Protein model quality assessment method and apparatus, computer device, and storage medium

Abstract

Embodiments of the present application relate to the technical field of artificial intelligence, and in particular to a protein model quality assessment method and apparatus, a computer device, and a storage medium. The method comprises: receiving a PDB file inputted by a user terminal, wherein the PDB file comprises features of each atom in protein and RNA structure scoring; constructing an Atom-ProteinQA model on the basis of the features of each atom in the protein and RNA structure scoring and a preset algorithm; receiving a quality assessment request sent by the user terminal, wherein the quality assessment request at least comprises protein structure data to be assessed; and inputting said protein structure data into the Atom-ProteinQA model to perform a quality assessment operation, to obtain a quality assessment result. In the present application, by using atomic-level LDDT scores from the protein and RNA structure scoring as supervision, accurate atomic-level IDDL score prediction is effectively performed, thereby allowing for atomic-level assessment of structural changes in the protein and RNA structure scoring.

Description

Protein model quality assessment method and device, computer device and storage medium TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, and in particular to a protein model quality assessment method and device, a computer device and a storage medium. BACKGROUND

[0002] Proteins are ubiquitous in almost all biological processes. Determining their structure and function helps to understand and potentially control these processes. However, although the determination of protein sequences is now a routine procedure, it is often very difficult to extract relevant functional knowledge of the system under study using this information. Indeed, the function of a protein depends on the combination of its chemical and mechanical properties, which are determined by its structure. Therefore, it is very important to identify the protein structure from its sequence, although it is a difficult task. Experimental structure identification is not feasible in all cases and is often very cumbersome and expensive. Therefore, in the past, computational methods have emerged that attempt to predict protein structures from their sequences. Most of these methods combine a sampling step of protein conformations, which produces protein conformations, with a model quality assessment step, which scores these conformations to select the protein conformation that is as close as possible to the native structure. Protein model quality assessment is an important and yet unsolved problem in structural bioinformatics.

[0003] There is a protein model quality assessment method that solves the scoring problem by using energy functions derived from general physical principles, such as CHARMM, AMBER, OPLS, GROMOS, derived from different known protein and RNA structure scoring sets, such as GOAP, RW, DFIRE, or both, such as Rosetta. They usually estimate a specific similarity score between the calculated model and the reference structure, such as LDDT, CAD, GDT, TMscore, etc. Recently, various learning-based methods have been developed. They make similarity predictions by using artificial neural networks, support vector machines, CNNs, LSTMs, GCNs, and 3D models.

[0004] However, the applicant has found that they cannot all predict atomic-level ProteinQA results because atomic inference for scoring all protein and RNA structures is very intensive computation. In addition, they cannot consider the topological and geometric relationships between atoms at the same time, resulting in the problem of low evaluation accuracy of traditional protein model quality assessment methods. SUMMARY

[0005] The embodiment of the application aims to provide a protein model quality evaluation method and device, computer equipment and a storage medium, to solve the problem of low evaluation accuracy of the traditional protein model quality evaluation method.

[0006] To solve the above technical problems, the embodiment of the application provides a protein model quality evaluation method, which adopts the technical scheme as follows:

[0007] Receiving the PDB file input by the user terminal, wherein the PDB file contains the characteristics of each atom in the protein and RNA structure scoring;

[0008] According to the characteristics of each atom in the protein and RNA structure scoring and the preset algorithm, an Atom-ProteinQA model is constructed;

[0009] Receiving the quality evaluation request sent by the user terminal, wherein the quality evaluation request at least includes the protein structure data to be evaluated;

[0010] Inputting the protein structure data into the Atom-ProteinQA model for quality evaluation operation to obtain the quality evaluation result.

[0011] Further, the preset algorithm first searches the neighbors of P n [i,j] with a radius of P n , then selects the adjacent atoms in P m and P n through a preset index, and calculates the respective distances of the adjacent atoms to P m [i,j] and P n [i,j], respectively. The difference between the single distances in P m and P n will be converted into a predefined score through a mapping function , and finally, the LDDT score on the atomic, residue and global levels is obtained by normalizing the total number of neighbors.

[0012] Further, the Atom-ProteinQA model includes a geometry perception module, and the step of inputting the protein structure data into the Atom-ProteinQA model for quality evaluation operation to obtain the quality evaluation result specifically includes the following steps:

[0013] Performing feature extraction operation on the protein structure data to obtain the voxelized Cartesian coordinates of each residue atom in the protein structure data;

[0014] According to the geometry perception module, the geometric features of the voxelized Cartesian coordinates are captured to generate a first atomic level prediction result.

[0015] Further, the Atom-ProteinQA model comprises a topology perception module, and the step of inputting the protein structure data into the Atom-ProteinQA model for quality evaluation operation to obtain a quality evaluation result specifically comprises the following steps:

[0016] According to the topology perception module, an atomic-level topology graph corresponding to the protein structure data is constructed;

[0017] According to the atomic-level topology graph and a message passing mechanism, a second atomic-level prediction result is output.

[0018] Further, the Atom-ProteinQA model comprises a geometry perception module, a topology perception module and a cross-model fusion module, and the step of inputting the protein structure data into the Atom-ProteinQA model for quality evaluation operation to obtain a quality evaluation result specifically comprises the following steps:

[0019] According to the geometry perception module, atomic feature data of the protein structure data is obtained;

[0020] According to the topology perception module, an initial topology graph of the protein structure data is obtained;

[0021] According to the cross-model fusion module, feature fusion operation is performed on the atomic features and the topology graph to obtain an updated topology graph;

[0022] According to the topology perception module and a message passing mechanism, quality prediction operation is performed on the updated topology graph to obtain a third atomic-level prediction result.

[0023] To solve the above technical problems, the embodiments of the present application also provide a protein model quality evaluation device, which adopts the following technical solutions:

[0024] A file acquisition module is configured to receive a PDB file input by a user terminal, wherein the PDB file contains features of each atom in protein and RNA structure scoring;

[0025] A model construction module is configured to construct an Atom-ProteinQA model according to the features of each atom in protein and RNA structure scoring and a preset algorithm;

[0026] A request acquisition module is configured to receive a quality evaluation request sent by a user terminal, wherein the quality evaluation request at least includes protein structure data to be evaluated;

[0027] a quality assessment module, configured to input the protein structure data into the Atom-Protein QA model to perform a quality assessment operation, and obtain a quality assessment result.

[0028] Further, the Atom-Protein QA model comprises a geometry perception module, and the quality assessment module comprises:

[0029] a feature extraction submodule, configured to perform a feature extraction operation on the protein structure data, and obtain voxelized Cartesian coordinates of each residue atom in the protein structure data;

[0030] a first prediction submodule, configured to generate a first atomic-level prediction result according to the geometry perception module capturing geometric features of the voxelized Cartesian coordinates.

[0031] Further, the Atom-Protein QA model comprises a topology perception module, and the quality assessment module comprises:

[0032] a topology graph construction submodule, configured to construct an atomic-level topology graph corresponding to the protein structure data according to the topology perception module;

[0033] a second prediction submodule, configured to output a second atomic-level prediction result according to the atomic-level topology graph and a message passing mechanism.

[0034] Further, the Atom-Protein QA model comprises a geometry perception module, and the calculation module comprises:

[0035] an atomic feature acquisition submodule, configured to acquire atomic feature data of the protein structure data according to the geometry perception module;

[0036] an initial topology graph acquisition submodule, configured to acquire an initial topology graph of the protein structure data according to the topology perception module;

[0037] a feature fusion submodule, configured to perform a feature fusion operation on the atomic features and the topology graph according to a cross-model fusion module, and obtain an updated topology graph;

[0038] a third prediction submodule, configured to perform a quality prediction operation on the updated topology graph according to the topology perception module and a message passing mechanism, and obtain a third atomic-level prediction result.

[0039] To solve the above technical problems, the embodiments of the present application further provide a computer device, which adopts the technical scheme as follows:

[0040] The protein model quality evaluation method comprises the following steps: receiving a PDB file input by a user terminal, wherein the PDB file contains the characteristics of each atom in protein and RNA structure scoring; constructing an Atom-ProteinQA model according to the characteristics of each atom in the protein and RNA structure scoring and a preset algorithm; receiving a quality evaluation request sent by the user terminal, wherein the quality evaluation request at least comprises protein structure data to be evaluated; inputting the protein structure data into the Atom-ProteinQA model for quality evaluation operation to obtain a quality evaluation result.

[0041] To solve the above technical problems, the embodiments of the present application also provide a computer readable storage medium, which adopts the technical scheme as follows:

[0042] The computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by a processor to implement the steps of the protein model quality evaluation method.

[0043] The present application provides a protein model quality evaluation method, comprising: receiving a PDB file input by a user terminal, wherein the PDB file contains the characteristics of each atom in protein and RNA structure scoring; constructing an Atom-ProteinQA model according to the characteristics of each atom in the protein and RNA structure scoring and a preset algorithm; receiving a quality evaluation request sent by the user terminal, wherein the quality evaluation request at least comprises protein structure data to be evaluated; inputting the protein structure data into the Atom-ProteinQA model for quality evaluation operation to obtain a quality evaluation result. Compared with the prior art, the present application uses the LDDT score of the atom level of the protein and RNA structure scoring as supervision, effectively performs accurate atom level IDDL score prediction, and evaluates the structure change of the protein and RNA structure scoring at an atom level. BRIEF DESCRIPTION OF DRAWINGS

[0044] In order to more clearly illustrate the schemes in the present application, the drawings needed in the description of the embodiments of the present application will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0045] Fig. 1 is an exemplary system architecture diagram to which the present application can be applied;

[0046] Fig. 2 is an implementation flowchart of the protein model quality evaluation method provided by the embodiments of the present application;

[0047] Fig. 3 is a schematic diagram of one embodiment of the Atom-ProteinQA model architecture provided by the embodiments of the present application;

[0048] Fig. 4 is a schematic diagram of one embodiment of the algorithm structure provided by the embodiments of the present application;

[0049] Fig. 5 is a schematic diagram of one embodiment of the model qualitative result comparison provided by the embodiments of the present application;

[0050] Figure 6 is a schematic diagram of the protein model quality assessment device provided in an embodiment of this application;

[0051] Figure 7 is a schematic diagram of the structure of an embodiment of a computer device according to this application. Detailed Implementation

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0053] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0054] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0055] As shown in Figure 1, the system architecture 100 may include a terminal device 101, a network 102, and a server 103. The terminal device 101 may be a laptop computer 1011, a tablet computer 1012, or a mobile phone 1013. The network 102 is used as a medium to provide a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0056] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.

[0057] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.

[0058] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.

[0059] It should be noted that the protein model quality assessment method provided in this application embodiment is generally executed by a server / terminal device, and correspondingly, the protein model quality assessment device is generally set in the server / terminal device.

[0060] It should be understood that the number of terminal devices, networks, and servers shown in Figure 1 is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0061] Referring again to Figure 2, a flowchart of an embodiment of the protein model quality assessment method according to this application is shown. The protein model quality assessment method includes steps S201, S202, S203, and S204.

[0062] In step S201, a PDB file input from a user terminal is received, wherein the PDB file contains the characteristics of each atom in the protein and RNA structure scoring;

[0063] In step S202, an Atom-ProteinQA model is constructed based on the characteristics of each atom in the protein and RNA structure scoring and a preset algorithm.

[0064] In step S203, a quality assessment request sent by a user terminal is received, wherein the quality assessment request includes at least the protein structure data to be assessed.

[0065] In step S204, the protein structure data is input into the Atom-ProteinQA model for quality assessment to obtain the quality assessment results.

[0066] In the embodiments of this application, in the model of this application, a complete set of protein and RNA structure scoring atoms is used. It is represented by three-dimensional coordinates and K-dimensional features. The K-dimensional features are the one-heat amino acid types of each atom. The goal of the Atom-ProteinQA model is to predict the global score (S) of a given bait protein. G ), shape is residue score (S) R ) and shape as Atomic score (S) A S R The score consists of M residues, S, representing the protein and RNA structure scores. A This includes N atom scores for protein and RNA structure scoring. Predicted scores, ranging from 0 to 1, measure the similarity between the decoy and the original structure. A higher quality protein and RNA structure scoring model is indicated by scores close to 1 (global score, residue score, and atom score). Quality decreases as scores approach 0. This can be described as follows:

[0067] As shown in Figure 3, the atomic-level ground truth is generated as supervision for atomic-level prediction. The first set of input features to the network is the voxelized Cartesian coordinates of each residue atom. The GP module takes the entire atom of the protein and RNA structure scores as input and generates atomic-level predictions. It focuses on capturing the geometric information of the protein and RNA structure scores and outputs fine-grained atomic features. In contrast, the TP module considers the topological relationships between the protein and RNA structure scores atoms, uses chemical bonds as edges to construct undirected graphs, and leverages message passing to aggregate atom neighborhood features. Cross-model feature fusion is performed between these two modules to enhance mutual representation. Finally, by combining the above components with multi-level supervision, Atom-ProteinQA can make accurate predictions at both the residue and atomic levels simultaneously.

[0068] During the training phase, atomic-level LDDT scores are calculated using native structures as the supervised objects. In the GP module, U-Net features based on sparse convolutions are updated and fed into the Topology Awareness (TP) module. For each protein and RNA structure score, a topology graph is constructed using the chemical bonds and distances within a threshold of the protein and RNA structure scores. Subsequently, the TP module is applied to update the graph features via message passing, thereby obtaining the final atomic-level predictions.

[0069] In some optional implementations of the embodiments of this application, the preset algorithm first searches for a radius of P. n Atom P n The neighbors of [i,j] are then selected by P using a preset index. m and P n The adjacent atoms in the matrix, and calculate the distance from each adjacent atom to P. m [i,j] and P nThe distances between [i,j], P m and P n The difference between individual distances will be determined by a mapping function. The scores are converted to predefined scores, and finally, LDDT scores at the atom, residue, and global levels are obtained by normalizing the total number of neighbors.

[0070] In this embodiment, we directly generate atomic-level labels as ground-based real-world supervision. In previous work, LDDT metrics have been most widely used for evaluating local accuracy. Specifically, it is calculated on all atomic pairs closer to a predefined threshold in the reference structure. Note that there is no open-source code for LDDT calculation, and Mariani et al.

[10] only provided an online server to generate residual-level LDDT scores. Therefore, we first input a PDB file, and the output contains features for each atom in the protein and RNA structure scoring. These features include 21 residue types (A, R, N, D, C, Q, E, G, H, I, L, K, M, F, P, S, T, W, Y, V, and unknown), 5 atom types (C, N, O, S, and unknown), and three coordinates for each atom. Subsequently, we adapted the original algorithm outlined in their paper, modifying it to generate atomic-level LDDT. As shown in Figure 4, in this algorithm, we use P... m , This represents the atomic coordinate set and its native structure for protein and RNA structure scoring models. Each protein and RNA structure score contains M residues, where the m-th residue includes N. m atom This application uses P[i,j] to represent the j-th atom of the i-th residue in the protein and RNA structural score P. More specifically, the algorithm first searches for a radius of P. n Atom P n The neighbors of [i,j] are selected, and the indices of these neighbors are preserved. Then, P is selected using a preset index. m and P n Adjacent atoms in the matrix, and calculate their values ​​to P. m [i,j] and P n The distances between [i,j] are calculated. Then, P... m and P n The difference between individual distances will be determined by a mapping function. This is converted to a predefined score. Finally, by normalizing the total number of neighbors, this application can obtain LDDT scores (i.e., ...) at the atom, residue, and global levels. and ).

[0071] This application provides a protein model quality assessment method, comprising: receiving a PDB file input from a user terminal, wherein the PDB file contains features of each atom in the protein and RNA structure scoring; constructing an Atom-ProteinQA model based on the features of each atom in the protein and RNA structure scoring and a preset algorithm; receiving a quality assessment request sent by the user terminal, wherein the quality assessment request includes at least the protein structure data to be assessed; inputting the protein structure data into the Atom-ProteinQA model for quality assessment, and obtaining a quality assessment result. Compared with the prior art, this application utilizes the LDDL scores at the atomic level of protein and RNA structure scoring as supervision, effectively performing accurate atomic-level IDDL score prediction, and assessing structural changes in protein and RNA structure scoring at an atomic level.

[0072] In some optional implementations of the embodiments of this application, the Atom-ProteinQA model includes a geometry sensing module. The step of inputting protein structure data into the Atom-ProteinQA model for quality assessment and obtaining the quality assessment result specifically includes the following steps:

[0073] Feature extraction is performed on the protein structure data to obtain the voxelized Cartesian coordinates of each residue atom in the protein structure data;

[0074] The first atomic-level prediction result is generated by capturing the geometric features of the voxelized Cartesian coordinates using the geometry perception module.

[0075] In this application embodiment, the geometry-aware (GP) module aims to capture geometric metric information of protein and RNA structure scoring structures. The protein and RNA structure scoring structures are first represented as 3D point clouds based on their atomic coordinates. Previous works have failed to learn effective geometric representations because they utilize conventional 3D-CNNs to extract features. However, normal voxelization in 3D space typically produces a large proportion of empty voxels, and convolving these empty voxels incurs a significant computational burden. Therefore, the resolution of the 3D voxels used in these methods is generally low, leading to the loss of geometric details in the protein and RNA structure scoring structures. In this paper, this application innovatively utilizes 3D sparse convolution to address the above problem. This application transforms the coordinates of the original atoms (i.e., the point cloud) into a sparse volumetric representation. Specifically, the 3D point cloud can be represented as x = {x k}={(p k ,f k The format is represented as )}, where p k f is the 3D coordinate of the k-th point. kThese are its corresponding features. First, this application moves all points to a local coordinate system with the geometric center as the origin. Then, this application rotates these points along randomly selected axes (x, y, or z axes) at random angles α ∈ [0, π). Point features {f k} remains unchanged; this application represents the transformed coordinates as follows: Subsequently, this application converts the transformed point cloud into a sparse voxel representation with a resolution of r:

[0076] in It is a rounding operation. yes Whether it belongs to the m-th voxel grid is a binary index. N m This is the number of points in the m-th voxel. This application averages the characteristics of these points. Only non-empty voxels (N) are retained. m These voxels (>0) are stored in a hash table. Then, the convolution operation is performed only on these non-empty voxels. In this way, this application can represent point clouds at a greater volumetric resolution while maintaining computational efficiency. The difference between traditional voxels using 3D-CNN and sparse voxels with sparse convolutions is illustrated.

[0077] In this embodiment, sparse representation effectively reduces memory usage and computation, which is particularly advantageous when handling higher resolutions and large point clouds. Finally, sparse voxels are fed into a sparse convolution-based U-Net, and the results are transformed back to the atomic level using nearest-neighbor interpolation. In this way, the present invention not only learns more efficient geometric representations and higher resolutions but also accelerates network computation.

[0078] In some optional implementations of the embodiments of this application, the Atom-ProteinQA model includes a topology-aware module. The step of inputting protein structure data into the Atom-ProteinQA model for quality assessment and obtaining the quality assessment result specifically includes the following steps:

[0079] An atomic-level topological graph corresponding to the protein structure data is constructed based on the topology sensing module;

[0080] The second atomic-level prediction result is output based on the atomic-level topology graph and the message passing mechanism.

[0081] In this embodiment, graph convolution is better suited for unstructured data than 3D-CNN, which excels at capturing geometric details through regular grids, and protein and RNA structure scoring is naturally unstructured. Therefore, this invention designs a topology-aware (TP) module that captures the topology of protein and RNA structure scoring by learning atomic-level chemical relationships, thereby enhancing features. Specifically, protein and RNA structure scoring molecules can be represented as graphs. in and These are the vertex set and the edge set, respectively. This invention uses the coordinates of atoms as nodes and constructs the edges of the graph using chemical bonds. Then, graph convolution is used to update the atomic features:

[0082] Where v i and This refers to the features of the i-th node before and after graph convolution. The i-th and j-th nodes (W) ij The convolution weights between nodes are calculated using the distance between the two atoms and the chemical bond. Specifically, B is weighted only when there is an actual chemical bond between the i-th and j-th nodes. ij It equals 1 only if it is true, otherwise it equals 0. Furthermore, this application sets the distance (D) between two nodes to 1. ij The model is characterized by […] and its spatial relationships are given. In the equations, […] represents concatenation operations. It is a learnable nonlinear transformation. This application uses the symbol N. i Let represent the adjacent index of node i.

[0083] By stacking multiple graph convolutional layers, this invention progressively enhances the feature representation of protein and RNA structural scoring atoms. After several layers, max pooling is used to aggregate the atomic features of each residue, yielding residue-level features. These residue-level features are further updated through more graph convolutional layers, ultimately generating residue-level LDDT scores. This invention does not output atomic-level predictions in the TP module because performing extensive message passing at the atomic level is computationally intensive and inefficient.

[0084] In some optional implementations of the embodiments of this application, the above-mentioned Atom-ProteinQA model includes a geometry-aware module, a topology-aware module, and a cross-model fusion module. The step of inputting protein structure data into the Atom-ProteinQA model for quality assessment and obtaining the quality assessment result specifically includes the following steps:

[0085] Atomic feature data of protein structure data are obtained from the geometric perception module;

[0086] The initial topology map of the protein structure data is obtained based on the topology sensing module;

[0087] The cross-model fusion module performs feature fusion operations on atomic features and topology graph to obtain an updated topology graph.

[0088] Based on the topology sensing module and message passing mechanism, a quality prediction operation is performed on the updated topology graph to obtain the third atomic-level prediction result.

[0089] In this embodiment, the Cross-Model Fusion (CMF) model enables feature interaction between the GP and TP modules. It aims to leverage skip connections from the GP module to the TP module at certain layer levels. In this way, the advantages of geometric representations can be integrated, and the residual prediction of the TP module can be improved through an end-to-end training protocol. Simultaneously, the atomic predictions of the GP module can also mutually enhance each other through backpropagation of the TP module. Specifically, for the features of the l-th sparse convolutional layer of the GP module (denoted as...),... This invention converts it to the same representation level as the l-th layer of the TP module. Here, features are transformed using nearest neighbor interpolation. The residue level of the first layer of the TP module is converted to atomic level, and the residue level of the second layer is converted to atomic level through residual max pooling. Then, in the feature... and TP module They are added element by element and then subjected to a nonlinear transformation to merge them.

[0090] The representation This represents element addition. It is a learnable transformation. Ultimately, the updated features... It will be used in layer l+1 of the TP module.

[0091] For training purposes, this invention uses the mean squared error (MSE) loss function to optimize Atom-ProteinQA, which is defined as the residue score predicted for protein and RNA structure scoring:

[0092] in This represents the true score of the nth atom. Indicates atomic characteristics, Indicate j th Each residue is truly worth dividing. j represents the bait protein th Atomic feature set of residues. Let θ represent the function of the model in this application, and let θ represent the learnable parameters.

[0093] In practical applications, Table 1 shows the performance results of this invention on CATH-2084 and Decoy-8000, with Atom-ProteinQA showing the best performance across all metrics. For non-learning-based methods, Ornate achieved the best performance. That is, without a data-driven approach, it cannot effectively capture the available information in the protein and RNA structure scoring structures. For learning-based methods, Vorocnn, by simply utilizing 3D-CNN, cannot achieve satisfactory results. Because it coarsely represents the atomic features of each residue in a single voxel, it cannot capture the geometry of local regions in the protein and RNA structure scoring. Although DeepAccNet uses additional supervision and labels for protein and RNA structure scoring refinement, they cannot predict scores more accurately than this application, especially at the local level, i.e., and. Accordingly, graph-based messaging alone does not yield optimal performance. Specifically, GNNRefine lacks the ability to extract fine-grained geometric features from the fine-grained protein and RNA structure scoring structures. This application also retrained the DeepUMQA model using the training dataset from Decoy-8000, which is the same dataset used to train the model in this application.

[0094] Furthermore, this application underwent quality assessment (QA) testing on the Decoy-8000 test set and the entire CASP14 dataset. The results show that Deep-UMQA performs exceptionally well in terms of global scores. However, the model in this application outperforms DeepUMQA in terms of residual-level scores. In contrast, Atom-ProteinQA achieves significant improvements across all scores, particularly in local metrics, through a combination of geometry-aware and chemical topology learning via GP and TP modules. It can be observed that Atom-ProteinQA outperforms ProteinGCN and DeepAccNet by 0.13 and 0.14 points, respectively, on CATH-2084. These significant gains primarily stem from jointly learned enhancement features obtained through geometry capture of high-resolution sparse voxels and chemical relation reference rings via atomic-level graphs. Figure 5 shows the visualization results of three real-world cases. A qualitative analysis of ProteinQA was performed on cases 2wcrA, 1d2zB, and 5hw3A in the Decoy-8000 test set, comparing the results of DeepAccNet, ProteinGCN, and Atom-ProteinQA. The gray line represents the true LDDT score, and the colored lines represent the results of different methods.

[0095] Table 1: Comparison of quantitative performance of the models (CATH-2084 and Decoy-8000 datasets)

[0096] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0097] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0098] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware with computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When executed, the program can include the processes of the embodiments of the above methods. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).

[0099] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0100] Referring further to Figure 6, as an implementation of the method shown in Figure 2 above, this application provides an embodiment of a protein model quality assessment device, which corresponds to the method embodiment shown in Figure 2, and the device can be specifically applied to various electronic devices.

[0101] As shown in Figure 6, the protein model quality assessment device 200 of this application embodiment includes:

[0102] The file acquisition module 210 is used to receive a PDB file input by the user terminal, wherein the PDB file contains the features of each atom in the protein and RNA structure scoring.

[0103] Model building module 220 is used to build an Atom-ProteinQA model based on the characteristics of each atom in the protein and RNA structure scoring and a preset algorithm;

[0104] The request acquisition module 230 is used to receive a quality assessment request sent by a user terminal, wherein the quality assessment request includes at least the protein structure data to be assessed.

[0105] The quality assessment module 240 is used to input protein structure data into the Atom-ProteinQA model for quality assessment and obtain quality assessment results.

[0106] In this embodiment, a protein model quality assessment device 200 is provided, comprising: a file acquisition module 210 for receiving a PDB file input by a user terminal, wherein the PDB file contains features of each atom in the protein and RNA structure scoring; a model construction module 220 for constructing an Atom-ProteinQA model based on the features of each atom in the protein and RNA structure scoring and a preset algorithm; a request acquisition module 230 for receiving a quality assessment request sent by the user terminal, wherein the quality assessment request includes at least the protein structure data to be assessed; and a quality assessment module 240 for inputting the protein structure data into the Atom-ProteinQA model for quality assessment to obtain a quality assessment result. Compared with the prior art, this application utilizes the LDDL scores at the atomic level of protein and RNA structure scoring as supervision to effectively and accurately predict IDDL scores at the atomic level, and assesses structural changes in protein and RNA structure scoring at an atomic level.

[0107] In some optional implementations of the embodiments of this application, the above-mentioned Atom-ProteinQA model includes a geometry sensing module, and the quality assessment module includes:

[0108] The feature extraction submodule is used to perform feature extraction operations on protein structure data to obtain the voxelized Cartesian coordinates of each residue atom in the protein structure data.

[0109] The first prediction submodule is used to generate the first atomic-level prediction result based on the geometric features of the voxelized Cartesian coordinates captured by the geometry perception module.

[0110] In some optional implementations of the embodiments of this application, the above-mentioned Atom-ProteinQA model includes a topology-aware module, and the quality assessment module includes:

[0111] The topology graph construction submodule is used to construct an atomic-level topology graph corresponding to the protein structure data based on the topology-aware module.

[0112] The second prediction submodule is used to output the second atomic-level prediction result based on the atomic-level topology graph and the message passing mechanism.

[0113] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to Figure 7 for details; Figure 7 is a basic structural block diagram of the computer device according to an embodiment of this application.

[0114] The computer device 300 includes a memory 310, a processor 320, and a network interface 330 that are interconnected via a system bus. It should be noted that only the computer device 300 with components 310-330 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the computer device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0115] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0116] The memory 310 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 310 may be an internal storage unit of the computer device 300, such as the hard disk or memory of the computer device 300. In other embodiments, the memory 310 may also be an external storage device of the computer device 300, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Of course, the memory 310 may also include both internal storage units and external storage devices of the computer device 300. In this embodiment, the memory 310 is typically used to store the operating system and various application software installed on the computer device 300, such as computer-readable instructions for protein model quality assessment methods. Furthermore, the memory 310 can also be used to temporarily store various types of data that have been output or will be output.

[0117] In some embodiments, the processor 320 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 320 is typically used to control the overall operation of the computer device 300. In this embodiment, the processor 320 is used to execute computer-readable instructions stored in the memory 310 or to process data, such as executing computer-readable instructions for the protein model quality assessment method.

[0118] The network interface 330 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 300 and other electronic devices.

[0119] The computer device provided in this application uses the LDDT scores at the atomic level of protein and RNA structure scoring as supervision to effectively and accurately predict IDDL scores at the atomic level, and to evaluate changes in protein and RNA structure scoring at an atomic level.

[0120] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by at least one processor to cause the at least one processor to perform the steps of the protein model quality assessment method described above.

[0121] The computer-readable storage medium provided in this application utilizes the LDDT scores at the atomic level of protein and RNA structure scoring as supervision to effectively and accurately predict IDDL scores at the atomic level, and to evaluate changes in protein and RNA structure scoring at an atomic level.

[0122] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0123] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to provide a more thorough and comprehensive understanding of the disclosure of this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application.

Claims

1. A method for assessing the quality of a protein model, characterized in that, Includes the following steps: Receive a PDB file input from a user terminal, wherein the PDB file contains features of each atom in the protein and RNA structure scoring; The Atom-ProteinQA model is constructed based on the characteristics of each atom in the protein and RNA structure scoring and a preset algorithm. Receive a quality assessment request sent by a user terminal, wherein the quality assessment request includes at least the protein structure data to be assessed; The protein structure data is input into the Atom-ProteinQA model for quality assessment, and the quality assessment results are obtained.

2. The protein model quality assessment method according to claim 1, characterized in that, The preset algorithm first searches for a radius of P. n Atom P n The neighbors of [i,j] are then selected by P using a preset index. m and P n The adjacent atoms in the matrix, and calculate the distance from each of the adjacent atoms to P. m [i,j] and P n The distances between [i,j], P m and P n The difference between individual distances will be determined by a mapping function. The scores are converted to predefined scores, and finally, LDDT scores at the atom, residue, and global levels are obtained by normalizing the total number of neighbors.

3. The protein model quality assessment method according to claim 1, characterized in that, The Atom-ProteinQA model includes a geometry sensing module. The step of inputting the protein structure data into the Atom-ProteinQA model for quality assessment and obtaining the quality assessment result specifically includes the following steps: Feature extraction is performed on the protein structure data to obtain the voxelized Cartesian coordinates of each residue atom in the protein structure data. The geometric features of the voxelized Cartesian coordinates are captured by the geometric perception module to generate the first atomic-level prediction result.

4. The protein model quality assessment method according to claim 1, characterized in that, The Atom-ProteinQA model includes a topology-aware module. The step of inputting the protein structure data into the Atom-ProteinQA model for quality assessment and obtaining the quality assessment result specifically includes the following steps: An atomic-level topology map corresponding to the protein structure data is constructed based on the topology sensing module. The second atomic-level prediction result is output based on the atomic-level topology graph and the message passing mechanism.

5. The protein model quality assessment method according to claim 1, characterized in that, The Atom-ProteinQA model includes a geometry-aware module, a topology-aware module, and a cross-model fusion module. The step of inputting the protein structure data into the Atom-ProteinQA model for quality assessment and obtaining the quality assessment result specifically includes the following steps: Atomic feature data of the protein structure data are obtained based on the geometric sensing module; The initial topology map of the protein structure data is obtained based on the topology sensing module; The cross-model fusion module performs feature fusion operations on the atomic features and the topology graph to obtain an updated topology graph. The updated topology graph is subjected to quality prediction operation based on the topology-aware module and message passing mechanism to obtain the third atomic-level prediction result.

6. A protein model quality assessment device, characterized in that, include: The file acquisition module is used to receive PDB files input by the user terminal, wherein the PDB files contain the features of each atom in the protein and RNA structure scoring; The model building module is used to build an Atom-ProteinQA model based on the characteristics of each atom in the protein and RNA structure scoring and a preset algorithm. The request acquisition module is used to receive a quality assessment request sent by a user terminal, wherein the quality assessment request includes at least the protein structure data to be assessed. The quality assessment module is used to input the protein structure data into the Atom-ProteinQA model for quality assessment and obtain the quality assessment results.

7. The protein model quality assessment device according to claim 6, characterized in that, The Atom-ProteinQA model includes a geometry-aware module, and the quality assessment module includes: The feature extraction submodule is used to perform feature extraction operations on the protein structure data to obtain the voxelized Cartesian coordinates of each residue atom in the protein structure data. The first prediction submodule is used to generate a first atomic-level prediction result based on the geometric features captured by the geometric perception module of the voxelized Cartesian coordinates.

8. The protein model quality assessment device according to claim 6, characterized in that, The Atom-ProteinQA model includes a topology-aware module, and the quality assessment module includes: The topology graph construction submodule is used to construct an atomic-level topology graph corresponding to the protein structure data based on the topology sensing module. The second prediction submodule is used to output the second atomic-level prediction result based on the atomic-level topology graph and the message passing mechanism.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, it implements the steps of the protein model quality assessment method as described in any one of claims 1 to 5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the protein model quality assessment method as described in any one of claims 1 to 5.

Images