Molecular skeleton hopping processing method and its device, medium, electronic device and computer program

By generating and mapping atomic latent vectors to spatial distributions, the method achieves novel drug molecules by overcoming the limitations of rule-based molecular skeleton hopping, enhancing efficiency and reducing costs.

JP7595772B2Active Publication Date: 2024-12-06TENCENT TECHNOLOGY (SHENZHEN) CO LTD +1
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2023538910
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2021-03-10
Filing Date
2022-02-28
Publication Date
2024-12-06
Estimated Expiration
2042-02-28

AI Technical Summary

Technical Problem

Existing molecular skeleton hopping methods in medicinal chemistry are limited by rule-based approaches, leading to a lack of novelty in generated drug molecules due to adherence to the design mindset of medicinal chemistry experts.

Method used

A method involving the generation of atomic latent vectors from drug molecules, followed by atomic masking to obtain backbone and side chain latent vectors, and mapping these vectors to spatial distributions to achieve target hopping degrees, enabling the automatic creation of novel drug molecules.

Benefits of technology

This approach breaks the design constraints of medicinal chemistry, resulting in highly novel drug molecules with improved efficiency and reduced labor and time costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007595772000040
    Figure 0007595772000040
  • Figure 0007595772000041
    Figure 0007595772000041
  • Figure 0007595772000042
    Figure 0007595772000042
Patent Text Reader

Abstract

A processing method for molecular scaffold hopping, its device, computer-readable medium, electronic device, and computer program product are disclosed. The processing method for molecular scaffold hopping includes the steps of generating an atom latent vector corresponding to a reference drug molecule based on a structure of a connection graph corresponding to the reference drug molecule, performing an atom masking process on the atom latent vector to obtain a scaffold latent vector and a side chain latent vector contained in the atom latent vector, generating a target scaffold latent vector having a target hopping degree for the scaffold latent vector based on the spatial distribution of the scaffold latent vector, and generating a drug molecule after hopping based on the target scaffold latent vector and the side chain latent vector.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] (CROSS-REFERENCE TO RELATED APPLICATIONS) The present examples claim priority to a Chinese patent application filed with the China Patent Office on March 10, 2021, bearing application number 202110260343.1, the entire contents of which are incorporated by reference into the present examples.

[0002] The present application relates to the field of computer and communication technology, and in particular to a molecular backbone hopping processing method and an apparatus, medium, electronic device and computer program product thereof. [Background technology]

[0003] Scaffold hopping (also called “scaffold hopping”) is a very important tool for medicinal chemistry design, whose main objectives are to modify conventional molecular structures, to substitute local structures of complex natural products, and / or to improve the pharmacokinetic properties of molecules by modifying the molecular scaffold.

[0004] The technical solutions of backbone hopping in the related art are mainly pharmacophore-based models, molecular shape similarity search, etc. However, all of these technical solutions are generated based on rules, so it is difficult to escape from the design mind of medicinal chemistry experts, and therefore the novelty of the molecules after hopping is lacking. Summary of the Invention [Problem to be solved by the invention]

[0005] The present embodiment provides a molecular backbone hopping process that improves the novelty of newly generated drug molecules, and an apparatus, medium and electronic device for the process.

[0006] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by the practice of the invention. [Means for solving the problem]

[0007] An embodiment of the present application provides a processing method for molecular scaffold hopping, the method including the steps of generating an atomic latent vector corresponding to a reference drug molecule based on a structure of a connection graph corresponding to the reference drug molecule; performing an atomic masking process on the atomic latent vector to obtain a scaffold latent vector and a side chain latent vector contained in the atomic latent vector; generating a target scaffold latent vector having a target hopping degree for the scaffold latent vector based on the spatial distribution of the scaffold latent vector; and generating a drug molecule after hopping based on the target scaffold latent vector and the side chain latent vector.

[0008] An embodiment of the present application provides a processing device for molecular scaffold hopping, the device comprising: a first generation unit configured to generate an atomic latent vector corresponding to a reference drug molecule based on a structure of a connection graph corresponding to the reference drug molecule; a first processing unit configured to perform an atomic masking process on the atomic latent vector to obtain a scaffold latent vector and a side chain latent vector contained in the atomic latent vector; a second generation unit configured to generate a target scaffold latent vector having a target hopping degree for the scaffold latent vector based on the spatial distribution of the scaffold latent vector; and a third generation unit configured to generate a drug molecule after hopping based on the target scaffold latent vector and the side chain latent vector.

[0009] An embodiment of the present application provides a computer readable medium having stored thereon a computer program, the computer program causing a processor to perform the method of processing molecular scaffold hopping as described in the above embodiments.

[0010] An embodiment of the present application provides an electronic device comprising one or more processors and a storage device storing one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors realize the molecular scaffold hopping processing method described in the above-mentioned embodiment.

[0011] The present embodiment provides a computer program product or computer program including computer instructions stored in a computer readable storage medium, the computer instructions being read by a processor of a computing device from the computer readable storage medium and executed to cause the computing device to perform the molecular scaffold hopping processing method provided by the various exemplary embodiments above. Effect of the Invention

[0012] In the technical solution provided by some embodiments of the present application, an atomic masking process is performed on the atomic latent vector corresponding to a drug molecule to obtain a backbone latent vector and a side chain latent vector, and then a target backbone latent vector having a target hopping degree is generated based on the spatial distribution of the backbone latent vector, and a drug molecule after hopping is generated based on the target backbone latent vector and the side chain latent vector. Thus, by mapping the backbone latent vector to the spatial distribution, the generated target backbone latent vector can jump out of the design mindset of medicinal chemistry experts, thus achieving excellent novelty, and can be automatically performed by electronic devices, improving the efficiency of backbone hopping.

[0013] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. [Brief description of the drawings]

[0014] [Figure 1]FIG. 1 illustrates a schematic diagram of an exemplary system architecture to which the technical solutions of the present embodiments can be applied. [Diagram 2] 1 shows a flowchart of a molecular backbone hopping process according to an embodiment of the present invention. [Figure 3A] 1 shows a flowchart for generating an atomic latent vector corresponding to a reference drug molecule according to an embodiment of the present application. [Figure 3B] 1 shows a flowchart of performing an atom masking process on an atom potential vector corresponding to a reference drug molecule according to an embodiment of the present disclosure. [Figure 3C] 1 shows a flowchart for generating a target skeleton latent vector with a specified hopping degree according to an embodiment of the present disclosure. [Figure 4A] 1 shows a schematic structural diagram of a machine learning model according to an embodiment of the present application. [Figure 4B] 2 shows a schematic diagram of the processing steps of a graph encoder according to an embodiment of the present invention; [Diagram 5] 1 shows a schematic diagram of an atomic masking and readout function according to an embodiment of the present invention. [Figure 6] 1 shows a schematic diagram of a distance representation method according to an embodiment of the present invention; [Figure 7] 1 shows a schematic diagram of a decoder process according to an embodiment of the present invention. [Figure 8] FIG. 1 shows a schematic diagram of a process for generating backbone and side chain latent vectors through a model, according to an embodiment of the present invention. [Figure 9] 1 shows a schematic diagram of a model decoding process according to an embodiment of the present invention; [Figure 10] FIG. 1 shows a schematic diagram of a skeleton hopping method according to an embodiment of the present invention. [Figure 11] FIG. 2 shows a block diagram of a processing device for molecular backbone hopping according to an embodiment of the present invention. [Figure 12] FIG. 1 shows a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, exemplary embodiments will be described more comprehensively with reference to the drawings. However, it is understood that the exemplary embodiments can be implemented in various forms and are not limited to the examples described herein; on the contrary, these embodiments are provided to make the present invention more comprehensive and complete, and to comprehensively convey the concept of the exemplary embodiments to those skilled in the art.

[0016] Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to fully understand the embodiments of the present invention. However, those skilled in the art will recognize that the technical solutions of the present invention may be implemented without one or more of the specific details, or may employ other methods, components, devices, steps, etc. In other cases, well-known methods, devices, implementations, or operations are not shown or described in detail to avoid obscuring aspects of the present invention.

[0017] The block diagrams shown in the drawings are merely functional entities which do not necessarily correspond to physically separate entities, i.e. these functional entities may be implemented in the form of software, in one or more hardware modules or integrated circuits, in different networks and / or processor and / or microcontroller devices.

[0018] The flowcharts shown in the drawings are merely illustrative and do not necessarily include all contents and operations / steps, and do not necessarily need to be performed in the order described. For example, some operations / steps may be decomposed, and some operations / steps may be merged or partially merged, so that the actual execution order may be changed according to the actual situation.

[0019] It should be noted that "plurality" as referred to in this specification refers to two or more than two. "And / or" describes a relationship between related objects, for example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the related objects before and after it are in an "or" relationship.

[0020] The technical solution according to the embodiment of the present application relates to a technology such as machine learning in artificial intelligence, specifically, to the application of machine learning technology to a technical solution of drug molecule scaffold hopping.

[0021] Before introducing the molecular backbone hopping technical solution of the embodiment of the present application, the processing solutions in the related art will be introduced first. The backbone hopping technical solutions proposed in the related art are mainly based on pharmacophore-based models, molecular shape-based searches, fingerprint-based chemical similarity searches and machine learning algorithms.

[0022] Here, the pharmacophore-based model simulates the active conformation of the ligand molecule through conformational search and molecular stacking, i.e., retains the molecular backbone of the characteristic atoms required for activity. The most important characteristic of a pharmacophore is that it has a molecular interaction characteristic shared by a group of active molecules, so the pharmacophore is an abstract concept rather than representing a real molecule or a group of chemical groups. The characteristics of the pharmacophore include hydrogen bond acceptors and donors, positive and negative charge interactions, hydrophobic interactions, aromatic ring interactions, etc. When such pharmacophore characteristics are capable of hopping from one molecule to another, i.e., when the reference molecule and the test molecule have the same pharmacophore characteristics, backbone hopping can be realized. A similar technical proposal is a protein structure-based drug design method, which realizes backbone hopping by expressing the interactions between a small molecule and the residues of a binding site in a protein as a vector and searching for corresponding molecules with the same characteristic vector in a compound library.

[0023] The search based on molecular shape mainly considers the volume in the molecular space to perform similarity search, and it is hoped that the binding action with the target protein can be maintained and backbone replacement can be realized. The problems with this technology and other search technology are that the search time is very long, it is limited by the existing chemical space, and it can only be searched in the existing compound library, and there are many false positive molecules, making it difficult to ensure the activity of the molecule.

[0024] With the development of artificial intelligence technology, especially with the application of artificial intelligence technology to molecular generation, the ability of drug development has been accelerated. The greatest advantage of molecular generation is to generate brand new molecules, directly realize the de novo design of drug molecules, and expand the existing molecular space. The AI ​​algorithm-based molecular generation method proposed in the related technology places excessive emphasis on the restructuring ability and correctness of molecules, which is difficult to meet the actual needs of pharmaceutical companies. For example, pharmaceutical companies want to modify existing molecules and break out of the existing structure while maintaining the activity of the molecule. However, although the technical proposals in the related technology can meet the requirement of maintaining activity, they are all generated based on rules, so it is difficult to break out of the design mind of medicinal chemistry experts, which causes the novelty of the newly generated molecules to be lacking.

[0025] In view of the above problems, the present embodiment provides a new molecular skeleton hopping processing technical solution, by mapping skeleton latent vectors into spatial distribution, so that the generated target skeleton latent vectors can jump out of the design mindset of medicinal chemistry experts, thus achieving excellent novelty, and can be automatically performed by electronic devices, thereby reducing labor costs and time costs. The technical solution of the present embodiment is described in detail below.

[0026] FIG. 1 shows a schematic diagram of an exemplary system architecture to which the technical solutions of the present embodiments can be applied.

[0027] 1, the system architecture 100 may include a terminal 110, a network 120, and a server 130. Here, the terminal 110 and the server 130 are connected via the network 120.

[0028] In the present embodiment, the terminal 110 may be, but is not limited to, a smartphone, a tablet PC, a laptop, a desktop computer, a smart speaker, or a smart watch. The network 120 may be a communication medium of various connection types that can provide a communication link between the terminal 110 and the server 130, such as a wired communication link, a wireless communication link, or an optical fiber cable, and the present embodiment is not limited thereto. The server 130 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0029] 1 are merely illustrative. Any number of terminals 110, networks 120, and servers 130 may be included depending on the needs of the implementation. For example, the server 130 may be a server cluster composed of multiple servers.

[0030] In the present embodiment, the user can use the terminal 110 to submit a reference drug molecule, i.e., a reference drug molecule that requires backbone hopping processing, to the server 130 via the network 120, and identify the backbone that needs hopping of the reference drug molecule. However, it is not necessary to identify the backbone that needs hopping of the reference drug molecule. After obtaining the reference drug molecule, the server 130 can convert the structure of the reference drug molecule into the structure of a connection graph, and then generate an atomic latent vector corresponding to the reference drug molecule according to the structure of the connection graph corresponding to the reference drug molecule.

[0031] After generating an atomic latent vector corresponding to a reference drug molecule, the server 130 can perform an atomic masking process on the atomic latent vector to obtain a backbone latent vector and a side chain latent vector contained in the atomic latent vector. To realize the backbone hopping process, a target backbone latent vector having a target hopping degree for the backbone latent vector can be generated based on the spatial distribution of the backbone latent vector, and then a drug molecule after hopping can be generated based on the target backbone latent vector and the above-mentioned obtained side chain latent vector.

[0032] The technical solution of the present embodiment maps the skeletal latent vector into a spatial distribution, allowing the generated target skeletal latent vector to break out of the design mindset of medicinal chemistry experts, thus achieving excellent novelty, and allowing automatic implementation by equipment, thereby reducing labor and time costs.

[0033] It should be noted that the molecular scaffold hopping processing method provided by the embodiment of the present application is generally executed by the server 130, and correspondingly, the molecular scaffold hopping processing device is generally installed in the server 130. However, in other embodiments of the present application, the terminal device may also have a function similar to that of the server, so as to execute the molecular scaffold hopping processing technical solution provided by the embodiment of the present application.

[0034] In the present embodiment, the user submits a reference drug molecule, which is a molecule that requires backbone hopping processing, through the terminal 110, and identifies the backbone that needs to be hopped of the reference drug molecule. However, it is not necessary to identify the backbone that needs to be hopped of the reference drug molecule. After obtaining the reference drug molecule, the terminal 110 can convert the structure of the reference drug molecule into a structure of a connection graph, and then generate an atomic latent vector corresponding to the reference drug molecule according to the structure of the connection graph corresponding to the reference drug molecule.

[0035] After generating an atomic latent vector corresponding to a reference drug molecule, the terminal 110 performs an atomic masking process on the atomic latent vector to obtain a skeleton latent vector and a side chain latent vector contained in the atomic latent vector. To realize the skeleton hopping process, a target skeleton latent vector having a target hopping degree for the skeleton latent vector can be generated based on the spatial distribution of the skeleton latent vector, and then a post-hopping drug molecule can be generated based on the target skeleton latent vector and the above-mentioned obtained side chain latent vector.

[0036] The implementation details of the technical solutions according to the embodiments of the present application are described below.

[0037] Fig. 2 shows a flow chart of a molecular skeleton hopping processing method according to an embodiment of the present application, which can be executed by an electronic device having a computing function, for example, by the server 130 shown in Fig. 1. As shown in Fig. 2, the molecular skeleton hopping processing method includes at least steps S210 to S240, and the detailed description is as follows:

[0038] In step S210, an atom latent vector corresponding to the reference drug molecule is generated based on the structure of the connection graph corresponding to the reference drug molecule.

[0039] In the present embodiment, the reference drug molecule is a molecule that requires a backbone hopping process, and the connection graph structure corresponding to the reference drug molecule is a connection graph structure obtained based on the structural transformation of the reference drug molecule.

[0040] For example, the structure of the connectivity graph corresponding to one drug molecule is

number

[0041] In this embodiment, after generating a connection graph structure corresponding to the reference drug molecule, an atomic latent vector corresponding to the reference drug molecule is generated based on the connection graph structure corresponding to the reference drug molecule. Specifically, as shown in FIG. 3A, the process includes the following steps S310a, S320a, and S330a, and the detailed description is as follows:

[0042] In step S310a, node information for each node in the connected graph structure is determined by a graph encoder based on the node features and edge features included in the connected graph structure.

[0043] In the present embodiment, for convenience of explanation, the following description will be given by taking one node v and one node w in the structure of the connected graph as an example. The node feature of node v is x v The node feature of node w is expressed as x w The edge feature between node v and node w is expressed as e vw The node information of node v is expressed as m v The latent vector of node v is expressed as hv Then, the process of calculating node information for each node in the structure of the connected graph by the graph encoder (including multiple cascaded hidden layers) may include:

[0044] The node feature (x) of the first node in the connected graph structure (the first node can be any node in the connected graph structure, e.g., node v) v ), the node feature (x w ), and edge information in the first hidden layer (hidden layer t) between the first node and another node other than the second node (e.g., node k) in the adjacent nodes of the first node (

number

number

[0045] Next, the edge information in the first hidden layer between the first node and other nodes

number

number

number

number

[0046] In the present embodiment,

number

number

[0047] In the above formula (1),

number

number

number

number

[0048] In the present embodiment,

number

number

[0049] In the above formula (2), g t (·) represents the update process, which may be a simple accumulation or averaging, or a Gated Recurrent Unit (GRU) computational form. If it is a GRU computational form,

number

number

[0050] In the present embodiment,

number

number

[0051] In the above equation (3), τ(·) represents the rectified linear unit (ReLU; also called normalized linear unit function); W represents the parameter that needs to be learned; cat(·) represents concatenating two vectors to form a longer vector, for example, concatenating a 3-dimensional vector with a 5-dimensional vector to obtain an 8-dimensional vector.

[0052] In this embodiment, the node information of node v is m vAssuming that, the node information m v can be calculated by the following formula (4):

number

[0053] In the above formula (4),

number

[0054] In step S320a, a latent vector for each node is generated based on the node information of each node and the node features of each node.

[0055] In the embodiment of the present invention, for example, the node information m of node v v After obtaining the node information m v and node feature x of node v v Based on this, the latent vector h of node v is v can be generated, for example, by the following formula (5):

number

[0056] In the above formula (5), τ(·) represents the ReLU function; W a represents the parameter that needs to be learned; cat(·) represents the concatenation of two vectors to form one longer vector.

[0057] In step S330a, an atom latent vector corresponding to the reference drug molecule is generated based on the latent vector of each node and the atoms contained in the reference drug molecule.

[0058] In one embodiment, after the latent vector of each node is obtained, the atomic latent vector corresponding to the reference drug molecule can be represented in the form of a matrix, that is, the atomic latent vector corresponding to the reference drug molecule can be represented by arranging the latent vector of each node into a matrix in row and column format (such as an H matrix).

[0059] It should be noted that the above-mentioned equations (1) to (5) are merely examples, and in other embodiments of the present application, new calculation equations can be obtained by further appropriately modifying the above-mentioned equations (1) to (5) (for example, increasing the multiple, decreasing the multiple, increasing by a fixed value, decreasing by a fixed value, etc.).

[0060] Next, as shown in FIG. 2, in step S220, an atom masking process is performed on the atom latent vector corresponding to the reference drug molecule to obtain a backbone latent vector and a side chain latent vector contained in the atom latent vector.

[0061] In the present embodiment, after the atomic latent vector corresponding to the reference drug molecule is obtained, the process of performing an atomic masking process on the atomic latent vector corresponding to the reference drug molecule may include steps S310b, S320b, and S330b, as shown in FIG. 3B, and the detailed description is as follows:

[0062] In step S310b, a bit vector corresponding to a reference drug molecule is determined, the length of the bit vector is the same as the number of atoms contained in the reference drug molecule, and the bit values ​​corresponding to the backbone atoms in the bit vector are first values.

[0063] For example, in the above embodiment, the first value may be, for example, 1, i.e., if an atom belongs to a backbone atom, the corresponding one bit value in the bit vector is 1; if an atom belongs to a side chain atom, the corresponding one bit value in the bit vector is 0. In this case, the bit vector is represented by a matrix S sca can be expressed as:

number

[0064] In the above formula (6),

number

number

[0065] In the present embodiment, the aforementioned bit vector can be preset, and indicates which atoms in the reference drug molecule belong to backbone atoms and which atoms belong to side chain atoms. For example, according to the backbone that needs to be hopped (replaced) in the reference drug molecule, searching and matching can be performed in the reference drug molecule in the manner of structure search to determine the backbone atoms and side chain atoms in the reference drug molecule, and then the aforementioned bit vector can be obtained.

[0066] In the present embodiment, the bit vector may be preset according to a set of skeleton determination rules, which may include multiple requirements, such as requirements for the number of heavy atoms in the skeleton, requirements for the number of skeleton rings, etc., and are not limited thereto in the present embodiment. In the case of a drug molecule, the skeleton part in the drug molecule can be automatically detected according to the skeleton determination rules, and then the bit vector can be generated based on the skeleton part and the part other than the skeleton part (i.e., the side chain part) in the drug molecule.

[0067] In step S320b, a filtering process is performed on the atom latent vector corresponding to the reference drug molecule based on the bit vector to obtain a latent vector of backbone atoms and a latent vector of side chain atoms.

[0068] In the present embodiment, the atomic potential vector corresponding to the reference drug molecule (i.e., the potential vector of the original atom) is Hnode Then, the potential vector of the backbone atom selected from the atomic potential vector corresponding to the reference drug molecule is

number

number

[0069] In step S330b, a multi-head attention process is performed on the latent vectors of the backbone atoms to obtain backbone latent vectors, and a multi-head attention process is performed on the latent vectors of the side-chain atoms to obtain side-chain latent vectors.

[0070] In the present embodiment, a multi-headed attention mechanism is used to determine scores (i.e., weights) corresponding to the latent vectors of each atom (backbone atoms and side-chain atoms) and to calculate the backbone latent vectors and the side-chain latent vectors based on the scores (i.e., weights).

[0071] For example, the atomic potential vector corresponding to the reference drug molecule (i.e., the potential vector of the original atom) is H node Assuming that the skeleton latent vector Z sca is represented by the following formula (7), and the side chain potential vector Z sc is expressed as the following formula (8):

number

[0072] In the above equations (7) and (8), the softmax(·) function realizes the function of the multi-head attention mechanism; W1 and W2 are both learnable parameters;

number

[0073] Of course, the above-mentioned equations (7) and (8) are merely examples, and in other embodiments of the present application, the above-mentioned equations (7) and (8) can be further appropriately modified (e.g., increasing the multiple, decreasing the multiple, increasing by a fixed value, decreasing by a fixed value, etc.) to obtain a new calculation formula.

[0074] As shown in FIG. 2, in step S230, a target skeleton latent vector having a target hopping degree (also referred to as a specified hopping degree) with respect to the skeleton latent vector is generated based on the spatial distribution of the skeleton latent vector.

[0075] In the present embodiment, the spatial distribution of the skeleton latent vector may be a Gaussian mixture distribution, a von Mises-Fisher Mixture (vMFM) distribution, etc. The following takes the Gaussian mixture distribution as an example:

[0076] In the present embodiment, a plurality of skeleton clusters may be pre-defined, and the cluster center of each skeleton cluster in the plurality of skeleton clusters is adapted to a Gaussian mixture distribution. For example, the plurality of skeleton clusters are obtained by clustering the skeletons of existing molecules through a skeleton clustering algorithm (i.e., a clustering algorithm), and the cluster center of one skeleton cluster is adapted to a Gaussian mixture distribution, so that all skeleton latent vectors contained in one skeleton cluster belong to a spatial distribution corresponding to the Gaussian mixture distribution.

[0077] In this case, after the skeleton latent vector is obtained, a first distance between the skeleton latent vector and the cluster center of each skeleton cluster can be determined, and then a target skeleton cluster to which the skeleton of the reference drug molecule belongs is determined based on the first distance, and further a mixed Gaussian distribution to which the skeleton latent vector belongs is determined based on the cluster center of the target skeleton cluster.

[0078] For example, the cluster center of the m-th skeleton cluster is (μ m ,σ m ), where μ m represents the cluster center, and σ m In general, the backbone latent vector corresponding to the i-th drug molecule is denoted by Z sca,i Then, the backbone latent vector Z sca,i and the distance d between the cluster center of the mth skeleton cluster i can be shown as equation (9):

number

[0079] The skeleton latent vector Z corresponding to the i-th drug molecule calculated from the above formula (9) sca,i and the distance d between the cluster center of the mth skeleton cluster i Based on this, we obtain the scaffold latent vector Z corresponding to the cluster center and the i-th drug molecule. sca,i The closest target skeleton cluster (c i ) can be selected, and the target skeleton cluster c i The cluster center (μ i ,σ i ) to determine the Gaussian mixture distribution to which the skeleton cluster vector belongs.

[0080] After the spatial distribution of the scaffold latent vectors corresponding to the reference drug molecules is obtained and the target scaffold clusters are determined, a target scaffold latent vector having a specified hopping degree relative to the scaffold latent vectors corresponding to the reference drug molecules can be generated according to needs, as described in detail below:

[0081] As shown in FIG. 3C, the process of generating a target skeleton latent vector having a target hopping degree in the embodiment of the present application may include step S310c and step S320c, and the detailed description is as follows:

[0082] In step S310c, a random sampling process is performed on the target skeleton cluster based on the target hopping degree to obtain an offset amount corresponding to the target hopping degree.

[0083] In step S320c, the backbone latent vector of the reference drug molecule is added to an offset amount corresponding to the target hopping degree to obtain a target backbone latent vector.

[0084] For example, the specified hopping degree may be hopping within the same skeleton cluster, hopping to an adjacent skeleton cluster, or hopping to a distant skeleton cluster, and the following describes these three hopping methods.

[0085] In the present embodiment, when the specified hopping degree is a first hopping degree, a first offset amount can be obtained by multiplying the variance of the target skeleton cluster with the first vector obtained by random sampling, and then adding the first offset amount to the skeleton latent vector corresponding to the reference drug molecule to generate a target skeleton latent vector. For example, the first hopping degree can be hopping within the same skeleton cluster.

[0086] Specifically, the target skeleton cluster is i and the backbone latent vector corresponding to the reference drug molecule is Z sca Then, the generated target skeleton latent vector having the first hopping degree can be expressed by the following equation (10):

number

[0087] In the above formula (10), Z new_sca represents the generated target skeleton latent vector; σ 2 (c i ) is the target skeleton cluster c irepresents the variance of a Gaussian mixture distribution that fits the cluster centers of N; N(0,1) represents random sampling based on a distribution with mean 0 and standard deviation 1.

[0088] In the embodiment of the present application, when the designated hopping degree is the second hopping degree, a first skeletal cluster whose distance from the cluster center of the target skeletal cluster is equal to or less than a first set value is selected from the multiple skeletal clusters, and then a second offset amount is generated based on the product of the variance of the first skeletal cluster and the second vector obtained by random sampling, the cluster center of the target skeletal cluster, and the cluster center of the first skeletal cluster, and the second offset amount is added to the skeletal latent vector corresponding to the reference drug molecule to generate a target skeletal latent vector. For example, the second hopping degree may be hopping to an adjacent skeletal cluster.

[0089] For example, the target skeleton cluster is c i The first skeletal cluster is denoted as c j and the backbone latent vector corresponding to the reference drug molecule is Z sca Then, the generated target skeleton latent vector having the second hopping degree can be expressed by the following equation (11):

number

[0090] In the above formula (11), Z new_sca represents the generated target skeleton latent vector; σ 2 (c j ) is the first skeletal cluster c j represents the variance of the Gaussian mixture distribution that matches the cluster centers of N(0,1) represents random sampling based on a distribution with mean 0 and standard deviation 1; i ) is the target skeleton cluster c i represents the center of μ(c j ) is the first skeletal cluster c j represents the center of μ(c k ) is a skeletal cluster c kδ' represents the first set value;π(·) represents one multi-nominal sampling;

number

[0091] In the embodiment of the present application, when the designated hopping degree is a third hopping degree, a second skeletal cluster whose distance from the cluster center of the target skeletal cluster is equal to or greater than a second set value is selected from the multiple skeletal clusters, and then a third offset amount is generated according to the multiplication of the variance of the second skeletal cluster and the third vector obtained by random sampling, the cluster center of the target skeletal cluster, and the cluster center of the second skeletal cluster, and then the third offset amount is added to the skeletal latent vector corresponding to the reference drug molecule to generate a target skeletal latent vector. For example, the third hopping degree may be hopping to a distant skeletal cluster.

[0092] For example, the target skeleton cluster is c i and the second skeletal cluster is represented as c j and the backbone latent vector corresponding to the reference drug molecule is Z sca Then, the generated target skeleton latent vector with the third hopping degree can be expressed by the following equation (12):

number

[0093] In the above formula (12), Z new_sca represents the generated target skeleton latent vector; σ 2 (c j ) is the second skeletal cluster c j represents the variance of the Gaussian mixture distribution that matches the cluster centers of N(0,1) represents random sampling based on a distribution with mean 0 and standard deviation 1;i ) is the target skeleton cluster c i represents the center of μ(c j ) is the second skeletal cluster c j represents the center of μ(c k ) is a skeletal cluster c k Δ represents the second set value;π(·) represents one multi-nominal sampling;

number

[0094] It should be noted that the above-mentioned equations (9) to (12) are merely examples, and in other embodiments of the present application, new calculation equations can be obtained by further appropriately modifying the above-mentioned equations (9) to (12) (for example, increasing the multiple, decreasing the multiple, increasing by a fixed value, decreasing by a fixed value, etc.).

[0095] Subsequently, as shown in FIG. 2, in step S240, a post-hopping drug molecule is generated based on the target backbone latent vector and the side chain latent vector.

[0096] In the present embodiment, the backbone latent vector in the reference drug molecule can be replaced with the target backbone latent vector and combined with the side chain latent vector to obtain the drug molecule after hopping.

[0097] In the present embodiment, the target and target activity value of the specified reference drug molecule can also be obtained, and then a drug molecule after hopping is generated according to the target backbone latent vector, the side chain latent vector, the target and target activity value of the reference drug molecule. In the technical solution of this embodiment, the activity of the generated drug molecule can be limited by the target and target activity value of the reference drug molecule.

[0098] In the embodiment of the present application, after generating the drug molecule after hopping, the generated drug molecule can be further subjected to a selection process.For example, based on the drug molecule after hopping, a molecular filtering process of physicochemical properties can be carried out to obtain the drug molecule with drug-like properties, and then obtain the eutectic structure corresponding to the reference drug molecule, and the drug molecule with drug-like properties is docked into the eutectic structure, and based on the binding mode between the drug molecule with drug-like properties and the eutectic structure, the drug molecule that does not match the eutectic structure is removed to obtain the selected drug molecule; and based on the docking of the selected drug molecule and the eutectic structure, the synthesis and verification of the compound can be carried out.

[0099] For example, the eutectic structure corresponding to the reference drug molecule may be a eutectic structure in which the reference drug molecule exists or a eutectic structure in which a homologous compound of the reference drug molecule exists. A drug molecule that does not match the eutectic structure may be a drug molecule whose structure after docking is significantly unreasonable.

[0100] In the embodiment of the present application, the relevant processing in the above embodiment can be performed through a machine learning model. In the process of training the machine learning model, the technical solution of the embodiment of the present application puts forward a technical solution to generate a loss function according to the cross entropy loss and the predicted loss of the machine learning model for the sample molecule. The following describes the methods of obtaining the cross entropy loss and the predicted loss respectively:

[0101] In the present embodiment, in the calculation of the cross-entropy loss, a sample skeleton latent vector corresponding to the sample molecule can be obtained, and at the same time, a plurality of skeleton clusters (these plurality of skeleton clusters may be the same as the plurality of skeleton clusters used for processing the reference drug molecule described above) can be obtained, and the cluster center of each skeleton cluster in these plurality of skeleton clusters is fitted to a mixed Gaussian distribution. Then, a second distance between the sample skeleton latent vector of the sample molecule and the cluster center of each skeleton cluster is determined, and the skeleton cluster to which the sample skeleton of the sample molecule belongs is determined based on the second distance, and a distance-based cross-entropy loss is generated based on the distance between the sample skeleton latent vector and the cluster center of the skeleton cluster to which the sample skeleton belongs.

[0102] In addition, the technical scheme for obtaining the sample skeleton latent vector corresponding to the sample molecule is the same as the technical scheme for obtaining the skeleton latent vector corresponding to the reference drug molecule, so no further description will be given. At the same time, the formula for calculating the second distance between the sample skeleton latent vector of the sample molecule and the cluster center of each skeleton cluster can also be calculated according to the above-mentioned formula (9).

[0103] For example, taking the above-mentioned formula (9) as an example (since the relevant calculation formulas and processing methods for the sample molecule and the reference drug molecule are the same, the distance between the scaffold latent vector corresponding to the reference drug molecule and the cluster center of the scaffold cluster, and the distance between the scaffold latent vector corresponding to the sample molecule and the cluster center of the scaffold cluster can be calculated using formula (9)), the distance between the sample scaffold latent vector corresponding to the i-th drug molecule (here, it can be understood as the i-th sample molecule) and the cluster center of the scaffold cluster to which the sample scaffold belongs is expressed as d i If we assume that the expression is expressed as d i We can improve the accuracy of model training by adding a certain bias based on:

number

[0104] In equation (13), d adj,i , d i represents the distance after applying the bias based on onehot(·) represents the one-hot coding function; c i where denotes the scaffold cluster to which the sample scaffold of the i-th sample molecule belongs; δ denotes the parameter.

[0105] In the present embodiment, d adj,i After obtaining the distance-based cross entropy loss L cls can be generated:

number

[0106] In equation (14), σ m represents the standard deviation of the Gaussian mixture distribution that matches the cluster center of the m-th skeletal cluster.

[0107] In the present embodiment, the machine learning model includes a decoder, and after a sample backbone latent vector and a sample side chain latent vector corresponding to a sample molecule are obtained by the machine learning model, the sample backbone latent vector, the sample side chain latent vector and a target molecule corresponding to the sample molecule are input into the decoder, and then a predicted loss of the machine learning model is calculated based on the output of the decoder and the target molecule.

[0108] It should be noted that the target molecule is the desired product molecule after the sample molecule is processed. The technical solution for obtaining the sample backbone latent vector and the sample side chain latent vector corresponding to the sample molecule is similar to the technical solution for obtaining the backbone latent vector and the side chain latent vector corresponding to the reference drug molecule, so the detailed description will not be repeated here.

[0109] In the present embodiment, after the cross-entropy loss and the predicted loss of the machine learning model are calculated, a loss function of the machine learning model can be generated based on the cross-entropy loss and the predicted loss of the machine learning model, and the parameters of the machine learning model are adjusted based on the loss function. For example, the loss function L of the machine learning model can be generated by the following formula (15):

number

[0110] In the above formula (15), L recon represents the predicted loss of the machine learning model; β represents a hyperparameter for adjusting the weight between the two losses.

[0111] The goal of training a machine learning model is to minimize the loss function L, among which the cross entropy loss L cls The purpose of the setting is to ensure as much as possible that after training the machine learning model, each skeletal latent vector determined by the machine learning model is near the center of the skeletal cluster to which it belongs; the prediction loss L recon The purpose of the setting is to ensure as much as possible that after the machine learning model is trained, the machine learning model can find an excellent target backbone latent vector, and further ensure that a drug molecule that meets the conditions is obtained.

[0112] After training the machine learning model, the reference drug molecule can be processed according to the machine learning model to obtain the hopping drug molecule. In order to easily understand the technical solution of the present embodiment, the implementation details of the technical solution of the present embodiment are described in detail below in combination with Figures 3 to 10:

[0113] As shown in FIG. 4A, when performing the process of molecular scaffold hopping through a machine learning model, the model structure may include a graph encoder 401, an atom masking and readout function, a Gaussian mixture distribution (GM) fitting process, and a decoder 402. Here, the graph encoder is mainly used to generate an atom latent vector corresponding to a drug molecule; the atom masking and readout function part is mainly used to obtain a scaffold latent vector and a side chain latent vector through the atom masking process; the Gaussian mixture distribution fitting process is used to realize a Gaussian mixture distribution of the scaffold latent vector to realize the process of different hopping degrees; the decoder is used to output the drug molecule obtained after the hopping process. The following will explain these parts in detail respectively:

[0114] In the present embodiment, the graph encoder includes a Directed Message Passing Neural Network (D-MPNN), which is a graph convolutional neural network. The graph convolutional neural network operates directly on the graph structure that contains the chemical structure, and unlike the fingerprint representation, which assigns a single fixed-length feature vector to the molecule, the graph structure representation assigns a feature vector to each bond and atom in the chemical structure.

[0115] In simple terms, D-MPNN can be understood as a multi-step neural network, where each step is essentially a feed-forward neural network that generates a set of latent representations that are used as input for the next step. The main part of D-MPNN is the message-passing step, which utilizes the local substructure of the molecular diagram to update the latent vectors. After the message-passing step, the latent vectors from all edges are aggregated into a single fixed-length latent vector, which is fed into the feed-forward neural network to generate predictions. As shown in Figure 4B, each connection is represented by a pair of directed edges, and messages from the orange directional connection in (a) (i.e., 3 → 2 and 4 → 2 in (a)) are used to inform the update of the hidden state of the red directional connection (i.e., 2 → 1 in (a)). Messages from the green directional connection in (b) (i.e., 5 → 1 in (b)) are used to inform the update of the hidden state of the purple directional connection (i.e., 1 → 2 in (b)). Figure 4B (c) shows the update function of the hidden representation form of the red directed combination in (a) (i.e., 2 → 1 in (a)), which is an iterative process and can be repeated multiple times (e.g., 5 times). Concat, shown in Figure 4B, is a policy in deep learning that can effectively handle input samples of variable size.

[0116] For a drug molecule, before being input to the graph encoder, the drug molecule can be first converted into a connected graph structure with corresponding chemical bonds and atomic attributes on the edges and points. The connected graph structure corresponding to such a drug molecule is:

number

[0117] In the present embodiment, as shown in FIG. 5, the atom masking and readout function portion is mainly used to obtain latent vector representations of all atoms (i.e., latent vectors of atoms corresponding to drug molecules) and then obtain latent vector representations of the backbone and side chains, i.e., backbone latent vectors and side chain latent vectors, by masking and readout to atoms.

[0118] For example, the atom masking process can be performed based on a bit vector, and the length of the bit vector is the same as the number of atoms contained in the drug molecule. For example, the bit vector can be represented by the above formula (6).

[0119] In the present embodiment, the readout function is used to obtain the backbone latent vector and the side chain latent vector, and further, in the present embodiment, a selective self-attention mechanism is used. The atomic latent vector of one drug molecule is expressed as H node Assuming this, the backbone latent vector and the side chain latent vector can be calculated by the above-mentioned formulas (7) and (8), respectively.

[0120] In the present embodiment, the Gaussian mixture fitting process mainly realizes the Gaussian mixture distribution of the skeleton latent vector to realize the processing of different hopping degrees. For the side chain, Gaussian distribution fitting can be performed, or the distribution hypothesis can be not performed. In the present embodiment, in order to better maintain the side chain, the AutoEncoder method can be adopted to process the hidden space of the side chain without performing the Gaussian distribution hypothesis.

[0121] In the present embodiment, the skeletons of existing molecules can be divided into M different skeleton clusters in advance by a skeleton clustering algorithm. The existing molecules may be sample molecules used to train a machine learning model, or may be molecules selected from a molecular library, and these molecules are not limited to drug molecules. In the skeleton hidden space, it is desirable that points of the same skeleton cluster are close to each other and points of different skeleton clusters are far from each other, so that the cluster centers of the M hidden spaces are defined as (μ m ,σ m ), where μ m represents the cluster center, and σ m represents the standard deviation. Furthermore, by the above formula (9), the skeleton latent vector Z sca,i and the distance d between the cluster center of the mth skeleton cluster i At the same time, the distance d i After the distance is calculated, the distance-based cross entropy loss L cls can be calculated.

[0122] In the present embodiment, the decoder can adopt a SMILES decoder, i.e., decode the hidden layer representation into a SMILES rather than a graph. SMILES can be understood as a spanning tree of a graph unfolded according to rules, and each drug molecule can have a corresponding canonical SMILES, so it is reasonable for the decoder to use SMILES. As shown in FIG. 7, the decoder can follow a teacher forcing mode, the operation principle of which is to use the real data (Ground Truth) of the training data set as the input x(t+1) at the next time point at time t in the training process, instead of using the output of the model at the previous time point. 701 in FIG. 7 is the input part of the real data, and 702 is the output part of the model.

[0123] In the present embodiment, loss reconstruction (i.e., loss prediction) is performed based on the final output result of the encoder and the correct answer (i.e., actual data), and L recon Meanwhile, the loss function of the model includes the reconstruction loss and the cross entropy loss, which can be seen from the above equation (15).

[0124] After the model is trained, the model can be used to generate a molecule, and in the process of generating a molecule, a molecule needs to be input as a reference drug molecule, and the reference drug molecule is a drug molecule that needs backbone replacement. In addition, the backbone that needs replacement in the reference drug molecule can also be represented, and after the structure of the reference drug molecule is converted into the structure of a connected graph and input into the model, the model can obtain the backbone latent vector and the side chain latent vector corresponding to the reference drug molecule, as shown in Figure 8, the function of the graph encoder 801 is the same as that of the graph encoder 401 in Figure 4A, and the processing process is similar to the relevant content introduced in the above-mentioned embodiment, and detailed description will not be repeated.

[0125] In the present embodiment, after obtaining the skeleton latent vector, the molecule generation process is slightly different from the model training process. In the molecule generation process, when the target skeleton latent vector is obtained, a resampling process is required. The decoding process is as shown in Figure 9. The function of the decoder 901 is the same as that of the decoder 402 in Figure 4A. Here, the side chain latent vector is not changed, and no sampling process is performed. The skeleton latent vector is in a mixed Gaussian distribution state through model training, and such a distribution state facilitates the skeleton hopping process.

[0126] In the present embodiment, according to the hopping degree, the hopping method can be divided into three types: hopping within the same scaffold cluster (scaffold crawling), hopping to adjacent scaffold clusters (scaffold hopping), and hopping to distant scaffold clusters (scaffold leaping). As shown in Figure 10, scaffold crawling is the most minor hopping, and the change of the molecule after hopping is the smallest. Mainly, the scaffold latent vector is sampled from the scaffold cluster of the same reference drug molecule (cluster 1001 in Figure 10), and the target scaffold latent vector corresponding to the new sampling point (i.e., the newly generated scaffold latent vector) can be expressed by the above-mentioned formula (10).

[0127] Scaffold hopping is a large hopping, and the change in the molecular skeleton after hopping is large. The skeleton latent vector is sampled from the adjacent skeleton cluster of the reference drug molecule (cluster 1002 in Figure 10), and the target skeleton latent vector corresponding to the new sampling point (i.e., the newly generated skeleton latent vector) can be expressed by the above-mentioned equation (11).

[0128] Scaffold leaping is the maximum hopping, and the molecular skeleton change after hopping is very large. The skeleton latent vector is sampled from a cluster far away from the skeleton cluster of the reference drug molecule (cluster 1003 in Figure 10), and the target skeleton latent vector corresponding to the new sampling point (i.e., the newly generated skeleton latent vector) can be expressed by the above-mentioned equation (12).

[0129] Next, the side chain latent vector and the target backbone latent vector are obtained, while the activity conditions such as the pIC50 value corresponding to the target and expected value of the reference drug molecule need to be input into the model, as shown in Figure 9. After obtaining the three vectors, the side chain latent vector, the target backbone latent vector, and the activity conditions, the model can generate a new post-hopping drug molecule through the SMILES decoder.

[0130] In the present embodiment, the drug molecules generated after backbone hopping can go through the following two stages of selection: the first stage is molecular filtering based on physicochemical properties, and can be selected according to Lipinski's Rule of Five, etc., with the aim of ensuring that all molecules in the subsequent evaluation have drug-like properties; the second stage is preparing ligands for drug-like molecules that meet the requirements of physicochemical properties, and then entering the subsequent molecular docking stage, with the aim of selecting drug-like molecules that have real and strong binding ability with targets.

[0131] For example, the crystal structure for molecular docking can be searched from the Protein Data File (PDB: Protein Data Bank) database, for example, a co-crystal structure in which a reference drug molecule or its homologous compound exists can be selected, and at the same time, it is necessary to ensure that the selected crystal structure has high resolution and the structure of the protein near the binding pocket is complete. During docking, the protein is first prepared in the molecular docking software, and then the molecule is docked into the prepared crystal structure to judge the accuracy of the structure according to the binding mode, and the molecular binding mode in the co-crystal structure is also used as a template for molecular docking to analyze whether the binding mode of the molecule generated by AI is reasonable. In the technical solution of this embodiment, molecules that are obviously structurally unreasonable are removed by virtual selection, and then a more accurate molecular docking is performed for all structures retained in the previous step. Then, the obtained binding patterns are re-scored by a 3D-CNN (Convolutional Neural Network), and molecules with a 3D-CNN score of at least 0.8 (the numbers are for illustrative purposes only) and the binding patterns of important active sites are not lost are selected, and then the compound synthesis and verification can be performed.

[0132] In the above embodiment, the graph encoder may be a Dual-MPNN. However, the SMILES decoder may be replaced with various natural language processing decoders, such as a grammar-variational autoencoder (VAE), a syntax directed VAE (SD-VAE), and the decoding part of a transformer.

[0133] According to the technical solutions of the above embodiments of the present application, by mapping the skeleton latent vectors into a spatial distribution, the generated target skeleton latent vectors can jump out of the design mindset of medicinal chemistry experts, thus achieving excellent novelty, and can be automatically performed by electronic devices, thereby reducing labor and time costs.

[0134] The following describes an embodiment of the apparatus of the present application, which can be used to implement the molecular framework hopping processing method in the above embodiment of the present application. For details not disclosed in the embodiment of the apparatus of the present application, reference can be made to the embodiment of the molecular framework hopping processing method in the above embodiment of the present application.

[0135] FIG. 11 shows a block diagram of a molecular scaffold hopping processing device according to an embodiment of the present application, which can be installed in a device having a computing function, such as the server 130 shown in FIG.

[0136] As shown in FIG. 11, the molecular scaffold hopping processing device 1100 of the embodiment of the present application includes a first generation unit 1102, a first processing unit 1104, a second generation unit 1106, and a third generation unit 1108.

[0137] Here, the first generation unit 1102 is configured to generate an atomic latent vector corresponding to a reference drug molecule based on a structure of a connected graph corresponding to the reference drug molecule; the first processing unit 1104 is configured to perform an atomic masking process on the atomic latent vector to obtain a backbone latent vector and a side chain latent vector contained in the atomic latent vector; the second generation unit 1106 is configured to generate a target backbone latent vector having a target hopping degree for the backbone latent vector based on the spatial distribution of the backbone latent vector; and the third generation unit 1108 is configured to generate a drug molecule after hopping based on the target backbone latent vector and the side chain latent vector.

[0138] In some embodiments, based on the above-mentioned technical proposal, the nodes in the structure of the connected graph represent atoms of the reference drug molecule; the first generation unit 1102 is configured as follows: based on the node features and edge features contained in the structure of the connected graph, determine node information of each node in the structure of the connected graph through a graph encoder, where the node features represent atomic features of the reference drug molecule, and the edge features represent features between atoms of the reference drug molecule; based on the node information of each node and the node features of each node, generate a latent vector of each node; based on the latent vector of each node and the atoms contained in the reference drug molecule, generate an atomic latent vector corresponding to the reference drug molecule.

[0139] In some embodiments, based on the above technical solutions, the first generation unit 1102 is configured as follows: the graph encoder includes a plurality of cascaded hidden layers; and determines information in a second hidden layer between the first node and the second node based on node features of a first node in the structure of the connected graph, node features of a second node in the structure of the connected graph, and edge information in the first hidden layer between the first node and other nodes other than the second node in the adjacent nodes of the first node, where the first node is any node in the structure of the connected graph, and the second node is any node in the structure of the connected graph. an arbitrary adjacent node of a node, the second hidden layer being a next hidden layer of the first hidden layer; determining edge information in the second hidden layer between the first node and the second node based on edge information in the first hidden layer between the first node and the other node and information in the second hidden layer between the first node and the second node; where the edge information in the initial hidden layer between two nodes in the structure of the connected graph is obtained based on the node features of one of the two nodes and the edge features between the two nodes; obtaining node information of each node by adding up corresponding edge information in the multiple hidden layers of each node.

[0140] In some embodiments, based on the above-mentioned technical proposal, the first processing unit 1104 is configured to: determine a bit vector corresponding to the reference drug molecule, where the length of the bit vector is the same as the number of atoms contained in the reference drug molecule, and the bit value corresponding to a backbone atom in the bit vector is a first value; perform a sorting process on the atom latent vector corresponding to the reference drug molecule based on the bit vector to obtain the backbone atom latent vector and the side chain atom latent vector; perform a multi-head attention process on the backbone atom latent vector to obtain the backbone latent vector, and perform a multi-head attention process on the side chain atom latent vector to obtain the side chain latent vector.

[0141] In some embodiments, based on the above-mentioned technical proposal, the first processing unit 1104 is further configured to: obtain a plurality of skeletal clusters, and the cluster center of each skeletal cluster in the plurality of skeletal clusters is matched to a mixed Gaussian distribution; determine a first distance between the skeletal latent vector and the cluster center of each skeletal cluster, and determine a target skeletal cluster to which the skeletal of the reference drug molecule belongs based on the first distance; and determine a mixed Gaussian distribution to which the skeletal latent vector belongs according to the cluster center of the target skeletal cluster.

[0142] In some embodiments, based on the above-mentioned technical solutions, the second generation unit 1106 is configured as follows: perform a random sampling process on the target skeleton cluster based on the target hopping degree to obtain an offset amount corresponding to the target hopping degree; perform an addition process between the skeleton latent vector and the offset amount corresponding to the target hopping degree to obtain the target skeleton latent vector.

[0143] In some embodiments, based on the above-mentioned technical solution, the second generation unit 1106 is configured to, when the target hopping degree is a first hopping degree, perform an integration process of the variance of the target skeleton cluster and a first vector obtained by random sampling to obtain a first offset amount, and use the first offset amount as an offset amount corresponding to the first hopping degree, where the first hopping degree represents hopping within the same skeleton cluster.

[0144] In some embodiments, based on the above-mentioned technical solutions, the second generation unit 1106 is configured as follows: when the target hopping degree is a second hopping degree, select a first skeletal cluster from the multiple skeletal clusters, and the distance between the cluster center of the first skeletal cluster and the cluster center of the target skeletal cluster is less than or equal to a first set value, and the second hopping degree indicates hopping to an adjacent skeletal cluster; generate a second offset amount based on the variance of the first skeletal cluster and the multiplication of a second vector obtained by random sampling, and the cluster center of the target skeletal cluster and the cluster center of the first skeletal cluster, and use the second offset amount as the offset amount corresponding to the second hopping degree.

[0145] In some embodiments, based on the above-mentioned technical solutions, the second generation unit 1106 is configured as follows: if the specified hopping degree is a third hopping degree, select a second skeletal cluster from the multiple skeletal clusters, and the distance between the cluster center of the second skeletal cluster and the cluster center of the target skeletal cluster is greater than or equal to a second set value; generate a third offset amount based on the multiplication of the variance of the second skeletal cluster and a third vector obtained by random sampling, and the cluster center of the target skeletal cluster and the cluster center of the second skeletal cluster, and use the third offset amount as the offset amount corresponding to the third hopping degree.

[0146] In some embodiments, based on the above-mentioned technical proposal, the third generation unit 1108 is configured to obtain the target and target activity value of the specified reference drug molecule, and generate a drug molecule after hopping based on the target backbone latent vector, the side chain latent vector, the target and the target activity value of the reference drug molecule.

[0147] In some embodiments, based on the above-mentioned technical solutions, the apparatus 1100 further includes a second processing unit, which is configured to: after generating the hopping drug molecule, perform a molecular filtration process of physicochemical properties based on the hopping drug molecule to obtain a drug molecule with drug-like properties; obtain a eutectic structure corresponding to the reference drug molecule, and dock the drug molecule with drug-like properties into the eutectic structure; remove drug molecules that do not match the eutectic structure based on the binding mode between the drug molecule with drug-like properties and the eutectic structure to obtain a selected drug molecule; and perform compound synthesis and verification based on the docking of the selected drug molecule and the eutectic structure.

[0148] In some embodiments, based on the above-mentioned technical proposal, the processing method of molecular skeleton hopping is realized by a machine learning model, and the apparatus 1100 further includes a third processing unit, which is configured as follows: obtain a sample skeleton latent vector corresponding to a sample molecule; obtain a plurality of skeleton clusters, and the cluster center of each skeleton cluster in the plurality of skeleton clusters conforms to a mixed Gaussian distribution; determine a second distance between the sample skeleton latent vector of the sample molecule and the cluster center of each skeleton cluster, and determine the skeleton cluster to which the sample skeleton of the sample molecule belongs based on the second distance; generate a distance-based cross-entropy loss based on the distance between the sample skeleton latent vector and the cluster center of the skeleton cluster to which the sample skeleton belongs; generate a loss function of the machine learning model based on the cross-entropy loss and the predicted loss of the machine learning model for the sample molecule, and adjust parameters of the machine learning model based on the loss function.

[0149] In some embodiments, based on the above-mentioned technical proposal, the machine learning model includes a decoder, and the third processing unit is further configured to obtain a sample backbone latent vector and a sample side chain latent vector corresponding to the sample molecule through the machine learning model, and then input the sample backbone latent vector, the sample side chain latent vector and a target molecule corresponding to the sample molecule into the decoder; and determine the predicted loss based on the output of the decoder and the target molecule.

[0150] FIG. 12 is a schematic structural diagram of a computer system suitable for implementing an electronic device according to an embodiment of the present invention.

[0151] It should be noted that the computer system 1200 of the electronic device shown in FIG. 12 is merely an example and does not limit the functions and scope of use of the embodiments of the present application.

[0152] As shown in Fig. 12, the computer system 1200 includes a central processing unit (CPU) 1201, which can execute various appropriate operations and processes, such as the methods described in the above embodiments, according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage portion 1208 into a random access memory (RAM) 1203. The RAM 1203 also stores various programs and data required for system operation. The CPU 1201, the ROM 1202, and the RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to the bus 1204.

[0153] To the I / O interface 1205, there are connected an input section 1206 including a keyboard, a mouse, etc., an output section 1207 including a cathode ray tube (CRT), a liquid crystal display (LCD), a speaker, etc., a storage section 1208 including a hard disk, etc., and a communication section 1209 including a network interface card such as a LAN (Local Area Network) card and a modem. The communication section 1209 executes communication processing via a network such as the Internet. A driver 1210 is also connected to the I / O interface 1205 as necessary. A removable medium 1211 such as a disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc. is installed in the driver 1210 as necessary, and a computer program read from the driver 1210 is installed in the storage section 1208 as necessary.

[0154] In particular, according to an embodiment of the present application, the above processes described with reference to the flowcharts can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product including a computer program loaded on a computer-readable medium, the computer program including a computer program for performing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from a network via the communication portion 1209 and / or installed from a removable medium 1211. When the computer program is executed by the CPU 1201, various functions limited to the system of the present application are performed.

[0155] In addition, the computer-readable medium shown in the embodiment of the present application may be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. The computer-readable storage medium may be, for example, but is not limited to, an electric, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of the computer-readable storage medium include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may be used as a data signal contained in baseband or propagated by part of a carrier, carrying a computer-readable computer program therein. Such a propagated data signal may take a number of forms, including, but not limited to, an electromagnetic signal, an optical signal, or any suitable combination of the above. A computer-readable signal medium may be any computer-readable medium other than a computer-readable storage medium, which may transmit, propagate, or transmit a program used by or in conjunction with an instruction execution system, apparatus, or device. A computer program contained in a computer-readable medium may be transmitted in any suitable medium, including, but not limited to, wireless, wired, or any suitable combination of the above.

[0156] The flowcharts and block diagrams in the drawings illustrate possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. Here, each block in the flowchart or block diagram may represent a module, program segment, or part of code, including one or more executable instructions for implementing a certain logical function. It should be noted that in some alternative implementations, the functions described in the blocks may occur in a different order from that described in the drawings. For example, two blocks shown in succession may actually be executed substantially in parallel, or may be executed in reverse order depending on the functions involved. It should be noted that each block in the block diagram or flowchart, and combinations of blocks in the block diagram or flowchart, may be realized in a dedicated hardware-based system that executes a certain function or operation, or may be realized in a combination of dedicated hardware and computer instructions.

[0157] The units according to the embodiments of the present application may be implemented in the form of software or hardware, and the described units may be installed in a processor, where the names of these units do not necessarily constitute limitations on the units themselves.

[0158] In another aspect, the present embodiment further provides a computer-readable medium, which may be included in the electronic device described in the above embodiment, or may exist independently without being incorporated in the electronic device. The computer-readable medium stores one or more programs, and when the one or more programs are executed by one of the electronic devices, the electronic device realizes the method described in the above embodiment.

[0159] It should be noted that, although the above detailed description refers to several modules or units of an apparatus for performing operations, such division is not essential. In fact, according to an embodiment of the present application, the features and functions of two or more of the modules or units described above may be embodied in one module or unit. Conversely, the features and functions of one of the modules or units described above may be further divided so as to be further embodied by multiple modules or units.

[0160] From the above description of the embodiments, those skilled in the art can understand that the exemplary embodiments described herein may be realized by software, or may be realized by combining necessary hardware with software. Therefore, the technical solutions of the embodiments of the present application may be embodied in the form of a software product, and the software product may be stored on a non-volatile storage medium (which may be a CD-ROM, a U disk, a removable hard disk, etc.) or a network, which includes some instructions for making a computing device (such as a personal computer, a server, a touch terminal, or a network device) execute the method according to the embodiments of the present application.

[0161] Those skilled in the art may easily conceive other embodiments of the present application after considering the specification and practicing the embodiments of the present invention disclosed herein. This application is intended to cover any modifications, uses, or adaptations of the present invention. These modifications, uses, or adaptations follow the general principles of the present disclosure and include common knowledge or customary technical means well known in the art that are not disclosed in the present disclosure. The specification and examples should be considered as exemplary only, with the true scope and spirit of the invention being pointed out in the following claims.

[0162] It should be understood that the present application is not limited to the specific configurations shown in the above drawings, and various modifications and changes can be made without departing from the scope of the present application. [Explanation of symbols]

[0163] 100 System Architecture 110 Terminal 120 Network 130 Server 701 Actual data input part Output part of the 702 model 1001,1002,1003 Cluster 1100 Processing device for molecular skeleton hopping 1102 First Generation Unit 1104 First Processing Unit 1106 Second Generation Unit 1108 3rd Generation Unit 1200 Computer Systems 1201 Central Processing Unit (CPU) 1202 Read-Only Memory (ROM) 1203 Random Access Memory (RAM) 1204 Bus 1205 Input / Output (I / O) Interface 1206 Input section 1207 Output part 1208 Memory part 1209 Communication section 1210 Driver 1211 Removable Media

Claims

1. 1. A method for processing molecular backbone hopping performed by electronic equipment, comprising: generating an atomic latent vector corresponding to a reference drug molecule based on a structure of a connectivity graph corresponding to the reference drug molecule, wherein nodes in the structure of the connectivity graph represent atoms of the reference drug molecule, and the structure of the connectivity graph includes node features representing atomic features of the reference drug molecule and edge features representing features between atoms of the reference drug molecule; performing an atom masking process on the atom latent vector to obtain a backbone latent vector and a side chain latent vector included in the atom latent vector; generating a target skeleton latent vector having a target hopping degree with respect to the skeleton latent vector based on a spatial distribution of the skeleton latent vector; generating a post-hopping drug molecule based on the target backbone potential vector and the side chain potential vector; Methods for handling molecular backbone hopping.

2. The step of generating an atomic latent vector corresponding to a reference drug molecule based on a structure of a connected graph corresponding to the reference drug molecule, determining node information for each node in the connected graph structure via a graph encoder based on the node features and the edge features included in the connected graph structure; generating a latent vector for each of the nodes based on the node information of each of the nodes and the node features of each of the nodes; generating an atomic latent vector corresponding to the reference drug molecule based on the latent vector of each node and the atoms contained in the reference drug molecule; 2. The method of claim 1 for treating molecular backbone hopping.

3. the graph encoder includes a plurality of cascaded hidden layers; The step of determining node information for each node in the structure of the connected graph through a graph encoder based on the node features and the edge features included in the structure of the connected graph includes: determining information in a second hidden layer between the first node and the second node based on node characteristics of the first node in the connected graph structure, node characteristics of the second node in the connected graph structure, and edge information in a first hidden layer between the first node and other nodes other than the second node in adjacent nodes of the first node, wherein the first node is any node in the connected graph structure, the second node is any adjacent node of the first node in the connected graph structure, and the second hidden layer is a hidden layer next to the first hidden layer; determining edge information in a second hidden layer between the first node and the second node based on edge information in a first hidden layer between the first node and the other node and information in a second hidden layer between the first node and the second node, where edge information in an initial hidden layer between two nodes in the structure of the connected graph is obtained based on node features of one of the two nodes and edge features between the two nodes; adding corresponding edge information in the multiple hidden layers of each node to obtain node information of each node; The method of claim 2 , wherein the molecular backbone hopping is carried out.

4. The step of performing an atom masking process on the atom latent vector to obtain a backbone latent vector and a side chain latent vector included in the atom latent vector includes: determining a bit vector corresponding to the reference drug molecule, the length of the bit vector being the same as the number of atoms in the reference drug molecule, and bit values ​​corresponding to backbone atoms in the bit vector being first values; performing a filtering process on the atom latent vectors corresponding to the reference drug molecule based on the bit vector to obtain the backbone atom latent vectors and the side chain atom latent vectors; performing a multi-head attention process on the latent vectors of the backbone atoms to obtain the backbone latent vectors, and performing a multi-head attention process on the latent vectors of the side chain atoms to obtain the side chain latent vectors.

2. The method of claim 1 for treating molecular backbone hopping.

5. Before generating a target skeleton latent vector having a target hopping degree for the skeleton latent vector based on the spatial distribution of the skeleton latent vector, the processing method of the molecular skeleton hopping includes: obtaining a plurality of skeletal clusters, wherein a cluster center of each skeletal cluster in the plurality of skeletal clusters fits a Gaussian mixture distribution; determining a first distance between the scaffold latent vector and a cluster center of each of the scaffold clusters, and determining a target scaffold cluster to which the scaffold of the reference drug molecule belongs based on the first distance; and determining a Gaussian mixture distribution to which the skeleton latent vector belongs according to a cluster center of the target skeleton cluster.

2. The method of claim 1 for treating molecular backbone hopping.

6. The step of generating a target skeleton latent vector having a target hopping degree with respect to the skeleton latent vector based on the spatial distribution of the skeleton latent vector includes: performing a random sampling process on the target skeleton cluster based on the target hopping degree to obtain an offset amount corresponding to the target hopping degree; and adding the skeleton latent vector and an offset amount corresponding to the target hopping degree to obtain the target skeleton latent vector. The method of claim 5 for molecular backbone hopping.

7. The step of performing a random sampling process on the target skeleton cluster based on the target hopping degree to obtain an offset amount corresponding to the target hopping degree includes: When the target hopping degree is a first hopping degree, a variance of the target skeleton cluster is multiplied with a first vector obtained by random sampling to obtain a first offset amount, and the first offset amount is used as an offset amount corresponding to the first hopping degree, where the first hopping degree represents hopping within the same skeleton cluster. The method of claim 6 for molecular backbone hopping.

8. The step of performing a random sampling process on the target skeleton cluster based on the target hopping degree to obtain an offset amount corresponding to the target hopping degree includes: selecting a first skeletal cluster from the plurality of skeletal clusters when the target hopping degree is a second hopping degree, where a distance between a cluster center of the first skeletal cluster and a cluster center of the target skeletal cluster is less than or equal to a first set value, and the second hopping degree indicates hopping to an adjacent skeletal cluster; multiplying the variance of the first skeleton cluster by a second vector obtained by random sampling, generating a second offset amount based on a cluster center of the target skeleton cluster and a cluster center of the first skeleton cluster, and using the second offset amount as an offset amount corresponding to the second hopping degree. The method of claim 6 for molecular backbone hopping.

9. The step of performing a random sampling process on the target skeleton cluster based on the target hopping degree to obtain an offset amount corresponding to the target hopping degree includes: selecting a second skeletal cluster from the plurality of skeletal clusters when the target hopping degree is a third hopping degree, wherein a distance between a cluster center of the second skeletal cluster and a cluster center of the target skeletal cluster is equal to or greater than a second set value; multiplying the variance of the second skeleton cluster by a third vector obtained by random sampling; generating a third offset amount based on the cluster center of the target skeleton cluster and the cluster center of the second skeleton cluster; and using the third offset amount as an offset amount corresponding to the third hopping degree. The method of claim 6 for molecular backbone hopping.

10. The step of generating a post-hopping drug molecule based on the target backbone potential vector and the side chain potential vector comprises: obtaining a target and a target activity value for the designated reference drug molecule; generating a post-hopping drug molecule based on the target backbone latent vector, the side chain latent vector, the target of the reference drug molecule, and the target activity value; 2. The method of claim 1 for treating molecular backbone hopping.

11. After generating a post-hopping drug molecule based on the target backbone potential vector and the side chain potential vector, the molecular backbone hopping processing method includes: performing a molecular filtering process of physicochemical properties based on the drug molecules after hopping to obtain drug molecules with drug-like properties; obtaining a eutectic structure corresponding to the reference drug molecule and docking the drug molecule with drug-like properties into the eutectic structure; removing drug molecules that do not match the eutectic structure based on a binding mode between the drug molecules having the drug-like properties and the eutectic structure to obtain selected drug molecules; and performing compound synthesis and validation based on docking of the selected drug molecule with the eutectic structure.

2. The method of claim 1 for treating molecular backbone hopping.

12. The molecular backbone hopping process is realized by a machine learning model; The molecular skeleton hopping treatment method includes: obtaining sample skeletal latent vectors corresponding to sample molecules and obtaining a plurality of skeletal clusters, wherein a cluster center of each skeletal cluster in the plurality of skeletal clusters conforms to a Gaussian mixture distribution; determining a second distance between the sample skeleton latent vector of the sample molecule and a cluster center of each skeleton cluster, and determining a skeleton cluster to which the sample skeleton of the sample molecule belongs based on the second distance; generating a distance-based cross-entropy loss based on the distance between the sample skeleton latent vector and a cluster center of a skeleton cluster to which the sample skeleton belongs; generating a loss function for the machine learning model based on the cross-entropy loss and a predicted loss of the machine learning model for the sample molecules; and adjusting parameters of the machine learning model based on the loss function. A process for molecular backbone hopping according to any one of claims 1 to 11.

13. the machine learning model includes a decoder; The molecular skeleton hopping treatment method includes: After obtaining a sample backbone latent vector and a sample side chain latent vector corresponding to the sample molecule through the machine learning model, inputting the sample backbone latent vector, the sample side chain latent vector, and a target molecule corresponding to the sample molecule into the decoder; determining the predicted loss based on an output of the decoder and the target molecule.

13. The method of claim 12, wherein the molecular backbone hopping is

14. 1. An apparatus for processing molecular scaffold hopping, comprising: a first generating unit configured to generate an atomic latent vector corresponding to a reference drug molecule based on a structure of a connectivity graph corresponding to the reference drug molecule, wherein nodes in the structure of the connectivity graph represent atoms of the reference drug molecule, and the structure of the connectivity graph includes node features representing atomic features of the reference drug molecule and edge features representing features between atoms of the reference drug molecule; a first processing unit configured to perform an atom masking process on the atom latent vector to obtain a backbone latent vector and a side chain latent vector included in the atom latent vector; a second generating unit configured to generate a target skeleton latent vector having a target hopping degree with respect to the skeleton latent vector based on a spatial distribution of the skeleton latent vector; a third generation unit configured to generate a post-hopping drug molecule based on the target backbone potential vector and the side chain potential vector; A molecular scaffold hopping processing device comprising:

15. A computer readable medium having stored thereon a computer program causing a processor to carry out the method of processing molecular backbone hopping according to any one of claims 1 to 13.

16. An electronic device, one or more processors; and a memory device storing one or more programs, the one or more programs, when executed by the one or more processors, causing the one or more processors to perform the molecular scaffold hopping processing method described in any one of claims 1 to 13.

17. A computer program configured to cause a computer to carry out the method for processing molecular backbone hopping according to any one of claims 1 to 13.

Citation Information

Patent Citations

  • Systems and methods for applying a convolutional network to spatial data

    CN108140131A

  • Method and system for calculating similarities between compound molecules and applications thereof

    CN108205613A

  • Method and equipment for generating keywords

    CN111209468A

  • Drug design cloud computing process control system and method based on virtual reality

    CN112201301A

  • Use of feature point pharmacophores (fepops)

    WO2005081158A2