Data screening method, computing device, storage medium and program product

Abstract

Embodiments of the invention provide a data screening method, a computing device, a storage medium and a program product. The data screening method comprises the steps of obtaining at least one item molecule attribute information and a verified molecule set corresponding to a target item; determining a molecule set according to the item molecule attributes and the verified molecule set; then acquiring a set access frequency and a molecular set score corresponding to each molecular set, and acquiring a molecular set weight corresponding to each molecular set according to the set access frequency and the molecular set score; determining a target sample molecule set according to the molecule set weight of each molecule set, and generating a project scoring model for the target project based on the target sample molecule set; and determining a target molecule in a molecule database according to the project scoring model. The molecules for training the scoring model are determined by obtaining the weight corresponding to each molecule set, so that the problem of poor scoring model effect caused by excessive invalid molecules when the scores for training the molecules are obtained can be avoided.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of computer technology, and in particular to a data filtering method. Background Technology

[0002] With the development of computational chemistry and bioinformatics technologies, and given that drugs are composed of multiple molecules, virtual screening methods—which involve selecting molecules from a large pool of molecules to construct drugs—are increasingly widely used in drug development. These methods can significantly improve the efficiency of new drug discovery and reduce the time and cost of experimental screening. However, this also brings some new challenges, requiring continuous optimization and improvement of existing techniques.

[0003] Currently, using pre-trained predictive models to screen large-scale molecular libraries is computationally expensive because the computational cost increases superlinearly with the size of the molecular library. This results in significant computational resources and time consumption during large-scale virtual screening. Furthermore, inherent model errors reduce the accuracy of the screening. Therefore, a more efficient and accurate data screening method is needed to address these shortcomings. Summary of the Invention

[0004] In view of the above, this specification provides a data filtering method, a data filtering method applied to cloud devices, and also relates to a data filtering device, a computing device, a computer-readable storage medium, and a computer program product, in order to solve the technical defects existing in the prior art.

[0005] According to a first aspect of the embodiments of this specification, a data filtering method is provided, comprising:

[0006] Obtain at least one project's molecular attribute information and the set of verified molecules corresponding to the target project;

[0007] Based on the molecular attribute information of each project and the verified molecular set, at least one molecular set is determined in the molecular database, wherein the molecular set is a set of molecules that have not been verified for the target project;

[0008] Obtain the set access count and set score for each set, and obtain the set weight for each set based on the set access count and set score. The set access count is the number of molecules that are checked in the process of determining the target sample set, and the set score is the number of target feature molecules that are hit by the corresponding set during the process of determining the target sample set. The target feature molecules are molecules related to the target item.

[0009] Based on each molecular set and its corresponding molecular set weight, a target sample molecular set is determined, and a project scoring model for the target project is generated based on the target sample molecular set.

[0010] At least one target molecule is identified in the molecular database based on the project scoring model.

[0011] According to a second aspect of the embodiments of this specification, a data filtering method is provided, applied to a cloud device, comprising:

[0012] The device receives a data filtering instruction sent by a terminal device, wherein the data filtering instruction includes at least one project molecular attribute information, a target project, and a set of verified molecules corresponding to the target project;

[0013] Based on the molecular attribute information of each project and the verified molecular set, at least one molecular set is determined in the molecular database, wherein the molecular set is a set of molecules that have not been verified for the target project;

[0014] Obtain the set access count and set score for each set, and obtain the set weight for each set based on the set access count and set score. The set access count is the number of molecules that are checked in the process of determining the target sample set, and the set score is the number of target feature molecules that are hit by the corresponding set during the process of determining the target sample set. The target feature molecules are molecules related to the target item.

[0015] Based on each molecular set and its corresponding molecular set weight, a target sample molecular set is determined, and a project scoring model for the target project is generated based on the target sample molecular set.

[0016] Based on the project scoring model, at least one target molecule is identified in the molecular database, and each target molecule is sent to the terminal device.

[0017] According to a third aspect of the embodiments of this specification, a computing device is provided, comprising:

[0018] Memory and processor;

[0019] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above-described data filtering method.

[0020] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the data filtering method described above.

[0021] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, including a computer program / instructions that, when executed by a processor, implement the steps of the data filtering method described above.

[0022] One embodiment of this specification implements the following: acquiring molecular attribute information of at least one project and a set of verified molecules corresponding to the target project; determining at least one molecular set in a molecular database based on the molecular attribute information of each project and the set of verified molecules, wherein the molecular set is a set of molecules that have not been verified for the target project; acquiring the set access count and molecular set score corresponding to each molecular set, and obtaining the molecular set weight corresponding to each molecular set based on the set access count and molecular set score, wherein the set access count is the number of molecules verified in the process of determining the target sample molecular set, and the molecular set score is the number of target feature molecules hit during the process of determining the target sample molecular set, and the target feature molecules are molecules related to the target project; determining the target sample molecular set based on each molecular set and the molecular set weight corresponding to each molecular set, and generating a project scoring model for the target project based on the target sample molecular set; and determining at least one target molecule in the molecular database based on the project scoring model.

[0023] By applying the scheme of the embodiments in this specification, multiple molecular sets are divided based on the molecular attribute information related to the target project. This increases the diversity between the molecular sets, thereby improving efficiency and accuracy in subsequent screening and analysis. Subsequently, by utilizing the molecular set weights of each molecular set, the molecules used to train the scoring model are determined. This avoids the problem of poor scoring model performance caused by obtaining too many invalid molecules when acquiring training molecular scores. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a data filtering method provided in one embodiment of this specification;

[0025] Figure 2 This is a schematic diagram illustrating the statistical distribution of molecular attribute information of a project, provided in one embodiment of this specification.

[0026] Figure 3 This is a schematic diagram of a molecular set search tree provided in one embodiment of this specification;

[0027] Figure 4 This is a flowchart illustrating a data filtering method applied to a cloud device, as provided in one embodiment of this specification.

[0028] Figure 5 This is an architecture diagram of a data filtering system provided in one embodiment of this specification;

[0029] Figure 6 This is a flowchart illustrating the processing steps of a drug molecule screening method provided in one embodiment of this specification;

[0030] Figure 7 This is a schematic diagram of the structure of a data filtering device provided in one embodiment of this specification;

[0031] Figure 8 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0032] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0033] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0034] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0035] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0036] First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0037] Fluorophenyl (Fsp3): This is an indicator describing the proportion of sp3-hybridized carbon atoms in a molecule. An sp3-hybridized carbon atom is formed by four different atoms or groups of atoms bonded together, typically forming a tetrahedral geometry. The value of fsp3 ranges from 0 to 1, representing the proportion of sp3-hybridized carbon atoms in the molecule. A higher value indicates a greater proportion of sp3-hybridized carbon atoms in the molecule.

[0038] Partition Coefficient (logP): This is the logarithm of a molecule's lipid-water partition coefficient. It describes the molecule's tendency to partition between the lipid (oily) phase and the aqueous phase. The higher the logP, the more likely the molecule is to partition into the lipid phase, while the lower the logP, the more likely the molecule is to partition into the aqueous phase.

[0039] Molecular weight (MW): This is the molecular weight of a molecule, which is the sum of the relative atomic masses of all its atoms. It is usually expressed in Daltons (Da).

[0040] Total Polar Surface Area (TPSA): This is the polar surface area of ​​a molecule, referring to the total surface area of ​​all hydrogen bond donors and acceptors on the molecule's surface. It is usually expressed in square angstroms. Units.

[0041] Chiral center count: This is the number of chiral centers in a molecule. A chiral center is a carbon atom, and at least three of its four ligands are distinct. A molecule with a chiral center can exist in two or more chiral isomers.

[0042] Hydrogen Bond Donor Count (HBD): This is the number of functional groups (hydrogen bond acceptors) in a molecule that can provide hydrogen bonds. HBD represents the number of hydrogen bond acceptors, which are usually negatively charged atoms such as nitrogen, oxygen, and fluorine, and can interact with protons through hydrogen bonds.

[0043] Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH clustering) is a clustering algorithm for processing large datasets. This algorithm progressively compresses the data by building a tree structure, preserving important clustering information. BIRCH clustering is particularly suitable for scenarios with limited memory, as it can efficiently process large-scale data within a limited memory space. The algorithm first builds a compact tree structure, and then prunes and clusters the tree as needed, thereby achieving efficient data clustering analysis.

[0044] K-means clustering is a commonly used clustering algorithm that divides a dataset into K clusters, each represented by its mean. The algorithm iteratively assigns data points to clusters with the nearest mean, then updates the mean position until convergence. K-means clustering is suitable for most numerical datasets, and is computationally simple and fast. However, it is highly sensitive to the initial cluster selection and requires pre-setting the number of clusters.

[0045] Hierarchical clustering is a clustering algorithm that constructs a hierarchical structure by progressively merging or splitting data points to form a tree-like cluster structure. The algorithm is divided into two types: bottom-up (agglomerated hierarchical clustering) and top-down (split hierarchical clustering). Hierarchical clustering does not require a pre-defined number of clusters and can provide hierarchical relationships in the data, but it has higher computational complexity and is suitable for exploring the internal structure of data and multi-level analysis.

[0046] Molecular docking: a computational chemistry approach used to predict how small molecules (often drug molecules) interact with biomolecules (often proteins) to understand their binding patterns and affinities.

[0047] Wet experiments: Real-world experiments conducted in a laboratory using actual substances (e.g., chemical reagents, biological specimens, or organisms). Used in drug discovery to detect the activity of small molecules and proteins in pocket binding.

[0048] Morgan fingerprints are bitmap fingerprints used to represent molecular structures. They generate fixed-length binary vectors by traversing all atoms and their neighbors within a molecule. This method captures the topological features of molecules based on atomic connections, aiding in similarity searches and molecular classification. Morgan fingerprints are commonly used in cheminformatics to improve the efficiency and accuracy of searching molecular databases.

[0049] This specification provides a data filtering method, a data filtering method applied to cloud devices, and also relates to a data filtering device, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0050] The benefits of molecular screening lie in its ability to effectively identify molecules with the potential to be used in developing targeted drugs. Early target identification and high-throughput screening can rapidly narrow down the pool of candidate molecules, thereby improving the efficiency and success rate of drug development. In this process, computational chemistry methods such as molecular docking and virtual screening can predict the interactions between small molecules and target drugs, reducing the workload of laboratory experiments. However, current challenges include the high computational complexity of molecular docking and low sample efficiency. Existing general affinity prediction models struggle to achieve ideal accuracy in large-scale virtual screening, and computational costs increase significantly with the size of the molecular library. Furthermore, traditional virtual screening processes suffer from numerous false positives (subsequent experiments show a low degree of interaction with the target drug, but the prediction shows a high degree of interaction) and false negatives (subsequent experiments show a high degree of interaction with the target drug, but the prediction shows a low degree of interaction), leading to reduced reliability of the screening results. The problem of low sample efficiency is also significant, as traditional methods often require a large number of samples to accurately estimate the value of each strategy, which not only increases computational costs but also affects the practicality and scalability of the algorithm.

[0051] See Figure 1 , Figure 1 A flowchart of a data filtering method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0052] Step 102: Obtain at least one project molecular attribute information and the set of verified molecules corresponding to the target project.

[0053] In practical applications, the target project is a specific research objective, the validated molecule set is a collection of molecules whose relationship with the target project has been experimentally verified, and the project molecule attribute information is molecular attribute information that has a significant impact on the activity of molecules in the molecular database.

[0054] For example, the target project can be understood as a specific research objective or project, such as drug development for a disease, development of environmental pollutant detection technology, development of new energy materials, etc. This specification does not impose any limitations on the comparison. The validated molecule set can be understood as molecules that have been validated for the target project. For example, if the target project is drug development for disease A, the validated molecule set refers to each molecule whose activity (i.e., the probability of producing an effect by binding to the biological target) against disease A has been validated; similarly, if the target project is development of detection technology for environmental pollutant A, the validated molecule set refers to each molecule whose degree of compatibility against environmental pollutant A has been tested, etc. This specification does not impose any limitations on the comparison.

[0055] Project molecular attribute information can be understood as the molecular features and parameters that significantly influence molecule activity in a molecular database. Due to the large amount of molecular attribute information, various molecular attribute information is used to determine which to use as criteria for dividing the molecule set. Considering that the number of molecules in the validated molecule set for the target project is insufficient to represent the most important molecular features in the molecular database, calculations were performed on the validated molecule set for multiple target projects, identifying the molecular attributes and their significant impact on molecule activity. Subsequently, based on the correlation and importance of each molecular attribute with the activity of molecules and multiple target projects, as well as the relationships between various molecular attributes, a predetermined number of representative project molecular attribute information items are determined as the project molecular attribute information. Molecular attribute information can be understood as the properties possessed by molecules in a physical dimension.

[0056] It should be noted that if the number of molecules in the validated molecular set for the target project has reached a number that can reflect molecular characteristics that are important to the activity of molecules in the molecular database, the molecular attribute information of the project can also be molecular attribute information that is highly correlated with the target project.

[0057] In one embodiment provided in this specification, based on the fact that tens of millions of molecules in the molecular database have been validated and analyzed with multiple different projects, the molecular properties that are more important to the activity of the molecules were identified as fsp3 (Fraction of sp3-hybridized carbon atoms), logP (Partition Coefficient), MW (Molecular Weight), TPSA (Topological Polar Surface Area), chiral center count, and HBD (Hydrogen Bond Donor).

[0058] By dividing molecules into multiple sets based on molecular attributes that are important to their activity, the differences between these sets can be increased, which in turn helps to improve efficiency and accuracy in subsequent screening and analysis.

[0059] Step 104: Based on the molecular attribute information of each project and the verified molecular set, determine at least one molecular set in the molecular database, wherein the molecular set is a set of molecules that have not been verified for the target project.

[0060] In practical applications, a molecular set is a collection of molecules that share the same characteristics in terms of molecular attribute information for each project.

[0061] For example, the specific methods for determining at least one molecular set in the molecular database based on the molecular attribute information of each item can be as follows: a molecular similarity-based method can be used to perform similarity matching on molecules using their structural fingerprints or descriptors, thereby selecting at least one molecular set; a machine learning algorithm, such as a clustering algorithm, can be used to group the molecules in the molecular database using the molecular attribute information of each item, and select at least one molecular set based on the attribute characteristics of each molecule; a rule-based filtering method can also be used, setting different value ranges for the molecular attribute information of each item to filter out molecular sets in the set where the molecular attribute information of each molecule in the set falls within the same value range, etc. This specification does not impose any limitations on these methods.

[0062] By grouping molecules with similar characteristics into sets, screening efficiency can be improved and computational resources can be reduced during molecular screening. This helps to concentrate resources and efforts on the molecules most likely to succeed, thus accelerating the subsequent drug discovery process.

[0063] Furthermore, based on the molecular attribute information of each project and the verified molecular set, at least one molecular set is determined in the molecular database, including:

[0064] A first molecular database is obtained based on the verified molecule set, wherein the first molecular database is a molecular database composed of molecules other than those in the verified molecule set.

[0065] Determine at least two project molecular attribute ranges corresponding to each project's molecular attribute information, where the project molecular attribute range information is the value range of the project molecular attribute information;

[0066] Obtain the molecular project attribute values ​​corresponding to the molecular attribute information of each molecule in the first molecular database for each project;

[0067] Based on the molecular attribute range information of each project and the molecular project attribute values ​​corresponding to each molecule in the first molecular database, at least one molecular set is determined.

[0068] In practical applications, the first molecular database is the complete molecular database excluding the verified molecule set. The project molecular attribute range information is the multiple value ranges corresponding to the molecular attribute information of a certain project. The molecular project attribute value is the value of a certain molecule for the molecular attribute information of a certain project.

[0069] For example, the project molecular attribute range information can be understood as the range information for the values ​​corresponding to the project molecular attribute information, based on the statistical distribution of molecules in the first molecular database within the corresponding project molecular attribute information. (Refer to...) Figure 2 , Figure 2This diagram illustrates the statistical distribution of item molecular attribute information according to one embodiment of the present specification. It includes item molecular attribute information 1 and item molecular attribute information 2. In the first molecular database, 10% of the molecules corresponding to item molecular attribute information 1 have a molecular attribute value <= 0.2; 40% have a molecular attribute value between (0.2, 0.3); 25% have a molecular attribute value between (0.3, 0.4); 10% have a molecular attribute value between (0.4, 0.5); 5% have a molecular attribute value between (0.5, 0.6); and 5% have a molecular attribute value > 0.6. Similarly, in the first molecular database, 10% of the molecules corresponding to item molecular attribute information 2 have a molecular attribute value < 3. 5% of the items corresponding to sub-attribute information 2 have an attribute value of [3, 3.5). 20% of the items corresponding to sub-attribute information 2 have an attribute value of [3.5, 4). 50% of the items corresponding to sub-attribute information 2 have an attribute value of [4, 4.5). 10% of the items corresponding to sub-attribute information 2 have an attribute value of [4.5, 5). 5% of the items corresponding to sub-attribute information 2 have an attribute value >= 5. Therefore, we can determine that the range of sub-attribute information 1 is (-∞, 0.2], (0.2, 0.3], (0.3, 0.4], (0.4, 0.5], (0.5, 0.6], (0.6, +∞). The range of sub-attribute information 2 is (-∞, 0.3), [0.3, 3.5), [3.5, 4), [4, 4.5), [4.5, 5), [5, +∞).

[0070] It should be noted that the method of determining the molecular set can be understood as dividing the molecular item attribute values ​​corresponding to the molecular attribute information of each item in the first molecular database according to the attribute range information of each item's molecular attribute information. Continuing with the previous example, it includes 36 molecular sets. Taking molecular set 1 as an example, the range of molecular item attribute values ​​corresponding to item molecular attribute information 1 is (-∞, 0.2], and the range of molecular item attribute values ​​corresponding to item molecular attribute information 2 is (-∞, 0.3); taking molecular set 2 as an example, the range of molecular item attribute values ​​corresponding to item molecular attribute information 1 is (-∞, 0.2], and the range of molecular item attribute values ​​corresponding to item molecular attribute information 2 is [0.3, 3.5); taking molecular set 36 as an example, the range of molecular item attribute values ​​corresponding to item molecular attribute information 1 is (0.6, +∞), and the range of molecular item attribute values ​​corresponding to item molecular attribute information 2 is [5, +∞), and so on. The range of molecular item attribute values ​​corresponding to molecules in the other molecular sets will not be elaborated here.

[0071] By using multiple ranges of molecular attribute information that are important to the activity of molecules related to the target project, multiple molecular sets can be divided, thereby increasing the differences between the molecular sets and improving the efficiency and accuracy of subsequent screening and analysis.

[0072] Step 106: Obtain the set access count and set score corresponding to each molecular set, and obtain the set weight corresponding to each molecular set based on the set access count and set score. The set access count is the number of molecules verified in the process of determining the target sample molecular set, and the set score is the number of target feature molecules hit by the corresponding molecular set in the process of determining the target sample molecular set. The target feature molecules are molecules related to the target project.

[0073] In practical applications, the molecular set score is the number of target feature molecules identified after the molecules in the molecular set are verified during the molecular set search process, the molecular set weight is the comprehensive evaluation of the molecular set during the molecular set search process, and the target feature molecules are molecules related to the target project.

[0074] For example, the number of times a set is accessed can be understood as information characterizing the importance and degree of exploration of that molecular set throughout the search tree.

[0075] It should be noted that the specific methods for obtaining the number of times each molecular set is accessed and the score of each molecular set, and for obtaining the weight of each molecular set based on these data, can be as follows: First, calculate the number of times each molecular set is accessed by statistically analyzing its access frequency during the simulated screening process, and then determine the weight by combining this with a weighted average of the molecular set scores. Second, analyze historical data of each molecular set's performance, use a regression model to predict future set scores, and then calculate the weight. Third, use a molecular set search tree to obtain the number of times each molecular set is accessed and the score of each molecular set, and then generate accurate molecular set weights, etc. This manual does not impose any restrictions on these methods.

[0076] By obtaining the set access count and set score corresponding to each molecular set, the set weight of each molecular set can be obtained, which can effectively improve the efficiency and accuracy of the screening process. This reflects the importance of each molecular set in the actual screening and helps to prioritize the better-performing molecular sets in subsequent molecular screening, reducing the waste of resources on irrelevant sets.

[0077] Furthermore, obtain the set visit count and set score for each molecular set, including:

[0078] Obtain the target molecular set search tree corresponding to the target project, wherein the molecular set search tree includes a root node, at least two child nodes and at least two leaf nodes, the root node is the starting point of the molecular set search tree, the child nodes are nodes derived from the root node or other child nodes according to the project molecular attribute information, and the leaf nodes are the end nodes of the search tree. The root node represents the starting node for determining the molecular set corresponding to the target project, the child nodes represent the decision results based on the project molecular attribute range information, and the leaf nodes represent the molecular sets corresponding to the decision results based on the decision results of the molecular attribute range information of their respective projects.

[0079] The number of times each molecular set is accessed and the molecular set score are determined based on the node score and the number of times each leaf node and its corresponding parent node are accessed in the search tree of the target molecular set. The node score is the number of target feature molecules that are accessed and hit in the molecular set corresponding to the node.

[0080] In practical applications, the molecular ensemble search tree (MAS) is a tree-like data structure corresponding to a target project, used to explore and evaluate the potential of multiple molecular ensembles during drug discovery. It considers the different attributes and characteristics of each molecular ensemble during the search process, progressively optimizing the decision path through continuous selection, expansion, simulation, and backpropagation steps. The node visit count of a leaf node in the MAS represents the number of validation molecules obtained by that leaf node during the molecular ensemble search process, and the node score of a leaf node represents the number of molecules with the target characteristics among the validation molecules of that leaf node during the molecular ensemble search process.

[0081] It should be noted that the node score is the number of target feature molecules accessed and hit in the molecular set corresponding to the node. The molecular set corresponding to the node can be understood as the molecular set included after the node makes a decision based on the molecular attribute range information of the project. Each leaf node corresponds to a molecular set. Therefore, the node score of the child node can be further understood as the sum of the molecular scores of each leaf node corresponding to the node. The node score of the leaf node can be understood as the number of molecules related to the target project among the verified molecules in the molecular set corresponding to the leaf node.

[0082] For example, the method for obtaining the set weight corresponding to each molecular set based on the set access count and molecular set score determined by the leaf nodes and their corresponding parent nodes in the molecular set search tree is shown in Formula 1 in one embodiment of this specification:

[0083]

[0084] UCB i w represents the set weights corresponding to the set of molecules. i w represents the number of times the leaf node of the molecule set is visited. i-1 Let n be the number of times the leaf node of the molecule set is visited, and its corresponding parent node is visited. i The node score is the number of leaf nodes corresponding to the numerator set.

[0085] By identifying molecular sets through a molecular set search tree, and calculating the weight of the molecular set based on the node score and node visit count of the leaf node corresponding to the leaf node and its corresponding parent node, the balance between exploration and utilization of the molecular set can be accurately reflected. This enables effective strategy selection in complex decision-making environments, improves computational efficiency, and enhances the correlation between the selected molecular set and the target project.

[0086] Further, obtaining the target molecule set search tree corresponding to the target project includes:

[0087] Determine at least two project molecular attribute ranges corresponding to each project's molecular attribute information, where the project molecular attribute range information is the value range of the project molecular attribute information;

[0088] Obtain the initial molecular set search tree corresponding to the molecular attribute range information of each project, and initialize the initial molecular set search tree according to the verified molecular set;

[0089] The set of molecules to be searched is determined based on each set of molecules, and the node score and the number of times each node is visited are obtained in the initial molecular set search tree.

[0090] Repeat the following steps until the search tree iteration stops: Based on the node score and node visit count of each node in the initial molecular set search tree, determine at least one current search molecule in the set of molecules to be searched; update the node score and node visit count of the leaf nodes and ancestor nodes corresponding to each current search molecule; update the set of molecules to be searched based on the current search molecule; wherein, the initial molecular set search tree obtained when the search tree iteration stops is reached is the target molecular set spanning tree.

[0091] In practical applications, the set of molecules to be searched is the set of molecules required to iterate over each node in the molecular set search tree, the current search molecule is the molecule used to iterate over each node in the molecular set search tree in the current iteration round, and the search tree iteration stopping condition is the condition for stopping the iteration of the search tree.

[0092] For example, by identifying at least one current search molecule in a single iteration, multiple current search molecules can be used simultaneously to iterate the molecular set search tree in a single iteration. This accelerates the iteration process of the molecular set search tree, thereby improving the efficiency of molecular screening when screening molecular databases containing a large number of molecules. By processing the evaluation of multiple molecular sets in parallel, a larger search space can be covered in the same amount of time. Furthermore, by computing and updating the scores and access counts of each node in parallel, potential high-value candidate molecule sets can be found more quickly, thereby optimizing the decision path, improving the screening efficiency and accuracy of target molecules, and ultimately increasing the success rate of drug discovery.

[0093] It's important to note that identifying a molecule in the search set and updating the node scores and visit counts of each node in the molecule search tree can be understood as dynamically adjusting these scores and visit counts. This dynamic adjustment of the node scores and visit counts then dynamically modifies the confirmation strategy for the current molecule, enabling more effective evaluation and screening of potential candidate molecules. This process optimizes the decision path based on the visit count and score of each node, ensuring that existing high-value nodes are fully utilized while exploring new potential high-value nodes, thereby improving the efficiency and accuracy of the entire molecule screening process.

[0094] Furthermore, obtaining the initial molecular set search tree corresponding to the molecular attribute range information of each project, and initializing the initial molecular set search tree according to the verified molecular set, includes:

[0095] The child nodes corresponding to the molecular set search tree are determined based on the molecular attribute range information of each project.

[0096] The node score and node visit count of each node in the molecular set search tree are initialized based on the verified molecular set.

[0097] Specifically, each child node in the molecular set search tree is obtained by splitting according to the molecular attribute range information. The molecular set search tree includes multiple layers of nodes. The sibling nodes are obtained by obtaining multiple corresponding nodes according to the multiple molecular attribute range information corresponding to the molecular attribute information of the corresponding project at the same parent node. The number of layers in the molecular set search tree is the number of project molecular attribute information plus 1 (that is, adding a starting root node).

[0098] Structural reference of molecular set search tree Figure 3 , Figure 3 This diagram illustrates a molecular set search tree according to one embodiment of the present specification. In this embodiment, the target project corresponds to 6 project molecular attribute information, and each project molecular attribute information corresponds to 6 molecular attribute range information. First, the molecular set search tree has a root node that initiates the molecular set search. Then, according to project molecular attribute information 1, 6 child nodes corresponding to the root node are obtained. Each child node corresponds to a different molecular attribute range information according to project molecular attribute information 1. Next, according to project molecular attribute information 2, 6 child nodes corresponding to the aforementioned 6 child nodes are obtained. Each child node corresponds to a different molecular attribute range information according to project molecular attribute information 2, and so on. 6 child nodes are obtained according to project molecular attribute 6. 6 There are 10 leaf nodes, and each leaf node corresponds to a different set of molecules.

[0099] It should be noted that initializing the node scores and node visit counts of each node in the molecular set search tree based on the validated molecule set can be understood as initializing the node scores and node visit counts of each node in the molecular set search tree using the validated molecule set (prior knowledge of the activity / inactivity of the target molecules). Specifically, the molecular attribute values ​​of molecules in the validated molecule set determine which leaf node they will be assigned to, and this initializes the node scores and node visit counts of the leaf node and its ancestor nodes. The specific method for initializing node scores and node visit counts is as follows: the visit count of a leaf node is initialized to the sum of the non-target feature molecules and target feature molecules at that leaf node; the node score of a leaf node is initialized to the number of target feature molecules at that leaf node; the node visit count of an ancestor node is the sum of the visit counts of all its child nodes; and the node score of an ancestor node is the sum of the scores of all its child nodes.

[0100] It is important to note that when the project's molecular attribute range information, i.e., the project's molecular attribute information, is not related to the target project (i.e., the project's molecular attribute information corresponds to the molecular database, not the target project), the child nodes corresponding to the molecular set search tree can be determined when the molecular set is determined based on the target project's attribute information. If the molecular set and the child nodes corresponding to the molecular set search tree have already been determined based on the project's molecular attribute information when executing the target project, then when executing the data filtering method corresponding to the target project, the molecular set search tree corresponding to the molecular database can be directly obtained. This molecular set search tree also includes a root node, at least two child nodes, and at least two leaf nodes. It is only necessary to initialize the node scores and node access counts corresponding to each node in the molecular set search tree using the verified molecular set of the target project.

[0101] Initializing the molecular search tree with a validated molecule set allows for an initial state setup based on already validated data, ensuring high accuracy and effectiveness from the outset. The validated molecule set provides real experimental data and evaluation results, ensuring the search tree is built upon existing scientific knowledge and experimental findings. This initialization method helps reduce blind exploration during the search process, improving the efficiency and success rate of molecular screening, and also enabling faster identification of potentially high-value candidate molecules, thereby enhancing the overall efficiency and accuracy of the molecular screening process.

[0102] Furthermore, the set of molecules to be searched is determined based on each set of molecules, including:

[0103] Obtain at least one cluster molecular attribute information, wherein the cluster molecular attribute information is molecular attribute information other than the molecular attribute information of each item;

[0104] Based on the molecular attribute information of each cluster, at least one molecule to be searched is determined for each molecular set, wherein the molecular set to be searched includes at least one molecule to be searched.

[0105] In practical applications, cluster molecular attribute information is used to obtain the molecular attribute information of the set of molecules to be searched in the molecular set, and the molecules to be searched are the molecules used to perform molecular set search tree iteration.

[0106] For example, since the molecular attribute values ​​corresponding to each item's molecular attribute information for each molecule in the molecular set are all within the same range, it is necessary to use clustering molecular attribute information other than the item's molecular attribute information to cluster the molecules in the molecular set, and obtain the molecules at each cluster center as the search molecules for subsequent search tree iteration. The methods for clustering molecules can include BIRCH clustering (Balanced Iterative Reducing and Clustering using Hierarchies), K-means clustering, and hierarchical clustering, etc., and this specification does not impose any limitations on this.

[0107] Considering that molecular databases are typically in the hundreds of billions, even if multiple sets of molecules to be searched are divided based on the molecular attribute information of each project, each set of molecules to be searched usually contains a large number of molecules, and the molecules also have similar characteristics. Therefore, in order to avoid the problem of search tree iteration bias caused by repeatedly selecting molecules with similar attributes from a certain set of molecules during the iteration of the search tree, it is necessary to determine at least one representative molecule to be searched in each set of molecules.

[0108] Furthermore, based on the node score and node visit count corresponding to each node in the initial molecule set search tree, at least one current search molecule is determined in the set of molecules to be searched, including:

[0109] Based on the node score and node visit count of each node in the molecular set search tree, determine the current search weight corresponding to each leaf node;

[0110] Based on the current search weight corresponding to each leaf node, at least one current search leaf node is determined;

[0111] At least one current search molecule is determined from among the search molecules corresponding to each current search leaf node.

[0112] In practical applications, the current search weight represents the importance of the leaf node in the current iteration process, and the current search leaf node is the leaf node that needs to be visited in the current iteration process.

[0113] For example, the current search weight of each leaf node can be obtained by calculating the exploration utilization score of each leaf node and then normalizing it based on the exploration utilization scores of all leaf nodes. Alternatively, it can be obtained by calculating the exploration utilization scores of each leaf node and its corresponding ancestor node, obtaining the product of the search utilization scores of each leaf node and its ancestor node, and then normalizing the product of all nodes. This specification does not impose any restrictions on this. It should be noted that the exploration utilization score of a node can be understood as a balance evaluation score reflecting the molecular set between exploration and utilization. This search utilization score takes into account the currently known performance (utilization) and the potential for future exploration (exploration), and can effectively evaluate and screen high-potential molecular sets in this iteration.

[0114] In one embodiment provided in this description, the node exploration utilization score is as shown in Formula 2:

[0115]

[0116] Where Nodei represents the i-th child node, and Nodei-1 represents the parent node of the i-th child node; This represents the number of times the molecule is visited by the i-th child node. This represents the numerator fraction of the i-th child node. The number of times the parent node of the i-th child node is visited can be seen as the number of target feature molecules found; T represents the temperature factor, and a temperature curve T that is inversely proportional to the number of iterations can be defined to control whether the overall search tends to explore (T value > 1) or exploit (T value < 1).

[0117] It should be noted that determining at least one current search leaf node based on the current search weight of each leaf node can be achieved by allocating the number of search molecules corresponding to each leaf node to each leaf node according to the current search weight of each leaf node. If the total number of search molecules is less than the number of leaf nodes, then the leaf node with the larger current search weight is determined as the current search node. Alternatively, a preset threshold for the number of search molecules for each leaf node can be used, and the number of search molecules corresponding to each leaf node can be allocated to each leaf node according to the current search weight and the threshold. If the total number of search molecules is insufficient to allocate a number of search molecules greater than the threshold to each leaf node, then the leaf node with the larger current search weight is determined as the current search node, and so on. This specification does not impose any restrictions on this method.

[0118] It is important to note that determining at least one current search molecule among the search molecules corresponding to the current search leaf node can be achieved by removing search molecules that have been verified in previous iterations from the search molecules corresponding to the current search leaf node, and then randomly selecting several search molecules determined according to the previous weights from the removed search molecules as the current search molecules.

[0119] By calculating the exploration scores of each node and determining the current search leaf node for the current iteration, it is possible to effectively...

[0120] Furthermore, based on each current search molecule, update the node score and node visit count of the corresponding leaf node and the corresponding ancestor node of each leaf node, including:

[0121] Determine the target current search molecule, wherein the target current search molecule is any one of the current search molecules;

[0122] Increment the node visit count of the leaf node and each ancestor node corresponding to the current search molecule of the target by 1;

[0123] If the target current search molecule is a target feature molecule, increment the node score of the leaf node and each ancestor node corresponding to the target current search molecule by 1.

[0124] In practical applications, the target molecule currently being searched is the molecule being validated to iterate through the molecule set search tree. Validating the target molecule to determine if it is a target feature molecule can be done through methods such as molecular docking, wet experiments, etc., and this specification does not impose any restrictions on this approach.

[0125] For example, if the target current search molecule is not the target feature molecule, it means that there is no correspondence between the target current search molecule and the target item. Therefore, only when the target current search molecule is the target feature molecule, the node score of the leaf node and each ancestor node corresponding to the target current search molecule is increased by 1, that is, the degree of association between the leaf node and its corresponding ancestor nodes and the target item is increased by 1. Otherwise, no operation is performed.

[0126] By determining whether the currently searched molecule is a target feature molecule related to the target project, and updating the node score and node visit count of each node in the search tree, the accuracy and efficiency of the search tree can be effectively improved. This ensures that the search process is more targeted, focusing on the set of high-potential molecules related to the target project, thereby optimizing resource allocation and experimental design. It not only reduces the number of invalid searches but also accelerates the screening process of candidate molecules, helping to advance drug discovery and development more quickly.

[0127] Furthermore, the stopping condition for the search tree iteration includes:

[0128] The preset number of iteration rounds has been reached; and / or,

[0129] The number of target feature molecules in the historical search molecule set reaches a preset threshold, wherein the historical search molecule set is obtained based on the current search molecules in each iteration round.

[0130] In practical applications, the historical search molecule set is the set of each current search molecule during the iteration process of the search tree.

[0131] By stopping the search tree iteration when the number of target feature molecules in each current search molecule reaches a preset threshold after a certain number of iterations and / or after a preset threshold in all iterations, computational resource consumption can be effectively controlled, overcomputation avoided, and a high-quality candidate molecule set obtained within a reasonable timeframe can be ensured, thereby improving the efficiency and effectiveness of the drug discovery process. Setting stopping conditions in advance allows for better planning of the experimental process, ensuring optimal decisions are made at critical junctures, and further improving the efficiency of molecule screening.

[0132] Step 108: Determine the target sample molecule set based on each molecule set and the corresponding molecule set weight, and generate a project scoring model for the target project based on the target sample molecule set.

[0133] In practical applications, the target sample molecule set is the molecule set used to train the project scoring model to obtain the project scoring model for the target project. The project scoring model is a model that predicts the degree of fit between the molecules and the target project.

[0134] For example, the item scoring model can be any model that can predict the molecular score of a molecule for a target item. For example, the input is the Morgan fingerprint of the molecule and the output is the predicted molecular score of the molecule for the target item. It can also be composed of three scoring models, each with the Morgan fingerprint of the molecule as input and outputting the poor molecular score, the more likely molecular score, and the better molecular score, etc. This specification does not impose any restrictions on this.

[0135] It should be noted that the method for generating the project scoring model based on the target sample molecular set can be as follows: the target sample molecular set can be identified by determining the molecular set that is most similar to it in the correspondence table between molecular sets and scoring models, and the scoring model corresponding to that molecular set can be determined as the project scoring model for the target project; or the initial scoring model can be trained based on the target sample molecular set to obtain the project scoring model for the target project, etc. This manual does not impose any restrictions on this.

[0136] Furthermore, based on each molecular set and its corresponding molecular set weights, the target sample molecular set is determined, including:

[0137] Based on the molecular set weights corresponding to each molecular set, at least one molecular set is determined as the initial sample molecular set, wherein the initial sample molecular set is a set of molecules used to obtain the project scoring model that has not yet been validated.

[0138] A reference sample molecule set is determined based on the verified molecule set and the historical search molecule set. The historical search molecule set is obtained based on the current search molecules in each iteration of the molecule set search tree. The reference sample molecule set is a set of molecules that have been verified and are used to obtain the project scoring model.

[0139] The target sample molecule set is determined based on the initial sample molecule set and the reference sample molecule set.

[0140] In practical applications, the initial sample molecular set is the molecular set with the higher molecular set weight among all molecular sets, and the reference sample molecular set is the molecular set that includes the validated molecular set and the historical search molecular set.

[0141] Since the validated molecule set and the historical search molecule set already include molecular scores representing each molecule for the target item, when determining the target sample molecule set for obtaining the item scoring model, in order to further improve the accuracy of the trained item scoring model, the validated molecule set and the historical search molecule set, which include molecular scores representing each molecule for the target item, are combined with the molecule set with higher weights in each molecule set. This can improve the accuracy of the trained item scoring model while reducing the decrease in molecule screening efficiency caused by obtaining molecular scores corresponding to the target item.

[0142] Furthermore, based on the target sample molecule set, a project scoring model is generated for the target project, including:

[0143] Obtain the sample molecule score corresponding to the target item in the target sample molecule set;

[0144] Based on each sample molecule and its corresponding sample molecule score, a project scoring model is obtained for the target project.

[0145] In practical applications, the molecular score represents the degree of fit between a molecule and the target project. A higher degree of fit indicates that the molecule is more likely to be the molecule required for the target project. Specifically, the molecular score for a molecule to the target project can be obtained through methods such as molecular docking, wet experiments, etc., to verify the molecule. This manual does not impose any restrictions on these methods.

[0146] It should be noted that since the target sample molecule set includes both validated molecule sets and historical search molecule sets, when obtaining the molecule scores of sample molecules through molecular docking, wet experiments, and other methods to verify molecules, only the molecule scores of molecules in the molecule sets with higher molecule set weights in the target sample molecule set can be obtained.

[0147] By obtaining the scores of each sample molecule in the target sample molecule set corresponding to the target item, a project scoring model for the target item can be trained. This allows the model to predict the molecular score of each molecule in the molecular database for the target item using only the project scoring model, thereby significantly reducing the workload and cost of experimental screening. This method not only improves screening efficiency but also enables rapid identification of high-potential candidate molecules at an early stage, providing a reliable basis for subsequent experimental verification, optimizing resource allocation, and accelerating the efficiency of molecule screening.

[0148] To further improve the accuracy of identifying target molecules based on the project scoring model, the project scoring model includes at least one quantile scoring model.

[0149] Based on the project scoring model, at least one target molecule is identified in the molecular database, including:

[0150] A second molecular database is determined based on the target sample molecule set, wherein the second molecular database is a molecular database composed of molecules other than those in the target sample molecule set.

[0151] Based on the quantile scoring model, at least one quantile score corresponding to each molecule in the second molecular database is obtained.

[0152] Based on the quantile fractions of each molecule in the second molecular database, at least one target molecule is determined in the second molecular database.

[0153] In practical applications, quantile scoring models are regression models based on different quantiles, used to evaluate the scores of molecules under different performance conditions. These models predict the molecular scores that a molecule might obtain under poor, moderate, and good performance conditions by predicting different performance quantiles of the input molecule, thus providing a basis for the screening and optimization of target molecules. The quantile score is the range of scores predicted for a molecule under the corresponding conditions of each quantile scoring model.

[0154] By using quantile scores from different quantile scoring models, we can determine the performance of molecules in different contexts, thereby guiding further screening and optimization processes.

[0155] In one embodiment provided in this specification, the target molecule is determined by three scoring models: a first quantile scoring model, a second quantile scoring model, and a third quantile scoring model. The first quantile scoring model is a deep learning model that obtains the molecular score of the input molecule that is poor, and the first quantile score is the molecular score of the input molecule that is poor. The second quantile scoring model is a deep learning model that obtains the molecular score of the input molecule that is more likely, and the second quantile score is the molecular score of the input molecule that is more likely. The third quantile scoring model is a deep learning model that obtains the molecular score of the input molecule that is better, and the second quantile score is the molecular score of the input molecule that is better.

[0156] In this embodiment, the first quantile scoring model can be understood as a low quantile scoring model, used to evaluate molecules performing poorly. The obtained first quantile score can be understood as the molecular score that an input molecule might obtain under poor conditions. The sample molecules used to train the first quantile scoring model are molecules whose performance is at a poor quantile (e.g., a poor quantile of 0.2 corresponds to a sample molecule with a score in the bottom 20%). This first quantile score can determine the score of the input molecule under poor conditions, thereby further improving the accuracy of identifying target molecules.

[0157] The second quantile scoring model can be understood as a median scoring model, used to evaluate molecules performing at a moderate level. The obtained second quantile score can be interpreted as the molecular score that the input molecule might obtain under moderate conditions. Sample molecules used to train the second quantile scoring model are molecules whose performance is at a median quantile (e.g., a median quantile of 0.5 corresponds to a score in the middle 50%). This second quantile score can determine the score of the input molecule under moderate conditions, thereby enabling further screening and optimization to ensure the accuracy of target molecules under moderate performance conditions.

[0158] The third quantile scoring model can be understood as a high quantile scoring model, used to evaluate molecules under good performance conditions. The obtained third quantile score can be understood as the molecular score that the input molecule might obtain under good conditions. Sample molecules used to train the third quantile scoring model are molecules that perform at a good quantile (e.g., a good quantile of 0.8 corresponds to samples with scores in the top 20%). This third quantile score can determine the score of the input molecule under good conditions, thereby allowing for further screening and optimization to ensure the accuracy of the target molecule under good performance conditions.

[0159] It should be noted that the scoring model can be expressed as Formula 3:

[0160] Q Y (τ|X)=X·β(τ)…Formula 3

[0161] Where τ is the quantile value, Q Y (τ|X) is the fraction of Y at the τth quantile given X, and β(τ) is the regression coefficient of the numerator fraction corresponding to the quantile. These coefficients vary with τ.

[0162] The formula for calculating the loss value of each scoring model is Formula 4:

[0163]

[0164] Where, ρ τ (u) is the quantile loss function, where u is...

[0165] Taking molecular scoring using three quantile scoring models as an example, in one embodiment, the selected quantiles are 0.2, 0.5, and 0.8. That is, the quantile corresponding to the first quantile scoring model is 0.2, the second quantile scoring model is 0.5, and the third quantile scoring model is 0.8. The independent variable is the Morgan fingerprint of the molecule, and the dependent variable is the score predicted by the docking model. When τ = 0.5, that is, the second quantile scoring model predicts the median of the dependent variable. The model with τ = 0.2 (i.e., the first quantile scoring model) predicts the lower quantile (20th percentile) of the dependent variable given the independent variable, reflecting the conditional limit, i.e., there is an 80% probability that the actual observed value will be higher than this predicted value. The model with τ = 0.8 (i.e., the third quantile scoring model) predicts the higher quantile (80th percentile) of the dependent variable given the independent variable, reflecting the conditional upper limit, i.e., there is an 80% probability that the actual observed value will be lower than this predicted value. By using the predicted scores from the three quantile models, the probability distribution of a molecule's score can be obtained. Then, by setting different increasing low quantile thresholds, middle quantile thresholds, and quantile thresholds, molecules with a first quantile score higher than the low quantile threshold (i.e., a higher lower limit), a second quantile score higher than the middle quantile threshold (i.e., a score that is consistently higher than the threshold), and a third quantile score higher than the high quantile threshold (i.e., a higher upper limit) can be selected as target molecules.

[0166] By using predicted scores from multiple quantile models, a comprehensive assessment of molecules at different performance levels can be achieved. This multi-level assessment approach provides more comprehensive and detailed analytical results, helping to identify candidate molecules that perform well in various possible scenarios, thereby optimizing the screening process and improving the success rate and efficiency of drug development.

[0167] Step 110: Identify at least one target molecule in the molecular database according to the project scoring model.

[0168] In practical applications, the method of finding at least one target molecule in the molecular database based on the project scoring model can be as follows: obtain the molecular scores of each data in the molecular database excluding molecules in the target sample molecule set through the project scoring model, then sort the obtained molecular scores, and determine the molecules with higher molecular scores by a predetermined number as the target molecules.

[0169] Since the target molecules can be understood as molecules that are predicted to be closely related to the target project, the most promising candidate molecules can be screened and fed back, which can reduce the number of subsequent experiments in the laboratory.

[0170] By applying the scheme of the embodiments in this specification, multiple molecular sets are divided based on the molecular attribute information related to the target project. This increases the diversity between the molecular sets, thereby improving efficiency and accuracy in subsequent screening and analysis. Subsequently, a molecular set search tree is used to determine the molecular set weights of each set, and then these weights are used to determine the molecules used to train the scoring model. This reduces the number of molecular score acquisition steps when obtaining scores for training molecules, thus reducing the application cost of molecular screening. Furthermore, determining the molecules used to train the scoring model through molecular set weights reduces the number of invalid molecules while minimizing the application cost of molecular screening, thereby improving the performance of the trained scoring model. Ultimately, this approach increases the accuracy of molecular screening while reducing its application cost.

[0171] Corresponding to the above method embodiments, this specification also provides embodiments of a data filtering method applied to cloud devices, see [link to documentation]. Figure 4 , Figure 4 A flowchart is shown of a data filtering method for cloud devices according to an embodiment of this specification, which specifically includes the following steps.

[0172] Step 402: Receive a data filtering instruction sent by the terminal device, wherein the data filtering instruction includes at least one project molecular attribute information, a target project, and a set of verified molecules corresponding to the target project.

[0173] Step 404: Based on the molecular attribute information of each project and the verified molecular set, determine at least one molecular set in the molecular database, wherein the molecular set is a set of molecules that have not been verified for the target project.

[0174] Step 406: Obtain the set access count and set score corresponding to each set, and obtain the set weight corresponding to each set based on the set access count and set score. The set access count is the number of molecules verified in the process of determining the target sample set, and the set score is the number of target feature molecules hit by the corresponding set in the process of determining the target sample set. The target feature molecules are molecules related to the target project.

[0175] Step 408: Determine the target sample molecule set based on each molecule set and the corresponding molecule set weight, and generate a project scoring model for the target project based on the target sample molecule set.

[0176] Step 410: Determine at least one target molecule in the molecular database according to the project scoring model, and generate data screening results based on each target molecule.

[0177] Step 412: Send the data filtering results to the terminal device.

[0178] The above is an illustrative scheme of a data filtering method applied to a cloud device according to this embodiment. It should be noted that the technical solution of the data filtering method applied to the cloud device belongs to the same concept as the technical solution of the data filtering method described above. For details not described in detail in the technical solution of the data filtering method applied to the cloud device, please refer to the description of the technical solution of the data filtering method described above.

[0179] The scheme implemented in this specification involves obtaining molecular attribute information related to the target project from the validated molecule set for the target project included in the data filtering instruction sent by the terminal device. This allows for the division into multiple molecule sets, increasing the diversity between them and improving efficiency and accuracy in subsequent filtering and analysis. A molecular set search tree is then used to determine the molecular set weights of each set, followed by using these weights to determine the molecules used to train the scoring model. This reduces the number of molecule score acquisition steps, thus lowering the application cost of molecular filtering. Furthermore, determining the molecules used for training the scoring model by using molecular set weights reduces the number of invalid molecules while minimizing the application cost of molecular filtering, thereby improving the performance of the trained scoring model. This results in increased accuracy in molecular filtering while reducing application costs, ultimately improving the efficiency of the terminal device in performing further operations on the target project based on the filtered target molecules.

[0180] See Figure 5 , Figure 5 This specification illustrates an architecture diagram of a data filtering system provided in one embodiment of the specification. The data filtering system may include a client 100 and a server 200.

[0181] Client 100 is used to send data filtering instructions to server 200;

[0182] Server 200 is used to receive data filtering instructions, wherein the data filtering instructions include at least one item molecular attribute information, a target item, and a set of verified molecules corresponding to the target item; determine at least one molecular set in a molecular database based on the item molecular attribute information and the set of verified molecules, wherein the molecular set is a set of molecules that have not been verified for the target item; obtain the set access count and molecular set score corresponding to each molecular set, and obtain the molecular set weight corresponding to each molecular set based on the set access count and molecular set score, wherein the set access count is the number of molecules verified in the process of determining the target sample molecular set, and the molecular set score is the number of target feature molecules hit in the process of determining the target sample molecular set, and the target feature molecules are molecules related to the target item; determine the target sample molecular set according to each molecular set and the molecular set weight corresponding to each molecular set, and generate an item scoring model for the target item based on the target sample molecular set; determine at least one target molecule in the molecular database according to the item scoring model, and generate data filtering results based on each target molecule; and send the data filtering results to client 100.

[0183] Client 100 is also used to receive data filtering results sent by server 200.

[0184] By applying the scheme of the embodiments in this specification, the validated molecule set for the target project included in the data filtering instruction sent by the client is used to obtain project molecular attribute information related to the target project. This allows for the division of multiple molecule sets, increasing the diversity between them and improving efficiency and accuracy in subsequent filtering and analysis. Subsequently, a molecule set search tree is used to determine the molecule set weights of each set, and then these weights are used to determine the molecules used to train the scoring model. This reduces the number of molecule score acquisition steps when obtaining scores for training molecules, thus reducing the application cost of molecule filtering. Furthermore, determining the molecules used to train the scoring model by using molecule set weights reduces the number of invalid molecules while minimizing the application cost of molecule filtering, thereby improving the performance of the trained scoring model. This improves the accuracy of molecule filtering while reducing its application cost, ultimately increasing the efficiency of the client-side in performing further operations on the target project based on the filtered target molecules.

[0185] The data filtering system may include multiple clients 100 and a server 200. Clients 100 can be referred to as edge devices, and server 200 can be referred to as cloud devices. Multiple clients 100 can establish communication connections through server 200. In the data filtering scenario, server 200 is used to provide data filtering services among multiple clients 100. Each client 100 can act as a sender or receiver, communicating through server 200.

[0186] Users can interact with server 200 through client 100 to receive data sent by other clients 100, or send data to other clients 100, etc. In a data filtering scenario, users can publish data streams to server 200 through client 100, server 200 can generate data filtering results based on the data stream, and push the data filtering results to other clients that have established communication.

[0187] In this system, client 100 and server 200 establish a connection via a network. The network provides the medium for communication between client 100 and server 200. The network can include various connection types, such as wired or wireless communication links or fiber optic cables. Data transmitted by client 100 may need to undergo encoding, transcoding, compression, or other processing before being published to server 200.

[0188] Client 100 can be a browser, an app (application), a web application such as an H5 (HyperText Markup Language 5) application, a lightweight application (also known as a mini-program), or a cloud application. Client 100 can be developed based on the software development kit (SDK) of the corresponding service provided by server 200, such as a real-time communication (RTC) SDK. Client 100 can be deployed on electronic devices and depends on the device or certain apps on the device to run. Electronic devices may have displays and support information browsing, such as personal mobile terminals like mobile phones, tablets, and personal computers. Various other types of applications can also be configured on electronic devices, such as human-computer interaction applications, model training applications, text processing applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, and social media platform software.

[0189] Server 200 may include servers providing various services, such as servers providing communication services to multiple clients, servers supporting backend training of models used on clients, and servers processing data sent by clients. It should be noted that server 200 can be implemented as a distributed server cluster composed of multiple servers, or as a single server. The server can also be a server in a distributed system, or a server integrated with blockchain. The server can also be a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology.

[0190] It is worth noting that the data filtering method provided in the embodiments of this specification is generally executed by the server. However, in other embodiments of this specification, the client may also have similar functions to the server, thereby executing the data filtering method provided in the embodiments of this specification. In other embodiments, the data filtering method provided in the embodiments of this specification may also be executed jointly by the client and the server.

[0191] Combined with appendix Figure 6 Taking the application of the data screening method provided in this specification in drug molecule screening as an example, the data screening method will be further explained. Among other things, Figure 6 A flowchart illustrating the processing steps of a drug molecule screening method provided in one embodiment of this specification is shown, specifically including the following steps.

[0192] Step 602: Determine the 6-layer molecular set search tree constructed from the molecular attribute range information of each item corresponding to the molecular database.

[0193] It should be noted that each level in the molecular set search tree is split according to the molecular attribute information and the range information of one of the project's molecular attributes. The range information of the project's molecular attribute information is determined based on the statistical information of each molecule in the molecular library, which corresponds to six ranges of molecular attribute information for each project. In other words, the constructed molecular set search tree includes six ranges. 6 Leaf nodes.

[0194] Step 604: Based on the range of molecular attribute information corresponding to the six items in each leaf node of the molecular set search tree, divide the molecules in the molecular database into six categories. 6 Each set of molecules corresponds to its respective leaf node.

[0195] Step 606: Obtain the set of validated molecules corresponding to the target project.

[0196] Step 608: Determine the corresponding leaf node based on the project molecular attribute information of each molecule in the verified molecular set, and initialize the node score of each leaf node to the number of target feature molecules in the corresponding molecule in the verified molecular set, and initialize the node access count to the number of molecules in the corresponding molecule in the verified molecular set.

[0197] It should be noted that the target feature molecules are molecules related to the aforementioned target projects.

[0198] Step 610: Initialize the node score and node visit count of each node in the molecular set search tree (excluding leaf nodes) to the sum of the node scores and node visit counts of their corresponding child nodes, respectively.

[0199] Step 612: Aggregate the molecules in the molecular set corresponding to each leaf node according to the clustering molecular attribute information other than the project molecular attribute information, determine the molecules of each aggregation center as the search molecules corresponding to the molecular set of each leaf node, and combine each search molecule to form the search molecule set.

[0200] Step 614: Calculate the search value score of each node based on its corresponding node score, the number of times it is visited, and the number of times its parent node is visited. Calculate the search weight of each node based on the search value scores of its sibling nodes.

[0201] Step 616: The product of the search weight corresponding to each leaf node and the search weight of its corresponding ancestor node is taken as the search weight corresponding to each leaf node.

[0202] Step 618: Allocate 10,000 search slots to multiple leaf nodes according to the search weights corresponding to each leaf node, determine the search quantity corresponding to each leaf node, and determine the current search molecule with the corresponding search quantity in the molecule set to be searched in the molecule set corresponding to each leaf node as the current search molecule set.

[0203] Step 620: Determine whether each molecule in the current search molecule set is a target feature molecule, and increase the node score of each leaf node by the number of target feature molecules in each current search molecule. The node visit count is increased to the number of the corresponding current search molecules.

[0204] Step 622: Initialize the node score and node visit count of each node in the molecular set search tree (excluding leaf nodes) to the sum of the node scores and node visit counts of their corresponding child nodes, respectively.

[0205] Step 624: Remove each current search molecule from the set of molecules to be searched, add the current search molecule to the history of search molecules, and increment the iteration count by 1.

[0206] Step 626: Determine whether the number of target feature molecules in the historical search molecules has reached the preset threshold for the number of target feature molecules, or whether the number of iterations has exceeded the threshold for the number of iterations. If yes, proceed to step 628; otherwise, proceed to step 614.

[0207] Step 628: Calculate the weight of each leaf node based on its node score, the number of times it is visited, and the number of times its corresponding parent node is visited.

[0208] Step 630: Determine the target leaf node according to the weight of each leaf node, and determine the set of molecules in the molecular set corresponding to each target leaf node as the initial sample molecular set.

[0209] Step 632: Obtain the molecular score of each molecule in the initial sample molecule set for the target project, and combine the molecules corresponding to the above-mentioned verified molecule set, historical search molecule set and initial sample molecule set to form the target sample molecule set.

[0210] Step 634: Use each molecule in the target sample molecule set and its corresponding molecule score as training samples to train and obtain a scoring model.

[0211] Step 636: Use a scoring model to predict the scores of molecules in the molecular library other than the target sample molecule set mentioned above, and obtain the scores of each molecule.

[0212] Step 638: Determine the target molecules based on the scores of each molecule, and use each target molecule and the set of target sample molecules as the molecular screening results.

[0213] By applying the scheme of the embodiments in this specification, multiple molecular sets are divided based on molecular attribute information related to the target drug. This increases the diversity between the molecular sets, thereby improving efficiency and accuracy in subsequent screening and analysis. Subsequently, a molecular set search tree is used to determine the molecular set weights of each set, and then these weights are used to determine the molecules used to train the scoring model. This reduces the number of molecule score acquisition steps when obtaining scores for training molecules, thus reducing the application cost of molecular screening. Furthermore, determining the molecules used to train the scoring model through molecular set weights reduces the number of invalid molecules while minimizing the application cost of molecular screening, thereby improving the performance of the trained scoring model. Ultimately, this improves the accuracy of molecular screening for the target drug while reducing the application cost of molecular screening.

[0214] Corresponding to the above method embodiments, this specification also provides embodiments of a data filtering device. Figure 7 A schematic diagram of a data filtering device according to one embodiment of this specification is shown. Figure 7 As shown, the device includes:

[0215] The first acquisition module 702 is configured to acquire at least one project molecular attribute information and a set of verified molecules corresponding to the target project.

[0216] The first determining module 704 is configured to determine at least one molecular set in the molecular database based on the molecular attribute information of each project and the verified molecular set, wherein the molecular set is a set of molecules that have not been verified for the target project;

[0217] The second acquisition module 706 is configured to acquire the set access count and set score corresponding to each set of molecules, and to obtain the set weight corresponding to each set of molecules based on the set access count and set score. The set access count is the number of molecules verified in the process of determining the target sample set of molecules, and the set score is the number of target feature molecules hit by the access of the corresponding set of molecules in the process of determining the target sample set of molecules. The target feature molecules are molecules related to the target item.

[0218] The third acquisition module 708 is configured to determine the target sample molecule set based on each molecule set and the molecule set weights corresponding to each molecule set, and generate a project scoring model for the target project based on the target sample molecule set.

[0219] The second determining module 710 is configured to determine at least one target molecule in the molecular database according to the project scoring model.

[0220] Optionally, the first determining module 704 is further configured to:

[0221] A first molecular database is obtained based on the verified molecule set, wherein the first molecular database is a molecular database composed of molecules other than those in the verified molecule set.

[0222] Determine at least two project molecular attribute ranges corresponding to each project's molecular attribute information, where the project molecular attribute range information is the value range of the project molecular attribute information;

[0223] Obtain the molecular project attribute values ​​corresponding to the molecular attribute information of each molecule in the first molecular database for each project;

[0224] Based on the molecular attribute range information of each project and the molecular project attribute values ​​corresponding to each molecule in the first molecular database, at least one molecular set is determined.

[0225] Optionally, the second acquisition module 706 is further configured to:

[0226] Obtain the target molecular set search tree corresponding to the target project, wherein the target molecular set search tree includes a root node, at least two child nodes and at least two leaf nodes, the root node is the starting point of the molecular set search tree, the child nodes are nodes derived from the root node or other child nodes according to the project molecular attribute information, and the leaf nodes are the end nodes of the search tree. The root node represents the starting node for determining the molecular set corresponding to the target project, the child nodes represent the decision results based on the project molecular attribute range information, and the leaf nodes represent the molecular set corresponding to the decision results based on the decision results of the molecular attribute range information of each of their respective projects.

[0227] The number of times each molecular set is accessed and the molecular set score are determined based on the node score and the number of times each leaf node and its corresponding parent node are accessed in the search tree of the target molecular set. The node score is the number of target feature molecules that are accessed and hit in the molecular set corresponding to the node.

[0228] Optionally, the second acquisition module 706 is further configured to:

[0229] Determine at least two project molecular attribute ranges corresponding to each project's molecular attribute information, where the project molecular attribute range information is the value range of the project molecular attribute information;

[0230] Obtain the initial molecular set search tree corresponding to the molecular attribute range information of each project, and initialize the initial molecular set search tree according to the verified molecular set;

[0231] The set of molecules to be searched is determined based on each set of molecules, and the node score and the number of times each node is visited are obtained in the initial molecular set search tree.

[0232] Repeat the following steps until the search tree iteration stops: Based on the node score and node visit count of each node in the initial molecular set search tree, determine at least one current search molecule in the set of molecules to be searched; update the node score and node visit count of the leaf nodes and ancestor nodes corresponding to each current search molecule; update the set of molecules to be searched based on the current search molecule; wherein, the initial molecular set search tree obtained when the search tree iteration stops is reached is the target molecular set spanning tree.

[0233] Optionally, the second acquisition module 706 is further configured to:

[0234] The child nodes corresponding to the molecular set search tree are determined based on the molecular attribute range information of each project.

[0235] The node score and node visit count of each node in the molecular set search tree are initialized based on the verified molecular set.

[0236] Optionally, the second acquisition module 706 is further configured to:

[0237] Obtain at least one cluster molecular attribute information, wherein the cluster molecular attribute information is molecular attribute information other than the molecular attribute information of each item;

[0238] Based on the molecular attribute information of each cluster, at least one molecule to be searched is determined for each molecular set, wherein the molecular set to be searched includes at least one molecule to be searched.

[0239] Optionally, the second acquisition module 706 is further configured to:

[0240] Based on the node score and node visit count of each node in the molecular set search tree, determine the current search weight corresponding to each leaf node;

[0241] Based on the current search weight corresponding to each leaf node, at least one current search leaf node is determined;

[0242] At least one current search molecule is determined from among the search molecules corresponding to each current search leaf node.

[0243] Optionally, the second acquisition module 706 is further configured to:

[0244] Determine the target current search molecule, wherein the target current search molecule is any one of the current search molecules;

[0245] Increment the node visit count of the leaf node and each ancestor node corresponding to the current search molecule of the target by 1;

[0246] If the target current search molecule is a target feature molecule, increment the node score of the leaf node and each ancestor node corresponding to the target current search molecule by 1.

[0247] Optionally, the second acquisition module 706 is further configured to:

[0248] The preset number of iteration rounds has been reached; and / or,

[0249] The number of target feature molecules in the historical search molecule set reaches a preset threshold, wherein the historical search molecule set is obtained based on the current search molecules in each iteration round.

[0250] Optionally, the third acquisition module 708 is further configured to:

[0251] Based on the molecular set weights corresponding to each molecular set, at least one molecular set is determined as the initial sample molecular set, wherein the initial sample molecular set is a set of molecules used to obtain the project scoring model that has not yet been validated.

[0252] A reference sample molecule set is determined based on the verified molecule set and the historical search molecule set. The historical search molecule set is obtained based on the current search molecules in each iteration of the molecule set search tree. The reference sample molecule set is a set of molecules that have been verified and are used to obtain the project scoring model.

[0253] The target sample molecule set is determined based on the initial sample molecule set and the reference sample molecule set.

[0254] Optionally, the third acquisition module 708 is further configured to:

[0255] Obtain the sample molecule score corresponding to the target item in the target sample molecule set;

[0256] Based on each sample molecule and its corresponding sample molecule score, a project scoring model is obtained for the target project.

[0257] Optionally, the project scoring model includes at least one quantile scoring model;

[0258] The second determining module 710 is further configured to:

[0259] A second molecular database is determined based on the target sample molecule set, wherein the second molecular database is a molecular database composed of molecules other than those in the target sample molecule set.

[0260] Based on the quantile scoring model, at least one quantile score corresponding to each molecule in the second molecular database is obtained.

[0261] Based on the quantile fractions of each molecule in the second molecular database, at least one target molecule is determined in the second molecular database.

[0262] By applying the scheme of the embodiments in this specification, multiple molecular sets are divided based on the molecular attribute information related to the target project. This increases the diversity between the molecular sets, thereby improving efficiency and accuracy in subsequent screening and analysis. Subsequently, a molecular set search tree is used to determine the molecular set weights of each set, and then these weights are used to determine the molecules used to train the scoring model. This reduces the number of molecular score acquisition steps when obtaining scores for training molecules, thus reducing the application cost of molecular screening. Furthermore, determining the molecules used to train the scoring model through molecular set weights reduces the number of invalid molecules while minimizing the application cost of molecular screening, thereby improving the performance of the trained scoring model. Ultimately, this approach increases the accuracy of molecular screening while reducing its application cost.

[0263] The above is an illustrative scheme of a data filtering device according to this embodiment. It should be noted that the technical solution of this data filtering device and the technical solution of the data filtering method described above belong to the same concept. For details not described in detail in the technical solution of the data filtering device, please refer to the description of the technical solution of the data filtering method described above.

[0264] Figure 8 A structural block diagram of a computing device 800 according to one embodiment of this specification is shown. The components of the computing device 800 include, but are not limited to, a memory 810 and a processor 820. The processor 820 is connected to the memory 810 via a bus 830, and a database 850 is used to store data.

[0265] The computing device 800 also includes an access device 840, which enables the computing device 800 to communicate via one or more networks 860. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 840 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0266] In one embodiment of this specification, the above-described components of the computing device 800 and Figure 8 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 8 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0267] The computing device 800 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 800 can also be a mobile or stationary server.

[0268] The processor 820 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the aforementioned data filtering method and the data filtering method applied to cloud devices.

[0269] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device belongs to the same concept as the data filtering method and the data filtering method applied to cloud devices described above. For details not described in detail in the technical solution of the computing device, please refer to the description of the data filtering method and the data filtering method applied to cloud devices described above.

[0270] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described data filtering method and the data filtering method applied to a cloud device.

[0271] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium belongs to the same concept as the data filtering method and the data filtering method applied to cloud devices described above. Details not described in detail in the technical solution of the storage medium can be found in the descriptions of the data filtering method and the data filtering method applied to cloud devices described above.

[0272] An embodiment of this specification also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the above-described data filtering method and the steps of the data filtering method applied to a cloud device.

[0273] The above is an illustrative example of a computer program according to this embodiment. It should be noted that the technical solution of this computer program belongs to the same concept as the data filtering method and the data filtering method applied to cloud devices described above. Details not described in detail in the computer program's technical solution can be found in the descriptions of the data filtering method and the data filtering method applied to cloud devices described above.

[0274] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0275] The computer instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0276] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0277] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0278] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A data filtering method, comprising: Obtain at least one project's molecular attribute information and the set of verified molecules corresponding to the target project; Based on the molecular attribute information of each project and the verified molecular set, at least one molecular set is determined in the molecular database, wherein the molecular set is a set of molecules that have not been verified for the target project; Obtain the set access count and set score for each set, and obtain the set weight for each set based on the set access count and set score. The set access count is the number of molecules that are checked in the process of determining the target sample set, and the set score is the number of target feature molecules that are hit by the corresponding set during the process of determining the target sample set. The target feature molecules are molecules related to the target item. Based on each molecular set and its corresponding molecular set weight, a target sample molecular set is determined, and a project scoring model for the target project is generated based on the target sample molecular set. At least one target molecule is identified in the molecular database based on the project scoring model.

2. The method of claim 1, wherein at least one molecular set is determined in the molecular database based on the molecular attribute information of each project and the verified molecular set, comprising: A first molecular database is obtained based on the verified molecule set, wherein the first molecular database is a molecular database composed of molecules other than those in the verified molecule set. Determine at least two project molecular attribute ranges corresponding to each project's molecular attribute information, where the project molecular attribute range information is the value range of the project molecular attribute information; Obtain the molecular project attribute values ​​corresponding to the molecular attribute information of each molecule in the first molecular database for each project; Based on the molecular attribute range information of each project and the molecular project attribute values ​​corresponding to each molecule in the first molecular database, at least one molecular set is determined.

3. The method as described in claim 1, wherein obtaining the set access count and set score corresponding to each set includes: Obtain the target molecular set search tree corresponding to the target project, wherein the target molecular set search tree includes a root node, at least two child nodes and at least two leaf nodes, the root node is the starting point of the molecular set search tree, the child nodes are nodes derived from the root node or other child nodes according to the project molecular attribute information, and the leaf nodes are the end nodes of the search tree. The root node represents the starting node for determining the molecular set corresponding to the target project, the child nodes represent the decision results based on the project molecular attribute range information, and the leaf nodes represent the molecular set corresponding to the decision results based on the decision results of the molecular attribute range information of each of their respective projects. The number of times each molecular set is accessed and the molecular set score are determined based on the node score and the number of times each leaf node and its corresponding parent node are accessed in the search tree of the target molecular set. The node score is the number of target feature molecules that are accessed and hit in the molecular set corresponding to the node.

4. The method as described in claim 3, wherein obtaining the target molecule set search tree corresponding to the target project includes: Determine at least two project molecular attribute ranges corresponding to each project's molecular attribute information, where the project molecular attribute range information is the value range of the project molecular attribute information; Obtain the initial molecular set search tree corresponding to the molecular attribute range information of each project, and initialize the initial molecular set search tree according to the verified molecular set; The set of molecules to be searched is determined based on each set of molecules, and the node score and the number of times the node is visited are obtained in the search tree of the initial set of molecules. Repeat the following steps until the search tree iteration stops: Based on the node score and node visit count of each node in the initial molecular set search tree, determine at least one current search molecule in the set of molecules to be searched; update the node score and node visit count of the leaf nodes and ancestor nodes corresponding to each current search molecule; update the set of molecules to be searched based on the current search molecule; wherein, the initial molecular set search tree obtained when the search tree iteration stops is reached is the target molecular set search tree.

5. The method as described in claim 4, obtaining an initial molecular set search tree corresponding to the molecular attribute range information of each project, and initializing the initial molecular set search tree according to the verified molecular set, includes: The child nodes corresponding to the molecular set search tree are determined based on the molecular attribute range information of each project. The node score and node visit count of each node in the molecular set search tree are initialized based on the verified molecular set.

6. The method of claim 4, wherein determining the set of molecules to be searched based on each set of molecules includes: Obtain at least one cluster molecular attribute information, wherein the cluster molecular attribute information is molecular attribute information other than the molecular attribute information of each item; Based on the molecular attribute information of each cluster, at least one molecule to be searched is determined for each molecular set, wherein the molecular set to be searched includes at least one molecule to be searched.

7. The method as described in claim 4, wherein determining at least one current search molecule in the set of molecules to be searched based on the node score and node visit count corresponding to each node in the initial molecule set search tree, includes: Based on the node score and node visit count of each node in the molecular set search tree, determine the current search weight corresponding to each leaf node; Based on the current search weight corresponding to each leaf node, at least one current search leaf node is determined; At least one current search molecule is determined from among the search molecules corresponding to each current search leaf node.

8. The method as described in claim 4, wherein updating the node score and node visit count of the leaf node corresponding to each current search molecule and the ancestor node corresponding to each leaf node based on each current search molecule includes: Determine the target current search molecule, wherein the target current search molecule is any one of the current search molecules; Increment the node visit count of the leaf node and each ancestor node corresponding to the current target search molecule by 1; If the target current search molecule is a target feature molecule, increment the node score of the leaf node and each ancestor node corresponding to the target current search molecule by 1.

9. The method of claim 4, wherein the search tree iteration stopping condition includes: The preset number of iteration rounds has been reached; and / or, The number of target feature molecules in the historical search molecule set reaches a preset threshold, wherein the historical search molecule set is obtained based on the current search molecules in each iteration round.

10. The method according to any one of claims 1-9, wherein determining the target sample molecule set based on each molecule set and the molecule set weights corresponding to each molecule set includes: Based on the molecular set weights corresponding to each molecular set, at least one molecular set is determined as the initial sample molecular set, wherein the initial sample molecular set is a set of molecules used to obtain the project scoring model that has not yet been validated. A reference sample molecule set is determined based on the verified molecule set and the historical search molecule set. The historical search molecule set is obtained based on the current search molecules in each iteration of the molecule set search tree. The reference sample molecule set is a set of molecules that have been verified and are used to obtain the project scoring model. The target sample molecule set is determined based on the initial sample molecule set and the reference sample molecule set.

11. The method according to any one of claims 1-9, wherein generating a project scoring model for the target project based on the target sample molecule set, comprising: Obtain the sample molecule score corresponding to the target item in the target sample molecule set; Based on each sample molecule and its corresponding sample molecule score, a project scoring model is obtained for the target project.

12. The method of claim 11, wherein the project scoring model comprises at least one quantile scoring model; Based on the project scoring model, at least one target molecule is identified in the molecular database, including: A second molecular database is determined based on the target sample molecule set, wherein the second molecular database is a molecular database composed of molecules other than those in the target sample molecule set. Based on the quantile scoring model, at least one quantile score corresponding to each molecule in the second molecular database is obtained. Based on the quantile fractions of each molecule in the second molecular database, at least one target molecule is determined in the second molecular database.

13. A data filtering method applied to cloud devices, comprising: The device receives a data filtering instruction sent by a terminal device, wherein the data filtering instruction includes at least one project molecular attribute information, a target project, and a set of verified molecules corresponding to the target project; Based on the molecular attribute information of each project and the verified molecular set, at least one molecular set is determined in the molecular database, wherein the molecular set is a set of molecules that have not been verified for the target project; Obtain the set access count and set score for each set, and obtain the set weight for each set based on the set access count and set score. The set access count is the number of molecules that are checked in the process of determining the target sample set, and the set score is the number of target feature molecules that are hit by the corresponding set during the process of determining the target sample set. The target feature molecules are molecules related to the target item. Based on each molecular set and its corresponding molecular set weight, a target sample molecular set is determined, and a project scoring model for the target project is generated based on the target sample molecular set. Based on the project scoring model, at least one target molecule is identified in the molecular database, and data screening results are generated based on each target molecule. The data filtering results are sent to the terminal device.

14. A computing device, comprising: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1-13.

15. A computer-readable storage medium storing a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-13.

16. A computer program product comprising a computer program / instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-13.

Images