A method and device for predicting drug properties based on artificial intelligence

Abstract

The present invention relates to the technical field of drug screening, and specifically discloses a drug property prediction method and device based on artificial intelligence, comprising: S1: collecting structural data and relevant property labels of drug molecules, ensuring the accuracy and completeness of data, identifying various functional groups in drug molecules and describing them; S2: extracting drug molecular structural features and introducing the characteristics of functional groups, screening and selecting the extracted feature vectors according to a property prediction model to improve the model processing efficiency; the present invention uses two different dimensional feature extractions to fuse multi-dimensional features to describe drug molecular features, so that information on drug analysis can be fully and accurately extracted and the advantages of the two can be fully combined, and at the same time, the frequency of the initial frequent unilateral edge is used to set staged execution operations so that frequent subgraph mining can be performed adaptively, thereby reducing the spatiotemporal complexity of mining frequent subgraphs, so that the prediction results of drug properties through a trained model are more accurate.

Description

Technical Field

[0001] The present invention belongs to the technical field of drug screening, and in particular relates to a drug property prediction method and device based on artificial intelligence. Background Art

[0002] Drug discovery is the process of discovering new drug candidates, involving knowledge in fields such as medicine, biology, and pharmacology. Researchers generally select targets based on the biochemical mechanisms involved in disease conditions, and then use candidate drugs discovered in academic, pharmaceutical, or biotechnology research laboratories to test interactions with drug targets. Once a drug is confirmed to be effective against a target, it is validated by examining its effects on related diseases. There may be as many as 50,000 to 100,000 potential drug candidates for each disease, and these molecules need to be rigorously screened. With the emergence of interdisciplinary thinking, the application of artificial intelligence in the field of drug discovery has attracted the interest of a large number of researchers. By introducing artificial intelligence technology in the preparation of new drug compounds, the properties of compound molecules can be predicted more objectively and accurately, thereby determining the efficacy and safety of predicted samples for specific diseases, which not only accelerates the drug development cycle but also improves research efficiency.

[0003] In recent years, with the development of pharmacokinetics, the efficiency of drug screening has gradually improved. However, testing a new compound with potential medicinal value requires in vivo testing and experiments, which takes a lot of time and cost. Therefore, there is an urgent need for an accurate, efficient and low-cost way to quickly identify whether a molecule has side effects on organisms and whether it can become a potential drug.

[0004] Therefore, it is necessary to propose a drug property prediction method and device based on artificial intelligence to at least partially solve the problems existing in the prior art.

[0005] The above information disclosed in this background technology is only used to increase the understanding of the background technology of the present invention and therefore, it may include information that does not constitute the prior art known to ordinary technicians in this field. Summary of the invention

[0006] The purpose of the present invention is to provide a method and device for predicting drug properties based on artificial intelligence to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A drug property prediction method based on artificial intelligence, comprising:

[0009] S1: Collect the structural data and related property labels of drug molecules, ensure the accuracy and completeness of the data, identify the various functional groups in drug molecules and describe them;

[0010] Generate a molecular graph structure based on the structural data and property labels of drug molecules, find the location of functional groups and functional group relationship information by searching for specific substructures in the molecular graph structure, and generate a molecular subgraph structure in combination with the functional group characteristic description;

[0011] S2: Extract the structural features of drug molecules and introduce the characteristics of functional groups. Screen and select the extracted feature vectors according to the property prediction model to improve the model processing efficiency.

[0012] Features are extracted from the molecular graph structure and the molecular subgraph structure respectively, and the molecular graph structure features and the molecular subgraph structure features are combined to construct a comprehensive feature vector. In this process, the molecular subgraph structure features are subjected to data dimensionality reduction through principal component analysis to reduce the dimensionality and redundant information of the molecular subgraph structure features.

[0013] S3: Input the selected feature vector into the trained property prediction model to predict the drug properties, obtain the property prediction results of the drug molecules, and interpret and analyze the prediction results;

[0014] The construction process of the property prediction model used is as follows:

[0015] Step 1: Data preprocessing: Collect and organize relevant drug data sets, extract useful elements and element relationship information by processing raw biochemical test data, and generate available drug map data;

[0016] Step 2: Feature structure discovery: Use the frequent subgraph mining algorithm to mine closed frequent subgraphs from the graph set as feature structures for the drug graph data obtained in step 1;

[0017] Step 3: Prediction model construction: Map the feature structure discovered in step 2 to the atlas to generate a feature space, and then use deep learning to build a property prediction model.

[0018] Preferably, in S1, the drug molecule data collected include the structure, biological activity data, pharmacophore information, etc. of the drug molecules, and various functional groups (such as hydroxyl groups, carbonyl groups) in the drug molecules are identified and described, including information such as position, type, and quantity, and the functional group information is converted into a representation that can be used for deep learning.

[0019] Preferably, in S2, a variable is defined for each functional group according to the functional group information identified in S1, and the variables are divided according to the features extracted from the molecular graph structure, and then the molecular graph structure features and variables belonging to the same action type are constructed into a feature vector as a comprehensive vector.

[0020] Preferably, in S2, feature extraction adopts the feature generation method of the property prediction model construction process to make the extracted features suitable for model processing, and in S3, the influence of functional group characteristics on drug properties is analyzed based on the prediction results to provide guidance for further drug design and optimization.

[0021] Preferably, the drug dataset in step 1 is used to divide the training set, validation set and test set for model training, parameter adjustment and evaluation, and the model is trained using the labeled drug dataset to learn the relationship between the structural characteristics and functional group characteristics of drug molecules and their properties.

[0022] Preferably, after training is completed, the trained model is evaluated using a validation set, and the performance of the model on unseen data is evaluated using a test set to test its accuracy and generalization ability in predicting drug properties, and the model is tuned and improved based on the analysis of the results.

[0023] Preferably, in step 3, a graph convolutional neural network is used to construct a property prediction model. In this process, the generated feature space is constructed as the initial feature vector of the node, and a graph convolution operation is defined based on the new feature space. In a manner similar to traditional graph convolution operations, the new feature space is substituted to propagate and update node features.

[0024] A drug property prediction device based on artificial intelligence, comprising at least one processor; a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a drug property prediction method based on artificial intelligence as described in any one of the above.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention uses feature extraction of two different dimensions to fuse multi-dimensional features to describe the characteristics of drug molecules, so that the information of drug analysis can be fully and accurately extracted and the advantages of the two can be fully combined to more accurately identify the properties of drug molecules. At the same time, the frequency of the initial frequent single edge is used to set the staged execution operation so that the frequent subgraph mining can be carried out adaptively, reducing the time and space complexity of mining frequent subgraphs, and then the key features screened are closer to the required ones, so that the prediction results of drug properties through the trained model are more accurate.

[0027] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flow chart of the prediction method of the present invention;

[0029] Figure 2 The same diagram of the present invention may have multiple different DFS Tree schematic diagrams. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Example

[0031] See also Figure 1-Figure 2 As shown, a drug property prediction method based on artificial intelligence includes:

[0032] S1: Collect the structural data and related property labels of drug molecules, ensure the accuracy and completeness of the data, identify the various functional groups in drug molecules and describe them;

[0033] Generate a molecular graph structure based on the structural data and property labels of drug molecules, find the location of functional groups and functional group relationship information by searching for specific substructures in the molecular graph structure, and generate a molecular subgraph structure in combination with the functional group characteristic description;

[0034] S2: Extract the structural features of drug molecules and introduce the characteristics of functional groups. Screen and select the extracted feature vectors according to the property prediction model to improve the model processing efficiency.

[0035] Features are extracted from the molecular graph structure and the molecular subgraph structure respectively, and the molecular graph structure features and the molecular subgraph structure features are combined to construct a comprehensive feature vector. In this process, the molecular subgraph structure features are subjected to data dimensionality reduction through principal component analysis to reduce the dimensionality and redundant information of the molecular subgraph structure features.

[0036] S3: Input the selected feature vector into the trained property prediction model to predict the drug properties, obtain the property prediction results of the drug molecules, and interpret and analyze the prediction results;

[0037] The construction process of the property prediction model used is as follows:

[0038] Step 1: Data preprocessing: Collect and organize relevant drug data sets, extract useful elements and element relationship information by processing raw biochemical test data, and generate available drug map data;

[0039] Step 2: Feature structure discovery: Use the frequent subgraph mining algorithm to mine closed frequent subgraphs from the graph set as feature structures for the drug graph data obtained in step 1;

[0040] Step 3: Prediction model construction: Map the feature structure discovered in step 2 to the atlas to generate a feature space, and then use deep learning to build a property prediction model.

[0041] In S2, feature extraction adopts the feature generation method of the property prediction model construction process to make the extracted features suitable for model processing;

[0042] In S3, the impact of functional group characteristics on drug properties is analyzed based on the prediction results to provide guidance for further drug design and optimization.

[0043] The drug dataset in step 1 is used to divide the training set, validation set, and test set for model training, parameter adjustment, and evaluation. The model is trained using the labeled drug dataset to learn the relationship between the structural characteristics and functional group characteristics of drug molecules and their properties.

[0044] After training, the validation set is used to evaluate the trained model, and the test set is used to evaluate the model's performance on unseen data, to test its accuracy and generalization ability in predicting drug properties, and to tune and improve the model based on the results.

[0045] In step 2, frequent subgraph mining is performed on samples with the same drug properties to obtain frequently occurring substructures, and then the obtained candidate substructures are pruned and denoised according to certain rules to find characteristic structures that affect or inspire drug properties. The core process is as follows:

[0046] (1) Traverse the edge set and calculate the frequency of each edge, delete the infrequent items, and retain the frequent single edges as the initial frequent subgraph;

[0047] (2) Generate a candidate subgraph and perform the rightmost path extension operation on the DFS Tree formed by the minimum DFS encoding of the initial subgraph, adding one edge at a time;

[0048] (3) Check whether the newly generated candidate subgraph encoding is the minimum encoding that satisfies the DFS lexicographic order. If it is, it will be retained waiting for the rightmost path to be expanded. If it is not, the corresponding candidate subgraph is a redundant subgraph and is deleted from the candidate subgraph to prune the expansion space;

[0049] (4) Calculate the frequency of candidate subgraphs, and generate frequent subgraphs by reverse encoding for those that meet the subgraph support;

[0050] (5) Repeat the above (2)-(4) process until no new candidate subgraphs or frequent subgraphs are generated.

[0051] DFS lexicographic order: Referring to the idea of ​​lexicographic sorting, the subscript is defined to indicate that the edge order relationship is the highest priority, the first vertex label is the second priority, the edge label is the third priority, and the second vertex label is the lowest priority. Figure 2 As shown in the figure, a produces b, c, and d after different traversals. Subscript marks are used to describe the discovery process. mean First discovered in , the edge sequence of b is linearly described as , , , , further adding data information to the edge sequence can get the corresponding DFS encoding , , , , a graph corresponds to multiple DFS Trees, so there will be multiple DFS codes accordingly. Here we sort them in DFS lexicographic order. Figure 2 The three DFS codes b, c, d , , To sort, first compare the encoding of the first side , , , the highest priority edge sequences are the same as , and then compare the first vertex label ,so Greater than and ,Compare and The second side code , , since the edge order is the same as the first vertex label, and the edge label ,so , and finally get the encoding sort .

[0052] Support setting: Based on the idea of ​​stage mining, the frequency of frequent unilateral As the upper bound of support, the frequent jump N is set to regulate the scope of mining in each stage, so the minimum support of the first stage is , and then the number of jumps C is used to control the number of stage mining, and the optimal setting value is obtained through parameter analysis and adjustment.

[0053] A drug property prediction device based on artificial intelligence, comprising at least one processor; a memory connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a drug property prediction method based on artificial intelligence such as any one of the above.

[0054] From the above, it can be seen that the present invention uses two different dimensions of feature extraction to fuse multi-dimensional features to describe the characteristics of drug molecules, so that the information of drug analysis can be fully and accurately extracted and the advantages of the two can be fully combined to more accurately identify the properties of drug molecules. At the same time, the frequency of the initial frequent single edge is used to set the staged execution operation so that the frequent subgraph mining can be carried out adaptively, which reduces the time and space complexity of mining frequent subgraphs, and then the key features screened are closer to the required ones, so that the prediction results of drug properties through the trained model are more accurate.

[0055] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0056] In the drawings of the embodiments disclosed in the present invention, only the structures involved in the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0057] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A drug property prediction method based on artificial intelligence, characterized in that: include: S1: Collect the structural data and related property labels of drug molecules, ensure the accuracy and completeness of the data, identify the various functional groups in drug molecules and describe them; Generate a molecular graph structure based on the structural data and property labels of drug molecules, find the location of functional groups and functional group relationship information by searching for specific substructures in the molecular graph structure, and generate a molecular subgraph structure in combination with the functional group characteristic description; S2: Extract the structural features of drug molecules and introduce the characteristics of functional groups. Screen and select the extracted feature vectors according to the property prediction model to improve the model processing efficiency. Features are extracted from the molecular graph structure and the molecular subgraph structure respectively, and the molecular graph structure features and the molecular subgraph structure features are combined to construct a comprehensive feature vector. In this process, the molecular subgraph structure features are subjected to data dimensionality reduction through principal component analysis to reduce the dimensionality and redundant information of the molecular subgraph structure features. S3: Input the selected feature vector into the trained property prediction model to predict the drug properties, obtain the property prediction results of the drug molecules, and interpret and analyze the prediction results; The construction process of the property prediction model used is as follows: Step 1: Data preprocessing: Collect and organize relevant drug data sets, extract useful elements and element relationship information by processing raw biochemical test data, and generate available drug map data; Step 2: Feature structure discovery: Use the frequent subgraph mining algorithm to mine closed frequent subgraphs from the graph set as feature structures for the drug graph data obtained in step 1; Step 3: Prediction model construction: Map the feature structure discovered in step 2 to the atlas to generate a feature space, and then use deep learning to build a property prediction model.

2. The method for predicting drug properties based on artificial intelligence according to claim 1, characterized in that: In S2, feature extraction adopts the feature generation method of the property prediction model construction process to make the extracted features suitable for model processing. In S3, the influence of functional group characteristics on drug properties is analyzed based on the prediction results to provide guidance for further drug design and optimization.

3. The method for predicting drug properties based on artificial intelligence according to claim 2, characterized in that: The drug dataset in step 1 is used to divide the training set, validation set and test set for model training, parameter adjustment and evaluation. The model is trained using the labeled drug dataset to learn the relationship between the structural characteristics and functional group characteristics of drug molecules and their properties.

4. The method for predicting drug properties based on artificial intelligence according to claim 3, characterized in that: After training, the validation set is used to evaluate the trained model, and the test set is used to evaluate the model's performance on unseen data, to test its accuracy and generalization ability in predicting drug properties, and to tune and improve the model based on the results.

5. A drug property prediction device based on artificial intelligence, characterized in that: It comprises at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a drug property prediction method based on artificial intelligence as described in any one of claims 1 to 4.

Images