A molecular property prediction method and related devices

Abstract

The application belongs to the technical field of molecular attribute prediction, and discloses a kind of molecular attribute prediction method and related equipment, the method includes: obtaining the fixation point trajectory and corresponding judgment result label when technical personnel judges the molecular attribute of organic material, form training data set;A attribute prediction model containing fixation point trajectory prediction submodel and attribute prediction submodel is trained using the data set;Wherein, fixation point trajectory prediction submodel is responsible for simulating the attention mode of experts to the key area of molecular structure diagram, and the attribute prediction submodel combines molecular structure diagram and the prediction information of fixation point trajectory prediction submodel to predict molecular attribute;Using the trained attribute prediction model to predict whether the organic material molecule to be measured has the target attribute;Therefore, the key features in the molecular structure can be more accurately identified, deep chemical reasoning is carried out, and the accuracy and robustness of complex organic material molecular attribute prediction are significantly improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of molecular property prediction, in particular to a molecular property prediction method and related equipment. BACKGROUND

[0002] In the field of material science, accurate prediction of molecular properties of organic materials is a key step in the development of new materials. Currently, experts (i.e. experienced technical personnel) can often exhibit stronger chemical intuition and better robustness than existing artificial intelligence models when judging the complex properties of organic material molecules, such as luminescent efficiency in photoelectric-related properties. This is because experts have accumulated a wealth of theoretical knowledge and experience through long-term practice, and can identify key structural regions by observing organic material molecular structure diagrams, and make relatively accurate judgments about the potential properties of organic material molecules accordingly.

[0003] Specifically, existing artificial intelligence models generally face the challenge of insufficient data when predicting the properties of organic material molecules. Due to the high cost of obtaining high-quality organic material molecular property data, the data set used for model training is often limited in size. In addition, existing artificial intelligence models are usually constructed based on mathematical statistical principles, lacking the ability to integrate and understand chemical theoretical knowledge. This means that the model, when making predictions, is difficult to combine local features and overall configurations of organic material molecular structure diagrams for deep chemical reasoning, as experts do. Therefore, existing artificial intelligence models often struggle to make accurate predictions when faced with complex organic material molecular property prediction tasks, and may even produce predictions that are inconsistent with chemical theory, thereby limiting their reliability and effectiveness in practical applications.

[0004] The existing technology needs to be improved in view of the above problems. SUMMARY

[0005] The purpose of the present application is to provide a molecular property prediction method and related equipment, aiming to solve the problems of insufficient data, lack of understanding of chemical theoretical knowledge, and difficulty in making accurate predictions when using artificial intelligence models to predict the properties of organic material molecules in the prior art.

[0006] In a first aspect, the present application provides a molecular property prediction method for predicting whether a target organic material molecule has a target property, characterized in that the method comprises the following steps:

[0007] A1. Obtain the gaze point trajectory of the technical personnel through the process of judging whether the organic material molecule has the target property by the organic material molecular structure diagram and the corresponding judgment result label, forming a training data set;

[0008] A2. training an attribute prediction model using the training dataset; the attribute prediction model comprises a gaze point trajectory prediction sub-model and an attribute prediction sub-model, the gaze point trajectory prediction sub-model is configured to predict a gaze point trajectory according to an organic material molecular structure diagram, the attribute prediction sub-model is configured to predict whether an organic material molecule has the target attribute according to the organic material molecular structure diagram and prediction information of the gaze point trajectory predicted by the gaze point trajectory prediction sub-model;

[0009] A3. predicting whether an organic material molecule to be tested has the target attribute using the trained attribute prediction model.

[0010] Preferably, step A1 comprises:

[0011] A101. obtaining a true case of whether an organic material molecule has a target attribute as a judgment result label of an organic material molecular structure diagram corresponding to the organic material molecule;

[0012] A102. obtaining gaze point trajectories and corresponding judgment results of a process in which a plurality of technical personnel judge whether the organic material molecule has the target attribute through the organic material molecular structure diagram of the organic material molecule;

[0013] A103. eliminating the gaze point trajectories in which the judgment results do not match the corresponding judgment result labels;

[0014] A104. eliminating the gaze point trajectories in which the number of gaze points deviating from the organic material molecular structure diagram exceeds a preset number threshold from the remaining gaze point trajectories; the gaze points deviating from the organic material molecular structure diagram refer to the gaze points not falling on the organic material molecular structure diagram;

[0015] A105. clustering the remaining gaze point trajectories to obtain a plurality of gaze point trajectory clusters, and eliminating the gaze point trajectories deviating too much from the center trajectory of the gaze point trajectory cluster from each gaze point trajectory cluster;

[0016] A106. combining each remaining gaze point trajectory and the corresponding judgment result label to form a sample, and adding the sample to the training dataset;

[0017] A107. performing steps A101-A106 for a plurality of different organic material molecules to obtain a final training dataset.

[0018] Preferably, step A104 comprises:

[0019] obtaining an envelope of the organic material molecular structure diagram;

[0020] expanding the envelope outward by a preset expansion distance;

[0021] counting a number of the gaze points falling outside the region surrounded by the expanded envelope line, to obtain a number of deviated gaze points;

[0022] if the number of deviated gaze points exceeds a preset number threshold, discarding the gaze point trajectory.

[0023] Preferably, step A105 comprises:

[0024] clustering the remaining gaze point trajectories to obtain a plurality of gaze point trajectory clusters;

[0025] for each gaze point trajectory cluster, obtaining a center trajectory of the gaze point trajectory cluster according to the gaze point trajectories of the gaze point trajectory cluster;

[0026] for each gaze point trajectory cluster, calculating a deviation degree between each gaze point trajectory of the gaze point trajectory cluster and the center trajectory;

[0027] discarding the gaze point trajectory with a deviation degree greater than a preset deviation degree threshold.

[0028] Preferably, step A2 comprises:

[0029] A201. training the gaze point trajectory predictor model according to the organic material molecular structure graphs and the corresponding gaze point trajectories in the training data set, so that the gaze point trajectory predictor model can predict gaze point trajectories according to organic material molecular structure graphs;

[0030] A202. freezing the gaze point trajectory predictor model;

[0031] A203. training the attribute predictor model according to the organic material molecular structure graphs and the corresponding judgment result labels in the training data set, in combination with the gaze point trajectory predictor model, so that the attribute predictor model can predict whether an organic material molecule has the target attribute according to an organic material molecular structure graph and the prediction information of the gaze point trajectory predictor model on the gaze point trajectory.

[0032] Preferably, step A203 comprises:

[0033] inputting the organic material molecular structure graph in the training data set into the gaze point trajectory predictor model, and extracting prediction information of the gaze point trajectory predictor model on the gaze point trajectory; the prediction information includes the gaze point trajectory predicted by the gaze point trajectory predictor model and / or hidden layer parameters generated by the gaze point trajectory predictor model in the prediction process;

[0034] inputting the organic material molecular structure graph in the training data set and the extracted prediction information into the attribute predictor model, to obtain an actual judgment result output by the attribute predictor model;

[0035] According to the actual judgment result and the corresponding judgment result label, a loss function is calculated, and model parameters of the attribute prediction sub-model are optimized based on the loss function.

[0036] Preferably, step A3 comprises:

[0037] A301. Obtain a molecular structure diagram of the to-be-tested organic material molecule;

[0038] A302. Input the molecular structure diagram into the gaze point trajectory prediction sub-model, and extract prediction information of the gaze point trajectory prediction sub-model on the gaze point trajectory;

[0039] A303. Input the molecular structure diagram and the prediction information into the attribute prediction sub-model to obtain a prediction result of whether the to-be-tested organic material molecule has the target attribute.

[0040] In a second aspect, the present application provides a molecular attribute prediction system for predicting whether a to-be-tested organic material molecule has a target attribute, which comprises a display, an eye tracker and a host computer.

[0041] The display is used to display an organic material molecule structure diagram under the control of the host computer.

[0042] The eye tracker is used to collect gaze point trajectories of a technical staff in the process of judging whether an organic material molecule has a target attribute through an organic material molecule structure diagram displayed by the display, and upload to the host computer.

[0043] The host computer is used to perform:

[0044] Obtain a judgment result of whether an organic material molecule corresponding to an organic material molecule structure displayed by the display has a target attribute by a technical staff, combine the gaze point trajectories uploaded by the eye tracker and the organic material molecule structure diagram corresponding to the gaze point trajectories, and form a training data set;

[0045] Train an attribute prediction model using the training data set; the attribute prediction model comprises a gaze point trajectory prediction sub-model and an attribute prediction sub-model, the gaze point trajectory prediction sub-model is used to predict a gaze point trajectory according to an organic material molecule structure diagram, and the attribute prediction sub-model is used to predict whether an organic material molecule has the target attribute according to the organic material molecule structure diagram and prediction information of the gaze point trajectory prediction sub-model on the gaze point trajectory;

[0046] Use the trained attribute prediction model to predict whether a to-be-tested organic material molecule has the target attribute.

[0047] In a third aspect, the present application provides an electronic device comprising a processor and a memory, wherein the memory stores a computer program executable by the processor, and the processor executes the computer program to perform the steps of the molecular property prediction method described above.

[0048] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to perform the steps of the molecular property prediction method described above.

[0049] Beneficial effects: The molecular property prediction method and related device provided by the present application construct a training data set containing the experience of technicians by obtaining the fixation point trajectory and judgment result label of the technicians when judging the molecular properties of organic materials. On this basis, an attribute prediction model containing a fixation point trajectory prediction sub-model and an attribute prediction sub-model is trained. The fixation point trajectory prediction sub-model can simulate the observation mode of technicians and predict the fixation point trajectory according to the molecular structure diagram; the attribute prediction sub-model combines the molecular structure diagram and the fixation point trajectory prediction information to predict whether the molecule has the target attribute. This method effectively solves the problem that the existing artificial intelligence model is not accurate in predicting the properties of complex organic material molecules due to insufficient data and lack of chemical intuition. By integrating the expert knowledge and judgment process of technicians into the model training, the method of the present application can more deeply understand the relationship between molecular structure and properties, thereby significantly improving the accuracy and robustness of molecular property prediction, overcoming the deficiency of existing models that are difficult to perform deep chemical reasoning, and providing a more reliable prediction tool for new material research and development. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A flowchart of the molecular property prediction method provided by the present application.

[0051] Figure 2 A schematic diagram of the molecular property prediction system provided by the present application.

[0052] Figure 3 A structural schematic diagram of the electronic device provided by the present application.

[0053] Figure 4 A schematic diagram of the molecular structure diagram of the organic material and the corresponding fixation point trajectory.

[0054] Figure 5 A schematic diagram of the attribute prediction model.

[0055] Label explanation: 1, display; 2, eye tracker; 3, upper computer; 301, processor; 302, memory; 303, communication bus. DETAILED DESCRIPTION

[0056] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present application.

[0057] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. Meanwhile, in the description of the present application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0058] Please refer to Figure 1 A molecular property prediction method in some embodiments of the present application is used to predict whether a to-be-tested organic material molecule has a target property, characterized in that the method comprises the following steps:

[0059] A1. Obtain the gaze point trajectory of the process in which a technician judges whether the organic material molecule has the target property through the organic material molecule structure diagram and the corresponding judgment result label, to form a training data set;

[0060] A2. Train a property prediction model using the training data set; the property prediction model comprises a gaze point trajectory prediction sub-model and a property prediction sub-model, the gaze point trajectory prediction sub-model is used to predict the gaze point trajectory according to the organic material molecule structure diagram, and the property prediction sub-model is used to predict whether the organic material molecule has the target property according to the organic material molecule structure diagram and the prediction information of the gaze point trajectory predicted by the gaze point trajectory prediction sub-model;

[0061] A3. Use the trained property prediction model to predict whether the to-be-tested organic material molecule has the target property.

[0062] The present application introduces the gaze point trajectory information of the technician to simulate the thinking process of the expert when judging the property of the organic material molecule, thereby effectively making up for the shortcomings of the existing model in chemical intuition and robustness, and significantly improving the accuracy and reliability of the prediction of the property of the complex organic material molecule.

[0063] The "molecular structure diagram of an organic material" refers to a graphical representation of the atomic connections and spatial arrangement in the molecule of an organic material, commonly used in the field of chemistry to help technicians visually understand the composition of the molecule.

[0064] The "target attribute" refers to a specific property that the molecule of an organic material is to be predicted, such as luminescence, electrical conductivity, stability, etc.

[0065] The "gaze point trajectory" refers to the path record of the technician's eyes moving and staying on the image when observing the molecular structure diagram of an organic material, reflecting the focus area and thought process of the technician, such as Figure 4 The schematic diagram showing the gaze point trajectory formed on the molecular structure diagram of an organic material is shown, where the horizontal coordinate u represents the pixel horizontal coordinate, and the vertical coordinate v represents the pixel vertical coordinate.

[0066] The "judgment result label" refers to the true judgment result given by the technician on whether the molecule of an organic material has the target attribute, usually a binary classification (such as "yes" or "no") or multi-classification label.

[0067] The "attribute prediction model" is a machine learning model whose purpose is to predict whether the molecule of an organic material has the target attribute based on the molecular structure diagram of the material. The model consists of two sub-models, namely the gaze point trajectory prediction sub-model and the attribute prediction sub-model.

[0068] The "gaze point trajectory prediction sub-model" is part of the attribute prediction model, whose function is to predict the gaze point trajectory that the technician may generate when observing the molecular structure diagram of an organic material based on the input of the structure diagram.

[0069] The "attribute prediction sub-model" is another part of the attribute prediction model, whose function is to combine the molecular structure diagram of an organic material and the prediction information provided by the gaze point trajectory prediction sub-model to finally predict whether the molecule of the material has the target attribute.

[0070] The implementation environment of the present application usually includes a data processing module, a model training module and a prediction module, which can run on one or more computer devices and exchange data through a network.

[0071] The present application proposes a molecular attribute prediction method, the core of which is to simulate the judgment process of technicians, thereby improving the accuracy of prediction.

[0072] Specifically, this method first requires acquiring the gaze trajectory and corresponding judgment label of technicians as they judge whether an organic material molecule possesses the target attribute based on its molecular structure diagram. This data forms the training dataset. In practice, eye-tracking devices can be used to record the eye movements of technicians while observing the organic material molecular structure diagrams, thus obtaining the gaze trajectory. For example, several experienced chemistry experts can be invited to view a series of organic material molecular structure diagrams on a monitor and judge whether each molecule possesses a specific target attribute. During this process, the eye tracker records the experts' gaze trajectories in real time. Simultaneously, the actual attribute label for each organic material molecule needs to be acquired as the judgment result label. This data collectively constitutes the training dataset for subsequent model training.

[0073] Furthermore, the resulting training dataset is used to train an attribute prediction model. This model is designed to include a gaze trajectory prediction sub-model and an attribute prediction sub-model. The gaze trajectory prediction sub-model is used to predict gaze trajectories based on the organic material molecular structure diagram. For example, a convolutional neural network (CNN)-based image processing model can be constructed as the gaze trajectory prediction sub-model, which can extract visual features from the input organic material molecular structure diagram and output the predicted gaze trajectory. The attribute prediction sub-model is then used to predict whether the organic material molecule possesses the target attribute based on the organic material molecular structure diagram and the gaze trajectory prediction information from the gaze trajectory prediction sub-model. For example, a multilayer perceptron (MLP) or another CNN model can be constructed as the attribute prediction sub-model, which receives the features of the organic material molecular structure diagram and the predicted gaze trajectory information as input and outputs the final attribute prediction result. During training, these two sub-models can be jointly trained or trained in stages to optimize the overall model performance.

[0074] Finally, the trained attribute prediction model is used to predict whether the tested organic material molecule possesses the target attribute. When predicting a new organic material molecule, its molecular structure diagram is first obtained and then input into the trained gaze trajectory prediction sub-model to obtain prediction information. Next, the molecular structure diagram and the prediction information are input together into the attribute prediction sub-model to finally obtain the prediction result of whether the tested organic material molecule possesses the target attribute.

[0075] The molecular property prediction method proposed in this application effectively solves the problem of lack of chemical intuition and robustness in existing artificial intelligence models by simulating the judgment process of technicians.

[0076] Specifically, when obtaining the gaze point trajectories of technicians and judgment result labels to form a training data set, the application introduces the experience knowledge of human experts. Traditional methods often rely only on molecular structure data and attribute labels for model training, ignoring the cognitive clues of experts in the judgment process. The application integrates these valuable cognitive information into the training data by collecting the gaze point trajectories of experts when observing the molecular structure diagram, so that the model can learn the key areas and reasoning paths that the experts pay attention to.

[0077] When training the attribute prediction model with the training data set, the application decomposes the attribute prediction model into a gaze point trajectory prediction sub-model and an attribute prediction sub-model. The gaze point trajectory prediction sub-model first predicts the gaze point trajectory according to the organic material molecular structure diagram, which is equivalent to simulating the process of the expert first identifying the key structure area when observing the molecular structure diagram. Subsequently, the attribute prediction sub-model combines the organic material molecular structure diagram and the predicted information to predict whether the organic material molecule has the target attribute. This staged prediction mechanism enables the model to not only extract features from the molecular structure diagram, but also use simulated expert attention information for deeper chemical reasoning. Compared with traditional models that directly predict attributes from molecular structure diagrams, the model of the application can better understand the intrinsic relationship between molecular structure and attributes, thereby improving the accuracy and robustness of prediction.

[0078] Thus, when predicting whether the target organic material molecule has the target attribute using the trained attribute prediction model, the model of the application can provide a prediction result closer to the expert judgment. This method not only improves the accuracy of prediction, but also provides interpretability for the prediction result, because the predicted gaze point trajectory can indicate the molecular structure region that the model pays attention to when making a judgment, which has important guiding significance for the research and optimization of new materials.

[0079] The molecular attribute prediction method proposed in the application has significant advantages and innovations compared to existing technologies. Traditional existing molecular attribute prediction methods usually directly input the molecular structure diagram into the deep learning model for end-to-end prediction. Although this method can achieve attribute prediction to some extent, it lacks the inherent understanding of chemical theoretical knowledge and the integration of expert experience, and its prediction results often show limitations when facing complex attributes, especially in the case of insufficient data, the generalization ability and robustness of the model are poor.

[0080] The core innovation of the present application is to introduce the gaze point trajectory information of technical personnel and construct a two-stage prediction architecture including a gaze point trajectory prediction sub-model and an attribute prediction sub-model. By obtaining the gaze point trajectory of the technical personnel when judging the molecular attribute, the present application integrates the cognitive process and attention mechanism of human experts into the model training. This enables the gaze point trajectory prediction sub-model to learn and simulate the key areas that the experts focus on when observing the molecular structure graph, thereby providing more instructive feature information for the attribute prediction sub-model.

[0081] For example, when predicting the light-emitting efficiency of an organic material molecule, the traditional model may only learn from global features, while the model of the present application can focus on specific functional groups or conjugated systems in the molecule that are closely related to the light-emitting performance through the predicted gaze point trajectory. This focus on key areas enables the attribute prediction sub-model to perform more accurate chemical reasoning, thereby significantly improving the prediction accuracy. In addition, this method also provides interpretability for the prediction results, as the predicted gaze point trajectory can intuitively show the molecular structure region on which the model relies when making a judgment, which is of great value for chemists to understand the model decision-making process and design new materials. Therefore, the present application not only improves the prediction performance, but also enhances the credibility and practicality of the model, providing a more intelligent and efficient solution for attribute prediction of organic material molecules.

[0082] In some embodiments, step A1 comprises:

[0083] A101. Obtain the true situation of whether the organic material molecule has the target attribute as the judgment result label of the organic material molecular structure graph corresponding to the organic material molecule;

[0084] A102. Obtain the gaze point trajectories and corresponding judgment results of a plurality of technical personnel judging whether the organic material molecule has the target attribute through the organic material molecular structure graph of the organic material molecule;

[0085] A103. Discard the gaze point trajectories that do not match the corresponding judgment result labels among the judgment results;

[0086] A104. For the remaining gaze point trajectories, discard the gaze point trajectories whose number of deviated gaze points exceeds a preset number threshold; the deviated gaze point refers to a gaze point that does not fall on the organic material molecular structure graph;

[0087] A105. Cluster the remaining gaze point trajectories to obtain a plurality of gaze point trajectory clusters, and discard the gaze point trajectories that deviate too much from the center trajectory of the gaze point trajectory cluster from each gaze point trajectory cluster;

[0088] A106. Grouping each remaining gaze point trajectory and the corresponding judgment result label into a sample, and adding the sample to the training data set;

[0089] A107. Performing steps A101-A106 for a plurality of different organic material molecules to obtain a final training data set.

[0090] Specifically, step A101 aims to obtain the true situation of whether the organic material molecule has the target attribute, which is usually obtained through experimental verification or authoritative database query as a benchmark for subsequent data cleaning and model training. The true situation is used as the judgment result label corresponding to the organic material molecule structure graph.

[0091] In step A102, gaze point trajectories of multiple technical personnel when observing the organic material molecule structure graph and making judgments, and their respective judgment results are obtained by equipment such as an eye tracker. This step aims to collect multi-source, multi-angle artificial judgment data.

[0092] Among them, step A103 is used for the first layer of screening of the preliminary obtained data. Specifically, if the judgment result of the technical personnel is inconsistent with the true situation (i.e. the judgment result label) obtained in step A101, it is considered that the gaze point trajectory of the technical personnel may be misleading or erroneous, and therefore the inconsistent gaze point trajectory is removed. The purpose is to ensure that the gaze point trajectories in the training data set correspond to the correct judgment result label, and to improve the reliability of the data.

[0093] Further, step A104 aims to remove trajectories containing a large number of noise or invalid gaze points. Deviation gaze points refer to gaze points that do not fall on the organic material molecule structure graph, such as the technical personnel's line of sight briefly moving away from the molecule structure graph area during observation. By counting the number of deviation gaze points in the gaze point trajectory and comparing it with the pre-set number threshold, low-quality gaze point trajectories generated due to inattention or misoperation can be effectively identified and removed. The purpose is to remove gaze point information unrelated to the molecule structure graph itself, so that the remaining gaze point trajectories are more focused on the key areas of the molecule structure.

[0094] On this basis, step A105 performs clustering processing on the gaze point trajectories that have passed the preliminary screening. Through clustering, trajectories with similar gaze patterns can be grouped into one class, forming multiple gaze point trajectory clusters. For each gaze point trajectory cluster, a center trajectory is calculated, which can be understood as a representative pattern of the gaze point trajectories in the cluster. Subsequently, gaze point trajectories that deviate too much from the center trajectory are removed from the cluster. The purpose of this step is to further remove outliers or outliers within the cluster, ensuring that the trajectories within each gaze point trajectory cluster have high consistency and representativeness, thereby refining more common expert gaze patterns.

[0095] In step A106, each remaining gaze point trajectory and the corresponding judgment result label after the above multiple screening and optimization are combined to form a high-quality sample, which is added to the training data set. Finally, step A107 ensures the richness and diversity of the training data set, and by repeatedly performing steps A101 to A106 for a plurality of different organic material molecules, a comprehensive training data set containing a plurality of molecular types and attribute judgment cases can be constructed.

[0096] The scheme of the present application effectively solves the problems of noise, inconsistency and low quality that may exist in the original data by fine management of the construction process of the training data set. Specifically, step A103 eliminates gaze point trajectories with incorrect judgments by comparing the judgment results of technicians with the true situation, thereby avoiding the model from learning from incorrect human judgments. Step A104 ensures the effectiveness of the gaze point trajectory by identifying and eliminating a large number of gaze points deviating from the molecular structure diagram, so that the model can focus more on the key areas of the molecular structure itself. In addition, step A105 further refines the commonality of expert gaze patterns by clustering and eliminating outliers within the cluster, reduces the bias caused by individual differences or accidental factors, so that the training data set can more accurately reflect the core focus and decision path of technicians when judging the molecular attributes. It is precisely due to these data cleaning and optimization steps that the training data set has higher quality and representativeness.

[0097] Through the above technical scheme, the present application can significantly improve the quality and reliability of the training data set. Through multiple screening and optimization, inaccurate, inconsistent or noisy gaze point trajectories are excluded, so that the training data set can more accurately reflect the real and effective gaze patterns and decision logic of technicians when judging the attributes of organic material molecules. Therefore, the attribute prediction model trained using the high-quality training data set will have significantly improved learning efficiency and prediction accuracy, and the generalization ability of the model will be enhanced, so that it can more reliably predict whether the to-be-tested organic material molecule has the target attribute.

[0098] In some possible implementations, step A104 can include:

[0099] Obtaining an envelope of the molecular structure diagram of the organic material;

[0100] Expanding the envelope outward by a preset expansion distance;

[0101] Counting the number of gaze points of the gaze point trajectory that fall outside the area surrounded by the expanded envelope to obtain the number of deviating gaze points;

[0102] If the number of deviating gaze points exceeds a preset number threshold, the gaze point trajectory is eliminated.

[0103] Specifically, the envelope of the organic material molecular structure diagram can be understood as a smallest convex polygon or a smallest rectangular region that can completely enclose the molecular structure diagram. The purpose is to provide a clear boundary for subsequent expansion operations.

[0104] Further, the outward expansion of the envelope refers to uniform expansion in each direction of the envelope by a preset expansion distance, forming a larger region than the original envelope. For example, the expansion distance can be set according to the accuracy of the eye tracker, the screen resolution, or an empirical value. The purpose is to provide a certain tolerance space for the fixation point to avoid misjudgment as a deviation from the fixation point due to slight deviations.

[0105] Wherein, the number of fixation points falling outside the region enclosed by the expanded envelope is to judge the position of each fixation point in the fixation point trajectory. If its coordinates are outside the region enclosed by the expanded envelope (including the envelope itself), it is counted as a deviation from the fixation point. Thus, the total number of deviations from the fixation point in a fixation point trajectory can be obtained. In practical applications, if the number of deviations from the fixation point exceeds a preset number threshold (which can be set according to actual needs), it is considered that the fixation point trajectory contains too much noise or invalid information and should be excluded from the training data set. The preset number threshold can be adjusted according to actual needs and data characteristics, for example, it can be set as a percentage of the total number of fixation points in the trajectory. The purpose is to ensure the quality of the training data set.

[0106] The scheme of the present application introduces the envelope of the organic material molecular structure diagram and expands it to construct an effective fixation region with a certain tolerance. When the fixation point trajectory of the technician is collected, even if there is a slight eye movement error or the fixation point slightly exceeds the actual boundary of the molecular structure diagram, as long as it still falls within the region enclosed by the expanded envelope, it will not be immediately judged as a deviation from the fixation point. Only when the fixation point deviates significantly from the expanded region, it will be accurately identified as a deviation from the fixation point. Thus, by counting the number of fixation points falling outside the expanded region and comparing it with the preset number threshold, those fixation point trajectories that truly contain a large amount of noise or invalid information can be more accurately and robustly identified and excluded, thereby avoiding the misdeletion of valid data due to too strict boundary judgment or the retention of too much noise data due to too loose boundary judgment.

[0107] By the technical solution, the gaze point trajectories containing excessive noise in the training data set can be more accurately identified and removed. Compared with simply judging whether the gaze point falls on the molecular structure diagram, the expanded envelope is used as the judgment basis to provide a reasonable fault tolerance space for the gaze point, effectively reducing the misjudgment rate caused by the accuracy limitation of the eye tracker or the difference in gaze habits of the technicians. This significantly improves the quality of the training data set, and further helps to train a more stable and accurate attribute prediction model, thereby improving the reliability of the model in predicting the molecular attributes of organic materials.

[0108] Further, the step A105 can include:

[0109] clustering the remaining gaze point trajectories to obtain a plurality of gaze point trajectory clusters;

[0110] For each gaze point trajectory cluster, a center trajectory of the gaze point trajectory cluster is obtained according to the gaze point trajectories of the gaze point trajectory cluster.

[0111] For each gaze point trajectory cluster, a deviation degree between each gaze point trajectory of the gaze point trajectory cluster and the center trajectory is calculated.

[0112] The gaze point trajectory with a deviation degree greater than a preset deviation degree threshold is removed.

[0113] Wherein, clustering the remaining gaze point trajectories can be understood as grouping the gaze point trajectories with similar characteristics or patterns. For example, K-means, DBSCAN, hierarchical clustering and other algorithms can be used to cluster the gaze point trajectories, and the purpose is to classify the trajectories with similar attention areas or scanning paths formed by experts in the judgment process into a category for subsequent fine selection.

[0114] Further, for each gaze point trajectory cluster, a center trajectory of the gaze point trajectory cluster is obtained according to the gaze point trajectories of the gaze point trajectory cluster. The center trajectory can be regarded as a representative trajectory of the trajectory cluster, which can reflect the common pattern or average behavior of the gaze point trajectories in the cluster. For example, the center trajectory can be obtained by calculating the average trajectory, the median trajectory of all trajectories in the cluster, or selecting the trajectory with the smallest sum of distances from all other trajectories in the cluster. The purpose is to provide a benchmark for subsequent deviation degree calculation.

[0115] In practical applications, for each gaze point trajectory cluster, the deviation degree between each gaze point trajectory in the gaze point trajectory cluster and the center trajectory is calculated. The deviation degree is a quantitative index for measuring the difference between a single gaze point trajectory in the cluster and the center trajectory of the cluster. For example, the deviation degree can be calculated by Dynamic Time Warping (DTW) distance, Euclidean distance, Fréchet distance, etc. The purpose is to quantify the degree of conformity of each trajectory to the representative pattern of the cluster.

[0116] Thus, the gaze point trajectory with a deviation degree greater than the preset deviation threshold is removed. The preset deviation threshold is a preset value for determining whether a gaze point trajectory is considered an outlier or noise in the cluster. When the deviation degree of a gaze point trajectory exceeds the threshold, it is considered that the trajectory is too different from the overall pattern of the cluster and should be removed. The threshold can be determined by experience, statistical analysis or by cross-validation, etc. The purpose is to remove abnormal trajectories that are not representative in the cluster and further purify the training data set.

[0117] The scheme of the present application introduces a center trajectory and deviation degree calculation to finely screen the clustered gaze point trajectories. Specifically, after similar gaze point trajectories are classified into different gaze point trajectory clusters, a representative center trajectory is determined for each cluster to provide a clear reference for subsequent quality evaluation. Then, by quantitatively calculating the deviation degree between each gaze point trajectory in the cluster and the center trajectory, the "normal" degree of each trajectory can be objectively evaluated. It is precisely due to this quantitative deviation degree evaluation that it is possible to accurately identify and remove abnormal trajectories that are significantly inconsistent with the mainstream pattern in the cluster based on a preset deviation threshold. This mechanism effectively avoids errors that may be caused by subjective judgment or rough screening, ensuring high consistency and representativeness within each gaze point trajectory cluster.

[0118] Through the above technical scheme, the present application can finely and accurately screen the gaze point trajectories in the training data set. By calculating the deviation degree of the gaze point trajectory and the center trajectory and setting a threshold, abnormal trajectories with a large deviation degree in the cluster can be effectively identified and removed, thereby significantly improving the purity and quality of the training data set. This helps to ensure that the attribute prediction model learns more representative and reliable expert gaze patterns during training, thereby improving the accuracy and robustness of the model in predicting the properties of organic material molecules and avoiding prediction bias caused by noise or outliers in the training data.

[0119] Specifically, step A2 includes:

[0120] A201. According to the organic material molecular structure graph and the corresponding gaze point trajectory in the training data set, the gaze point trajectory predictor model is trained, so that the gaze point trajectory predictor model can predict the gaze point trajectory according to the organic material molecular structure graph;

[0121] A202. Freeze the gaze point trajectory predictor model;

[0122] A203. According to the organic material molecular structure graph and the corresponding judgment result label in the training data set, the attribute predictor model is trained in combination with the gaze point trajectory predictor model, so that the attribute predictor model can predict whether the organic material molecule has the target attribute according to the organic material molecular structure graph and the prediction information of the gaze point trajectory predicted by the gaze point trajectory predictor model.

[0123] Specifically, step A201 aims to first train the gaze point trajectory predictor model independently. In this phase, the sub-model is input with the organic material molecular structure graph in the training data set, and learns with the corresponding real gaze point trajectory as the supervision signal. The goal is to enable the gaze point trajectory predictor model to accurately extract and predict the gaze point trajectory of the technician from the molecular structure graph.

[0124] Wherein, step A202 refers to fixing the model parameters of the gaze point trajectory predictor model after it completes the training and achieves the expected prediction performance, and no longer allowing it to update in the subsequent training process. This ensures that the gaze point trajectory predictor model provides stable and high-quality gaze point trajectory prediction information during the training phase of the attribute predictor model.

[0125] In practical applications, step A203 is carried out after the gaze point trajectory predictor model is frozen. In this phase, the attribute predictor model is trained, and its input includes the organic material molecular structure graph in the training data set and the gaze point trajectory prediction information provided by the frozen gaze point trajectory predictor model. The attribute predictor model takes the corresponding judgment result label as the supervision signal, and learns how to combine the molecular structure graph and the gaze point trajectory prediction information to accurately predict whether the organic material molecule has the target attribute.

[0126] The solution of the present application effectively solves the training stability problem by decomposing the training process of the attribute prediction model into two sequentially performed stages and introducing the mechanism of freezing the gaze point trajectory prediction sub-model. First, the gaze point trajectory prediction sub-model is independently trained, enabling it to focus on learning the ability to accurately predict the gaze point trajectory from the organic material molecular structure graph. The training of this stage ensures that the gaze point trajectory prediction sub-model can provide high-quality intermediate features or prediction results. Subsequently, during the training of the attribute prediction sub-model, the gaze point trajectory prediction sub-model is frozen, and its parameters are no longer updated. Thus, the attribute prediction sub-model can be trained on the basis of a stable upstream model that has learned effective gaze point information, thereby avoiding mutual interference and optimization difficulties that may occur during joint training. The attribute prediction sub-model can focus more on learning how to combine molecular structure graph information with stable gaze point trajectory prediction information to accurately make the final attribute judgment.

[0127] Through the above technical solution, the training process of the attribute prediction model is optimized, significantly improving the stability and efficiency of the training. Since the gaze point trajectory prediction sub-model has been fully trained and frozen before the training of the attribute prediction sub-model, the attribute prediction sub-model can learn on a more reliable feature basis, thereby avoiding training fluctuations caused by unstable output of the gaze point trajectory prediction sub-model. This enables the attribute prediction sub-model to converge more effectively and ultimately improves the prediction accuracy and robustness of the entire attribute prediction model for the target attributes of organic material molecules.

[0128] In some preferred embodiments, the present application is implemented as follows:

[0129] As a specific implementation, in step A201, the gaze point trajectory prediction sub-model can be constructed as a sequence-to-sequence model based on convolutional neural network (CNN) and recurrent neural network (RNN). Among them, the CNN part is used to extract spatial features from the organic material molecular structure graph, and the RNN part (such as long short-term memory network LSTM or gated recurrent unit GRU) is used to predict the gaze point trajectory sequence according to the extracted feature sequence. During training, mean square error (MSE) can be used as the loss function to minimize the difference between the predicted gaze point trajectory and the real gaze point trajectory. Specifically, in step A202, when the gaze point trajectory prediction sub-model is trained, all trainable parameters (such as convolution kernel weights, RNN unit weights, etc.) are set to an un-updatable state. This can be achieved by setting the `requires_grad` attribute of the corresponding layer to `False` in a deep learning framework such as TensorFlow or PyTorch. Further, in step A203, the attribute prediction sub-model can be designed as a multi-modal fusion network. The network receives two inputs: one is the original organic material molecular structure graph, and the other is the prediction information from the frozen gaze point trajectory prediction sub-model. The prediction information can include the predicted gaze point trajectory sequence itself, or the hidden layer parameters generated by the gaze point trajectory prediction sub-model during prediction (such as the final hidden state or attention weight of RNN). The attribute prediction sub-model can include a separate CNN branch to process the molecular structure graph and a separate network branch to process the gaze point trajectory prediction information. The output features of the two branches are then concatenated or fused through an attention mechanism and input to a fully connected layer for binary classification (e.g., with or without target attributes). During training, the cross-entropy loss function can be used, and the gradient descent optimizer (such as Adam) can be used to update the parameters of the attribute prediction sub-model.

[0130] Preferably, step A203 can include:

[0131] inputting the organic material molecular structure graph in the training data set into the gaze point trajectory prediction sub-model, and extracting prediction information of the gaze point trajectory prediction sub-model for the gaze point trajectory; the prediction information includes the gaze point trajectory predicted by the gaze point trajectory prediction sub-model and / or the hidden layer parameters generated by the gaze point trajectory prediction sub-model during prediction;

[0132] inputting the organic material molecular structure graph in the training data set and the extracted prediction information into the attribute prediction sub-model to obtain the actual judgment result output by the attribute prediction sub-model;

[0133] According to the actual judgment result and the corresponding judgment result label, a loss function is calculated, and model parameters of the attribute prediction sub-model are optimized based on the loss function.

[0134] Specifically, the prediction information can be understood as any information that can help attribute prediction generated by the gaze point trajectory prediction sub-model when processing the organic material molecular structure graph. For example, it can include the gaze point trajectory predicted by the gaze point trajectory prediction sub-model according to the organic material molecular structure graph. The trajectory can be represented in the form of a heat map, a coordinate sequence, or a probability distribution, and is used to indicate the area or path that the human expert may focus on when judging the molecular attribute. In addition, the prediction information can also include the hidden layer parameters generated by the gaze point trajectory prediction sub-model in the prediction process. These hidden layer parameters are usually intermediate representations of the model for abstracting and encoding input features, and they can contain richer and deeper semantic information than the final predicted trajectory, and can reflect the attention points and feature extraction process of the model on the molecular structure graph.

[0135] In practical applications, the organic material molecular structure graph in the training data set and the extracted prediction information are input into the attribute prediction sub-model, aiming to provide more comprehensive input for the attribute prediction sub-model. After receiving these inputs, the attribute prediction sub-model will output an actual judgment result about whether the organic material molecule has the target attribute based on its internal logic and parameters. The actual judgment result is usually a classification label.

[0136] Further, in order to optimize the performance of the attribute prediction sub-model, a loss function needs to be calculated according to the actual judgment result and the corresponding judgment result label. The loss function is used to quantify the difference or error between the model prediction result and the true label. For example, for a binary classification problem, a binary cross-entropy loss function can be used; for a multi-classification problem, a cross-entropy loss function can be used. Based on the loss function, the model parameters of the attribute prediction sub-model are iteratively adjusted and optimized through a backpropagation algorithm and an optimizer (for example, gradient descent, Adam, SGD, etc.), to minimize the loss function value, so that the prediction result of the attribute prediction sub-model is closer to the true judgment result label.

[0137] The scheme of the present application explicitly extracts the prediction information of the gaze point trajectory prediction sub-model, and inputs it into the attribute prediction sub-model together with the original organic material molecular structure graph, so that the attribute prediction sub-model can learn more comprehensive information about the molecular structure graph, and the performance of the attribute prediction sub-model can be improved. Figure 1As input to the attribute prediction sub-model, the gaze point trajectory prediction sub-model can make full use of the "attention" or "reasoning path" of human experts in judging molecular attributes during the learning process. This mechanism effectively transfers the experience and knowledge of human experts to the attribute prediction sub-model in a computable form, thereby bridging the semantic gap that may exist when predicting attributes based solely on molecular structure graphs. By calculating the loss function between the actual judgment result and the true label, and optimizing the model parameters of the attribute prediction sub-model accordingly, the attribute prediction sub-model can continuously learn and improve, enhancing its prediction ability and making its internal decision-making mechanism more in line with the judgment logic of human experts.

[0138] Through the above technical solution, the attribute prediction sub-model can not only learn features from the organic material molecular structure graph itself, but also integrate the prediction information provided by the gaze point trajectory prediction sub-model, which simulates the focus points of human experts. This significantly enhances the learning efficiency and prediction accuracy of the attribute prediction sub-model. This combination allows the model to better understand the key areas or features in the molecular structure related to the target attribute, thereby improving the reliability and interpretability of the prediction results. In addition, by iteratively optimizing the model parameters based on the loss function, the attribute prediction sub-model can continuously adapt and learn, further enhancing its generalization ability and robustness, enabling it to make more accurate and expert-judgment-compliant attribute predictions when faced with new organic material molecules to be tested.

[0139] Specifically, step A3 includes:

[0140] A301. Obtain the molecular structure graph of the organic material molecule to be tested;

[0141] A302. Input the molecular structure graph into the gaze point trajectory prediction sub-model and extract the prediction information of the gaze point trajectory prediction sub-model for the gaze point trajectory;

[0142] A303. Input the molecular structure graph and the prediction information into the attribute prediction sub-model to obtain the prediction result of whether the organic material molecule to be tested has the target attribute.

[0143] Specifically, in step A301, the molecular structure graph of the organic material molecule to be tested can be obtained in various forms, such as through chemical structure database retrieval, user manual input, image recognition technology to extract from literature, etc. The molecular structure graph is the basic input information for subsequent prediction.

[0144] In step A302, when the molecular structure diagram is input into the gaze point trajectory prediction sub-model, the sub-model generates a series of data representing potential gaze regions or paths, i.e., prediction information, based on the knowledge obtained from its training. The prediction information can include the predicted gaze point trajectory itself, or the hidden layer parameters generated by the gaze point trajectory prediction sub-model during the prediction process, which can reflect the sub-model's deep understanding and feature extraction results of the molecular structure diagram.

[0145] Further, in step A303, the attribute prediction sub-model receives the original information of the molecular structure diagram and the prediction information obtained from the gaze point trajectory prediction sub-model. By combining these two types of information, the attribute prediction sub-model can more comprehensively and accurately understand the features of the molecular structure diagram and their association with the target attribute. Finally, the attribute prediction sub-model outputs a judgment result, such as a binary classification result (yes / no with the target attribute), indicating the likelihood of the target attribute of the organic material molecule under test.

[0146] The scheme of the present application first inputs the molecular structure diagram of the organic material molecule under test into the gaze point trajectory prediction sub-model to simulate the visual focus of human experts during the judgment process, thereby extracting prediction information related to the human cognitive process. Subsequently, the original information of the molecular structure diagram and the extracted prediction information are jointly input into the attribute prediction sub-model. This dual input mechanism enables the attribute prediction sub-model to not only utilize the features of the molecular structure diagram itself, but also combine the simulated human expert gaze point information, thereby incorporating the "explanatory" or "attention" mechanism of expert experience in the prediction process. As a result, the attribute prediction model can more comprehensively and deeply understand the complex relationship between the molecular structure and the target attribute, improving the accuracy and interpretability of the prediction.

[0147] Through the above technical scheme, in predicting whether the organic material molecule under test has the target attribute, the trained gaze point trajectory prediction sub-model and attribute prediction sub-model can be fully utilized. Specifically, by obtaining the molecular structure diagram of the organic material molecule under test and inputting it into the gaze point trajectory prediction sub-model to obtain prediction information, and then jointly inputting the molecular structure diagram and prediction information into the attribute prediction sub-model, accurate prediction of the molecular attribute can be achieved. This method not only improves the accuracy of prediction, but also introduces the gaze point trajectory prediction sub-model, making the prediction process have certain interpretability, which helps technical personnel understand the basis for the model's judgment, thereby enhancing the practicality and reliability of the model.

[0148] Reference Figure 2 The present application provides a molecular attribute prediction system for predicting whether an organic material molecule under test has a target attribute, which comprises a display 1, an eye tracker 2, and an upper computer 3.

[0149] The display 1 is used to display the molecular structure diagram of the organic material under the control of the host computer 3;

[0150] The eye tracker 2 is used to collect the fixation point trajectory of the technical personnel in the process of judging whether the organic material molecule has the target attribute through the molecular structure diagram of the organic material displayed by the display 1, and upload it to the host computer 3;

[0151] The host computer 3 is used to perform:

[0152] Obtain the judgment result of the technical personnel on whether the organic material molecule corresponding to the organic material molecular structure displayed by the display 1 has the target attribute, combine the fixation point trajectory uploaded by the eye tracker 2 and the corresponding organic material molecular structure diagram, and form a training data set (for details, refer to the previous step A1);

[0153] Train the attribute prediction model using the training data set; the attribute prediction model includes a fixation point trajectory prediction sub-model and an attribute prediction sub-model, the fixation point trajectory prediction sub-model is used to predict the fixation point trajectory according to the organic material molecular structure diagram, and the attribute prediction sub-model is used to predict whether the organic material molecule has the target attribute according to the organic material molecular structure diagram and the prediction information of the fixation point trajectory predicted by the fixation point trajectory prediction sub-model (for details, refer to the previous step A2);

[0154] Use the trained attribute prediction model to predict whether the to-be-tested organic material molecule has the target attribute (for details, refer to the previous step A3).

[0155] Wherein, the display 1 can be any display device that can clearly present images, such as liquid crystal display (LCD), organic light emitting diode display (OLED) or projector. Its main function is to provide a visual interface for technical personnel to observe and analyze the molecular structure diagram of organic material.

[0156] Wherein, the eye tracker 2 can be an eye tracking device based on infrared light reflection principle, or an eye tracking device based on video image processing. It is placed in a suitable position when the technical personnel observe the display 1, so as to accurately capture the eye movement data of the technical personnel.

[0157] Wherein, the technical personnel can directly input their judgment results (such as "yes" or "no") on the host computer 3 through keyboard, mouse or other input devices, or convert oral judgments into text data through voice recognition system.

[0158] Please refer to Figure 3A structural schematic diagram of an electronic device provided for an embodiment of the present application, the present application provides an electronic device, comprising: a processor 301 and a memory 302, the processor 301 and the memory 302 are interconnected and communicate with each other through a communication bus 303 and / or other forms of connection mechanism (not marked), the memory 302 stores a computer program executable by the processor 301, when the electronic device runs, the processor 301 executes the computer program to execute the molecular attribute prediction method in any optional implementation manner of the above-mentioned embodiments, to realize the following functions: obtaining the gaze point trajectory and the corresponding judgment result label of the process that the technical personnel judge whether the organic material molecule has the target attribute through the organic material molecular structure diagram, forming a training data set; training an attribute prediction model using the training data set; the attribute prediction model comprises a gaze point trajectory prediction sub-model and an attribute prediction sub-model, the gaze point trajectory prediction sub-model is used for predicting the gaze point trajectory according to the organic material molecular structure diagram, and the attribute prediction sub-model is used for predicting whether the organic material molecule has the target attribute according to the organic material molecular structure diagram and the prediction information of the gaze point trajectory predicted by the gaze point trajectory prediction sub-model; using the trained attribute prediction model, whether the target attribute of the organic material molecule to be tested has the target attribute is predicted.

[0159] The embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to perform the molecular attribute prediction method in any optional implementation manner of the above embodiment, so as to realize the following functions: obtaining a gaze point trajectory and a corresponding judgment result label in a process in which a technical personnel judges whether an organic material molecule has a target attribute through an organic material molecular structure diagram, and forming a training data set; training an attribute prediction model by using the training data set; the attribute prediction model comprises a gaze point trajectory prediction sub-model and an attribute prediction sub-model, the gaze point trajectory prediction sub-model is used for predicting the gaze point trajectory according to the organic material molecular structure diagram, and the attribute prediction sub-model is used for predicting whether the organic material molecule has the target attribute according to the organic material molecular structure diagram and prediction information of the gaze point trajectory predicted by the gaze point trajectory prediction sub-model; and predicting whether a to-be-tested organic material molecule has the target attribute by using the trained attribute prediction model. The computer readable storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0160] The above merely describes the embodiments of the present application but is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method of molecular property prediction for predicting whether a molecule of an organic material under test has a target property, characterized by, The method comprises the following steps: A1. Obtain the gaze point trajectory of a technician in the process of judging whether the organic material molecule has a target attribute through the organic material molecule structure diagram and the corresponding judgment result label, and form a training data set; A2. Train an attribute prediction model using the training data set; the attribute prediction model comprises a gaze point trajectory prediction sub-model and an attribute prediction sub-model, the gaze point trajectory prediction sub-model is used to predict the gaze point trajectory according to the organic material molecule structure diagram, and the attribute prediction sub-model is used to predict whether the organic material molecule has the target attribute according to the organic material molecule structure diagram and the prediction information of the gaze point trajectory predicted by the gaze point trajectory prediction sub-model; A3. Use the trained attribute prediction model to predict whether the to-be-tested organic material molecule has the target attribute; Step A1 comprises: A101. Obtain the true situation of whether the organic material molecule has a target attribute as the judgment result label of the organic material molecule structure diagram corresponding to the organic material molecule; A102. Obtain the gaze point trajectory of a plurality of technicians in the process of judging whether the organic material molecule has a target attribute through the organic material molecule structure diagram of the organic material molecule and the corresponding judgment result; A103. Remove the gaze point trajectory that does not match the judgment result label; A104. For the remaining gaze point trajectories, remove the gaze point trajectory whose number of deviated gaze points exceeds a preset number threshold; the deviated gaze point refers to a gaze point that does not fall on the organic material molecule structure diagram; A105. Cluster the remaining gaze point trajectories to obtain a plurality of gaze point trajectory clusters, and remove the gaze point trajectory that deviates greatly from the center trajectory of the gaze point trajectory cluster from each gaze point trajectory cluster; A106. Form a sample by combining each remaining gaze point trajectory and the corresponding judgment result label, and add the sample to the training data set; A107. Perform steps A101-A106 for a plurality of different organic material molecules to obtain a final training data set.

2. The method of predicting molecular properties according to claim 1, wherein, Step A104 comprises: Obtain the envelope of the organic material molecule structure diagram; Expand the envelope outward by a preset expansion distance; Count the number of gaze points of the gaze point trajectory that fall outside the area surrounded by the expanded envelope to obtain the number of deviated gaze points; If the number of deviated gaze points exceeds a preset number threshold, remove the gaze point trajectory.

3. The method of predicting molecular properties according to claim 1, wherein, Step A105 comprises: Cluster the remaining gaze point trajectories to obtain a plurality of gaze point trajectory clusters; For each gaze point trajectory cluster, obtain the center trajectory of the gaze point trajectory cluster according to the gaze point trajectory of the gaze point trajectory cluster; For each gaze point trajectory cluster, calculate the deviation degree between each gaze point trajectory of the gaze point trajectory cluster and the center trajectory; Remove the gaze point trajectory whose deviation degree is greater than a preset deviation degree threshold.

4. The method of claim 1, wherein, Step A2 comprises: A201. Training the gaze point trajectory predictor model according to the organic material molecular structure diagram and the corresponding gaze point trajectory in the training data set, so that the gaze point trajectory predictor model can predict the gaze point trajectory according to the organic material molecular structure diagram; A202. Freezing the gaze point trajectory predictor model; A203. Training the attribute prediction sub-model according to the organic material molecular structure diagram and the corresponding judgment result label in the training data set, in combination with the gaze point trajectory predictor model, so that the attribute prediction sub-model can predict whether the organic material molecule has the target attribute according to the organic material molecular structure diagram and the prediction information of the gaze point trajectory predicted by the gaze point trajectory predictor model.

5. The method of claim 4, wherein the molecular property is selected from the group consisting of: solubility, stability, bioavailability, and toxicity. Step A203 includes: inputting the organic material molecular structure diagram in the training data set into the gaze point trajectory predictor model, and extracting the prediction information of the gaze point trajectory predicted by the gaze point trajectory predictor model; the prediction information includes the gaze point trajectory predicted by the gaze point trajectory predictor model and / or the hidden layer parameters generated by the gaze point trajectory predictor model in the prediction process; inputting the organic material molecular structure diagram in the training data set and the extracted prediction information into the attribute prediction sub-model to obtain the actual judgment result output by the attribute prediction sub-model; calculating a loss function according to the actual judgment result and the corresponding judgment result label, and optimizing the model parameters of the attribute prediction sub-model based on the loss function.

6. The method of claim 1, wherein, Step A3 includes: A301. Obtain the molecular structure diagram of the organic material molecule to be tested; A302. Input the molecular structure diagram into the gaze point trajectory predictor model and extract the prediction information of the gaze point trajectory predicted by the gaze point trajectory predictor model; A303. Input the molecular structure diagram and the prediction information into the attribute prediction sub-model to obtain the prediction result of whether the organic material molecule to be tested has the target attribute.

7. A molecular property prediction system for predicting whether a molecule of an organic material under test has a target property based on the molecular property prediction method of any one of claims 1-6. The system includes a display, an eye tracker and a host computer; The display is used to display the organic material molecular structure diagram under the control of the host computer; The eye tracker is used to collect the gaze point trajectory of the technical personnel in the process of judging whether the organic material molecule has the target attribute through the organic material molecular structure diagram displayed by the display, and upload it to the host computer; The host computer is used to perform: obtain the judgment result of the technical personnel on whether the organic material molecule corresponding to the organic material molecular structure displayed by the display has the target attribute, combine the gaze point trajectory uploaded by the eye tracker and the corresponding organic material molecular structure diagram to form a training data set; train the attribute prediction model using the training data set; the attribute prediction model includes a gaze point trajectory predictor model and an attribute prediction sub-model, the gaze point trajectory predictor model is used to predict the gaze point trajectory according to the organic material molecular structure diagram, and the attribute prediction sub-model is used to predict whether the organic material molecule has the target attribute according to the organic material molecular structure diagram and the prediction information of the gaze point trajectory predicted by the gaze point trajectory predictor model; Using the trained attribute prediction model, predicting whether a molecule of an organic material to be tested has the target attribute.

8. An electronic device, comprising: A computer program product comprising a computer readable storage medium having computer readable program code embodied therewith, the computer readable program code adapted to be executed by a computer to implement the steps of the method of any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, A computer program product comprising a computer readable storage medium having computer readable program code embodied therewith, the computer readable program code adapted to be executed by a computer to implement the steps of the method of any one of claims 1-6.

Images