Odor prediction method, device and equipment, readable storage medium and program product

Abstract

The invention relates to a smell prediction method, device and equipment, a readable storage medium and a program product. The method comprises the following steps: acquiring a molecular data set corresponding to a to-be-detected object; the molecular data set comprises chemical molecules included in the to-be-detected object and physical and chemical information corresponding to the chemical molecules; performing odor molecule prediction based on the physical and chemical information corresponding to each chemical molecule by the logistic regression model, and determining an odor molecule set; the odor molecule set comprises at least one odor molecule; performing data conversion based on the molecular structure information corresponding to the odor molecules in the odor molecule set to obtain odor molecule graph structure data; performing odor type prediction by the graph neural network model according to the odor molecule graph structure data to obtain an odor type prediction result of odor molecules in the odor molecule set; and determining an odor prediction result of the to-be-detected object based on the odor type prediction result of the odor molecules in the odor molecule set. By adopting the method, odor prediction can be efficiently and accurately realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of odor detection, and in particular to an odor prediction method, device, equipment, readable storage medium and program product. BACKGROUND

[0002] With the development of odor detection technology, odor detection technology is also increasingly used in the tobacco industry. In the tobacco industry, the odor sensory quality of cigarettes is one of the core values. Cigarettes are composed of thousands of chemical molecules, but only a part of the chemical molecules can be perceived by the human olfactory system.

[0003] In related technologies, identifying the odor type of tobacco odor needs to rely on expensive and inefficient physical and chemical experiments, such as gas chromatography-olfactometry, which cannot meet the demand of large-scale and rapid screening. Therefore, there is an urgent need in the art for an efficient and accurate odor prediction method. SUMMARY

[0004] Therefore, it is necessary to provide an efficient and accurate odor prediction method, device, equipment, readable storage medium and program product to solve the above technical problems.

[0005] In a first aspect, the present application provides an odor prediction method, comprising:

[0006] obtaining a molecular data set corresponding to a to-be-detected object; the molecular data set includes chemical molecules included in the to-be-detected object and physical and chemical information corresponding to each of the chemical molecules;

[0007] performing odor molecule prediction based on the physical and chemical information corresponding to each of the chemical molecules by a logistic regression model to determine an odor molecule set; the odor molecule set includes at least one odor molecule;

[0008] performing data conversion based on molecular structure information corresponding to the odor molecules in the odor molecule set to obtain odor molecule graph structure data;

[0009] performing odor type prediction according to the odor molecule graph structure data by a graph neural network model to obtain an odor type prediction result of the odor molecules in the odor molecule set;

[0010] determining an odor prediction result of the to-be-detected object based on the odor type prediction result of the odor molecules in the odor molecule set.

[0011] In one embodiment, the training process of the logistic regression model comprises:

[0012] obtaining physical and chemical information corresponding to a sample molecule and a sample label of the sample molecule; wherein the sample label is used to indicate whether the sample molecule belongs to an odor molecule;

[0013] predicting, by a first base model, an odor molecule of the sample molecule based on physical and chemical information corresponding to the sample molecule, to obtain an odor molecule prediction result of the sample molecule;

[0014] determining a first loss based on a sample label of the sample molecule and the odor molecule prediction result;

[0015] adjusting parameters of the first base model based on the first loss, to obtain the logistic regression model.

[0016] In one of the embodiments, the training process of the graph neural network model comprises:

[0017] obtaining graph structure data of a sample odor molecule and a reference odor type label;

[0018] predicting, by a second base model, an odor type of the sample odor molecule based on the graph structure data of the sample odor molecule, to obtain an odor type prediction result of the sample odor molecule;

[0019] calculating a loss based on the odor type prediction result of the sample odor molecule and the reference odor type label, to determine a second loss;

[0020] adjusting parameters of the second base model based on the second loss, to obtain the graph neural network model.

[0021] In one of the embodiments, the physical and chemical information corresponding to the chemical molecule comprises at least one of a lipid-water partition coefficient, a boiling point, and a vapor pressure.

[0022] In one of the embodiments, the data conversion based on the molecular structure information corresponding to the odor molecules in the odor molecule set comprises:

[0023] determining atoms and chemical bonds included in the odor molecules based on the molecular structure information corresponding to the odor molecules in the odor molecule set;

[0024] obtaining odor molecule graph structure data corresponding to the odor molecules in the odor molecule set, by taking the atoms of the odor molecules as nodes and the chemical bonds as edges.

[0025] In one of the embodiments, the determination of the odor prediction result of the to-be-tested object based on the odor type prediction results of the odor molecules in the odor molecule set comprises:

[0026] weighting the odor type prediction results of the odor molecules in the odor molecule set based on preset weights of the odor types, to obtain the odor prediction result of the to-be-tested object.

[0027] In a second aspect, the present application also provides an odor prediction device, comprising:

[0028] an acquisition module configured to acquire a molecular data set corresponding to a to-be-tested object, wherein the molecular data set comprises chemical molecules included in the to-be-tested object and physical and chemical information corresponding to each of the chemical molecules;

[0029] a screening module configured to perform odor molecule prediction based on the physical and chemical information corresponding to each of the chemical molecules by using a logistic regression model, and determine an odor molecule set, wherein the odor molecule set comprises at least one odor molecule;

[0030] a data processing module configured to perform data conversion based on molecular structure information corresponding to the odor molecules in the odor molecule set, and obtain odor molecule graph structure data;

[0031] an odor prediction module configured to perform odor type prediction based on the odor molecule graph structure data by using a graph neural network model, and obtain an odor type prediction result of the odor molecules in the odor molecule set;

[0032] an output module configured to determine an odor prediction result of the to-be-tested object based on the odor type prediction result of the odor molecules in the odor molecule set.

[0033] In a third aspect, the present application also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0034] In a fourth aspect, the present application also provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the above method when executed by a processor.

[0035] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, and the computer program implements the steps of the above method when executed by a processor.

[0036] The aforementioned odor prediction method, apparatus, equipment, readable storage medium, and program product first use a logistic regression model to perform preliminary screening using the physicochemical information of chemical molecules. This allows for the selection of a set of odor molecules that may influence odor from a massive pool of chemical molecules. Then, a graph neural network model is used to predict the odor type using the graph structure data of odor molecules, determining the odor type prediction result for each odor molecule. Through the graph structure data of odor molecules, the graph neural network model can directly learn the topological structure information of molecules, thereby more accurately capturing the intrinsic relationship between molecular structure and odor type, improving the accuracy of odor type prediction. Finally, the odor type prediction results of the target object are determined by combining the odor type prediction results of each odor molecule. In this process, the logistic regression model screens odor molecules that have an odor influence, avoiding the waste of resources caused by subsequent odor type analysis of molecules without odor influence, thus improving odor prediction efficiency. Simultaneously, the graph structure data preserves the topological structure information of chemical molecules, improving the accuracy of odor type prediction for chemical molecules, and consequently improving the accuracy of odor prediction results for the target object, achieving efficient and accurate odor prediction for the target object. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a diagram illustrating the application environment of the odor prediction method in one embodiment;

[0039] Figure 2 This is a flowchart illustrating an odor prediction method in one embodiment;

[0040] Figure 3 This is a schematic diagram of the training process of a logistic regression model in one embodiment;

[0041] Figure 4 This is a schematic diagram of the application process of the odor prediction method in one embodiment;

[0042] Figure 5 This is a schematic diagram illustrating the application process of a logistic regression model in one embodiment;

[0043] Figure 6 This is a schematic diagram illustrating the validation results of a logistic regression model in one embodiment;

[0044] Figure 7This is a schematic diagram of the model processing of a graph neural network model in one embodiment;

[0045] Figure 8 This is a schematic diagram of the odor prediction process for vanillin molecules in one embodiment;

[0046] Figure 9 This is a structural block diagram of an odor prediction device in one embodiment;

[0047] Figure 10 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0049] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the embodiments, or any combination of multiple embodiments.

[0050] The odor prediction method provided in this application embodiment can be applied to, for example... Figure 1 The application environment shown is illustrated. Terminal 101 communicates with server 102 via a network. A data storage system can store the data that server 102 needs to process. The data storage system can be integrated onto server 102, or it can be located in the cloud or on another network server.

[0051] Users can input relevant information about the object to be tested through terminal 101, such as the chemical molecules included in the object, the physicochemical information of the chemical molecules, and the molecular structure information. Server 102 can deploy various models, such as logistic regression model and graph neural network model.

[0052] Users can initiate an odor prediction request through terminal 101. The odor prediction request may include an address link to the molecular dataset of the object to be tested. After receiving the odor prediction request, server 102 obtains the molecular dataset corresponding to the object to be tested through the address link in the odor prediction request. The molecular dataset includes the chemical molecules included in the object to be tested and the corresponding physicochemical information of each chemical molecule. An odor molecule set is determined by a logistic regression model based on the corresponding physicochemical information of each chemical molecule. The odor molecule set includes at least one odor molecule. Data conversion is performed based on the molecular structure information corresponding to the odor molecules in the odor molecule set to obtain odor molecule graph structure data. An odor type prediction is performed by a graph neural network model based on the odor molecule graph structure data to obtain the odor type prediction result of the odor molecules in the odor molecule set. Based on the odor type prediction result of the odor molecules in the odor molecule set, the odor prediction result of the object to be tested is determined. Finally, server 102 sends the odor prediction result of the object to be tested to terminal 101, which then displays it to the user.

[0053] Terminal 101 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, drones, low-altitude aircraft, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, and projection equipment. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. Server 102 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0054] In one exemplary embodiment, such as Figure 2 As shown, an odor prediction method is provided, which can be applied to... Figure 1 Taking the server in the example, the explanation includes the following steps 201 to 205. Wherein:

[0055] Step 201: Obtain the molecular dataset corresponding to the object to be tested; the molecular dataset includes the chemical molecules included in the object to be tested and the corresponding physicochemical information of each chemical molecule.

[0056] The object to be tested is the object whose odor needs to be predicted; for example, the object to be tested can be cigarettes, spices, cosmetics, etc.

[0057] In some embodiments, the chemical composition of the test object may be analyzed in advance to determine the chemical molecules included in the test object. Then, the characteristics of each chemical molecule may be analyzed to determine the physicochemical information corresponding to the chemical molecule.

[0058] In some embodiments, the molecular dataset corresponding to the test object can be pre-collected. It is understood that the test object is obtained by combining multiple chemical molecules. Therefore, a general chemical molecule dataset can be constructed by pre-collecting the physicochemical information of multiple chemical molecules. In this way, when performing odor prediction, it is only necessary to determine the chemical molecules included in the test object. Then, the corresponding physicochemical information is queried from the general chemical molecule dataset to obtain the molecular dataset of the test object, thereby improving the efficiency of odor prediction.

[0059] Physicochemical information corresponding to a chemical molecule refers to the characteristic information of the chemical molecule at the physical or chemical level, such as boiling point, molecular mass, molecular structure, density, solubility, etc. It can be understood that the physicochemical information corresponding to a chemical molecule does not need to include all the physicochemical characteristics of the chemical molecule, but only the physicochemical characteristics that may affect the odor, such as volatility, molecular structure, etc.

[0060] In some embodiments, the physicochemical information corresponding to the chemical molecules includes at least one of lipid-water partition coefficient, boiling point, and vapor pressure; wherein, lipid-water partition coefficient refers to the concentration ratio of chemical molecules when thermodynamic equilibrium is reached between the lipid phase and the aqueous phase; vapor pressure refers to the pressure exerted by gas molecules generated by liquid evaporation or solid sublimation on the container wall or other materials.

[0061] Furthermore, the three physicochemical parameters of lipid-water partition coefficient, boiling point, and vapor pressure are closely related to the volatility of molecules and their ability to interact with olfactory receptors. Therefore, odor molecules can be classified and predicted based on these three physicochemical parameters.

[0062] Step 202: The logistic regression model predicts odor molecules based on the physicochemical information corresponding to each chemical molecule to determine the odor molecule set; the odor molecule set includes at least one odor molecule.

[0063] Logistic regression is a neural network model used to solve classification problems. In this embodiment, it can efficiently classify the chemical molecules included in the test object into two categories: odor molecules and non-odor molecules.

[0064] Odor molecules refer to chemical molecules that can influence the odor characteristics of the test object; non-odor molecules refer to molecules that do not contribute to the odor of the test object. It is understandable that the number of chemical molecules that can be included in the test object is enormous. However, not all chemical molecules can influence the odor characteristics of the test object. If we directly predict the odor type for each chemical molecule, it will consume a lot of computing resources and affect the processing efficiency. Whether a chemical molecule can become an odor molecule depends on whether the chemical molecule has certain specific physicochemical properties, such as the ability to bind to olfactory receptors and the characteristics of a volatile chemical structure. Therefore, chemical molecules can be screened through physicochemical information to screen out possible odor molecules, thereby avoiding the need to predict the odor category for all chemical molecules and improving the efficiency of odor prediction.

[0065] In some embodiments, the logistic regression model is pre-trained; furthermore, the model can be trained in a supervised manner. For example, the training process may include: firstly, constructing a training dataset containing a large number of labeled samples, where each sample is explicitly labeled as an odor molecule or a non-odor molecule; during training, the model takes the feature vector of the sample as input and optimizes the loss function of logistic regression through gradient descent, so that the model learns the decision boundary that can distinguish between the two types of molecules. The core training objective is: when the input is a sample belonging to the odor molecule, the probability output by the model approaches 1, that is, it is predicted as an odor molecule; and when the input is a non-odor molecule, the probability output approaches 0, that is, it is predicted as a non-odor molecule, thus achieving accurate classification of odor molecules and non-odor molecules in the test object.

[0066] It is understandable that the odor molecules included in the odor molecule concentration are all selected from the chemical molecules included in the test object.

[0067] Step 203: Based on the molecular structure information corresponding to the odor molecules in the odor molecule set, perform data conversion to obtain odor molecule diagram structure data.

[0068] Molecular structure information refers to a set of data that can accurately describe the interatomic connections, spatial arrangement, and geometric features within a chemical molecule; it includes atom type, bond type, bond length, bond angle, and molecular configuration.

[0069] Graph structure data is a data representation form with nodes and edges as core elements, used to describe complex relationships or topological structures between entities; in the embodiments of this application, the odor molecule graph structure data uses atoms as nodes and the chemical bonds between atoms as edges.

[0070] In some embodiments, data transformation can be achieved through a pre-written transformation script; specifically, the data transformation process may include: extracting atom type and chemical bond information from molecular structure information through regular expressions, then constructing a molecular graph structure with nodes as atoms and edges as chemical bonds, and further standardizing node features (such as adding atomic number, charge, etc.) and edge features (such as adding bond type, bond length, etc.) to finally obtain odor molecular graph structure data.

[0071] In some embodiments, during the construction of odor molecule graph structure data, the physicochemical information corresponding to odor molecules can also be fused to obtain odor molecule graph structure data; specifically, the physicochemical information corresponding to odor molecules can be added as node features to the corresponding nodes; in this way, the odor molecule graph structure data can include not only the topological information of chemical molecules, but also their corresponding physicochemical information; further improving the information richness of the odor molecule graph structure data.

[0072] Step 204: The graph neural network model predicts the odor type based on the odor molecule graph structure data, and obtains the odor type prediction results of the odor molecules in the odor molecule set.

[0073] The graph neural network model is a model used to process graph structure data. Furthermore, in this embodiment, the graph neural network model is used to predict the odor type based on the odor molecule graph structure data of the odor molecules, and obtain the odor type prediction result of the odor molecules.

[0074] In some embodiments, the graph neural network model is pre-trained; specifically, the training process includes: providing sample graph structure data of sample odor molecules and odor type labels of sample odor molecules; inputting the sample graph structure data and corresponding odor type labels into the graph neural network model; and having the graph neural network model learn the potential association between odor type and molecular topology through the sample graph structure data and corresponding odor type labels. Thus, after the training termination condition is met, the obtained graph neural network model has the ability to predict odor type based on the odor molecule graph structure data of odor molecules.

[0075] In some embodiments, the odor type prediction result of an odor molecule can be the specific odor type of the odor molecule, such as odor type prediction result being odor type a, or odor type prediction result being odor type b, etc.; in other embodiments, the odor type prediction result can be the probability distribution of an odor molecule among multiple preset odor types, such as preset odor types including odor type a, odor type b and odor type c, then the odor type prediction result can be: the probability of odor type a is 30%, the probability of odor type b is 30%, and the probability of odor type c is 40%.

[0076] Step 205: Based on the odor type prediction results of the odor molecules in the odor molecule cluster, determine the odor prediction results of the object to be tested.

[0077] In some embodiments, the odor type prediction results of each odor molecule in the odor molecule set can be combined to obtain the odor prediction result of the object to be tested; for example, if the odor type of odor molecule 1 in the odor molecule set is a, and the odor type of odor molecule 2 in the odor molecule set is b, then the odor prediction result of the object to be tested is a combination of types (a, b).

[0078] In other embodiments, considering that different odor types may vary in intensity under the same conditions, weights can be set for each odor type based on its intensity. The odor type prediction results of each odor molecule in the odor molecule set are then weighted, and the odor prediction result of the object to be tested is determined based on the weighted result. For example, if odor type a is stronger than odor type b, the weight of odor type a can be preset to 2, and the weight of odor type b to 0.5. Based on the odor type prediction results of each odor molecule in the odor molecule set, 40% of the odor molecules are identified as odor type a, and 60% as odor type b. Multiplying by the corresponding weights, the odor contribution rate of odor type a is 80%, and the odor contribution rate of odor type b is 30%. Therefore, odor type a, which has the highest odor contribution rate, can be determined as the odor prediction result of the object to be tested. It should be noted that the above examples are merely illustrative and do not constitute a limitation on this solution.

[0079] In the aforementioned odor prediction method, a preliminary screening is first performed using the physicochemical information of chemical molecules through a logistic regression model. This allows for the selection of odor molecules that may influence odor from a vast pool of chemical molecules. Next, a graph neural network model is used to predict odor type using the graph structure data of odor molecules. This determines the odor type prediction result for each odor molecule. By using the graph structure data of odor molecules, the graph neural network model can directly learn the topological structure information of molecules, thereby more accurately capturing the intrinsic relationship between molecular structure and odor type, and improving the accuracy of odor type prediction. Finally, the odor type prediction results of the target object are determined by combining the odor type prediction results of each odor molecule. In this process, the logistic regression model screens odor molecules that have an odor influence, avoiding the waste of resources caused by subsequent odor type analysis of molecules without odor influence, thus improving the efficiency of odor prediction. Simultaneously, the graph structure data preserves the topological structure information of chemical molecules, improving the accuracy of odor type prediction for chemical molecules, and consequently improving the accuracy of odor prediction results for the target object, achieving efficient and accurate odor prediction for the target object.

[0080] In one exemplary embodiment, such as Figure 3As shown, Figure 3 The diagram illustrates the training process of the logistic regression model according to an embodiment of this application. The training process of the logistic regression model includes steps 301 to 304. Wherein:

[0081] Step 301: Obtain the physicochemical information corresponding to the sample molecule and the sample label of the sample molecule; wherein, the sample label is used to indicate whether the sample molecule belongs to the odor molecule.

[0082] For example, if a sample molecule is an odor molecule, the corresponding sample label is "1"; if a sample molecule is not an odor molecule, the corresponding sample label is "0".

[0083] In some embodiments, sample molecules and their corresponding physicochemical information can be obtained from a public chemical database (PubChem).

[0084] Step 302: The first base model performs odor molecule prediction based on the physicochemical information corresponding to the sample molecules to obtain the odor molecule prediction results of the sample molecules.

[0085] The first pedestal model is a pre-trained model with classification capabilities. It is understandable that by using sample molecules to train the first pedestal model, it can achieve better performance in the classification task of odor molecules.

[0086] Specifically, the odor molecule prediction results include the probability that a sample molecule belongs to an odor molecule.

[0087] Step 303: Determine the first loss based on the sample label of the sample molecules and the prediction results of the odor molecules.

[0088] The first loss is used to measure the distance between the sample label of the sample molecule and the predicted result of the odor molecule. Furthermore, the first loss can be calculated by a loss function, such as the cross-entropy loss function or the cosine similarity loss function.

[0089] Step 304: Adjust the parameters of the first base model based on the first loss to obtain the logistic regression model.

[0090] Specifically, the parameters of the first base model are iteratively adjusted in the direction of minimizing the first loss using the gradient descent algorithm until the first training termination condition is met; further, the first training termination condition can be either the number of iterations reaching a threshold or the first loss being less than a loss threshold.

[0091] In the above embodiments, the first pedestal model is trained specifically for the odor molecule classification task by using sample molecules, so that the final logistic regression model can more accurately meet the needs of the odor molecule classification task, thereby effectively improving the accuracy of odor molecule classification.

[0092] In some embodiments, the training process of the graph neural network model includes the following:

[0093] Obtain the graphical structure data of sample odor molecules and reference odor type labels.

[0094] The reference odor type label is used to indicate the reference odor type of the sample odor molecules; furthermore, the reference odor type label can be represented by one-heat encoding.

[0095] In some embodiments, the graphical structure data of sample odor molecules can be obtained by preprocessing the chemical structure information of the sample odor molecules.

[0096] The second pedestal model predicts the odor type based on the graph structure data of the sample odor molecules, and obtains the odor type prediction results of the sample odor molecules.

[0097] The second pedestal model is a pre-trained model with graph data processing capabilities.

[0098] In some embodiments, the odor type prediction results of sample odor molecules may include the probability distribution of sample odor molecules under multiple preset odor types.

[0099] The second loss is determined by calculating the odor type prediction results of the sample odor molecules and the reference odor type label.

[0100] In some embodiments, the second loss is used to measure the distance between the predicted odor type of the sample odor molecule and the reference odor type label; further, the second loss can be calculated by a loss function, such as the cross-entropy loss function, the relative entropy loss function, etc.

[0101] The parameters of the second base model are adjusted based on the second loss to obtain the graph neural network model.

[0102] Specifically, the parameters of the second base model are iteratively adjusted in the direction of minimizing the second loss using the gradient descent algorithm until the second training termination condition is met. Furthermore, the second training termination condition can be that the number of iterations reaches a threshold, or that the second loss is less than a loss threshold, etc.

[0103] In the above embodiments, the second pedestal model is trained using the graph structure data of sample odor molecules, so that the trained graph neural network model can learn the deep mapping relationship between molecular structure and complex odor categories, thereby effectively improving the accuracy of the graph neural network model in predicting odor types.

[0104] In some embodiments, the process of constructing odor molecular graph structure data using molecular structure information includes the following steps: data conversion based on the molecular structure information corresponding to odor molecules in the odor molecule set to obtain odor molecular graph structure data, including:

[0105] Based on the molecular structure information corresponding to the odor molecules in the odor molecule set, the atoms and chemical bonds included in the odor molecules are determined.

[0106] In some embodiments, atoms and chemical bonds included in molecular structure information can be extracted using regular expressions; alternatively, molecular structure information can be traversed through a search to determine the atoms and chemical bonds included therein. The specific implementation method is not limited here.

[0107] Using atoms of odor molecules as nodes and chemical bonds as edges, we obtain the odor molecule graph structure data corresponding to the odor molecules in the odor molecule set.

[0108] In some embodiments, nodes can also be attached with atomic features, such as atom type and valence state; similarly, edges can be attached with bond features, such as bond type and bond length.

[0109] In the above embodiments, odor molecule graph structure data is constructed based on the molecular structure information of odor molecules, so that the odor molecule graph structure data can reflect the spatial topological structure of odor molecules, thereby facilitating the graph neural network model to effectively capture the topological structure and functional group information of molecules based on the odor molecule graph structure data.

[0110] In some embodiments, the odor prediction result of the object to be tested is determined based on the odor type prediction result of the odor molecules in the odor molecule set, including:

[0111] Based on the preset weights of each odor type, the odor type prediction results of the odor molecules in the odor molecule set are weighted to obtain the odor prediction results of the object to be tested.

[0112] In some embodiments, the preset weight of odor type is determined based on the perceived intensity of odor type; further, the perceived intensity of odor type refers to the degree to which the human olfactory system accepts that odor type; it should be noted that the human olfactory system has different degrees of acceptance for different odor types.

[0113] For example, the weight of odor type a is preset to 1, the weight of odor type b to 2, and the weight of odor type c to 3. Based on the odor type prediction results of odor molecules in the odor molecule set, it is determined that odor molecules of odor type a account for 40%, odor molecules of odor type b account for 30%, and odor molecules of odor type c account for 30%. After multiplying by their respective weights, it is obtained that odor molecules of odor type a account for 40%, odor molecules of odor type b account for 60%, and odor molecules of odor type c account for 90%. That is, the odor prediction results of the object to be tested are: the odor contribution ratio of odor type a is 40 / (40+60+90), the odor contribution ratio of odor type b is 60 / (40+60+90), and the odor contribution ratio of odor type c is 70 / (40+60+90).

[0114] In this embodiment, the odor type prediction results of the odor molecules in the odor molecule set are weighted by the preset weights of various odor types to obtain the odor prediction results of the test object, so that the odor prediction results of the test object are more consistent with the actual olfactory perception.

[0115] In some embodiments, the odor prediction method provided in this application is used for cigarette odor prediction. Please refer to [link to relevant documentation]. Figure 4 , Figure 4 The following is a schematic diagram of the application process of the odor prediction method provided in the embodiments of this application, including steps 401 to 404:

[0116] Step 401: Input the cigarette molecule database. Specifically, provide the cigarette molecule database, which is a database containing a large number of known chemical molecules; this database not only contains the identification information of each molecule (such as name, CAS number, odor category, etc.), but more importantly, it contains its key physicochemical characteristics, including but not limited to lipid-water partition coefficient, boiling point, and vapor pressure.

[0117] Step 402, data processing. For example, the three physicochemical characteristics of a chemical molecule, namely the lipid-water partition coefficient X1, boiling point X2, and vapor pressure X3, are used as feature data of the chemical molecule to construct a feature vector (X1, X2, X3).

[0118] Step 403, rapid screening using logistic regression. For example, the constructed feature vector (X1, X2, X3) is standardized, transforming it into a distribution with a mean of 0 and a standard deviation of 1. This distribution is then input into a pre-trained logistic regression model. The logistic regression model calculates a linear prediction value (i.e., a linear prediction value) based on the standardized feature vector to indicate whether the chemical molecule belongs to the odorant molecule category. Where θ0 is the intercept term, and θ1, θ2, and θ3 correspond to the regression coefficients of the standardized features, the intercept term and regression coefficients are determined by the model parameter library, which is the model parameters determined during the training process of the logistic regression model. Finally, the probability that the chemical molecule belongs to the odor molecule is calculated based on the linear prediction value, i.e., probability p = 1 / (1 + e^(-θ0)). -z Obviously, the value of p ranges from 0 to 1.

[0119] If the value of p is greater than the threshold, the chemical molecule is considered a candidate odor molecule. If the chemical molecule is a candidate odor molecule, a graph structure can be constructed based on the structural data of the chemical molecule to obtain molecular graph data, and the odor type can be predicted based on the molecular graph data, i.e., step 404 is executed; otherwise, the chemical molecule is considered a non-odor active molecule, and the detection of the chemical molecule ends.

[0120] Step 404, Odor Type Prediction. Specifically, the graph data of the corresponding molecule is loaded by the graph neural network model. Graph convolutional layers are used as the core of message passing. Each graph convolutional layer is followed by an activation layer. Then, a global average pooling layer is used to obtain the graph-level representation of the entire molecule. A fully connected classifier outputs the multi-label odor classification probability.

[0121] The training process for the logistic regression model is as follows: Figure 5 As shown, Figure 5 The diagram illustrates the application process of the logistic regression model, including steps 501 to 508, wherein:

[0122] Step 501: Start training the logistic regression model. In this step, you can set the training parameters and optimization objectives.

[0123] Step 502, Data Preprocessing and Feature Engineering. For example, structured data can be constructed, including feature 1 (lipid-water partition coefficient), feature 2 (boiling point), feature 3 (vapor pressure), and label data, and the data can be standardized, such as by Z-score standardization.

[0124] Step 503: Construct the training dataset. For example, feature data and label data are combined to obtain a set of training data, and a training dataset is constructed based on multiple sets of training data.

[0125] Step 504, Model Optimization Training. Specifically, optimize the model parameters of the logistic regression model in the direction of gradient descent.

[0126] Step 505: Determine if the maximum number of iterations has been reached. If the maximum number of iterations has been reached, proceed to step 506; otherwise, return to step 504.

[0127] Step 506, Performance Evaluation and Validation. For example, the model's classification accuracy, recall, and other metrics can be validated.

[0128] Step 507, Model Saving. For example, the trained model parameters can be saved to a model parameter library.

[0129] Step 508, Model Testing. For example, the trained logistic regression model can be tested using a test set to verify its actual performance.

[0130] Specifically, a training dataset was constructed using the public chemical database (PubChem), which contains 1526 odor molecules (positive samples) and 269 odorless molecules (negative samples). Odor molecules were labeled as 1, and odorless molecules were labeled as 0. The lipid-water partition coefficient, boiling point, and vapor pressure were selected as the inputs to the logistic regression model.

[0131] The logistic regression model uses a binary logistic regression model, with a maximum number of iterations set to 400. The objective function is a binary cross-entropy loss function, the expression of which is as follows:

[0132] .

[0133] Where m is the number of training samples, y (i) h is the true label of the i-th sample. (i) It is the probability predicted by the model.

[0134] During model training, the objective is optimized using gradient descent, where the gradient is calculated using the following formula:

[0135] .

[0136] Where X is the feature matrix containing the bias term, y is the true label vector, and h is the probability vector predicted by the model.

[0137] After training, the model coefficients were determined as follows: the intercept term θ0 ranges from 0.3 to 4.0, the regression coefficient θ1 corresponding to the lipid-water partition coefficient ranges from 0.15 to 2.0, the regression coefficient θ2 corresponding to the boiling point ranges from -9.5 to 9.9, and the regression coefficient θ3 corresponding to the vapor pressure ranges from -1.8 to 1.0.

[0138] For example, taking the intercept term θ0 = 1.987540, the regression coefficient corresponding to the fat-water partition coefficient θ1 = 0.469737, the regression coefficient corresponding to the boiling point θ2 = -1.152015, and the regression coefficient corresponding to the vapor pressure θ3 = -0.178280, and setting the judgment threshold to 0.8, the model is tested, and the test results are as follows. Figure 6As shown, Figure 6 As shown in Figure a, the ROC curve (receiver operating characteristic curve) has the false positive rate (FPR) on the x-axis and the true positive rate (TPR) on the y-axis. The ROC curve shows a clear upper left convex shape, and the area under the curve (AUC) reaches 0.92, which is significantly higher than the diagonal baseline of the random classifier. This indicates that the model has excellent discriminative ability in the odor classification task and can effectively distinguish between odorous and odorless compound samples.

[0139] like Figure 6 As shown in b, Figure 6 Figure b presents the confusion matrix for the test, where rows represent the true labels and columns represent the predicted labels. Taking the second row as an example, for samples with the true label "odor" (i.e., the true label is odor molecules), 159 samples were predicted as "odorless" (i.e., the predicted label is non-odor molecules), and 1368 samples were predicted as "odor" (i.e., the predicted label is odor molecules). Based on the calculation results of the confusion matrix, the model's classification accuracy is 86.86%, and the precision is 94.67%, indicating that the vast majority of samples predicted as having an odor do indeed have an odor. The F1 score reaches 92.06%, comprehensively reflecting the good balance between precision and recall of the model.

[0140] like Figure 6 As shown in c, Figure 6 The histogram of characteristic coefficients of C shows the contribution of the three molecular descriptors to odor prediction: the regression coefficient of the lipid-water partition coefficient is 0.469737, indicating that compounds with higher lipid solubility are more likely to be predicted as having an odor; the regression coefficient of boiling point is -1.152015, which has the largest absolute value and is negative, indicating that compounds with higher boiling points are less likely to have an odor, and this feature is the most important negative indicator for odor prediction; the regression coefficient of vapor pressure is -0.178280, which is negative but has a relatively small impact, indicating that compounds with higher vapor pressure are slightly more likely to be odorless.

[0141] like Figure 6 As shown in d, Figure 6 The d in the figure provides a predicted probability histogram. The predicted probability histogram shows that the trained logistic regression model has high confidence in classifying most samples, and the predicted probability distribution does not show extreme clustering.

[0142] In some embodiments, the graph neural network model includes a GNN (Graph Neural Network) input layer, a global average pooling layer, a fully connected layer, and an output layer, such as... Figure 7 As shown, Figure 7 A schematic diagram of the model processing of a graph neural network model is shown.

[0143] The GNN input layer is used to receive graph structure data. The process of acquiring the graph structure data includes steps 601 to 603, wherein:

[0144] Step 601, Molecular Input. Specifically, input the digitized molecular structure information of the molecule.

[0145] Step 602, Graph Structure Conversion. Specifically, the molecular structure information of the molecules is converted into graph structure data.

[0146] Step 603: Determine nodes and edges. Specifically, extract node features and edge features from the graph structure data to be used as the input layer of the output GNN for odor type prediction.

[0147] Finally, please see Figure 8 Taking vanillin, a common flavoring molecule in cigarette smoke, as an example, the relevant physicochemical transport characteristics of vanillin were input into a logistic regression model. Its lipid-water partition coefficient was 1.21; boiling point was 285°C; and vapor pressure was... Taking the intercept term θ0 = 1.987540, the regression coefficients corresponding to the fat-water partition coefficient θ1 = 0.469737, the regression coefficients corresponding to the boiling point θ2 = -1.152015, and the regression coefficients corresponding to the vapor pressure θ3 = -0.178280, the linear prediction value can be calculated as z = 1.987540 + 0.469737 * 1.21 - 1.152015 * 285 - 0.178280 * 1.33 × 10⁻⁶. -5 ≈-325.7684; then the predicted probability p = 1 / (1+exp(-(-325.7684))) = 1 / (1+exp(325.7684)) ≈ 0.00, which is much smaller than the threshold of 0.8. Therefore, the logistic regression model determines that it is a non-odor active molecule and filters it out. Since the molecule did not pass the first-level logistic regression filter, it will not be sent to the subsequent graph neural network model for odor category prediction.

[0148] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0149] Based on the same inventive concept, this application also provides an odor prediction device for implementing the odor prediction method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more odor prediction device embodiments provided below can be found in the limitations of the odor prediction method described above, and will not be repeated here.

[0150] In one exemplary embodiment, such as Figure 9 As shown, Figure 9 A schematic diagram of an odor prediction device is provided. The odor prediction device 700 includes:

[0151] The acquisition module 701 is used to acquire the molecular dataset corresponding to the test object; the molecular dataset includes the chemical molecules included in the test object and the corresponding physicochemical information of each chemical molecule.

[0152] The screening module 702 is used to predict odor molecules based on the physicochemical information corresponding to each chemical molecule by a logistic regression model, and to determine the odor molecule set; the odor molecule set includes at least one odor molecule.

[0153] The data processing module 703 is used to perform data conversion based on the molecular structure information corresponding to the odor molecules in the odor molecule set, and obtain odor molecule diagram structure data.

[0154] The odor prediction module 704 is used to predict the odor type based on the odor molecule graph structure data by the graph neural network model, and obtain the odor type prediction results of the odor molecules in the odor molecule set.

[0155] The output module 705 is used to determine the odor prediction result of the object to be tested based on the odor type prediction result of the odor molecules in the odor molecule set.

[0156] In some embodiments, the odor prediction device 700 further includes a logistic regression model training module for acquiring physicochemical information corresponding to sample molecules and sample labels of sample molecules; wherein, the sample label is used to indicate whether the sample molecule belongs to an odor molecule; the first base model performs odor molecule prediction based on the physicochemical information corresponding to the sample molecule to obtain the odor molecule prediction result of the sample molecule; based on the sample label of the sample molecule and the odor molecule prediction result, a first loss is determined; the parameters of the first base model are adjusted based on the first loss to obtain a logistic regression model.

[0157] In some embodiments, the odor prediction device 700 further includes a graph neural network model training module for acquiring graph structure data of sample odor molecules and reference odor type labels; performing odor type prediction based on the graph structure data of sample odor molecules by a second pedestal model to obtain odor type prediction results of sample odor molecules; calculating a loss based on the odor type prediction results of sample odor molecules and reference odor type labels to determine a second loss; and adjusting the parameters of the second pedestal model based on the second loss to obtain a graph neural network model.

[0158] In one embodiment, the physicochemical information corresponding to the chemical molecule includes at least one of the following: lipid-water partition coefficient, boiling point, and vapor pressure.

[0159] In one embodiment, the data processing module 703 is specifically used to determine the atoms and chemical bonds included in the odor molecules based on the molecular structure information corresponding to the odor molecules in the odor molecule set; and to obtain the odor molecule graph structure data corresponding to the odor molecules in the odor molecule set by using the atoms of the odor molecules as nodes and the chemical bonds as edges.

[0160] In one embodiment, the output module 705 is specifically used to weight the odor type prediction results of the odor molecules in the odor molecule set based on the preset weights of each odor type, so as to obtain the odor prediction result of the object to be tested.

[0161] Each module in the aforementioned odor prediction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0162] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 10As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores odor prediction-related data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements an odor prediction method.

[0163] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0164] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method described above.

[0165] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0166] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method described above.

[0167] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0168] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0169] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0170] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An odor prediction method, characterized in that, The method includes: Obtain the molecular dataset corresponding to the object to be tested; the molecular dataset includes the chemical molecules included in the object to be tested and the physicochemical information corresponding to each chemical molecule; An odor molecule set is determined by predicting odor molecules based on the physicochemical information corresponding to each chemical molecule using a logistic regression model; the odor molecule set includes at least one odor molecule. Data conversion is performed based on the molecular structure information corresponding to the odor molecules in the odor molecule set to obtain odor molecule diagram structure data; The odor type prediction results of the odor molecules in the odor molecule set are obtained by using a graph neural network model to predict the odor type based on the odor molecule graph structure data; Based on the odor type prediction results of the odor molecules in the odor molecule set, the odor prediction result of the object to be tested is determined.

2. The method according to claim 1, characterized in that, The training process of the logistic regression model includes: Obtain the physicochemical information corresponding to the sample molecule and the sample label of the sample molecule; wherein, the sample label is used to indicate whether the sample molecule belongs to the category of odor molecules; The first pedestal model performs odor molecule prediction based on the physicochemical information corresponding to the sample molecules, and obtains the odor molecule prediction result of the sample molecules. Based on the sample label of the sample molecule and the prediction result of the odor molecule, the first loss is determined; The parameters of the first base model are adjusted based on the first loss to obtain the logistic regression model.

3. The method according to claim 1, characterized in that, The training process of the graph neural network model includes: Obtain the graphical structure data of sample odor molecules and reference odor type labels; The second pedestal model predicts the odor type based on the graph structure data of the sample odor molecules, and obtains the odor type prediction result of the sample odor molecules; Based on the odor type prediction results of the sample odor molecules and the reference odor type label, a loss calculation is performed to determine the second loss; The parameters of the second base model are adjusted based on the second loss to obtain the graph neural network model.

4. The method according to claim 1, characterized in that, The physicochemical information corresponding to the chemical molecules includes at least one of the following: lipid-water partition coefficient, boiling point, and vapor pressure.

5. The method according to claim 1, characterized in that, The process of converting data based on the molecular structure information corresponding to the odor molecules in the odor molecule set to obtain odor molecule diagram structure data includes: Based on the molecular structure information corresponding to the odor molecules in the odor molecule set, the atoms and chemical bonds included in the odor molecules are determined; Using the atoms of the odor molecules as nodes and the chemical bonds as edges, the odor molecule graph structure data corresponding to the odor molecules in the odor molecule set is obtained.

6. The method according to any one of claims 1-5, characterized in that, The determination of the odor prediction result of the object to be tested based on the odor type prediction result of the odor molecules in the odor molecule set includes: Based on the preset weights of each odor type, the odor type prediction results of the odor molecules in the odor molecule set are weighted to obtain the odor prediction result of the object to be tested.

7. An odor prediction device, characterized in that, The device includes: The acquisition module is used to acquire the molecular dataset corresponding to the test object; the molecular dataset includes the chemical molecules included in the test object and the physicochemical information corresponding to each chemical molecule; The screening module is used to predict odor molecules based on the physicochemical information corresponding to each chemical molecule using a logistic regression model, and to determine the odor molecule set; the odor molecule set includes at least one odor molecule. The data processing module is used to perform data conversion based on the molecular structure information corresponding to the odor molecules in the odor molecule set, and obtain odor molecule diagram structure data; The odor prediction module is used to predict the odor type by a graph neural network model based on the odor molecule graph structure data, and to obtain the odor type prediction result of the odor molecules in the odor molecule set; The output module is used to determine the odor prediction result of the object to be tested based on the odor type prediction result of the odor molecules in the odor molecule set.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images