A method and system for predicting and classifying drug molecular properties
By combining the characteristic fusion of molecular directed transmission information and Morgan molecular fingerprint information, the problems of insufficient number of target active molecules and poor performance of molecular descriptors in the prior art are solved, and more accurate prediction and classification of drug molecular properties are achieved.
Patent Information
- Application Number
- CN202210021248.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-10
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-01-10
AI Technical Summary
In the prediction of the molecular properties of drugs, the insufficient number of active molecules at the target site leads to poor performance of the QSAR model, and the molecular descriptor performs poorly when describing scaffold information, resulting in low screening and classification accuracy.
Two convolutional neural networks of different specifications are used to extract the molecular descriptor and molecular fingerprint respectively, and feature fusion is carried out in combination with molecular directed transmission information and Morgan molecular fingerprint information to achieve more accurate prediction of the molecular properties of the drug.
Through multi-dimensional feature extraction and fusion, the accuracy and classification accuracy of drug molecular properties prediction are improved, and are suitable for molecular descriptors and fingerprint information of different specifications.
Smart Images

Figure CN114530211B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of drug screening, and in particular to a method and system for predicting and classifying drug molecular properties. Background Art
[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.
[0003] Molecular property prediction is one of the most classic tasks in cheminformatics. As various diseases are becoming increasingly resistant to their drugs, it is necessary to screen out effective drugs for various diseases from large drug molecule libraries. With the development of deep neural network research, there are two traditional methods that combine deep learning: (1) operating on fixed molecular fingerprints commonly seen in traditional quantitative structure-activity relationship (QSAR) models. QSAR is based on the assumption that molecules with similar structures have similar biological activities. Given a certain number of targets, each target uses its known active molecules to establish a prediction model, and then uses the established model to screen all drugs to predict the interaction between drugs and targets. (2) Using graph convolutional networks to form molecular feature descriptors, subsequent feature extraction is performed on this basis. Both methods have produced substantial results and improved the accuracy of drug property prediction.
[0004] However, the inventors found that the prior art only has the following technical problems:
[0005] (1) If the number of known active molecules for a target is insufficient, the performance of the established QSAR model will be poor. Moreover, most QSAR models only predict the activity of one target, and the molecular fingerprint based on this model can only express less molecular information.
[0006] (2) Molecular descriptors obtained by common convolutional neural networks often perform poorly in describing molecular scaffold information, which means that such descriptors construct information in other datasets and their classification effect is reduced.
[0007] (3) Traditional methods ignore rich information in feature extraction, resulting in incomplete feature extraction and ultimately low screening and classification accuracy. Summary of the invention
[0008] In order to solve the above problems, the present invention proposes a method and system for predicting and classifying drug molecular properties, and proposes a classification method for molecular descriptors and molecular fingerprint information of different specifications, and adopts two convolutional neural networks of different specifications to extract features of molecular descriptors and molecular fingerprints respectively; adopts the fusion features of molecular directional transmission information and Morgan molecular fingerprint information to more accurately identify the molecular properties of drugs, thereby achieving accurate classification.
[0009] In order to achieve the above object, the present invention adopts the following technical solution:
[0010] In a first aspect, the present invention provides a method for predicting and classifying the molecular properties of a drug, comprising:
[0011] Obtaining a molecular map of the drug to be tested;
[0012] Extract the features of molecular descriptors and molecular fingerprints from the molecular graph to obtain molecular directional transfer information and Morgan molecular fingerprint information;
[0013] Feature fusion of molecular directional transfer information and Morgan molecular fingerprint information;
[0014] The trained classification model is used according to the fusion features to obtain the prediction results of the molecular properties of the drug.
[0015] As an optional implementation, the process of extracting features of molecular descriptors from molecular graphs to obtain molecular directional transmission information includes:
[0016] Initialize atomic hidden state;
[0017] According to the information transmission of the molecular graph, the atomic hidden state and atomic information are updated;
[0018] According to the updated atomic information and atomic hidden states, the atomic hidden states are summed to obtain the molecular directed transfer information.
[0019] As an optional implementation, the atomic hidden state and atomic information are respectively:
[0020]
[0021]
[0022] in, is the directional information from atom p to atom q at step s+1; is the directional hidden state from atom p to atom q at step s; is the directional hidden state from atom p to atom q at step s+1; N(p) is a set of adjacent atoms in the molecular graph; a p 、a jare the atomic characteristics of atom p and atom j respectively; M S and U S They are information function and atomic fixed-point update function, is the directed hidden state from atom j to atom p.
[0023] As an optional implementation, when updating the atomic hidden state and atomic information, an information function and an atomic fixed-point update function are used respectively. The information function and the atomic fixed-point update function are respectively:
[0024]
[0025]
[0026] Among them, T i is a learning matrix of hidden size c.
[0027] As an optional implementation, the process of extracting features of molecular fingerprints from molecular graphs to obtain Morgan molecular fingerprint information includes: acquiring molecular fingerprints, converting one-dimensional information of molecular fingerprints into two-dimensional information, and then performing feature extraction to obtain Morgan molecular fingerprint information.
[0028] As an optional implementation, the process of obtaining a molecular fingerprint includes:
[0029] Initialize the atoms in the molecular graph and assign integer identifiers to each heavy atom;
[0030] Taking each heavy atom as the center, the heavy atom circles are merged until the specified radius is reached, and a feature list is generated to obtain the molecular fingerprint.
[0031] As an optional implementation, the obtained molecular directed transfer information and Morgan molecular fingerprint information are flattened before feature fusion.
[0032] In a second aspect, the present invention provides a drug molecular property prediction and classification system, comprising:
[0033] an acquisition module, configured to acquire a molecular graph of a drug to be tested;
[0034] A feature extraction module is configured to extract features of molecular descriptors and molecular fingerprints from the molecular graph to obtain molecular directional transfer information and Morgan molecular fingerprint information;
[0035] A feature fusion module is configured to perform feature fusion on the molecular directed transfer information and the Morgan molecular fingerprint information;
[0036] The prediction classification module is configured to obtain the molecular property prediction result of the drug using the trained classification model according to the fusion features.
[0037] In a third aspect, the present invention provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in the first aspect is performed.
[0038] In a fourth aspect, the present invention provides a computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method described in the first aspect is performed.
[0039] Compared with the prior art, the present invention has the following beneficial effects:
[0040] The present invention proposes a method and system for predicting and classifying drug molecular properties. The method adopts two feature extraction networks of different dimensions to extract features of molecular descriptors and molecular fingerprints, and can fully and accurately extract feature information. The multi-dimensional features of molecular directional transmission information and Morgan molecular fingerprint information are fused in a connected and fused manner, and the advantages of the two are fully combined to more accurately identify the molecular properties of drugs, thereby more accurately achieving classification. Finally, the drug data set is classified through the trained network, and the classification results for molecular descriptors and molecular fingerprint information of different specifications are more accurate.
[0041] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0042] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.
[0043] Figure 1 A schematic diagram of a method for predicting and classifying molecular properties of drugs provided in Example 1 of the present invention;
[0044] Figure 2 A schematic diagram of molecular descriptor feature extraction provided in Example 1 of the present invention;
[0045] Figure 3 A schematic diagram of molecular fingerprint feature extraction provided in Example 1 of the present invention;
[0046] Figure 4 A schematic diagram for summarizing a data set provided in Example 1 of the present invention;
[0047] Figure 5(a)-5(d) A schematic diagram of ROC curve comparison of experimental results of some data sets provided in Example 1 of the present invention;
[0048] Figure 6This is a comparison diagram of the evaluation results provided in Example 1 of the present invention and the currently advanced classification model. DETAILED DESCRIPTION
[0049] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0050] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.
[0051] It should be noted that the terms used herein are only for describing specific embodiments, and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "include" and "have" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0052] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.
[0053] Example 1
[0054] like Figure 1 As shown, this embodiment provides a method for predicting and classifying drug molecular properties, which uses two convolutional neural networks of different specifications to extract features from molecular descriptors and molecular fingerprints respectively; and uses molecular directed transmission information and Morgan molecular fingerprint information to perform feature fusion, fully combining the advantages of the two, and more accurately identifying drug molecular properties, thereby achieving accurate classification.
[0055] Specifically include:
[0056] Obtaining a molecular map of the drug to be tested;
[0057] Extract the features of molecular descriptors and molecular fingerprints from the molecular graph to obtain molecular directional transfer information and Morgan molecular fingerprint information;
[0058] Feature fusion of molecular directional transfer information and Morgan molecular fingerprint information;
[0059] The trained classification model is used according to the fusion features to obtain the prediction results of the molecular properties of the drug.
[0060] This embodiment proposes a molecular information fusion neural network (MIFNN) structure to classify the effectiveness of drug molecules against various diseases, such as Figure 1 As shown in FIG. 1 , the molecular information fusion network includes a feature extraction part, a feature fusion part, and a feature classification part; the details are as follows:
[0061] The feature extraction part is to extract the features of molecular descriptors and molecular fingerprints through the Directed Message Transfer Neural Network (D-MPNN) and the Molecule Directed Information Feature Extraction Network (MDIFEN) respectively, and obtain the molecular directed transfer information and Morgan molecular fingerprint information;
[0062] The directed information transfer neural network is used to obtain molecular directed transfer information as a molecular descriptor; the directed information transfer neural network is a network structure running on the graph G, using atomic features a p The operation phase of the network structure is divided into two steps: generating molecular description phase and reading phase.
[0063] In the first stage, the information transmitted in the molecular graph is processed to generate symbols that can represent the molecular characteristics; and in the second stage, valuable molecular properties are predicted based on this symbol.
[0064] The formula for generating the molecular representation is as follows:
[0065]
[0066]
[0067] Among them, there are s steps in the first stage, and at a certain step s, there will be a hidden state of an atom p and atomic information Updated; M S and U S are information function and atomic fixed-point update function respectively; N(p) is a set of adjacent atoms in the graph G; is the directional hidden state from atom p to atom q at step s, is the directional hidden state from atom p to atom q at step s+1; is the directional information from atom p to atom q at step s; is the directional information from atom p to atom q at step s+1, which is a relatively independent state. There is not much relevant connection; p 、a j are the atomic characteristics of atoms p and j respectively; j is the other atoms near atom p except atom p, is the directed hidden state from atom j to atom p.
[0068] Before each information transmission operation, the hidden state information must be initialized. The initialized hidden state information is
[0069]
[0070] in, is the learning matrix, is the connection between atomic features and bond features, τ is the relu activation function;
[0071] The information function and atomic fixed-point update function are defined as:
[0072]
[0073]
[0074] Among them, T i is a learning matrix with hidden size c; T i ∈R c*c .
[0075] Add at each step A skip connection is provided for the original feature vector of the edge. Finally, we return to the atomic representation of the molecule and combine the generative formula of the molecule representation to enter the hidden state c of the atom p. p Update atomic information i p , specifically:
[0076]
[0077] c p =τ(T a cat(a p ,i p ));
[0078] in, is the hidden information of all atoms j connected to atom p; T a is a hidden matrix of size c, T a ∈[R c*c .
[0079] The molecular directed transfer information (DPI) is obtained by summing the hidden information, and the formula is as follows:
[0080]
[0081] After obtaining the molecular directional transfer information DPI as a molecular descriptor, the molecular directional transfer information is input into the molecular directional information feature extraction network for processing, such as Figure 2 As shown in the figure, after the molecular directional transmission information passes through the convolution layer, pooling layer, convolution layer, and pooling layer in sequence, the 300-bit molecular directional transmission information is processed to obtain 1200-bit feature information.
[0082] In this embodiment, Morgan molecular fingerprints are used simultaneously. Morgan fingerprints are circular fingerprints, also topological fingerprints, obtained by adjusting the standard Morgan algorithm. It can be roughly equivalent to extended connectivity fingerprints (ECFPs). These fingerprints have many advantages, such as fast calculation speed, no pre-defined (can represent an infinite number of different molecular features), can contain chirality information, each element in the fingerprint represents a specific substructure, can be easily analyzed and interpreted, and can be modified accordingly according to different needs. These fingerprints were originally intended to search for molecular features related to activity, rather than substructure searches.
[0083] The fingerprint generation process is roughly divided into the following steps:
[0084] (1) Atom initialization: assign an integer identifier to each heavy atom.
[0085] (2) Iterative update, taking each heavy atom as the center, merges the surrounding heavy atom circles until the specified radius is reached.
[0086] (3) Feature generation, operating on the substructure and generating a feature list.
[0087] In this embodiment, the number of bits of the required molecular fingerprint is set to 2048. After obtaining the corresponding fingerprint, it is converted from 2048*1 one-dimensional information to 32*64 two-dimensional information to be input into the molecular fingerprint feature extraction network (MFFEN), such as Figure 3 As shown in the figure, after passing through the convolution layer, pooling layer, convolution layer, and pooling layer in sequence, the Morgan molecular fingerprint information is obtained.
[0088] In this embodiment, a fusion layer is added after the feature extraction networks of two dimensions. The two networks extract different feature information through networks of different dimensions, namely, molecular directional transmission information and Morgan molecular fingerprint information. These information are accumulated and fused in the fusion layer. The obtained features are flattened before fusion. The formulas are as follows:
[0089]
[0090]
[0091] Among them, P i It is a 2D feature map of size m*n, a represents the eigenvalue, p i A 1D vector representing the flattened data.
[0092] The output of the fusion layer is used as the final complete feature information to wait for subsequent classification processing. The fusion formula is as follows:
[0093] f i T =FF(P i1 ,P i2 )=[a 111 ,...,a nn1 ,a 112 ,...,a nn2 ];
[0094] Among them, P i1 , P i2 Represent the obtained 1D feature vector and 2D feature vector respectively, and a represents a single eigenvalue in two different vectors. By adopting the flattening and feature fusion strategies, the model can obtain more comprehensive and accurate evaluation results.
[0095] In this embodiment, the molecular properties of the drug toxicity are predicted, and the probability of the drug's effectiveness against a certain type of disease can be analyzed based on the prediction results.
[0096] In this embodiment, since the complexity of the drug molecular feature data set studied and the molecular features obtained by the FC layer is relatively low, after comparative experiments of multiple classification models, the accuracy is highest when the support vector machine is used in the classification layer. Therefore, SVM is used to classify the features collected in the FC layer, and the kernel function uses RBF, the penalty parameter is set to 100, and the kernel parameter is 0.01.
[0097] When SVM is used as the classification layer for final classification, the result obtained is the distance from the experimental sample to the hyperplane, which is the basis for SVM classification. In this embodiment, the expected classification result is the probability of the effectiveness of the drug molecule for a certain disease, so the output of SVM needs to be processed, and the SVM output is converted into a posterior probability using the SigMoid fitting method. The advantage of the SigMoid fitting method is that the posterior probability can be well estimated while maintaining the sparsity of the SVM. The processing steps and formulas are as follows:
[0098]
[0099] Where A and B are the parameters to be fitted, and μ is the unthresholded output for sample x.
[0100] This example conducts a large number of experiments on public datasets such as HIV, BACE, and BBBP that are widely used in this field. Figure 4 The results obtained by the model are evaluated according to the evaluation index AUC (Area Under Curve) for classification results, and the evaluation effect is compared with the current advanced classification model. Figure 5(a)-5(d) and Figure 6 As shown, it is shown that the classification accuracy of this embodiment is higher, the classification effect is better, and it has certain practicality. This shows that the drug molecular property prediction and classification method based on deep learning convolutional neural network established in this embodiment is effective, which improves the calculation efficiency for the screening of effective target drugs in the subsequent large-scale drug molecule library and has certain practicality.
[0101] Example 2
[0102] This embodiment provides a drug molecular property prediction and classification system, including:
[0103] an acquisition module, configured to acquire a molecular graph of a drug to be tested;
[0104] A feature extraction module is configured to extract features of molecular descriptors and molecular fingerprints from the molecular graph to obtain molecular directional transfer information and Morgan molecular fingerprint information;
[0105] A feature fusion module is configured to perform feature fusion on the molecular directed transfer information and the Morgan molecular fingerprint information;
[0106] The prediction classification module is configured to obtain the molecular property prediction result of the drug using the trained classification model according to the fusion features.
[0107] It should be noted that the above modules correspond to the steps described in Example 1, and the examples and application scenarios implemented by the above modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above Example 1. It should be noted that the above modules, as part of the system, can be executed in a computer system such as a set of computer executable instructions.
[0108] In further embodiments, there is also provided:
[0109] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method described in Embodiment 1 is performed. For the sake of brevity, it will not be described in detail here.
[0110] It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general-purpose processors, digital signal processors DSP, application-specific integrated circuits ASIC, off-the-shelf programmable gate arrays FPGA or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0111] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.
[0112] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the method described in Example 1 is completed.
[0113] The method in Example 1 can be directly embodied as a hardware processor, or a combination of hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware. To avoid repetition, it is not described in detail here.
[0114] Those skilled in the art will appreciate that the units, i.e., algorithm steps, of the various examples described in conjunction with this embodiment can be implemented in electronic hardware or in a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0115] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.
Claims
1. A method for predicting and classifying drug molecular properties, characterized in that: include: Obtaining a molecular map of the drug to be tested; Extract the features of molecular descriptors and molecular fingerprints from the molecular graph to obtain molecular directional transfer information and Morgan molecular fingerprint information; The molecular descriptor is feature extracted by a directed information transfer neural network, and the obtained molecular directed transfer information is used as the molecular descriptor. The process of obtaining the molecular directed transfer information includes: initializing the atomic hidden state; updating the atomic hidden state and atomic information according to the information transfer of the molecular graph; and summing the atomic hidden state according to the updated atomic information and atomic hidden state to obtain the molecular directed transfer information; The process of obtaining a molecular fingerprint involves: Initialize the atoms in the molecular graph and assign integer identifiers to each heavy atom; Taking each heavy atom as the center, merge in the heavy atom circle until the specified radius is reached, generate a feature list, and obtain the molecular fingerprint; Obtaining the Morgan molecular fingerprint information specifically includes: obtaining the molecular fingerprint, converting the one-dimensional information of the molecular fingerprint into two-dimensional information, thereby performing feature extraction to obtain the Morgan molecular fingerprint information; The molecular directed transfer information and Morgan molecular fingerprint information are feature fused. Specifically, a fusion layer is added after the feature extraction networks of the two dimensions. The two networks extract different feature information through networks of different dimensions, namely, the molecular directed transfer information and Morgan molecular fingerprint information. These information are accumulated and fused in the fusion layer. The molecular property prediction results of the drug are obtained by using the trained classification model according to the fusion features. Specifically, a support vector machine is used to classify the features collected in the classification layer.
2. A method for predicting and classifying drug molecular properties according to claim 1, characterized in that: The atomic hidden state and atomic information are: in, is the directional information from atom p to atom q at step s+1; is the directional hidden state from atom p to atom q at step s; is the directional hidden state from atom p to atom q at step s+1; N(p) is a set of adjacent atoms in the molecular graph; a p、 a j are the atomic characteristics of atom p and atom j respectively; M S and U S They are information function and atomic fixed-point update function, is the directed hidden state from atom j to atom p.
3. A method for predicting and classifying drug molecular properties as claimed in claim 2, characterized in that: When updating the atomic hidden state and atomic information, the information function and atomic fixed-point update function are used respectively. The information function and atomic fixed-point update function are: Among them, T i is a learning matrix of hidden size c.
4. A method for predicting and classifying drug molecular properties according to claim 1, characterized in that: The obtained molecular directional transfer information and Morgan molecular fingerprint information are flattened before feature fusion.
5. A drug molecular property prediction and classification system, characterized in that: include: an acquisition module, configured to acquire a molecular graph of a drug to be tested; The feature extraction module is configured to extract features of the molecular descriptor and the molecular fingerprint from the molecular graph respectively, and obtain molecular directional transfer information and Morgan molecular fingerprint information; wherein the molecular descriptor is feature extracted by a directional information transfer neural network, and the obtained molecular directional transfer information is used as the molecular descriptor, and the process of obtaining the molecular directional transfer information includes: initializing the atomic hidden state; updating the atomic hidden state and the atomic information according to the information transfer of the molecular graph; and summing the atomic hidden state according to the updated atomic information and the atomic hidden state to obtain the molecular directional transfer information; The process of obtaining a molecular fingerprint involves: Initialize the atoms in the molecular graph and assign integer identifiers to each heavy atom; Taking each heavy atom as the center, merge in the heavy atom circle until the specified radius is reached, generate a feature list, and obtain the molecular fingerprint; Obtaining the Morgan molecular fingerprint information specifically includes: obtaining the molecular fingerprint, converting the one-dimensional information of the molecular fingerprint into two-dimensional information, thereby performing feature extraction to obtain the Morgan molecular fingerprint information; The feature fusion module is configured to perform feature fusion on the molecular directional transfer information and the Morgan molecular fingerprint information. Specifically, a fusion layer is added after the feature extraction networks of the two dimensions. The two networks respectively extract different feature information, namely, the molecular directional transfer information and the Morgan molecular fingerprint information, through networks of different dimensions. The information is accumulated and fused in the fusion layer. The prediction classification module is configured to obtain the prediction results of the molecular properties of the drug using the trained classification model according to the fusion features. Specifically, the support vector machine is used to classify the features collected in the classification layer.
6. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the method according to any one of claims 1 to 4 is completed.
7. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the method described in any one of claims 1 to 4.
