Device and method for measuring reliability of molecular structure prediction model

Abstract

The embodiment of the invention aims to provide a system, a computer program, a device and a method for measuring the reliability of a molecular structure prediction model. The method may comprise: a step of acquiring an image of a first molecular structure; acquiring a graph of the first molecular structure by using the molecular structure prediction model; a step of rendering the image of the first molecular structure on the basis of the map of the first molecular structure; and determining the reliability of the map of the first molecular structure on the basis of the image rendering result and the map of the first molecular structure.

Description

Technical Field

[0001] This disclosure relates to an apparatus and method for measuring the reliability of molecular structure prediction models. More specifically, this disclosure aims to provide convenience by providing the reliability of the results along with the predicted molecular structure provided by the molecular structure prediction model. Background Technology

[0002] A structural formula is a graphical representation of a chemical or molecular structure, expressing the arrangement of atoms in three-dimensional space. Structural formulas can also explicitly or implicitly represent the chemical bonds in a molecule. In particular, unlike molecular formulas, which can only use a limited set of symbols and have limited descriptive power, structural formulas provide geometric information about the molecular structure. For example, they can represent isomers with the same molecular formula but different atomic structures or arrangements.

[0003] In various documents, papers, and patents, structural formulas are often provided in the form of images. However, unlike text, images are difficult to retrieve, leading to the problem of finding documents containing the structural formula. Therefore, various methods for retrieving images of structural formulas are being developed. Models for extracting structural formulas from images are mainly used to create academic databases. If such academic databases contain erroneous data due to incorrect predictions, it becomes a fatal flaw for research. Therefore, there is a current need for a method that provides reliability information for the predicted structural formulas, distinguishes which predicted structural formulas can be judged as reliable, and stores this information in the database. Summary of the Invention

[0004] (The problem that the invention aims to solve)

[0005] This disclosure aims to provide a method and apparatus for predicting molecular structures using an image-based model, and to provide a reliability score when predicting molecular structures.

[0006] (The measures taken to solve the problem)

[0007] One embodiment of this disclosure provides a device and method for measuring the reliability of a molecular structure prediction model.

[0008] One embodiment of this disclosure aims to provide a system for measuring the reliability of a molecular structure prediction model. The system includes: a memory storing one or more instructions; and at least one processor executing the one or more instructions stored in the memory, the at least one processor executing the one or more instructions to perform the following operations: acquiring an image of a first molecular structure; acquiring a graph of the first molecular structure determined by the molecular structure prediction model; rendering an image of the first molecular structure based on the graph of the first molecular structure; and determining the reliability of the graph of the first molecular structure based on the image rendering result and the graph of the first molecular structure.

[0009] In one embodiment, the at least one processor may perform the following operations: based on a diagram of the first molecular structure, identify at least one of a first constituent element and a second constituent element; identify a first portion corresponding to the first constituent element from an image of the first molecular structure; identify a second portion corresponding to the second constituent element from an image of the first molecular structure; and perform image rendering by distinguishing the first portion and the second portion with different labels, wherein the first constituent element and the second constituent element may respectively include any one of a first atom, a second atom, a first chemical bond, and a second chemical bond.

[0010] In one embodiment, the molecular structure prediction model may include a first learning model, which may be a learning model trained to extract chemical table diagrams from molecular structure images as input.

[0011] In one embodiment, the at least one processor may take the image rendering result and a graph of the first molecular structure as input, and use a second learning model to output reliability.

[0012] In one embodiment, the second learning model may include: an image backbone model for extracting features of the image rendering result; a graph backbone model for extracting features of the graph of the first molecular structure; a feature connector for connecting the features of the image rendering result with the features of the graph of the first molecular structure; and a linear layer model that uses the output of the feature connector as input to determine reliability.

[0013] In one embodiment, the second learning model can be trained to compare the image rendering result with the first molecular structure. Figure 1 If the image rendering result is inconsistent with the graph of the first molecular structure, a first value is output; if the image rendering result is inconsistent with the graph of the first molecular structure, a second value is output.

[0014] In one embodiment, a map of a molecular structure with a reliability level above a predetermined level can be stored in a database.

[0015] One embodiment of this disclosure aims to provide a method for measuring the reliability of a molecular structure prediction model, which is executed by at least one processor. The method may include: acquiring an image of a first molecular structure; acquiring a graph of the first molecular structure using the molecular structure prediction model; rendering the image of the first molecular structure based on the graph of the first molecular structure; and determining the reliability of the graph of the first molecular structure based on the image rendering result and the graph of the first molecular structure.

[0016] In one embodiment, the step of rendering an image of the first molecular structure may include: identifying at least one of a first constituent element and a second constituent element based on the image of the first molecular structure; identifying a first portion corresponding to the first constituent element from the image of the first molecular structure; identifying a second portion corresponding to the second constituent element from the image of the first molecular structure; and rendering the image by distinguishing the first portion and the second portion with different markings, wherein the first constituent element and the second constituent element may each include any one of a first atom, a second atom, a first chemical bond, and a second chemical bond.

[0017] In one embodiment, the molecular structure prediction model may include a first learning model, which may be a learning model trained to extract chemical table diagrams from molecular structure images as input.

[0018] In one embodiment, the step of outputting the reliability of the graph of the first molecular structure may include: taking the image rendering result and the graph of the first molecular structure as input, and using a second learning model to output the reliability of the graph of the first molecular structure.

[0019] In one embodiment, the second learning model may include: an image backbone model for extracting features of the image rendering result; a graph backbone model for extracting features of the graph of the first molecular structure; a feature connector for connecting the features of the image rendering result with the features of the graph of the first molecular structure; and a linear layer model that uses the output of the feature connector as input to determine reliability.

[0020] In one embodiment, the second learning model can be trained to compare the image rendering result with the first molecular structure. Figure 1 If the image rendering result is inconsistent with the graph of the first molecular structure, a first value is output; if the image rendering result is inconsistent with the graph of the first molecular structure, a second value is output.

[0021] In one embodiment, graphs of molecular structures with a reliability value above a certain threshold can be stored in a database.

[0022] One embodiment of this disclosure relates to a program stored on a recording medium that causes a computer to execute the method involved in one embodiment of this disclosure.

[0023] One embodiment of this disclosure relates to a computer-readable recording medium containing a program that causes a computer to perform a method according to an embodiment of this disclosure.

[0024] One embodiment of this disclosure relates to a computer-readable recording medium that records a database used in one embodiment of this disclosure.

[0025] (The effect of the invention)

[0026] According to one embodiment of this disclosure, the reliability of the results of the prediction model can be provided along with information on the accuracy of the results of the prediction model, thereby helping the user to determine which information should be stored in the database. Attached Figure Description

[0027] Figure 1 This is a diagram illustrating a method for extracting molecular structural formulas from an image using a first learning model according to an embodiment of the present disclosure.

[0028] Figure 2 This is a flowchart of a method for providing reliable result values ​​of a first learning model using a second learning model according to an embodiment of the present disclosure.

[0029] Figure 3 This is a flowchart illustrating a method for measuring the reliability of a molecular structure prediction model according to an embodiment of this disclosure.

[0030] Figure 4a This is a diagram illustrating a method for measuring the reliability of a molecular structure prediction model when it makes an incorrect prediction, according to an embodiment of this disclosure.

[0031] Figure 4b This is a diagram illustrating a method for measuring the reliability of a molecular structure prediction model when it makes a correct prediction, according to an embodiment of this disclosure.

[0032] Figure 5 This is an example diagram of a second learning model involved in an embodiment of this disclosure.

[0033] Figure 6 This is a block diagram of a reliability measurement device for a molecular structure prediction model according to an embodiment of the present disclosure. Detailed Implementation

[0034] To clarify the technical concept of this disclosure, embodiments of this disclosure will be described in detail with reference to the accompanying drawings. In describing this disclosure, if a detailed description of a related well-known function or component is deemed to obscure the main point of this disclosure, such detailed description will be omitted. For components that have substantially the same function in the drawings, even if shown in different drawings, the same reference numerals and symbols will be used as much as possible. For ease of explanation, apparatus and methods will be described together where necessary. Each operation of this disclosure does not necessarily have to be performed in the order described; they may be performed in parallel, selectively, or individually.

[0035] The terminology used in the embodiments of this disclosure has been selected as widely used and common terms as possible, taking into account the functionality of this disclosure. However, this may change due to the intentions of those skilled in the art, precedents, the emergence of new technologies, etc. Furthermore, in specific cases, terms arbitrarily chosen by the applicant may also exist; in such cases, their meanings will be described in detail in the description section of the corresponding embodiments. Therefore, the terminology used in this specification should not be based solely on the name of the term, but rather on its meaning and its content throughout this disclosure.

[0036] Throughout this disclosure, unless the context clearly indicates otherwise, singular expressions may include plural expressions. Terms such as “comprising” or “having” are intended to indicate the presence of features, numbers, steps, actions, components, parts, or combinations thereof, and should not be construed as pre-excluding the possibility of the presence or addition of more than one other feature, number, step, action, component, part, or combination thereof. That is, throughout this disclosure, when a part “comprising” a component, unless otherwise stated, this means not excluding other components, but rather that other components may be further included.

[0037] Expressions such as "at least one" modify the entire list of components, rather than modifying each component in the list individually. For example, "at least one of A, B, and C" and "at least one of A, B, or C" mean only A, only B, only C, both A and B, both B and C, both A and C, all A, B, and C, or combinations thereof.

[0038] In addition, the terms “…part”, “…module”, etc., used in this disclosure refer to a unit that performs at least one function or action, which can be implemented by hardware, software, or a combination of hardware and software.

[0039] Throughout this disclosure, when a part is "connected" to other parts, this includes not only "direct connection" but also "electrical connection" through the intervention of other components. Furthermore, when a part "includes" a component, unless otherwise stated, this does not exclude other components but may further include them.

[0040] The phrase “configured to” as used throughout this disclosure may be used interchangeably with, for example, “suitable for”, “having the capacity to”, “designed to”, “adapted to”, “made to”, or “capable of”, depending on the context. The term “configured to” does not necessarily mean “specifically designed to” in terms of hardware. Rather, in some cases, the phrase “system configured to” may mean that the system can “perform” together with other devices or components. For example, “processor configured to perform A, B, and C” may refer to a dedicated processor (e.g., an embedded processor) for performing the respective operations, or a generic-purpose processor (e.g., a CPU or application processor) capable of performing the respective operations by executing one or more software programs stored in memory.

[0041] The artificial intelligence-related functions disclosed herein are implemented by a processor and memory. The processor may consist of one or more processors. In this case, the one or more processors may be general-purpose processors such as CPUs, APs, and DSPs (Digital Signal Processors); graphics-specific processors such as GPUs and VPUs (Vision Processing Units); or AI-specific processors such as NPUs. The one or more processors process input data under control according to predefined operating rules or AI models stored in memory. Alternatively, when the one or more processors are AI-specific processors, the AI-specific processors may be designed with hardware architectures dedicated to processing specific AI models.

[0042] The predefined operating rules or artificial intelligence models are characterized by being generated through training and learning. Here, "generated through training" means that predefined operating rules or artificial intelligence models, set to perform desired characteristics (or, objectives), are generated by training a base artificial intelligence model using a learning algorithm and multiple training data sets. This training can be performed on the device itself performing the artificial intelligence involved in this disclosure, or via a separate server and / or system. Examples of learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0043] Throughout this disclosure, the device may include, but is not limited to, servers, smartphones, tablet PCs, PCs, TVs, smart TVs, mobile phones, PDAs (personal digital assistants), speakers, laptops, media players, microservers, e-book object recognition devices, digital broadcasting object recognition devices, kiosks, MP3 players, digital cameras, robot vacuum cleaners, home appliances, other mobile or non-mobile computing devices, watches, glasses, hairbands, and rings with communication and data processing capabilities. In one embodiment, the device may run a system-related web-based or module-based application. For example, a device for measuring the reliability of a molecular structure prediction model may refer to a server on which a web-based application related to the system for measuring the reliability of the molecular structure prediction model may run. That is, the server may provide web services or software for measuring the reliability of the molecular structure prediction model.

[0044] The objective of one embodiment of this disclosure is to predict molecular structural formulas from images of molecular structures and to provide reliability for the predicted molecular structural formulas. However, this disclosure is not limited to embodiments of molecular structures and can certainly be applied to various technical fields of extracting image information from non-textual images.

[0045] Figure 1 This is a diagram illustrating a method for extracting molecular structural formulas from an image using a first learning model according to an embodiment of the present disclosure.

[0046] Reference Figure 1The apparatus, including a molecular structure prediction model, can identify an image 120 of a molecular structure from a file 110 and analyze the image 120 to extract the molecular structure corresponding to the image. In one embodiment, an apparatus may include a molecular structure prediction model and a reliability measurement model, or the molecular structure prediction model and the reliability measurement model may be included in different apparatuses, so that each apparatus can use the corresponding model to calculate the output value.

[0047] In one embodiment, the apparatus including a molecular structure prediction model can extract an image 120 of a molecular structure from a document 110. For example, the apparatus including a molecular structure prediction model can extract the image 120 of a molecular structure from the document 110 using panoptic segmentation technology. Additionally, the apparatus including a molecular structure prediction model can identify the atoms and chemical bonds contained in the image 120 of the molecular structure. In one embodiment, the apparatus including a molecular structure prediction model can analyze the types and positions of atoms, the types and positions of chemical bonds, etc., contained in the image 120 of the molecular structure. For example, it can identify vertices, atoms, superatoms, etc., implicitly representing carbon, and their positions, as well as the types of chemical bonds between them (e.g., single / double / triple bonds, up / down bonds, etc.). When analyzing the types of atoms or superatoms, OCR (Optical Character Recognition) technology can be used. In one embodiment, the apparatus including the molecular structure prediction model can determine a linear representation of the structural formula 130 corresponding to the image 120 of the molecular structure based on the identified types and positions of atoms, types and positions of chemical bonds, etc. The linear representation may include, but is not limited to, SMILES (Simplified Molecular Input Line Entry System), InChI (International Chemical Identifier), WLN (Wiswesser line notation), ROSDAL (Representation of Organic Structure Descriptions Arranged Linearly), SLN (SYBYL Line Notation), etc. Further, the apparatus including the molecular structure prediction model can determine a graph corresponding to the linear representation of the structural formula 130. The graph corresponding to the linear representation of the structural formula 130 includes a chemical table file (CT file) graph, for example, a Mol file graph.

[0048] When providing a linear representation of such a prediction (structural formula 130 or diagram), the conventional approach is to generate a chemical structural formula from the prediction result and measure the reliability if the generated structural formula is a structural formula that can actually bond. However, this method does not consider the actual image and outputs a high reliability score even in cases of incorrect prediction, leading to a high possibility of erroneous data accumulation. Therefore, a method is needed to provide a more accurate reliability score for the prediction model, and related content will refer to... Figures 2 to 5 This will be explained in more detail later.

[0049] Figure 2 This is a flowchart of a method for providing reliable result values ​​of a first learning model using a second learning model according to an embodiment of the present disclosure.

[0050] Reference Figure 2 Using the molecular structure prediction model 210, a graph 203 of the predicted first molecular structure can be obtained, which corresponds to the image 201 of the first molecular structure. The method for obtaining the graph 203 of the predicted first molecular structure can utilize a reference... Figure 1 The method of explanation.

[0051] In one embodiment, image rendering can be performed based on a graph 203 of a first molecular structure predicted by a molecular structure prediction model 210. Image rendering can be performed by identifying multiple constituent elements based on the graph of the first molecular structure, identifying portions corresponding to each of the multiple constituent elements from the image 201 of the first molecular structure, and then using different markers to distinguish and label the different constituent elements on the image 201 of the first molecular structure, which serves as the original image. Further, in order to identify portions corresponding to each of the multiple constituent elements from the image 201 of the first molecular structure, the positions of each of the multiple constituent elements can also be identified. The multiple constituent elements may include a first atom, a second atom, a third atom, a first chemical bond, a second chemical bond, a third chemical bond, etc. For example, referring to… Figure 2 To illustrate with an example, the reliability measurement device for the molecular structure prediction model can identify bromine (Br), four nitrogen atoms (N), eight vertices representing carbon (C), seven single bonds, five double bonds, and one triple bond based on Figure 203 of the first molecular structure. Elements can be represented by circles, and chemical bonds by line segments, but different colors are used for different elements or different chemical bonds. Based on this, the following can be obtained... Figure 2 The image rendering result is 205.

[0052] In one embodiment, the image rendering result 205 and the predicted first molecular structure graph 203 are input into the reliability model 230, and a reliability score 207 is obtained based on the similarity between the two input information. The more similar the input information of the two input values, the higher the reliability score 207 is obtained; the less similar they are, the lower the reliability score 207 is obtained.

[0053] Figure 3 This is a flowchart illustrating a method for measuring the reliability of a molecular structure prediction model according to an embodiment of this disclosure.

[0054] Reference Figure 3 In operation 310, the reliability measurement device of the molecular structure prediction model can acquire an image of the first molecular structure. In one embodiment, the reliability measurement device of the molecular structure prediction model can directly extract the image of the first molecular structure from a file to acquire the image of the first molecular structure, or it can acquire the extracted image of the first molecular structure from another device. That is, the reliability measurement device of the molecular structure prediction model or an external device can extract the image of the first molecular structure using artificial intelligence (AI) algorithms or predefined operating rules.

[0055] In operation 330, the reliability measurement device for the molecular structure prediction model can acquire a graph of the first molecular structure determined by the molecular structure prediction model. That is, the graph of the first molecular structure can be a graph corresponding to the image of the first molecular structure predicted by the molecular structure prediction model. In one embodiment, the molecular structure prediction model may include a first learning model trained to output a chemical table file (CT file) graph as input to an image of the molecular structure formula. The CT file graph may contain information about each atom of the molecule, the xyz coordinates of the atom, the chemical bond information between the atoms, etc. Similar to operation 310, the reliability measurement device for the molecular structure prediction model can directly acquire the graph of the first molecular structure determined by the molecular structure prediction model, or it can acquire it by receiving the graph of the first molecular structure determined by the molecular structure prediction model from an external device.

[0056] In operation 350, the reliability measurement device for the molecular structure prediction model can render an image of the first molecular structure based on a diagram of the first molecular structure. In one embodiment, the reliability measurement device for the molecular structure prediction model can identify multiple constituent elements based on the diagram of the first molecular structure, identify the parts corresponding to each of the multiple constituent elements from the image of the first molecular structure, and then distinguish the different constituent elements using different markers for image rendering. Here, the multiple constituent elements may include atoms, chemical bonds between atoms, etc. Distinguishing them using different markers may include distinguishing them using different colors or different shapes, but is not limited to these, and may also include various other marker forms that can distinguish them from each other.

[0057] For example, when the first constituent element is carbon and the second constituent element is nitrogen, the reliability measurement device of the molecular structure prediction model can identify carbon as the first constituent element and its position coordinates from the image of the first molecular structure, and identify nitrogen as the second constituent element and its position coordinates. Based on each position coordinate, it can identify the first part corresponding to the first constituent element and the second part corresponding to the second constituent element from the image of the first molecular structure, and then distinguish them by using different markers such as yellow for the first part and red for the second part, and perform image rendering.

[0058] For example, when the first constituent element is carbon, the second constituent element is nitrogen, the third constituent element is a single bond, and the fourth constituent element is a double bond, the reliability measurement device of the molecular structure prediction model can identify the carbon as the first constituent element and its position, the nitrogen as the second constituent element and its position, the single bond as the third constituent element and the double bond as the fourth constituent element from the image of the first molecular structure. Furthermore, based on the positions of each constituent element, the reliability measurement device can identify the first part corresponding to the first constituent element, the second part corresponding to the second constituent element, the third part corresponding to the third constituent element, and the fourth part corresponding to the fourth constituent element from the image of the first molecular structure. The first and second parts corresponding to the elements are marked with circles, but different colors are used to easily represent different elements; the third and fourth parts are marked with bars, but different colors are used to easily represent different chemical bonds, thereby enabling image rendering.

[0059] In operation 370, the reliability measurement device of the molecular structure prediction model can determine the reliability of the graph of the first molecular structure based on the image rendering result and the graph of the first molecular structure. In one embodiment, the reliability of the graph of the first molecular structure can be determined using a second learning model. The second learning model can be a learning model that takes the image rendering result and the graph of the first molecular structure as input and outputs a reliability value (e.g., a reliability score). In one embodiment, the learning model may include: an image backbone network model for extracting features from the image rendering result; a graph backbone network model for extracting features from the graph of the first molecular structure; a feature connection unit for connecting the extracted features from the image rendering result and the extracted features from the graph of the first molecular structure; and an extraction network model that extracts reliability by taking the output of the feature connection unit as input. Related content will be referred to Figure 5 This will be explained in more detail later.

[0060] In one embodiment, the second learning model can be trained to adapt the image rendering result to the first molecular structure. Figure 1 The first value is output when the image rendering result is consistent with the first molecular structure, and the second value is output when the image rendering result is inconsistent with the first molecular structure. For example, the second learning model can be trained during the learning phase to output a first value when the image rendering result is inconsistent with the first molecular structure. Figure 1 The system outputs 1 when the prediction is correct and 0 when it is incorrect. Therefore, in actual use after training, if the molecular structure prediction model correctly predicts the result, a value similar to 1 is extracted; if the prediction is incorrect, a value similar to 0 is output. The reliability measurement device for the molecular structure prediction model can determine the reliability of the graph of the first molecular structure based on the values ​​extracted by the second learning model. For example, reliability can be determined as a reliability value (or, a score).

[0061] According to one embodiment, the reliability measurement device for the molecular structure prediction model provides a predicted graph of the first molecular structure along with its reliability, thereby providing the accuracy of the predicted graph and aiding in determining whether to use it. Accordingly, the user or their device can determine to store the graph in the database only if the reliability is above a predetermined level (or value), thus ensuring that accurate data can be used in research and other applications.

[0062] Figure 4a This is a diagram illustrating a method for measuring the reliability of a molecular structure prediction model when it makes an incorrect prediction, according to an embodiment of this disclosure.

[0063] Reference Figure 4aThe molecular structure prediction model can predict the first molecular structure based on image 410 of the first molecular structure. For example, the first molecular structure may be a structure in which single and double bonds coexist. However, unlike humans, when a machine uses image 410 of the first molecular structure to predict the first molecular structure, it cannot be expected to achieve 100% accuracy. Therefore, a process is needed to verify whether the first molecular structure predicted based on image 410 of the first molecular structure is correctly predicted.

[0064] In one embodiment, the reliability measurement device for the molecular structure prediction model can acquire an image 410 of the first molecular structure and a diagram of the first molecular structure predicted by the first learning model. Furthermore, based on the diagram of the predicted first molecular structure, the reliability measurement device can identify the types of elements, their positions, the types of chemical bonds, and their positions, and render this information onto the image 410 of the first molecular structure. For example, if the first chemical bond 420 is identified as a double bond in the diagram of the first molecular structure, the reliability measurement device can use different colors to label the chemical bonds identified as double bonds and those identified as single bonds containing the first chemical bond 440 on the image 410 of the first molecular structure to generate an image rendering result 430. That is, the first chemical bond 420 is actually a single bond, but in the image rendering result, the first chemical bond 420 is labeled as having a double bond.

[0065] In one embodiment, the reliability measurement device for the molecular structure prediction model can determine the reliability of the first molecular structure image based on the image rendering result. For example, the reliability measurement device can identify the elemental and chemical bond portions by segmenting the image 410 of the first molecular structure, and identify whether they are the same or different elements, and whether they are the same or different chemical bonds. When the segmentation result of the image 410 of the first molecular structure is similar to the image rendering result 430, the reliability measurement device for the molecular structure prediction model can determine that the reliability is high; if they are not similar, the reliability can be determined to be low. Figure 4a In the example, in image rendering result 430, the first chemical bond 440 is identified as a double bond, but in image 410 of the first molecular structure, the first chemical bond 420 is identified as a single bond. Therefore, the reliability of the first molecular structure image can be determined to be low. Alternatively, the reliability score of the first molecular structure image is determined to be low.

[0066] Alternatively, in one embodiment, the reliability measurement device for the molecular structure prediction model can simultaneously utilize the image rendering results and the predicted graph of the first molecular structure to determine the reliability of the graph. According to one embodiment, the graph of the molecular structure is also used for reliability determination, thereby improving the accuracy of the determined reliability. Related content will be referred to... Figure 5This will be explained in more detail later.

[0067] Figure 4b This is a diagram illustrating a method for measuring the reliability of a molecular structure prediction model when it makes a correct prediction, according to an embodiment of this disclosure.

[0068] Reference Figure 4b The molecular structure prediction model can predict the second molecular structure based on image 450 of the second molecular structure. However, compared with the aforementioned... Figure 4a Similarly, as explained in the text, when a machine uses image 450 of the second molecular structure to predict the second molecular structure, unlike humans, 100% accuracy cannot be expected. Therefore, a process is needed to verify whether the second molecular structure predicted based on image 450 of the second molecular structure is correctly predicted.

[0069] In one embodiment, the reliability measurement device for the molecular structure prediction model can acquire an image 450 of the second molecular structure and a diagram of the second molecular structure predicted by the first learning model. Furthermore, based on the diagram of the predicted second molecular structure, the reliability measurement device can identify the types of elements, their positions, the types of chemical bonds, and the positions of chemical bonds, and render these onto the image 450 of the second molecular structure. For example, if in the diagram of the second molecular structure, chemical bonds other than the second chemical bond 460 and the third chemical bond 470 are labeled as single bonds, the second chemical bond 460 is labeled as a chemical bond extending forward of the paper plane, and the third chemical bond 470 is labeled as a chemical bond extending backward of the paper plane, the reliability measurement device can use different colors to label the second chemical bond 490, the third chemical bond 495, and the remaining chemical bonds on the image 450 of the second molecular structure to generate an image rendering result 480. Further, the reliability measurement device can also label the positions of the second and third chemical bonds to generate the image rendering result 480.

[0070] In one embodiment, the reliability measurement device for the molecular structure prediction model can determine the reliability of the second molecular structure image based on the image rendering result 480. For example, the device can identify the elemental and chemical bond portions by segmenting the image 450 of the second molecular structure, and determine whether they are the same or different elements, and whether they are the same or different chemical bonds. When the segmentation result of the image 450 of the second molecular structure is similar to the image rendering result 480, the device can determine that the reliability is high; if they are not similar, it can determine that the reliability is low. Figure 4bIn the example, since image rendering result 480 is similar to image 450 of the second molecular structure, the reliability of the image of the second molecular structure can be determined to be high. Alternatively, the reliability score of the image of the second molecular structure can be determined to be high.

[0071] Alternatively, in one embodiment, the reliability measurement device for the molecular structure prediction model can determine the reliability of the first molecular structure graph based on the image rendering results and the graph of the predicted first molecular structure. Related content will be referred to... Figure 5 This will be described in more detail later.

[0072] Figure 5 This is an example diagram of a second learning model involved in an embodiment of this disclosure.

[0073] Reference Figure 5 The second learning model 530 for measuring reliability may include: an image backbone model 540 for extracting features from the image rendering result 510; a graph backbone model 550 for extracting features from the graph 520 of the molecular structure; a feature connection unit 560 for connecting the extracted features of the image rendering result 510 and the extracted features of the graph 520 of the molecular structure; and a network model 570 that determines reliability using the output of the feature connection unit 560 as input. The network model 570 for determining reliability may be a model that classifies the connected features through linear layers.

[0074] In one embodiment, the second learning model 530 may be an artificial intelligence model that extracts a reliability value 580 by taking the image rendering result 510 and the molecular structure graph 520 as inputs. In this case, the image rendering result 510 becomes the input to the image backbone model 540, and the molecular structure graph 520 becomes the input to the graph backbone model 550, thus becoming inputs to different network models. The image backbone model 540 can extract features from the image rendering result 510, and the graph backbone model 550 can extract features from the molecular structure graph 520. Feature connection units 560 connect the features extracted from each backbone model, and the network model 570 used to determine reliability can determine a reliability value based on the extracted features. For example, a higher reliability outputs a higher value, and a lower reliability outputs a lower value.

[0075] Figure 6 This is a block diagram of a reliability measurement device for a molecular structure prediction model according to an embodiment of the present disclosure.

[0076] Reference Figure 6 The reliability measurement device 600 for the molecular structure prediction model may include a transceiver unit 610, a memory 620, a database 630, and a processor 640. However, Figure 6The components shown are not all the necessary components of the reliability measurement device 600 for the molecular structure prediction model. The reliability measurement device 600 for the molecular structure prediction model can be composed of components that are comparable to... Figure 6 The components shown can be implemented with more components, and can also be compared to Figure 6 The components shown are implemented with fewer components. Moreover, the transceiver unit 610, memory 620, and processor 640 can also be implemented with a single chip.

[0077] In one embodiment, the transceiver unit 610 can communicate with a terminal or other electronic device that is connected via a wired or wireless connection to the reliability measurement device 600 of the molecular structure prediction model. For example, the transceiver unit 610 can acquire images of the first molecular structure, diagrams of the first molecular structure determined by the molecular structure prediction model, etc., from other electronic devices.

[0078] The memory 620 can be installed and stored with programs such as applications and various types of data such as files. The processor 640 can access and utilize the data stored in the memory 620, or can store new data in the memory 620. Furthermore, the memory 620 can store more than one instruction. The processor 640 can execute more than one instruction stored in the memory.

[0079] The processor 640 controls the overall operation of the reliability measurement device 600 for the molecular structure prediction model and may include at least one processor such as a CPU or GPU. The processor 640 can control other components included in the reliability measurement device 600 to perform actions for making the reliability measurement device 600 for the molecular structure prediction model run. For example, the processor 640 may acquire an image of a first molecular structure and acquire a graph of the first molecular structure determined by the molecular structure prediction model, then render the image of the first molecular structure based on the graph of the first molecular structure, and further determine the reliability of the graph of the first molecular structure based on the image rendering result and the graph of the first molecular structure.

[0080] Database 630 can store various training data used to train the learning model. Furthermore, database 630 can store material information, phase information, simulation result information, etc., and in various embodiments, it can also store computational data output by the learning model. Figure 6 Although the illustration shows a molecular structure prediction model reliability measurement device 600 including a database 630, the database 630 can also be located externally. In this case, the database 630 can be connected to the molecular structure prediction model reliability measurement device 600 via wired or wireless connection.

[0081] Alternatively, the learning model can be implemented externally to the reliability measurement device 600 of the molecular structure prediction model (e.g., cloud-based) or included internally within the reliability measurement device 600 of the molecular structure prediction model.

[0082] An embodiment of this disclosure can also be implemented in the form of a recording medium containing computer-executable instructions, which may be computer-executable program modules. A computer-readable medium can be any available medium accessible by a computer, including all volatile and non-volatile media, removable and non-removable media. Furthermore, a computer-readable medium can include all computer storage media and communication media. Computer storage media includes all volatile and non-volatile, removable and non-removable media implemented by any method or technique for storing information such as computer-readable instructions, data structures, program modules, or other data. Communication media typically contain computer-readable instructions, data structures, or program modules, and include any information transmission medium.

[0083] The foregoing description of this disclosure is for illustrative purposes only, and those skilled in the art will understand that it can be readily modified into other specific forms without altering the technical concept or essential features of the invention. Therefore, it should be understood that the above embodiments are exemplary in all respects and not restrictive. For example, components described as a single form may be implemented in a distributed form, and similarly, components described as distributed may be implemented in a combined form.

[0084] The scope of this disclosure shall be determined by the appended claims, rather than by the specific description above, and shall be interpreted as including all changes or modifications derived from the meaning, scope and equivalents of the claims within the scope of this disclosure.

Claims

1. A system for measuring the reliability of molecular structure prediction models, characterized in that, have: A memory that stores more than one instruction; and At least one processor that executes the one or more instructions stored in the memory. The at least one processor executes the one or more instructions to perform the following operations: Obtain an image of the first molecular structure; Obtain a diagram of the first molecular structure determined by the molecular structure prediction model; Based on the diagram of the first molecular structure, the image of the first molecular structure is rendered. as well as Based on the image rendering result and the graph of the first molecular structure, the reliability of the graph of the first molecular structure is determined.

2. The system according to claim 1, wherein, The at least one processor performs the following operations: Based on the diagram of the first molecular structure, at least one of the first constituent element and the second constituent element is identified; Identify the first part corresponding to the first constituent element from the image of the first molecular structure; Identify the second part corresponding to the second constituent element from the image of the first molecular structure; as well as The image rendering is performed by using different markers to distinguish between the first part and the second part. The first constituent element and the second constituent element respectively include any one of the first atom, the second atom, the first chemical bond, and the second chemical bond.

3. The system according to claim 1, wherein, The molecular structure prediction model includes a first learning model. The first learning model is a learning model trained to extract chemical table images from molecular structure images as input.

4. The system according to claim 1, wherein, The at least one processor takes the image rendering result and the graph of the first molecular structure as input, and outputs reliability using a second learning model.

5. The system according to claim 4, wherein, The second learning model includes: An image backbone model is used to extract features from the rendered image. A graph backbone model is used to extract features of the graph of the first molecular structure; Feature connection portion, which is used to connect features of the image rendering result with features of the graph of the first molecular structure; and A linear layer model that uses the output of the feature connection as input to determine reliability.

6. The system according to claim 4, wherein, The second learning model is trained to output a first value when the image rendering result is consistent with the graph of the first molecular structure, and to output a second value when the image rendering result is inconsistent with the graph of the first molecular structure.

7. The system according to claim 1, wherein, Maps of molecular structures with a reliability level exceeding a predetermined threshold are stored in a database.

8. A method for measuring the reliability of a molecular structure prediction model, executed by at least one processor, the reliability measurement method comprising: Steps for obtaining an image of the first molecular structure; The steps of obtaining a graph of the first molecular structure using the molecular structure prediction model; The step of rendering an image of the first molecular structure based on the diagram of the first molecular structure; as well as The step of determining the reliability of the graph of the first molecular structure based on the image rendering result and the graph of the first molecular structure.

9. The reliability measurement method according to claim 8, wherein, The steps for rendering the image of the first molecular structure include: Based on the diagram of the first molecular structure, the step of identifying at least one of the first constituent element and the second constituent element; The step of identifying the first part corresponding to the first constituent element from the image of the first molecular structure; The steps of identifying the second part corresponding to the second constituent element from the image of the first molecular structure; and The step of rendering the image by using different markers to distinguish the first part and the second part. The first constituent element and the second constituent element respectively include any one of the first atom, the second atom, the first chemical bond, and the second chemical bond.

10. The reliability measurement method according to claim 8, wherein, The molecular structure prediction model includes a first learning model. The first learning model is a learning model trained to extract chemical table images from molecular structure images as input.

11. The reliability measurement method according to claim 8, wherein, The steps for ensuring the reliability of outputting the graph of the first molecular structure include: The steps involve taking the image rendering result and the graph of the first molecular structure as input, and using the second learning model to output the reliability of the graph of the first molecular structure.

12. The reliability measurement method according to claim 11, wherein, The second learning model includes: An image backbone model is used to extract features from the rendered image. A graph backbone model is used to extract features of the graph of the first molecular structure; Feature connection portion, which is used to connect the resulting features of the image rendering with the features of the graph of the first molecular structure; and A linear layer model that uses the output of the feature connection as input to determine reliability.

13. The reliability measurement method according to claim 11, wherein, The second learning model is trained to output a first value when the image rendering result is consistent with the graph of the first molecular structure, and to output a second value when the image rendering result is inconsistent with the graph of the first molecular structure.

14. The reliability measurement method according to claim 8, wherein, Maps of molecular structures with a reliability value above a certain threshold are stored in a database.

15. A program stored in a computer-readable recording medium that causes a computer to perform the reliability measurement method according to any one of claims 8 to 14.

Images