specified protein degradation targeting chimera, so that in the subsequent process of predicting the molecular structure, the prediction model can be used to achieve fast and accurate molecular fragment prediction, thereby improving the efficiency and accuracy of molecular fragment exploration. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation of the present disclosure. In the drawings:
[0025] FIG1 is a flow chart of a model training method provided in the present disclosure;
[0026] FIG2 is a schematic diagram of a process of recommending molecular structure information provided by the present disclosure;
[0027] FIG3 is a schematic diagram of the architecture of a drug molecule discovery system provided by the present disclosure;
[0028] FIG4 is a schematic diagram of a model training device provided by the present disclosure;
[0029] FIG5 is a schematic diagram of a molecular structure information recommendation device provided by the present disclosure;
[0030] FIG6 is a schematic structural diagram of an electronic device provided in the present disclosure corresponding to FIG1 or FIG2 . DETAILED DESCRIPTION
[0031] To make the objectives, technical solutions, and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below in conjunction with the specific embodiments of the present disclosure and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present disclosure.
[0032] The technical solutions provided by various embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
[0033] FIG1 is a flow chart of a model training method provided in the present disclosure, which includes the following steps.
[0034] S100: Obtain a dataset of protein degradation targeting chimeras.
[0035] The execution subject of the model training method provided in the present disclosure can be a terminal device such as a desktop computer, a laptop computer, or a server. For the sake of convenience, the present disclosure only uses the terminal device as the execution subject to illustrate the provided model training method.
[0036] In the present disclosure, a terminal device can obtain raw structural data of various protein degradation-targeting chimeras to construct a dataset for training a model. This data can be collected from an external network or obtained through file entry. Exemplarily, the raw data can include SMILES (Simplified Molecular-Input Line-Entry System) datasets of whole molecules and fragment molecules of protein degradation-targeting chimeras and protein degradation-targeting chimeras published online. SMILES datasets are one-dimensional structural representations of compounds.
[0037] S102: Constructing three-dimensional molecular graph information of a designated protein degradation targeting chimera based on the dataset of the protein degradation targeting chimera.
[0038] After obtaining the above dataset, the terminal device needs to search for data suitable for training the prediction model. That is to say, not all protein degradation targeting chimeras included in the dataset are suitable as samples for training. Some data may not have well-labeled molecular fragments, and some data may be "dirty data."
[0039] Therefore, the terminal device needs to search for protein degradation targeting chimeras that are suitable as training samples from the data set, that is, to determine the designated protein degradation targeting chimeras. A specific implementation method can be to create an extensible three-dimensional molecular graph structure data generator, clean, reconstruct and optimize the protein degradation targeting chimera compound data, thereby searching for the designated protein degradation targeting chimera as a training sample, and determining the three-dimensional molecular graph information of the designated protein degradation targeting chimera. The designated protein degradation targeting chimera obtained after cleaning, reconstructing and optimizing the protein degradation targeting chimera compound data has a logP value and PK value associated with the following invariant features and isovariant features. Among them, logP represents the oil-water partition coefficient of the compound, and the PK value represents the pharmacokinetic quality.
[0040] In the present disclosure, after the terminal device determines the data of multiple specified protein degradation targeting chimeras from the above-mentioned data set, it can further determine the characterization data of these specified protein degradation targeting chimeras, wherein the characterization data mentioned here may include: data used to characterize the atomic types within the specified protein degradation targeting chimera, the coordinates of each atom within the specified protein degradation targeting chimera in a specified coordinate system, the atomic mass vector of each atom within the specified protein degradation targeting chimera, the nuclear charge number vector of each atom within the specified protein degradation targeting chimera, the molecular mass of the specified protein degradation targeting chimera, the number of atoms within the specified protein degradation targeting chimera, the number of valence electrons of the specified protein degradation targeting chimera, the logP value of the specified protein degradation targeting chimera, the PK value of the specified protein degradation targeting chimera, the molecular structure characteristic data of the specified protein degradation targeting chimera, the pharmaceutical characteristic data of the specified protein degradation targeting chimera, matrix data used to characterize the molecular structure characteristics of the specified protein degradation targeting chimera, the coordinates of the anchor atoms selected in the specified protein degradation targeting chimera, the distance between the anchor atoms selected in the specified protein degradation targeting chimera, and other data. Those skilled in the art will appreciate that characterization data can be obtained through different calculation methods based on the original data. For example, the RDkit molecular information extraction algorithm can be used to obtain molecular physicochemical and pharmaceutical properties such as the number of atoms (number of nodes), molecular mass, number of valence electrons, logP value, PK value, and aromaticity within a specified protein degradation targeting chimera.
[0041] The coordinates of each atom within the designated protein degradation targeting chimera in a designated coordinate system and the atomic mass vector of each atom within the designated protein degradation targeting chimera can be determined with reference to international standards. For example, the atomic mass vector of each atom within the designated protein degradation targeting chimera can refer to the atomic mass vector of the atomic mass of each atom within the designated protein degradation targeting chimera according to the international standard.
[0042] The matrix data mentioned above for characterizing the molecular structural characteristics of the specified protein degradation targeting chimera may include an adjacency matrix for characterizing the specified protein degradation targeting chimera molecule, a bond order matrix for the specified protein degradation targeting chimera, a Coulomb matrix for characterizing the Coulomb forces between atoms in the specified protein degradation targeting chimera, and a distance matrix for the specified protein degradation targeting chimera.
[0043] The adjacency matrix of the designated protein degradation targeting chimera molecule is used to characterize whether the atoms in the designated protein degradation targeting chimera are bonded. For example, a bond between two atoms can be represented by "1", and no bond can be represented by "0".
[0044] The Coulomb matrix used to characterize the Coulomb forces between atoms in a given protein degradation targeting chimera can be determined using the following electrostatic interaction formula (1):
[0045] Among them, Z represents the nuclear charge number of the atom, C ij is used to represent the Coulomb force between atoms j and i, Used to represent the Euclidean distance encoding between atoms j and i.
[0046] The molecular structural characteristic data of the designated protein degradation targeting chimera mentioned above can be used to characterize the structural characteristics of the designated protein degradation targeting chimera, such as optical rotation and aromaticity.
[0047] After determining the aforementioned characterization data, the three-dimensional molecular graph information of the designated protein degradation-targeted chimera can be determined using at least a portion of the characterization data. As can be seen from the foregoing, since the present disclosure comprehensively considers various information regarding the designated protein degradation-targeted chimera when determining the three-dimensional molecular graph information of the designated protein degradation-targeted chimera, the ultimately determined three-dimensional molecular graph information is guaranteed to comprehensively characterize the characteristics of the designated protein degradation-targeted chimera, thereby ensuring the accuracy and rationality of the prediction results of the subsequent prediction model.
[0048] S104: Inputting the three-dimensional molecular graph information of the designated protein degradation targeting chimera into the prediction model to be trained, obtaining fragment information of molecular fragments that are combined with the designated protein degradation targeting chimera to have a preset drug function, and using the fragment information as target fragment information.
[0049] In the present disclosure, the above-mentioned prediction model is provided with an encoder and a decoder. Therefore, the prediction model ultimately predicts the molecular fragments that can be combined with the specified protein degradation targeting chimera to form a preset drug function based on the input three-dimensional molecular graph information. In fact, it can be regarded as a coding and decoding process.
[0050] Specifically, after the terminal device inputs the three-dimensional molecular graph information of the designated protein degradation-targeting chimera into the encoder of the prediction model to be trained, the encoder can determine an embedding vector for the interatomic distances within the designated protein degradation-targeting chimera and, based on this embedding vector, determine the three-dimensional molecular graph features of the designated protein degradation-targeting chimera. This three-dimensional molecular graph feature can then be input into the decoder of the prediction model to obtain fragment information of molecular fragments that combine with the designated protein degradation-targeting chimera to form a predetermined drug function, which serves as the target fragment information.
[0051] Therein, when determining the embedding vector of the interatomic distance within the designated protein degradation targeting chimera, it can be determined in a variety of ways, for example, when defining the distance between atoms as Under the premise of , the embedding vector can be determined by the Bessel form through the following formula (2-1) or (2-2):
[0052] Among them, e k (D ij ) represents the embedding vector of the distance between atom i and atom j contained in the protein degradation targeting chimera k, and c is used to represent the preset parameters.
[0053] For example, when defining the distance between atoms as Under the premise of , the embedding vector can be determined by the Gaussian form through the following formula (3):
[0054] Among them, σ is a preset parameter.
[0055] The embedding vector obtained by the above formula (2) or (3) can be further processed by the linear transformation formula (4) to obtain the embedding vector e of the interatomic distance ij : e ij =We · e(D ij )+b e (4)
[0056] Among them, W e is the weight parameter of the linear transformation, b e is the bias parameter of the linear transformation, e(D ij ) is produced by e k (D ij ), where the subscript k represents the basis vector label, which is the label of a complete set of bases.
[0057] Of course, in practical applications, the above embedding vectors can also be determined through other feasible methods, which will not be described one by one in this disclosure.
[0058] After obtaining the above embedding vector, it can be further processed through an attention mechanism to obtain more accurate three-dimensional molecular graph features. Specifically, the encoder can determine an attention weight for the above-mentioned designated protein degradation-targeted chimera. Then, based on the attention weight and the determined embedding vector, the encoder can determine the invariant features and isovariant features for the designated protein degradation-targeted chimera. Furthermore, based on the invariant features and isovariant features, the encoder can determine the three-dimensional molecular graph features of the designated protein degradation-targeted chimera.
[0059] The following is a specific example to describe the process of determining the three-dimensional molecular graph features of a specified protein degradation target chimera through the attention mechanism:
[0060] Through the three-dimensional molecular graph attention mechanism, the embedding forms of query (Q) and key (K) can be obtained as follows (5-1) and (5-2): i =W q s i +b q (5-1) K j =W k s j +b k (5-2)
[0061] Among them, Q i The Q feature matrix, K, used to represent atom i in a given protein degradation targeting chimera j A K-characteristic matrix representing atom j in a given protein degradation targeting chimera. i The invariant used to represent atom i in a given protein degradation targeting chimera, s i This invariant can be determined by the following formula (5-3): i =embed(Z i ) (5-3)
[0062] Among them, Z i To specify the charge number of atom i in the protein degradation targeting chimera, embed is used to convert the atomic charge number into a one-dimensional array. As can be seen from the above formula for determining the invariant of atom i, it is called an invariant because the charge number of atom i will never change. j The invariant used to represent the atom j in the designated protein degradation targeting chimera can also be determined by referring to the above formula. q To determine Q i The weight parameter of the feature matrix, b q To determine Q iThe bias parameter of the feature matrix. Similarly, W k To determine K j The weight parameter of the feature matrix, b k To determine K j Bias parameters for the feature matrix.
[0063] Then, we can construct an augmented matrix W a , get the output of attention weight, specifically refer to the following formula (6): Λ=[A ij , B ij , C ij ]=W a (Q i ×K j ×e ij ) (6)
[0064] In addition, the V feature matrix of the atoms in the specified protein degradation targeting chimera can be further determined. The V feature matrix can represent the values corresponding to the query and key in the attention mechanism. Among them, for any atom in the specified protein degradation targeting chimera, the atom can correspond to two forms of V feature matrices, one for determining the subsequent invariant features, and the other for determining the subsequent isovariant features. The V feature matrix used to determine the subsequent invariant features can be specifically referred to the following formula (7-1):
[0065] Among them, V s,j Used to represent the V feature matrix used to determine the subsequent invariant features, s j This is the invariant mentioned above for representing the atom j in the designated protein degradation targeting chimera, W ss To determine V s,j The weight parameter b is used to determine V s,j The bias parameters of . There is no special practical meaning, it can be regarded as V s,j This matrix is split into These three matrices are represented in this way mainly to determine the invariant features of atoms in the specified protein degradation targeting chimera in the subsequent process.
[0066] As for the V feature matrix used to determine the subsequent isovariable features, please refer to the following formula (7-2):
[0067] in, That is, the V feature matrix used to determine the isovariable features of the subsequent atom j, W vv To determine The weight parameter, bvv To determine The bias parameter of . is used to represent the vector determined by specifying the three-dimensional coordinates of atom j in the protein degradation targeting chimera.
[0068] After determining the above attention weights, it is necessary to further determine the invariant features and isovariant features of the designated protein degradation targeting chimera using the above attention weights.
[0069] Specifically, from the above content, we can know that the above V feature matrix is divided into two types, namely, one is V s,j , the other is Then, when determining the invariant features of a given protein degradation targeting chimera, the above attention weights can be determined for V s,j For weighting, please refer to the following formula (8-1):
[0070] As can be seen from the above formula, by combining the attention weight Λ determined above with the V of atom j in the specified protein degradation targeting chimera s,j By weighting, the invariant features of atom j in a given degradation-targeted chimera can be obtained.
[0071] It should be noted that in the above formula It can be understood as M s,ij This matrix is split into these three matrices. The reason for splitting it is that the invariant characteristics and isovariant characteristics of the specified protein degradation targeting chimera need to be determined in the subsequent process.
[0072] Furthermore, when determining the isovariate characteristics of a given protein degradation targeting chimera, the following formulas (8-2) and (8-3) can be used to determine:
[0073] in, is used to represent the vector determined by specifying the three-dimensional coordinates of atom i in the protein degradation targeting chimera, and is used to represent the vector determined by specifying the three-dimensional coordinates of atom j in the protein degradation targeting chimera.
[0074] In determining the above and After that, the isovariable characteristics of atom j in the designated protein degradation targeting chimera can be determined, specifically by the following formula (8-4):
[0075] Where N(i) represents the i-th atom.
[0076] After determining the invariant characteristics of each atom in the designated protein degradation targeting chimera, the invariant characteristics of each atom in the designated protein degradation targeting chimera can be further obtained. For details, please refer to the following formula (8-5):
[0077] Finally, the invariant features and isovariant features of each atom in the specified protein degradation targeting chimera can be expressed by the following formula (8-6):
[0078] From the above content, it can be seen that the determined invariant characteristics of the designated protein degradation targeting chimera actually characterize the molecular characteristics of the designated protein degradation targeting chimera that will not change with the molecular structure (because the invariant of each atom is actually determined by the charge number of each atom in the designated protein degradation targeting chimera, and the charge number of the atom is usually fixed), while the isovariant characteristics of the designated protein degradation targeting chimera can characterize the characteristics of the molecular structure of the designated protein degradation targeting chimera (because the isovariant characteristics are actually determined step by step based on the three-dimensional coordinates of each atom in the designated protein degradation targeting chimera).
[0079] Therefore, it can be understood that the subsequent decoder actually explores the characteristics of the molecular structure of the specified protein degradation targeting chimera and the molecular characteristics of the specified protein degradation targeting chimera that do not change with the molecular structure based on the isovariant characteristics and invariant characteristics of the specified protein degradation targeting chimera, to predict the atomic or molecular structure that can connect the specified protein degradation targeting chimera.
[0080] Furthermore, after determining the invariant features and isovariant features of the designated protein degradation targeting chimera, the three-dimensional molecular graph features of the designated protein degradation targeting chimera can be determined by message transmission and updating. Specifically, the invariant features and isovariant features are updated using Message, as shown in formula (9):
[0081] Among them, l is used to represent the number of updates, as well as That is, it represents the three-dimensional molecular graph features of the designated protein degradation targeting chimera at the lth and l+1th updates, and represents the invariant of atom i in the protein degradation targeting chimera at the time of the lth update, and It represents the equivalent variable of atom i in the protein degradation targeting chimera at the time of the lth update, It represents the invariant features and isovariant features of atom i in the protein degradation targeting chimera at the lth update.
[0082] Therefore, after l information propagation and updates, the three-dimensional molecular graph characteristics of the above-mentioned designated protein degradation targeting chimera are obtained. For example, l can be equal to 5.
[0083] In the present disclosure, the above-mentioned embedding vector can actually be determined before being input into the encoder. In this case, the determined embedding vector can actually be input into the encoder, and then the encoder uses the attention mechanism to determine the three-dimensional molecular graph features of the specified protein degradation targeting chimera.
[0084] As mentioned above, the process of predicting the target fragment based on the three-dimensional molecular graph features of the designated protein degradation targeting chimera can be viewed as the decoder predicting new atoms or new molecular structures that can be connected to the designated protein degradation targeting chimera based on the invariant features and isovariant features of each atom in the designated protein degradation targeting chimera. Therefore, when determining the above-mentioned three-dimensional molecular graph features, the decoder actually obtains the updated invariant features and isovariant features of each atom in the designated protein degradation targeting chimera through the information transfer mechanism to perform atom-by-atom prediction. The specific process can be understood as first predicting the first atom that can be connected to the designated protein degradation targeting chimera based on the designated protein degradation targeting chimera, and determining which atom in the designated protein degradation targeting chimera the first atom should be connected to. Then, based on the designated protein degradation targeting chimera and the predicted first atom, the decoder predicts the second atom that can be connected to the molecular structure composed of the designated protein degradation targeting chimera and the first atom, and predicts which atom in the molecular structure the second atom should be connected to, and so on, until a complete new molecular fragment is predicted.
[0085] This can be achieved through the following process.
[0086] After predicting a new atom, the invariant features of each atom contained in the molecular structure (i.e., the molecular structure composed of the designated protein degradation targeting chimera and the previously predicted atom) can be updated. Specifically, the update can be performed using the following formula (10):
[0087] in, It is used to represent the invariant characteristics of atom i in the above-mentioned molecular structure after the t-th atom is predicted. f1 can be an MLP (multi-layer perceptron) function.
[0088] Furthermore, the three-dimensional coordinates of the new atom can be predicted by the following formula (11):
[0089] in, are the predicted three-dimensional coordinates of the atoms, and is the reference coordinate, which can be It is regarded as the three-dimensional coordinates at the center of mass of the molecular structure (the molecular structure can refer to the above-mentioned designated protein degradation targeting chimera, or a new molecular structure composed after predicting t new atoms), and the Ker function can be a Gaussian function.
[0090] so, This can actually be thought of as a decoder that predicts the coordinates of new atoms.
[0091] We can further predict the edges between the new atom and the atoms in the molecular structure obtained after t-1 predictions. This can be achieved by the following formula (12):
[0092] in, Used to represent the latent variable features of the predicted new atoms, Used to represent the latent variable features of atom j in the molecular structure obtained after t-1 predictions. Used to represent the features and values of the predicted invariant features of each atom, the functions h and f2 can be MLP functions.
[0093] so, This can actually be regarded as a decoder for predicting edges.
[0094] pass and To predict the specific information of the new atom, it can be determined by the following formula (13):
[0095] The above formula (13) can be understood as determining the probability of the new atom in this case based on a three-dimensional coordinate of the new atom and a connection relationship of the new atom. In other words, the decoder can obtain multiple new atoms, each of which has its own coordinates and connection status with existing atoms. Therefore, the probability can be calculated according to formula (13), and the new atom with the highest probability is used as the predicted new atom to determine the specific information of the new atom (that is, the specific three-dimensional coordinates of the new atom and which atom the new atom is specifically connected to).
[0096] By using the above method and continuously iterating, a new molecular fragment is finally predicted.
[0097] It should be noted that, in the whole process of predicting new molecular fragments, the various formulas listed above are only used as examples to illustrate this process. In actual applications, new molecular fragments can also be predicted by other methods not limited to the above formulas. For example, when determining the above-mentioned isovariable features, M is actually used. s,ij In fact, you can also use the matrix data in M s,ij The above-mentioned isovariable features are determined as a whole. For another example, when determining the features of a three-dimensional molecular graph, the above-mentioned method of determining isovariable features is mentioned as using information transmission. However, in fact, the method of determining isovariable features can also be omitted, that is, the determined three-dimensional molecular graph features are not updated. Other examples are not listed here one by one.
[0098] In the present disclosure, the prediction model can actually predict the fragment information of multiple molecular fragments that can have the predicted drug function with the specified protein degradation targeting chimera based on the three-dimensional molecular graph information of the specified protein degradation targeting chimera, so a target fragment needs to be screened out from these molecular fragments later.
[0099] Therefore, in the present disclosure, after a terminal device inputs the three-dimensional molecular graph information of a designated protein degradation-targeting chimera into a prediction model to be trained, the prediction model will predict, based on the three-dimensional molecular graph information, the fragment information of multiple molecular fragments that combine with the designated protein degradation-targeting chimera to have a predetermined drug function, as candidate fragment information. The terminal device can then input this candidate fragment information into a preset reinforcement learning model, which then uses the reinforcement learning model to select the target fragment information from the candidate fragment information.
[0100] Among them, since the selected target fragment needs to meet certain characteristics that can be used as a drug, the above-mentioned reinforcement learning model can select the target fragment information from these candidate fragment information based on the logP value and PK value of the molecular structure.
[0101] Specifically, for each candidate fragment, the candidate fragment information is input into the aforementioned reinforcement learning model. The reinforcement learning model can then determine a score for the candidate fragment based on the molecular structure's logP value and PK value. After determining the scores for each candidate fragment, target fragment information can be selected based on the scores of the candidate fragments.
[0102] For example, the score of each candidate segment information may be determined by referring to the following formula (14):
[0103] Among them, λ(Y) is the function of logP value and PK value, Y represents the candidate fragment information, the specific value of target can be determined according to actual needs, and R(Y) is used to express the score of candidate fragment information Y.
[0104] S106: Training the prediction model according to the deviation between the plurality of predicted target fragment information and the fragment information of the tag molecule fragments corresponding to the plurality of designated protein degradation targeting chimeras.
[0105] After predicting the target fragment information, the deviation between each target fragment information and the fragment information of the corresponding tag molecule fragment corresponding to the designated protein degradation targeting chimera can be further predicted. The prediction model is then trained with minimizing this deviation as the optimization goal. The deviation includes the difference in size and weight between the two fragments in a fragment pair. A fragment pair includes a fragment corresponding to the predicted target fragment information and a corresponding tag molecule fragment.
[0106] It should be noted that in the present disclosure, the aforementioned prediction model and reinforcement learning model can actually be trained together using a joint training approach. That is, after determining the target fragment information, the loss value for the prediction model and the loss value for the reinforcement learning model are determined separately. Then, the two models are adjusted together with the optimization goal of minimizing the sum of these two loss values. Both loss values can actually be determined by the deviation between the target fragment information and the fragment information of the tag molecule corresponding to the designated protein degradation targeting chimera.
[0107] Of course, in practical applications, the prediction model and the reinforcement learning model can also be combined into one model. For example, a network layer that can implement reinforcement learning can be constructed in the above prediction model, and then the prediction model containing the reinforcement learning network layer can be trained using the above method.
[0108] Because protein degradation-targeting chimeric molecules can directly mediate the ubiquitination and degradation of pathogenic proteins, they are currently being used to develop drugs for the treatment of diseases including cancer, immune system disorders, and neurological diseases. In particular, in cancer treatment, compared to commonly used chemotherapy drugs with significant side effects, protein degradation-targeting chimeric molecules can eliminate this disadvantage by targeting proteins that are induced by cancer cells.
[0109] With the continuous development of fields such as artificial intelligence and machine learning and their continued penetration into drug research and development, drug research and development has become more complex than earlier quantitative structure-activity relationship modeling. Quantitative structure-activity relationship models based on artificial intelligence or machine learning rely on methods for representing molecular features and can be divided into three categories: molecular descriptors, molecular fingerprints, and molecular graphs. Molecular descriptors can effectively and quantitatively represent the physical, chemical, biological, and pharmaceutical properties of molecules; molecular fingerprints can be used to describe the similarity of intermolecular structures; and molecular graphs can treat atoms and chemical bonds within molecules as nodes and edges to characterize the structural characteristics of molecules. By combining the above three methods, developers can design effective methods for drug molecule design.
[0110] However, the current method cannot fully characterize the characteristics of the molecule, so the new molecular fragments finally obtained cannot be well combined with the protein degradation targeting chimera, thus failing to obtain a better drug molecular structure.
[0111] Therefore, the model training method provided by the present disclosure comprehensively refers to the characterization data of various molecular structures of the specified protein degradation targeting chimera in the process of determining the three-dimensional molecular graph information of the specified protein degradation targeting chimera, which enables the three-dimensional molecular graph information finally determined to comprehensively characterize the molecular structure characteristics of the specified protein degradation targeting chimera.
[0112] Moreover, when determining the three-dimensional molecular graph features of the designated protein degradation targeting chimera, since the determination is based on the invariant features and isovariant features of the designated protein degradation targeting chimera, this method can fully characterize the molecular structural characteristics of the designated protein degradation targeting chimera, thereby ensuring that the prediction model can subsequently perform accurate and reasonable predictions of new fragment information through the three-dimensional molecular graph features of the designated protein degradation targeting chimera.
[0113] After training the above prediction model, you can use it to predict molecular structure information and recommend molecular structure information. The specific process is shown in the figure below.
[0114] FIG2 is a schematic diagram of a process of recommending molecular structure information provided by the present disclosure.
[0115] S201: Obtain three-dimensional molecular map information of the original protein degradation targeting chimera.
[0116] In the present disclosure, a terminal device can receive a user-inputted molecular fragment structure prediction instruction and, through this prediction instruction, obtain the three-dimensional molecular graph information of the original protein degradation-targeted chimera. The determination of the three-dimensional molecular graph information here is substantially consistent with the determination of the three-dimensional molecular graph information in the aforementioned model training, and will not be described in detail here. The terminal device mentioned here can refer to a desktop computer, laptop computer, or other device.
[0117] S202: Inputting the three-dimensional molecular graph information of the original protein degradation targeting chimera into a pre-trained prediction model, so that the prediction model predicts the fragment information of the molecular fragments that are combined with the original protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information of the original protein degradation targeting chimera, and the prediction model is trained by the above-mentioned model training method.
[0118] The terminal device can input the three-dimensional molecular graph information of the original protein degradation targeting chimera into the prediction model deployed in the terminal device, and the prediction model will output the fragment information of the molecular fragments that are combined with the original protein degradation targeting chimera to have preset drug functions.
[0119] It should be pointed out that since the prediction model has been trained through supervised training in the above model training process, the reinforcement learning model is no longer needed in actual application. That is, the fragment information of the molecular fragments output by the prediction model is the final fragment information, and no further screening is required.
[0120] S203: According to the predicted fragment information of the molecular fragment, the fragment information of the molecular fragment is sent to the user.
[0121] After obtaining the fragment information of the molecular fragment through the prediction model, the fragment information of the molecular fragment can be recommended to the user. Of course, in actual application, the three-dimensional molecular graph information of the original protein degradation targeting chimera and the fragment information of the molecular fragment can also be stored in correspondence.
[0122] Of course, after determining the fragment information of the above-mentioned molecular fragment, the original protein degradation targeting chimera can be combined with the molecular fragment according to the fragment information, and then the information on the pharmaceutical, chemical and other properties of the new molecule can be tested and stored.
[0123] The present disclosure also provides a system for drug molecule exploration, as shown in FIG3 .
[0124] FIG3 is a schematic diagram of the architecture of a drug molecule discovery system provided by the present disclosure.
[0125] As shown in Figure 3, the system primarily consists of the following components: a storage subsystem, which stores the aforementioned datasets, as well as information about the molecular fragments predicted by the prediction model in actual applications, along with information about their pharmaceutical and chemical properties. A control subsystem, which predicts, based on the three-dimensional molecular graph of the original protein degradation-targeting chimera input into the subsystem, information about the molecular fragments that combine with the original degradation-targeting chimera to form drug-like molecules.
[0126] The control subsystem includes three units: molecular feature extraction unit, molecular generation unit and molecular screening unit. These three units are used to obtain three-dimensional molecular graph features, predict the fragment information of new molecular fragments, and screen out the final fragment information.
[0127] The system may also include a data optimizer for cleaning, reconstructing, and optimizing the structural data of the protein-degrading targeted chimera compound and transferring the processed data to the storage subsystem. Furthermore, the system may include an input control interface and an output control interface for interacting with the user.
[0128] The above are one or more implementation methods of the present disclosure. Based on the same idea, the present disclosure also provides a corresponding model training device and a molecular structure information recommendation device, as shown in Figures 4 and 5.
[0129] Figure 4 is a schematic diagram of a model training device provided by the present disclosure, including: an acquisition module 401, used to obtain a data set of protein degradation targeting chimeras; a construction module 402, used to construct three-dimensional molecular graph information of a specified protein degradation targeting chimera based on the data set of the protein degradation targeting chimera; a prediction module 403, used to input the three-dimensional molecular graph information of the specified protein degradation targeting chimera into the prediction model to be trained, obtain fragment information of molecular fragments that are combined with the specified protein degradation targeting chimera to have a preset drug function, and use the fragment information of the combined molecular fragments as target fragment information; a training module 404, used to train the prediction model based on the deviation between multiple predicted target fragment information and multiple fragment information of label molecular fragments corresponding to the specified protein degradation targeting chimeras.
[0130] Optionally, the construction module 402 is used to select data of a designated protein degradation targeting chimera from the data set; determine characterization data of the designated protein degradation targeting chimera based on the data of the designated protein degradation targeting chimera, the characterization data including: data for characterizing the type of atoms in the designated protein degradation targeting chimera, the coordinates of each atom in the designated protein degradation targeting chimera in a designated coordinate system, the atomic mass vector of each atom in the designated protein degradation targeting chimera, the nuclear charge number vector of each atom in the designated protein degradation targeting chimera, the molecular mass of the designated protein degradation targeting chimera, the number of atoms in the designated protein degradation targeting chimera, The invention relates to a method for preparing a three-dimensional molecular map of a protein degradation-targeted chimera, comprising: determining the number of valence electrons of the designated protein degradation-targeted chimera, the logP value of the designated protein degradation-targeted chimera, the PK value of the designated protein degradation-targeted chimera, the molecular structure characteristic data of the designated protein degradation-targeted chimera, the pharmaceutical characteristic data of the designated protein degradation-targeted chimera, the matrix data used to characterize the molecular structure characteristics of the designated protein degradation-targeted chimera, the coordinates of the anchor atoms selected in the designated protein degradation-targeted chimera, and the distances between the anchor atoms selected in the designated protein degradation-targeted chimera; and constructing the three-dimensional molecular map information of the designated protein degradation-targeted chimera according to the characterization data of the designated protein degradation-targeted chimera.
[0131] Optionally, the prediction model includes an encoder and a decoder; the prediction module 403 is specifically used to input the three-dimensional molecular graph information of the specified protein degradation targeting chimera into the encoder, so that the encoder determines the embedding vector of the atomic distance within the specified protein degradation targeting chimera, and determines the three-dimensional molecular graph features of the specified protein degradation targeting chimera based on the embedding vector; the three-dimensional molecular graph features are input into the decoder to obtain the fragment information of the molecular fragment that is combined with the specified protein degradation targeting chimera to have the preset drug function.
[0132] Optionally, the prediction module 403 is specifically used to determine, through the encoder, the attention weight for the specified protein degradation targeting chimera; through the encoder, determine, based on the attention weight and the embedding vector, the invariant features and isovariant features for the specified protein degradation targeting chimera; through the encoder, determine, based on the invariant features and the isovariant features, the three-dimensional molecular graph features of the specified protein degradation targeting chimera.
[0133] Optionally, the invariant feature characterizes a molecular feature of the designated protein degradation targeting chimera that does not change with the molecular structure, and the invariant feature is determined by the charge number of each atom in the designated protein degradation targeting chimera.
[0134] Optionally, the isovariant feature characterizes the molecular structural features of the designated protein degradation targeting chimera, and the isovariant feature is determined by the three-dimensional coordinates of each atom in the designated protein degradation targeting chimera.
[0135] Optionally, the prediction module 403 is specifically used to input the three-dimensional molecular graph information of the specified protein degradation targeting chimera into the prediction model to be trained, so that the prediction model predicts the fragment information of multiple molecular fragments that are combined with the specified protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information, as each candidate fragment information; and input the each candidate fragment information into a preset reinforcement learning model, so as to select the target fragment information from the each candidate fragment information through the reinforcement learning model.
[0136] Optionally, the prediction module 403 is specifically used to input each candidate fragment information into the reinforcement learning model, so that the reinforcement learning model determines the score of the candidate fragment information based on the logP value of the molecular structure and the PK value of the molecular structure; and selects the target fragment information from the candidate fragment information according to the score of each candidate fragment information.
[0137] Optionally, the deviation includes a size difference between two fragments in each fragment pair and a weight difference between two fragments in each fragment pair in a plurality of fragment pairs, and a fragment pair includes a fragment corresponding to the predicted target fragment information and a corresponding tag molecule fragment.
[0138] Figure 5 is a schematic diagram of a molecular structure information recommendation device provided by the present disclosure, including: an acquisition module 501, used to obtain three-dimensional molecular graph information of the original protein degradation targeting chimera; a prediction module 502, used to input the three-dimensional molecular graph information of the original protein degradation targeting chimera into a pre-trained prediction model, so that the pre-trained prediction model predicts the fragment information of the molecular fragment that is combined with the original protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information of the original protein degradation targeting chimera, and the pre-trained prediction model is trained by the model training method described in Figure 1 above; a recommendation module 503, used to send the fragment information of the molecular fragment to the user based on the predicted fragment information of the molecular fragment.
[0139] Optionally, the apparatus further comprises: a storage module 504, configured to store the three-dimensional molecular graph information of the original protein degradation targeting chimera and the predicted fragment information in correspondence thereto.
[0140] For example, the device described in FIG. 4 or FIG. 5 can be a terminal device such as a desktop computer or laptop computer, or a server or other dedicated device. The modules shown in FIG. 4 or FIG. 5 can be implemented as hardware, software, or a combination of hardware and software to achieve the described functions. These modules represent merely a division of logical functions; in actual implementation, they can be fully or partially integrated into a single physical entity or physically separated. Furthermore, these modules can be implemented entirely as software invoked by a processing element, entirely as hardware, or partially as software invoked by a processing element, while others can be implemented in hardware. For example, the determination module can be a separate processing element, integrated into a chip within the device, or stored as program code in the device's memory, invoked by a processing element to perform the functions of the determination module. The implementation of the other modules is similar. Furthermore, these modules can be fully or partially integrated or implemented independently. The processing element described herein can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each module above may be completed by an integrated logic circuit of hardware in a processor element or by instructions in the form of software.
[0141] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field programmable gate arrays (FPGAs). For another example, when a module is implemented by scheduling program code on a processing element, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).
[0142] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present disclosure is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid state disk (SSD)).
[0143] The present disclosure also provides a computer-readable storage medium storing a computer program that can be used to execute the model training method provided in FIG1 or the molecular structure information recommendation method provided in FIG2. The storage medium can be a non-volatile storage medium.
[0144] The present disclosure also provides a schematic structural diagram of an electronic device corresponding to FIG1 or FIG2, as shown in FIG6. As shown in FIG6, at the hardware level, the electronic device includes a processor, memory, and non-volatile memory, and may also include hardware required for other services, such as a network interface and an internal bus. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the model training method described in FIG1 or the molecular structure information recommendation method described in FIG2.
[0145] Of course, in addition to software implementation, this disclosure does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.
[0146] In the 1990s, technological improvements could be clearly distinguished as either hardware improvements (for example, improvements to circuit structures like diodes, transistors, and switches) or software improvements (improvements to process flows). However, with the advancement of technology, many process flow improvements today can now be considered direct improvements to hardware circuit structures. Designers almost always create the corresponding hardware circuit structure by programming the improved process flow into the hardware circuit. Therefore, it cannot be said that a process flow improvement cannot be implemented using hardware modules. For example, a programmable logic device (PLD), such as a field programmable gate array (FPGA), is an integrated circuit whose logical function is determined by user programming. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to hire a chip manufacturer to design and manufacture a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly done using "logic compiler" software. This is similar to the software compiler used when developing programs. Before compilation, the original code must also be written in a specific programming language, called a hardware description language (HDL). There is not just one HDL, but many, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art will also understand that by simply programming the method flow in one of these hardware description languages and then programming it into an integrated circuit, a hardware circuit that implements the logic method flow can be easily obtained.
[0147] The controller can be implemented in any suitable manner. For example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, logic gates, switches, application-specific integrated circuits (ASICs), programmable logic controllers, and embedded microcontrollers. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that in addition to implementing the controller in a purely computer-readable program code format, the controller can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered as structures within the hardware component. Or even, the devices for implementing various functions can be considered as both software modules that implement the method and structures within the hardware component.
[0148] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
[0149] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing the present disclosure, the functions of each unit can be implemented in the same or multiple software and / or hardware.
[0150] Those skilled in the art will appreciate that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0151] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.
[0152] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a product including an instruction device that implements the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.
[0153] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and / or one or more boxes in the block diagram.
[0154] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.
[0155] Memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.
[0156] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.
[0157] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.
[0158] Those skilled in the art will appreciate that embodiments of the present disclosure may be provided as methods, systems, or computer program products. Thus, the present disclosure may take the form of a fully hardware embodiment, a fully software embodiment, or an embodiment combining software and hardware. Furthermore, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0159] The present disclosure may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present disclosure may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communications network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.
[0160] The various embodiments of this disclosure are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between the other embodiments. In particular, the system embodiments are generally similar to the method embodiments, so their description is relatively simple. For relevant portions, refer to the description of the method embodiments.
[0161] The foregoing is merely an embodiment of the present disclosure and is not intended to limit the present disclosure. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure are intended to be included within the scope of the claims of the present disclosure.
Claims
1. A model training method, characterized in that: include: Obtain a dataset of protein degradation targeting chimeras; Constructing three-dimensional molecular graph information of a designated protein degradation targeting chimera according to the data set of the protein degradation targeting chimera; Inputting the three-dimensional molecular graph information of the designated protein degradation targeting chimera into the prediction model to be trained, obtaining the fragment information of the molecular fragments combined with the designated protein degradation targeting chimera to have the preset drug function, and using the fragment information of the combined molecular fragments as the target fragment information; The prediction model is trained according to the deviation between the multiple predicted target fragment information and the fragment information of the label molecule fragments corresponding to the multiple designated protein degradation targeting chimeras.
2. The method according to claim 1, characterized in that The step of constructing the three-dimensional molecular graph information of the designated protein degradation targeting chimera according to the data set of the protein degradation targeting chimera comprises: Selecting data of the designated protein degradation targeting chimera from the data set; According to the data of the designated protein degradation targeting chimera, characterization data of the designated protein degradation targeting chimera are determined, wherein the characterization data include: data for characterizing the type of atoms in the designated protein degradation targeting chimera, coordinates of each atom in the designated protein degradation targeting chimera in a designated coordinate system, atomic mass vectors of each atom in the designated protein degradation targeting chimera, nuclear charge number vectors of each atom in the designated protein degradation targeting chimera, molecular mass of the designated protein degradation targeting chimera, number of atoms in the designated protein degradation targeting chimera, number of valence electrons of the designated protein degradation targeting chimera, logP value of the designated protein degradation targeting chimera, PK value of the designated protein degradation targeting chimera, molecular structure characteristic data of the designated protein degradation targeting chimera, pharmaceutical characteristic data of the designated protein degradation targeting chimera, matrix data for characterizing the molecular structure characteristics of the designated protein degradation targeting chimera, coordinates of anchor atoms selected in the designated protein degradation targeting chimera, and distances between anchor atoms selected in the designated protein degradation targeting chimera; According to the characterization data of the designated protein degradation targeting chimera, the three-dimensional molecular map information of the designated protein degradation targeting chimera is constructed.
3. The method according to claim 1, characterized in that The prediction model includes an encoder and a decoder; the obtaining of the fragment information of the molecular fragments combined with the designated protein degradation targeting chimera to form a predetermined drug function, and using the fragment information of the combined molecular fragments as the target fragment information includes: The three-dimensional molecular graph information of the designated protein degradation targeting chimera is input into the encoder so that the encoder determines the embedding vector for the atomic distance within the designated protein degradation targeting chimera, and The embedding vector determines the three-dimensional molecular graph characteristics of the designated protein degradation targeting chimera; The three-dimensional molecular graph features are input into the decoder to obtain the fragment information of the molecular fragment that is combined with the designated protein degradation targeting chimera to have the preset drug function.
4. The method according to claim 3, characterized in that Determining the three-dimensional molecular graph features of the designated protein degradation targeting chimera according to the embedding vector comprises: Determining, by the encoder, an attention weight for the designated protein degradation targeting chimera; Determining, by the encoder, invariant features and isovariant features for the designated protein degradation targeting chimera according to the attention weights and the embedding vectors; The encoder determines the three-dimensional molecular graph features of the designated protein degradation targeting chimera based on the invariant features and the isovariant features.
5. The method according to claim 4, characterized in that The invariant feature characterizes the molecular feature of the designated protein degradation targeting chimera that does not change with the molecular structure, and the invariant feature is determined by the charge number of each atom in the designated protein degradation targeting chimera.
6. The method according to claim 4 or 5, characterized in that The isovariant feature characterizes the molecular structure feature of the designated protein degradation targeting chimera, and the isovariant feature is determined by the three-dimensional coordinates of each atom in the designated protein degradation targeting chimera.
7. The method according to claim 1, characterized in that The obtaining of the fragment information of the molecular fragments combined with the designated protein degradation targeting chimera to form a predetermined drug function, and using the fragment information of the combined molecular fragments as the target fragment information includes: Inputting the three-dimensional molecular graph information of the designated protein degradation targeting chimera into the prediction model to be trained, so that the prediction model predicts fragment information of multiple molecular fragments that are combined with the designated protein degradation targeting chimera to form a preset drug function as each candidate fragment information according to the three-dimensional molecular graph information; The candidate fragment information is input into a preset reinforcement learning model, so as to select the target fragment information from the candidate fragment information through the reinforcement learning model.
8. The method according to claim 7, characterized in that The step of inputting the candidate segment information into a preset reinforcement learning model to select the target segment information from the candidate segment information through the reinforcement learning model includes: For each candidate fragment information, the candidate fragment information is input into the reinforcement learning model, so that the reinforcement learning model determines the score of the candidate fragment information based on the logP value of the molecular structure and the PK value of the molecular structure; According to the score of each candidate segment information, the target segment information is selected from the candidate segment information. interest.
9. The method according to claim 1, characterized in that The deviation includes the difference in size between two fragments in each fragment pair among multiple fragment pairs and the difference in weight between two fragments in each fragment pair. A fragment pair includes a fragment corresponding to the predicted target fragment information and a corresponding label molecule fragment.
10. A method for recommending molecular structure information, characterized in that: include: Obtaining three-dimensional molecular map information of the original protein degradation targeting chimera; Inputting the three-dimensional molecular graph information of the original protein degradation targeting chimera into a pre-trained prediction model, so that the pre-trained prediction model predicts the fragment information of the molecular fragments that are combined with the original protein degradation targeting chimera to have a preset drug function based on the three-dimensional molecular graph information of the original protein degradation targeting chimera, wherein the pre-trained prediction model is trained by the method described in any one of claims 1 to 9 above; According to the predicted fragment information of the molecular fragment, the fragment information of the molecular fragment is sent to the user.
11. The method according to claim 10, characterized in that The method further comprises: The three-dimensional molecular graph information of the original protein degradation targeting chimera and the predicted fragment information are stored correspondingly.
12. A model training device, characterized in that: include: The acquisition module is used to obtain the dataset of protein degradation targeting chimeras; A construction module, for constructing three-dimensional molecular graph information of a designated protein degradation targeting chimera according to the data set of the protein degradation targeting chimera; A prediction module, used for inputting the three-dimensional molecular graph information of the designated protein degradation targeting chimera into the prediction model to be trained, obtaining the fragment information of the molecular fragments combined with the designated protein degradation targeting chimera to have a preset drug function, and using the fragment information of the combined molecular fragments as the target fragment information; The training module is used to train the prediction model according to the deviation between the multiple predicted target fragment information and the fragment information of the label molecule fragments corresponding to the multiple designated protein degradation targeting chimeras.
13. A molecular structure information recommendation device, characterized in that: include: An acquisition module, used to obtain three-dimensional molecular map information of the original protein degradation targeting chimera; A prediction module, used for inputting the three-dimensional molecular graph information of the original protein degradation targeting chimera into a pre-trained prediction model, so that the pre-trained prediction model predicts the fragment information of the molecular fragments combined with the original protein degradation targeting chimera to have a preset drug function according to the three-dimensional molecular graph information of the original protein degradation targeting chimera, wherein the pre-trained prediction model is trained by the method according to any one of claims 1 to 9 above; A recommendation module is used to recommend the fragments of the molecular fragments according to the predicted fragment information of the molecular fragments. Information sent to users.
14. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.
15. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to claim 10 or 11 is implemented.
16. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.
17. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to claim 10 or 11 is implemented.
Citation Information
Patent Citations
Drug and target interaction prediction method and device, equipment and storage medium
CN113160894A
Connector design method of protein degradation targeting chimera based on auto-encoder
CN114420197A
Universal method for drug design aiming at different target proteins
CN114464270A
Prediction processing method for binding postures of drug molecules and protein targets
CN115527607A
Specific target drug generation method and device based on graph neural network and MaxFlow platform
CN115762662A
Cited By
Method of training potential function model, computing device for atomic simulation, medium and product
CN122551922A
