Method, device and related equipment for predicting synergistic score of drug combination

Abstract

The application discloses a synergistic score prediction method and device of a drug combination, an intelligent terminal and a storage medium, and comprises the following steps: generating a meta-path instance set for a drug node in a heterogeneous drug-target interaction network according to a preset meta-path type, converting a meta-path instance in the meta-path instance set into a unified dimension meta-path instance, converting the unified dimension meta-path instance into a meta-path instance representation containing interaction information, and weighting and aggregating the meta-path instance representation according to an instance set type to obtain an embedding representation of the drug node based on different preset meta-path types; fusing the embedding representation according to an attention mechanism to obtain an overall meta-path embedding representation of the drug node; and feeding the overall meta-path embedding representation, drug chemical characteristic information and cell line information into a preset AE prediction module and a preset TE prediction module to obtain a synergistic score of the drug combination, thereby improving the accuracy of synergistic score prediction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and more particularly to a method, apparatus, smart terminal, and storage medium for predicting the synergistic effect score of drug combinations. Background Technology

[0002] Combination therapy is an important approach to treating complex diseases because using combinations of drugs that target different molecular mechanisms of the same cell can not only improve the overall therapeutic effect but also reduce the required concentration of each drug in the combination, thereby further reducing potential toxicity. However, due to the large size of the combination space, identifying new synergistic combinations is challenging. High-throughput screening methods based on deep learning for predicting the therapeutic effect (TE) or adverse effect (AE) of combination therapy have played an important role in drug discovery because they can reduce the cost of corresponding wet experiments by rapidly screening a wide range of high-confidence drug combinations.

[0003] In existing technologies, deep learning-based methods for predicting synergistic effects of drug combinations typically treat TE prediction and AE prediction as two separate tasks, ignoring the potential mechanistic commonalities between TE and AE, which leads to a decrease in the accuracy of synergistic effect score prediction.

[0004] Therefore, existing technologies still need improvement and development. Summary of the Invention

[0005] The main objective of this invention is to provide a method, device, smart terminal, and storage medium for predicting the synergistic effect score of drug combinations. This invention aims to address the problem in the prior art where TE prediction and AE prediction are treated as two independent tasks, ignoring the potential mechanistic commonalities between TE and AE, which leads to a decrease in the accuracy of synergistic effect score prediction.

[0006] To achieve the above objectives, the first aspect of the present invention provides a method for predicting the synergistic effect score of a drug combination, specifically, the method for predicting the synergistic effect score includes:

[0007] Acquire drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data;

[0008] A set of metapath instances is generated based on the drug nodes in the heterogeneous drug-target interaction network with the preset metapath type, and the metapath instances in the set of metapath instances are converted into unified dimension metapath instances to obtain a unified dimension metapath instance set.

[0009] The unified dimension metapath instance is converted into a metapath instance representation containing interaction information, and the metapath instance representation is weighted and aggregated according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types.

[0010] The embedding representations are fused according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node;

[0011] The metapath embedding representation, medicinal chemical feature information, and cell line information are fed into a preset AE prediction module and a preset TE prediction module to obtain the synergistic effect score of the drug combination.

[0012] Optionally, converting the unified dimension metapath instance into a metapath instance representation containing interaction information specifically includes: sending the unified dimension metapath instance set into a BiGRU aggregator of the corresponding instance set type to obtain a metapath instance representation containing interaction information based on a specific metapath.

[0013] Optionally, the step of weighted aggregation of the metapath instance representations according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types specifically includes: sending the metapath instance representations into the GAT extractor of the corresponding instance set type to obtain the embedding representation of the drug node based on the instance set type.

[0014] Optionally, the meta-path instance set includes: a first meta-path instance set, a second meta-path instance set, a third meta-path instance set, and a fourth meta-path instance set, wherein,

[0015] In the first metapath instance set, the starting drug node and the ending drug node of the metapath instance interact with the same target.

[0016] In the second metapath instance set, the starting drug node and the ending drug node of the metapath instance interact with two targets respectively, and the two targets interact with each other.

[0017] In the set of third-level metapath instances, the starting drug node and the ending drug node of the metapath instance interact with two targets, and the two targets interact with another target.

[0018] In the fourth metapath instance set, the synergistic score generated by the combined action of the starting drug node and the ending drug node of the metapath instance is higher than a certain threshold.

[0019] Optionally, converting the meta-path instances in the meta-path instance set into unified-dimensional meta-path instances to obtain a unified-dimensional meta-path instance set specifically includes: converting the meta-path instances in the four meta-path instance sets into tensors according to one-hot encoding of different node types, and unifying the feature lengths of each node in the tensors according to the transformation matrix of different node types to obtain four unified-dimensional meta-path instance sets.

[0020] Optionally, converting the unified dimension metapath instance into a metapath instance representation containing interaction information specifically includes: sending the four unified dimension metapath instance sets into four BiGRU aggregators of the corresponding instance set type, and converting the unified dimension metapath instances in the four unified dimension metapath instance sets into metapath instance representations containing interaction information.

[0021] Optionally, the step of weighted aggregation of the metapath instance representations according to the instance set type to obtain the embedding representations of the drug node based on different preset metapath types specifically includes: sending the metapath instance representations corresponding to the unified dimension metapath instances in the four unified dimension metapath instance sets into the GAT extractor of the corresponding instance set type, weighted aggregation of the metapath instance representations to obtain the embedding representations of four different metapaths.

[0022] A second aspect of the present invention provides a synergistic effect score prediction device for drug combinations, specifically, the synergistic effect score calculation device includes:

[0023] The module is used to acquire drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data.

[0024] The preprocessing module is used to generate a set of metapath instances for drug nodes in the heterogeneous drug-target interaction network according to a preset metapath type, and convert the metapath instances in the set of metapath instances into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set.

[0025] The processing module is used to convert the unified dimension metapath instance into a metapath instance representation containing interaction information, and to perform weighted aggregation of the metapath instance representation according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types.

[0026] The fusion module is used to fuse the embedding representations according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node;

[0027] The prediction module is used to send the overall metapath embedding representation, medicinal chemical feature information and cell line information into the preset AE prediction module and the preset TE prediction module, respectively, to obtain the synergistic effect score of the drug combination.

[0028] A third aspect of the present invention provides a smart terminal, the smart terminal including a memory, a processor, and a synergistic effect score prediction program for a drug combination stored in the memory and executable on the processor, wherein when the synergistic effect score prediction program for the drug combination is executed by the processor, it implements the steps of the synergistic effect score prediction method for a drug combination as claimed in any one of the claims.

[0029] A fourth aspect of the present invention provides a computer-readable storage medium storing a synergistic effect score prediction program for a drug combination, wherein the synergistic effect score prediction program for a drug combination, when executed by a processor, implements the steps of the synergistic effect score prediction method for a drug combination as claimed in any one of the claims.

[0030] As can be seen from the above, the synergistic effect score prediction method in the present invention includes: acquiring drug combination data; constructing a heterogeneous drug-target interaction network based on the drug combination data; generating a metapath instance set for drug nodes in the heterogeneous drug-target interaction network according to a preset metapath type; converting the metapath instances in the metapath instance set into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set; converting the unified-dimensional metapath instances into metapath instance representations containing interaction information; weighting and aggregating the metapath instance representations according to the instance set type to obtain the embedding representations of the drug nodes based on different preset metapath types; fusing the embedding representations according to an attention mechanism to obtain the overall metapath embedding representation of the drug nodes; and sending the overall metapath embedding representation, drug chemical feature information, and cell line information into a preset AE prediction module and a preset TE prediction module to obtain the synergistic effect score of the drug combination. Compared with the prior art, in this invention, the interaction information between drugs and targets is obtained by generating a set of metapath instances for drug nodes according to a preset metapath type. Furthermore, the unified dimension metapath instances are converted into metapath instance representations containing interaction information based on the preset metapath type. The metapath instance representations are then weighted and aggregated according to the instance set type to learn the interaction information between drugs and targets, i.e., the drug's mechanism of action. Through the added preset AE prediction module, the AE label information is effectively utilized based on backpropagation to optimize and adjust the model parameters related to TE prediction, thereby improving the accuracy of synergistic effect score prediction. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart illustrating a method for predicting the synergistic effect score of a drug combination according to an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of the structure of a synergistic effect score prediction device for drug combinations provided in an embodiment of the present invention;

[0034] Figure 3 This is a schematic diagram of the principle of a smart terminal provided in an embodiment of the present invention. Detailed Implementation

[0035] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of the invention. However, those skilled in the art will understand that the invention can be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods are omitted so as not to obscure the description of the invention with unnecessary detail.

[0036] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0037] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0038] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0041] High-throughput screening methods based on deep learning for predicting the therapeutic effect (TE) or adverse effect (AE) of drug combination therapy play a crucial role in drug discovery because they can effectively reduce the cost of corresponding wet experiments by rapidly screening a wide range of high-confidence drug combinations. In existing technologies, TE and AE prediction methods are often treated as independent tasks, neglecting the potential mechanistic commonalities between TE and AE. For example, relevant TE and AE of a drug can jointly reveal the drug's mechanism of action (MoA). Drug targets are the binding sites where drugs interact with the human body, including genes, receptors, enzymes, ion channels, transporters, and nucleic acids. The binding of drugs to these sites influences changes in biological events, thereby achieving the drug's therapeutic effect. Information on the interaction between drugs and targets is fundamental to modern drug discovery and development.

[0042] Existing models treat therapeutic efficacy (TE) prediction as a single task, inputting sample features and outputting the corresponding TE prediction results. The effectiveness of TE is typically quantified using therapeutic synergy scores, such as Bliss, Highest Single Agent (HSA), Loewe, and Zero Interaction Potency (ZIP). However, these methods rarely consider incorporating adverse event (AE) information, which shares mechanistic commonalities with TE. Superficially, the commonality lies in the fact that the features used for TE prediction can also be applied to AE prediction. At a deeper level, both TE and AE are measurable mechanistic changes caused by combined drug therapy, and both have been shown to be potentially important feature factors in mutual prediction. Therefore, the lack of AE information may lead to suboptimal datasets for the model, thus affecting the final TE prediction results. Furthermore, drug-target interaction information is generally constructed in the form of biological interaction networks, which are significantly heterogeneous from AE information, and effectively learning these two heterogeneous information sets presents challenges.

[0043] To address one or more problems in the prior art, the present invention provides a method for predicting the synergistic effect score of drug combinations, which helps to improve the accuracy of synergistic effect score prediction.

[0044] Exemplary methods

[0045] like Figure 1 As shown, this embodiment of the invention provides a method for predicting the synergistic effect score of a drug combination. Specifically, the synergistic effect score prediction method includes:

[0046] Step S100: Obtain drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data.

[0047] In this embodiment, drug combination data is acquired, and a heterogeneous drug-target interaction network is constructed based on the drug combination data. The network consists of two types of nodes: drug and target, and three different relationships between nodes. The heterogeneous drug-target interaction network includes three types of heterogeneous edges: drug-target interaction edges, target-target interaction edges, and drug-drug TE relationship edges.

[0048] Step S200: Generate a metapath instance set for drug nodes in the heterogeneous drug-target interaction network according to the preset metapath type, and convert the metapath instances in the metapath instance set into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set.

[0049] In this embodiment, based on the above heterogeneous edge types, nodes and prior knowledge, four meta-path types are designed for the drug node. A specific meta-path can be defined as a node sequence formed by a specific edge combination from the start node to the end node. The meta-path name and biological meaning of the meta-path are shown in Table 1, where D is the drug node and T is the target node.

[0050]

[0051]

[0052] Table 1

[0053] In this embodiment, based on predefined metapath types, four metapath instance sets (a metapath instance refers to a sequence of node data that follows a certain type of metapath definition) are generated for each drug node in the drug combination data of the heterogeneous drug-target interaction network. Each metapath instance in the instance set is then converted into a tensor with a uniform dimension. In this way, the drug's mechanism of action is explicitly simulated through metapaths.

[0054] Specifically, referring to Table 1, the metapath instance set includes: a first metapath instance set, a second metapath instance set, a third metapath instance set, and a fourth metapath instance set. In the first metapath instance set, the starting and ending drug nodes of the metapath instance interact with the same target. In the second metapath instance set, the starting and ending drug nodes of the metapath instance interact with two targets respectively, and the two targets interact with each other. In the third metapath instance set, the starting and ending drug nodes of the metapath instance interact with two targets respectively, and the two targets interact with another target. In the fourth metapath instance set, the synergistic effect score generated by the combined action of the starting and ending drug nodes of the metapath instance is higher than a certain threshold.

[0055] Furthermore, the meta-path instances in the four meta-path instance sets are converted into tensors based on one-hot encoding of different node types, and the feature lengths of each node in the tensors are unified according to the transformation matrix of different node types to obtain four unified-dimensional meta-path instance sets.

[0056] For example, taking the generation of a metapath instance set for a drug node D2 as an example, firstly, the heterogeneous drug-target interaction network is traversed, and four metapath instance sets (DTD, DTTD, DTTTD, DD) are generated according to the predefined metapath types. In each metapath instance set, the starting node of each metapath instance is any drug node, and the ending node is D2. Then, using node type-specific one-hot encoding (that is, one one-hot encoding set is used for each node type), all metapath instances in the four metapath instance sets are converted into tensors. At this time, since the number of nodes of each type in the heterogeneous network is different, the feature dimensions of each node generated based on one-hot encoding in each metapath instance are different. Therefore, the feature length of each node can be unified according to the node type-specific transformation matrix. For details, please refer to formula (1):

[0057] in, It is the new feature of the node after transformation, W t It is a transformation matrix specific to node type t. It is the unique hot feature of node i of node type t. Since the output dimension of each transformation matrix is ​​the same, four metapath instance sets with the same dimension can be obtained for drug node D2.

[0058] Step S300: Convert the unified dimension metapath instance into a metapath instance representation containing interaction information, and perform weighted aggregation of the metapath instance representation (overall representation) according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types.

[0059] In this embodiment, the unified dimension metapath instance set is fed into a BiGRU aggregator of the corresponding instance set type to obtain a metapath instance representation containing interaction information based on a specific metapath. The metapath instance representation is then fed into a GAT extractor of the corresponding instance set type to obtain an embedding representation of the drug node based on the instance set type.

[0060] Specifically, the four unified-dimensional meta-path instance sets are fed into four BiGRU aggregators of the corresponding instance set type, and the unified-dimensional meta-path instances in the four unified-dimensional meta-path instance sets are converted into meta-path instance representations containing interaction information based on specific meta-paths. That is, meta-path instances in different meta-path sets are converted into low-dimensional representations containing implicit interaction information of specific meta-paths.

[0061] For example, taking the drug node D2 mentioned above, the four meta-path instance sets (unified dimension meta-path instance sets) corresponding to drug node D2 with a unified dimension are sent into the four independent BiGRU aggregators. Through the aggregator, the overall representation of each meta-path instance (unified dimension meta-path instance) is obtained, and the representation of the ending drug node D2 in the instance is also obtained. This is used to generate the low-dimensional embedding of the corresponding drug node based on the specific class meta-path.

[0062] It should be noted that BiGRU is a bidirectional gated recurrent unit, which consists of two recurrent layers of GRU gated recurrent units with opposite information transmission. Since each metapath instance is a reversible sequence of node features, learning the instance sequence from the forward direction to extract its forward interaction information and learning from the backward direction to obtain its inverse interaction information both have practical biological significance. In this embodiment, BiGRU can better learn the bidirectional interaction information contained in each metapath instance.

[0063] Specifically, the meta-path instance representations corresponding to the unified dimension meta-path instances in the four unified dimension meta-path instance sets are fed into the GAT extractor of the corresponding instance set type, and the meta-path instance representations are weighted and aggregated to obtain the embedded representations of four different meta-paths.

[0064] In this embodiment, the metapath instance representations (overall representations) of all metapath instances in each metapath instance set of the drug node are sent to the independent GAT extractors corresponding to these instance set types. The metapath instance representations (overall representations) are weighted and aggregated according to different metapath types to obtain a set of embedding representations of the drug node based on different metapaths. That is, the GAT extractors are used to perform weighted summation on the newly generated low-dimensional representations in multiple metapath instance sets corresponding to each drug node based on different instance sets to obtain four embedding representations of the node specific to the metapath type.

[0065] For example, taking the drug node D2 mentioned above as an example, the meta-path instance representation (overall representation) corresponding to each meta-path instance in the four instance sets (DTD, DTTD, DTTTD, DD) of the drug node D2 is sent to the GAT extractor specific to the meta-path type.

[0066] It should be noted that GAT is a variant of the standard Graph Convolutional Networks (GCN), which can provide more accurate weights for the meta-path instance representation (overall representation) of each meta-path instance when aggregating information. However, since the training of the GAT extractor is unstable, the entire process needs to be repeated independently K times, and the results of the K times need to be concatenated.

[0067] Specifically, different GAT extractors can generate four independent representations for the D2 node based on the four meta-paths mentioned above, as detailed in formula (2):

[0068]

[0069] in, σ is the weight assigned to the overall representation of drug node j based on the i-th meta-path instance of meta-path m, and σ is the activation function. It is a trainable attention vector for the meta-path m. It represents the end of drug node j in the current metapath instance. It is the metapath instance representation (overall representation) of the current metapath instance. It refers to all metapath instances of node j in the metapath m instance set. It represents the end of drug j in a metapath m that starts with drug node k and ends with drug j. It is the overall representation of the instance from k to j in the instance set of the metapath m.

[0070] By concatenating the K results, a set of embedding representations of the drug node based on different meta-paths is obtained, as detailed in formula (3):

[0071]

[0072] in, Let σ be the embedding representation of drug node j on the metapath m, and σ be the activation function. It refers to all metapath instances of node j in the metapath m instance set. It is the weight assigned to the overall representation of drug node j based on the i-th metapath instance of metapath m. It is the metapath instance representation (overall representation) of the current metapath instance.

[0073] Step S400: The embedding representations are fused according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node.

[0074] In this embodiment, an attention mechanism is used to fuse the four metapath-specific embedding representations of the drug node to obtain a holistic metapath embedding representation that integrates comprehensive information from different metapaths.

[0075] For example, taking the drug node D2 mentioned above as an example, the embedding representations of D2 obtained in step S300 under the four meta-paths are fused using an attention mechanism to obtain the overall meta-path embedding representation of D2. For details, please refer to formulas (4), (5), (6) and (7):

[0076]

[0077]

[0078]

[0079]

[0080] in, V is the importance weight of metapath m when merging node type t. t It is a set of nodes of node type t. W is the trainable attention coefficient of node type t. t It is the metapath combination transformation matrix of node type t. b is the embedding representation of drug node i on metapath m. t It is a trainable bias vector of node type t.

[0081] Furthermore, These are the importance weights of the metapath m under node type t after normalization using the Softmax function. It is a mapping matrix of node type t. It is the global meta-path embedding representation of node i of node type t.

[0082] In step S500, the overall metapath embedding representation, drug chemical feature information, and cell line information are sent to the preset AE prediction module and the preset TE prediction module to obtain the synergistic effect score of the drug combination.

[0083] It should be noted that in this embodiment, the preset AE prediction module and the preset TE prediction module can be obtained through existing technologies. For example, the preset AE prediction module and the preset TE prediction module can be obtained by referring to the literature (Preuer, K., et al. DeepSynergy: predicting anti-cancer drugsynergy with Deep Learning. 2018; 34(9):1538-1546.), which will not be elaborated here.

[0084] In this embodiment, the overall metapath embedding representation, drug chemical information features, and cell line information corresponding to the drug combination to be tested (drug-drug-cell line) are sent to the preset AE prediction module to obtain the AE probability score, as detailed in formula (8):

[0085]

[0086] Among them, P ij AE σ is the probability score of AE, σ is the activation function, and W is the activation score of AE. AE It is a trainable parameter matrix, [...] represents the feature concatenation operation. It is the global meta-path embedding representation corresponding to drug i. These are the medicinal chemical information features corresponding to drug i. It is the global metapath embedding representation corresponding to drug j. It is the pharmaceutical chemical information feature corresponding to drug j.

[0087] The AE probability score, the overall metapath embedding representation corresponding to the drug combination to be tested (drug-drug-cell line), the drug chemical information features and the cell line information are sent to the preset TE prediction module to obtain the synergistic effect score of the drug combination to be tested, as shown in formula (9):

[0088]

[0089] Among them, DNN TE This is a deep neural network decoder, [...] represents the feature concatenation operation. It is the global meta-path embedding representation corresponding to drug i. These are the medicinal chemical information features corresponding to drug i. It is the global metapath embedding representation corresponding to drug j. These are the medicinal chemical information characteristics corresponding to drug j. It is cell line information, P ij AE It is the probability score of AE.

[0090] This invention describes the TE prediction task as a multi-task heterogeneous network synergistic fractional regression task. For drug-target interaction information, it draws on the concept of metapaths in the prior art, extracts reaction pathway information in the drug-target interaction network through a variety of predefined biologically meaningful metapaths, and learns this information through a graph convolutional network (GCN) based on a bidirectional gate recurrent unit (BiGRU) aggregator. Furthermore, for AE information, which is heterogeneous with the reaction information, an AE auxiliary task is added that shares the network backbone with the TE prediction main task. Based on backpropagation, the AE tag information is effectively utilized to optimize and adjust the model parameters related to TE prediction. In the prior art, since the interaction information between drugs and targets is constructed in the form of a biological interaction network, it has obvious heterogeneity with AE information. At the same time, it is difficult to effectively learn these two heterogeneous information. That is to say, since the form and content of drug target network data and AE data are different, it is impossible to integrate them. In this embodiment, the interaction information between drugs and targets is learned in the forward direction, and the AE information is learned in the reverse direction through the added AE module. The backpropagation optimizes the model parameters, thereby achieving effective learning of the two heterogeneous information.

[0091] In summary, this invention can effectively utilize interaction information and AE information while capturing the mechanistic commonalities between TE and AE, thereby potentially achieving better TE prediction performance than the standard single-task model that does not use AE information.

[0092] As can be seen from the above, the synergistic effect score prediction method provided in this embodiment of the invention includes: acquiring drug combination data; constructing a heterogeneous drug-target interaction network based on the drug combination data; generating a metapath instance set for drug nodes in the heterogeneous drug-target interaction network according to a preset metapath type; converting the metapath instances in the metapath instance set into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set; converting the unified-dimensional metapath instances into metapath instance representations containing interaction information; weighting and aggregating the metapath instance representations according to the instance set type to obtain the embedding representations of the drug nodes based on different preset metapath types; fusing the embedding representations according to an attention mechanism to obtain the overall metapath embedding representation of the drug nodes; and sending the overall metapath embedding representation, drug chemical feature information, and cell line information into a preset AE prediction module and a preset TE prediction module to obtain the synergistic effect score of the drug combination. Compared with the prior art, in this invention, the interaction information between drugs and targets is obtained by generating a set of metapath instances for drug nodes according to a preset metapath type. Furthermore, the unified dimension metapath instances are converted into metapath instance representations containing interaction information based on the preset metapath type. The metapath instance representations are then weighted and aggregated according to the instance set type to learn the interaction information between drugs and targets, i.e., the drug's mechanism of action. A preset AE prediction module is added, which effectively utilizes AE label information based on backpropagation to optimize and adjust the model parameters related to TE prediction, thereby improving the accuracy of synergistic effect score prediction.

[0093] Exemplary device

[0094] like Figure 2 As shown, corresponding to the above-mentioned method for predicting the synergistic effect score of drug combinations, this embodiment of the invention also provides a device for predicting the synergistic effect score of drug combinations, the device comprising:

[0095] Construction module 210 is used to acquire drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data;

[0096] Preprocessing module 220 is used to generate a set of metapath instances of drug nodes in the heterogeneous drug-target interaction network according to a preset metapath type, and convert the metapath instances in the set of metapath instances into unified dimension metapath instances to obtain a unified dimension metapath instance set.

[0097] Processing module 230 is used to convert the unified dimension metapath instance into a metapath instance representation containing interaction information, and to perform weighted aggregation of the metapath instance representation according to the instance set type, wherein the drug node is an embedded representation based on different preset metapath types.

[0098] The fusion module 240 is used to fuse the embedding representations according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node;

[0099] The prediction module 250 is used to send the overall metapath embedding representation, drug chemical feature information and cell line information into the preset AE prediction module and the preset TE prediction module to obtain the synergistic effect score of the drug combination.

[0100] Thus, drug combination data is acquired through a design and construction module, and a heterogeneous drug-target interaction network is constructed based on the drug combination data. A preprocessing module generates metapath instance sets for drug nodes in the heterogeneous drug-target interaction network according to preset metapath types, and converts the metapath instances in the metapath instance sets into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set. A processing module converts the unified-dimensional metapath instances into metapath instance representations containing interaction information, and performs weighted aggregation of the metapath instance representations according to the instance set type to obtain the embedding representations of the drug nodes based on different preset metapath types. A fusion module fuses the embedding representations according to an attention mechanism to obtain the overall metapath embedding representation of the drug nodes. A prediction module sends the overall metapath embedding representation, drug chemical feature information, and cell line information to preset AE prediction modules and preset TE prediction modules to obtain the synergistic effect score of the drug combination. This invention generates metapath instance sets for drug nodes through a preprocessing module to obtain drug-target interaction information. Furthermore, it converts the unified-dimensional metapath instances into metapath instance representations containing interaction information based on preset metapath types, and performs weighted aggregation of the metapath instance representations according to the instance set type to learn the drug-target interaction information, i.e., the drug's mechanism of action. By adding a preset AE prediction module, it effectively utilizes AE label information based on backpropagation to optimize and adjust the model parameters related to TE prediction, thereby improving the accuracy of synergistic effect score prediction.

[0101] This invention also provides a smart terminal, which includes a memory, a processor, and a synergistic effect score prediction program for drug combinations stored in the memory and executable on the processor. When the synergistic effect score prediction program for drug combinations is executed by the processor, it implements the steps of the synergistic effect score prediction method for any drug combination provided in this invention.

[0102] Based on the above embodiments, the present invention also provides a smart terminal, the principle block diagram of which can be as follows: Figure 3As shown. The intelligent terminal includes a processor and a memory. The memory of the intelligent terminal includes a synergistic effect score prediction program for drug combinations, and the memory provides an environment for the operation of the synergistic effect score prediction program for drug combinations. When the processor executes the synergistic effect score prediction program for drug combinations, it implements the steps of any of the above-described methods for predicting the synergistic effect score of drug combinations. It should be noted that the intelligent terminal may also include other functional modules or units, which are not specifically limited here.

[0103] Those skilled in the art will understand that Figure 3 The block diagram shown is merely a partial structural diagram related to the present invention and does not constitute a limitation on the smart terminal to which the present invention is applied. Specifically, the smart terminal may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0104] In one embodiment, when the synergy score prediction program for the drug combination is executed by the processor, it performs the following operational instructions:

[0105] Acquire drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data;

[0106] A set of metapath instances is generated based on the drug nodes in the heterogeneous drug-target interaction network with the preset metapath type, and the metapath instances in the set of metapath instances are converted into unified dimension metapath instances to obtain a unified dimension metapath instance set.

[0107] The unified dimension metapath instance is converted into a metapath instance representation containing interaction information, and the metapath instance representation is weighted and aggregated according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types.

[0108] The embedding representations are fused according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node;

[0109] The overall metapath embedding representation, medicinal chemical feature information, and cell line information are fed into a preset AE prediction module and a preset TE prediction module to obtain the synergistic effect score of the drug combination.

[0110] This invention also provides a computer-readable storage medium storing a camera-based underwater measurement parameter calibration program. When the camera-based underwater measurement parameter calibration program is executed by a processor, it implements the steps of any calibration method provided in this invention.

[0111] Model simulation analysis:

[0112] Experimental Environment and Data: To verify the validity of this application, a neural network model framework was built using Python 3.6 + PyTorch 1.10.2. During training, the Adam algorithm with a learning rate of 0.005 was used as the optimizer. The experiment was conducted in an environment with Linux 4.18, Intel Xeon Platinum 8360Y, 40G RAM, and NVIDIA A100-SXM.

[0113] In this embodiment, the dataset is randomly divided according to drug pairs. 80% of the samples corresponding to each drug pair are used as the training set, 10% as the validation set, and 10% as the test set. The model hyperparameters are adjusted using the validation set, and the actual performance is tested on the test set. Table 2 lists the important parameters in the model of this embodiment:

[0114]

[0115] Table 2

[0116] Experimental performance metrics: To verify the effectiveness of this application, the mean square error (MSE) and mean absolute error (MAE) were used to evaluate the predictive performance of the synergistic effect scores of different methods. The smaller the values ​​of MSE and MAE, the more accurate the model prediction.

[0117] To further demonstrate this application, the AE prediction module in this application is removed, and the input used by the AE prediction module during calculation is also removed. The structure of other modules remains unchanged. Compared with the embodiment of this application, all parts related to AE processing in the model are removed, and it is no longer possible to utilize the relevant information of AE.

[0118] In the experiments, the performance of the method of this application, a variant without the AE prediction module, and a representative deep learning-based TE prediction method, DeepSynergy, and its variant DeepSynergyAE (which adds AE label information as model input compared to DeepSynergy), was compared on a compiled dataset containing both TE and AE information. Table 3 shows that the performance metrics of the embodiment of this application are higher than other models, with MSE and MAE reaching 206.270 and 9.627, respectively. Furthermore, it can be found that the comparative method DeepSynergy cannot effectively utilize AE label information within its single-task framework, and the performance of DeepSynergyAE is worse than that of DeepSynergy.

[0119]

[0120]

[0121] Table 3

[0122] In summary, based on the assumption that the rational utilization of AE information can improve the accuracy of deep learning prediction methods for therapeutic synergy scores, the method of this invention uses a BiGRU aggregator-based GCN to extract drug-target interaction pathway information by learning biologically meaningful metapaths. Furthermore, an auxiliary AE prediction module optimizes the parameters of TE prediction based on backpropagation to further improve the performance of therapeutic synergy score prediction. Experimental results show that the method of this application with the added AE prediction module outperforms single-task variants without AE information and the single-task deep learning method DeepSynergy, which cannot effectively utilize AE tag information, on both MSE and MAE.

[0123] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the above device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this invention. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0124] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0125] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0126] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of the above modules or units is merely a logical functional division, and in actual implementation, it can be divided in other ways. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0127] If the integrated modules / units described above are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0128] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not mean that the essence of the corresponding technical solutions deviates from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for predicting the synergistic effect score of a drug combination, characterized in that, The method for predicting synergistic effect scores includes: Acquire drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data; A set of metapath instances is generated based on the drug nodes in the heterogeneous drug-target interaction network with the preset metapath type, and the metapath instances in the set of metapath instances are converted into unified dimension metapath instances to obtain a unified dimension metapath instance set. A specific metapath is defined as a sequence of nodes from the start node to the end node based on a specific edge combination; a metapath instance refers to a sequence of node data that follows a certain type of metapath definition; the metapath instance set includes: a first metapath instance set, a second metapath instance set, a third metapath instance set, and a fourth metapath instance set, wherein, In the first metapath instance set, the starting drug node and the ending drug node of the metapath instance interact with the same target. In the second metapath instance set, the starting drug node and the ending drug node of the metapath instance interact with two targets respectively, and the two targets interact with each other. In the set of third-level metapath instances, the starting drug node and the ending drug node of the metapath instance interact with two targets, and the two targets interact with another target. In the fourth metapath instance set, the synergistic effect score generated by the combined action of the starting drug node and the ending drug node of the metapath instance is higher than a certain threshold; the unified dimension metapath instance is converted into a metapath instance representation containing interaction information, and the metapath instance representation is weighted and aggregated according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types. The embedding representations are fused according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node; The overall metapath embedding representation, medicinal chemical feature information, and cell line information are fed into a preset AE prediction module and a preset TE prediction module to obtain the synergistic effect score of the drug combination.

2. The method for predicting the synergistic effect score of drug combinations according to claim 1, characterized in that, The step of converting the unified dimension metapath instance into a metapath instance representation containing interaction information specifically includes: sending the unified dimension metapath instance set into a BiGRU aggregator of the corresponding instance set type to obtain a metapath instance representation containing interaction information based on a specific metapath.

3. The method for predicting the synergistic effect score of drug combinations according to claim 1, characterized in that, The step of weighted aggregation of the metapath instance representations according to the instance set type to obtain the embedding representations of the drug node based on different preset metapath types specifically includes: The metapath instance representation is fed into the GAT extractor of the corresponding instance set type to obtain the instance set-based embedding representation of the drug node.

4. The method for predicting the synergistic effect score of drug combinations according to claim 1, characterized in that, The step of converting the metapath instances in the metapath instance set into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set specifically includes: The meta-path instances in the four meta-path instance sets are converted into tensors based on one-hot encoding of different node types, and the feature lengths of each node in the tensors are unified according to the transformation matrix of different node types to obtain four unified-dimensional meta-path instance sets.

5. The method for predicting the synergistic effect score of drug combinations according to claim 4, characterized in that, The step of converting the unified dimension metapath instance into a metapath instance representation containing interaction information specifically includes: The four unified dimension metapath instance sets are fed into four BiGRU aggregators of the corresponding instance set type, and the unified dimension metapath instances in the four unified dimension metapath instance sets are converted into metapath instance representations containing interaction information.

6. The method for predicting the synergistic effect score of drug combinations according to claim 5, characterized in that, The step of weighted aggregation of the metapath instance representations according to the instance set type to obtain the embedding representations of the drug node based on different preset metapath types specifically includes: The meta-path instance representations corresponding to the unified dimension meta-path instances in the four unified dimension meta-path instance sets are fed into the GAT extractor of the corresponding instance set type. The meta-path instance representations are weighted and aggregated to obtain the embedded representations of four different meta-paths.

7. A device for predicting the synergistic effect score of a drug combination, characterized in that, The synergy score calculation device includes: The module is used to acquire drug combination data and construct a heterogeneous drug-target interaction network based on the drug combination data. The preprocessing module is used to generate a set of metapath instances for drug nodes in the heterogeneous drug-target interaction network according to a preset metapath type, and convert the metapath instances in the set of metapath instances into unified-dimensional metapath instances to obtain a unified-dimensional metapath instance set. A specific metapath is defined as a sequence of nodes from the start node to the end node based on a specific edge combination; a metapath instance refers to a sequence of node data that follows a certain type of metapath definition; the metapath instance set includes: a first metapath instance set, a second metapath instance set, a third metapath instance set, and a fourth metapath instance set, wherein, In the first metapath instance set, the starting drug node and the ending drug node of the metapath instance interact with the same target. In the second metapath instance set, the starting drug node and the ending drug node of the metapath instance interact with two targets respectively, and the two targets interact with each other. In the set of third-level metapath instances, the starting drug node and the ending drug node of the metapath instance interact with two targets, and the two targets interact with another target. In the fourth metapath instance set, the synergistic score generated by the combined action of the starting drug node and the ending drug node of the metapath instance is higher than a certain threshold. The processing module is used to convert the unified dimension metapath instance into a metapath instance representation containing interaction information, and to perform weighted aggregation of the metapath instance representation according to the instance set type to obtain the embedding representation of the drug node based on different preset metapath types. The fusion module is used to fuse the embedding representations according to the attention mechanism to obtain the overall meta-path embedding representation of the drug node; The prediction module is used to send the overall metapath embedding representation, medicinal chemical feature information and cell line information into the preset AE prediction module and the preset TE prediction module, respectively, to obtain the synergistic effect score of the drug combination.

8. A smart terminal, characterized in that, The smart terminal includes a memory, a processor, and a synergistic effect score prediction program for drug combinations stored in the memory and executable on the processor. When executed by the processor, the synergistic effect score prediction program for drug combinations implements the steps of the synergistic effect score prediction method for drug combinations as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a synergistic effect score prediction program for a drug combination, which, when executed by a processor, implements the steps of the synergistic effect score prediction method for a drug combination as described in any one of claims 1-6.

Images