Complex protein structure determination method and device, electronic equipment and storage medium

Abstract

The invention relates to a complex protein structure determination method and device, electronic equipment and a storage medium, and the method comprises the steps: determining an electron density map of complex protein and a plurality of amino acid sequences forming the complex protein, and extracting supporting points located at a main chain in the electron density map; and matching the supporting points corresponding to the amino acid sequences in sequence according to a preset prediction sequence, and predicting the corresponding chain structures according to the amino acid sequences and the corresponding supporting points. After the chain structure corresponding to the amino acid sequence is determined each time, supporting point filtering is carried out according to the chain structure corresponding to the amino acid sequence. And under the condition that the chain structures corresponding to all the amino acid sequences are predicted, determining the protein structure of the complex protein according to the chain structure corresponding to each amino acid sequence. According to the method, the chain structure of each amino acid sequence can be accurately predicted in sequence based on the electron density map, and then the protein folding structure of the complex protein is jointly combined according to all the chain structures, so that the accuracy of the prediction result is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a method, device, electronic device and storage medium for determining a complex protein structure. Background Art

[0002] Using AI technology to predict protein structure is one of the hot applications in recent years. With the development of cryo-electron microscopy technology, structural modeling combined with cryo-electron microscopy 3D electron density maps is one of the development directions of AI prediction of protein structure. Among them, predicting multi-chain protein complexes involves many challenges. First, the total sequence length of many such proteins is very long. Directly inputting all of them into the model will cause memory overflow and take a lot of time. Second, for homologous multimers, it is often impossible to distinguish between chains, and two chains with the same sequence in the prediction results will appear twisted together or overlap in the middle. Summary of the invention

[0003] In view of this, the present disclosure proposes a method, device, electronic device and storage medium for determining the structure of a complex protein, aiming to accurately predict the structure of the complex protein.

[0004] According to a first aspect of the present disclosure, a method for determining a complex protein structure is provided, the method comprising:

[0005] Determining the electron density map of the complex protein and the multiple amino acid sequences constituting the complex protein;

[0006] Extracting supporting points in the electron density map, wherein the supporting points are pixel points located on the main chain of the complex protein in the electron density map;

[0007] Predicting the chain structure corresponding to each of the amino acid sequences in sequence according to a preset prediction order;

[0008] After determining the chain structure corresponding to the amino acid sequence each time, filtering the supporting points according to the chain structure corresponding to the amino acid sequence;

[0009] In response to predicting the chain structures corresponding to all the amino acid sequences, the protein structure of the complex protein is determined according to the chain structure corresponding to each of the amino acid sequences.

[0010] In a possible implementation manner, extracting support points in the electron density map includes:

[0011] The probability prediction model obtained through training predicts the confidence corresponding to each pixel point in the electron density map, wherein the confidence degree is used to characterize the probability that the corresponding pixel point belongs to the main chain of the complex protein;

[0012] A plurality of supporting points are obtained by screening according to the confidence level corresponding to each pixel point.

[0013] In a possible implementation, the step of obtaining a plurality of support points by screening according to the confidence level corresponding to each pixel point includes:

[0014] Obtaining a first preset number of pixel points with the highest confidence values ​​as candidate points;

[0015] Calculate the distance between each candidate point and other candidate points multiple times in an iterative manner;

[0016] The candidate points with low confidence between the two candidate points with the smallest distance are removed until a second preset number of candidate points are screened out, the iterative process is stopped, and the current candidate point is determined as a supporting point.

[0017] In a possible implementation, the prediction order is to predict the supporting points corresponding to each amino acid sequence in order from long to short.

[0018] In a possible implementation, the predicting the chain structure corresponding to each of the amino acid sequences includes:

[0019] According to a preset prediction order, the support point pair representation matrix is ​​determined multiple times according to the current support points in an iterative manner;

[0020] Determine the support point-residue pair representation matrix corresponding to the current amino acid sequence;

[0021] The corresponding chain structure is predicted based on the current amino acid sequence and the corresponding support point pair representation matrix and support point-residue pair representation matrix.

[0022] In a possible implementation, predicting the corresponding chain structure according to the current amino acid sequence and the corresponding support point pair representation matrix and support point-residue pair representation matrix includes:

[0023] The current amino acid sequence, the support point representation matrix corresponding to the amino acid sequence and the support point-residue pair representation matrix are input into the trained chain structure prediction model to predict the corresponding chain structure.

[0024] In a possible implementation, inputting the current amino acid sequence, the support point representation matrix and the support point-residue pair representation matrix corresponding to the amino acid sequence into the trained chain structure prediction model to predict the corresponding chain structure includes:

[0025] Inputting the amino acid sequence, the support point pair representation matrix and the support point-residue pair representation matrix into the chain structure prediction model, and processing the amino acid sequence through the pre-processing layer to obtain a residue pair representation matrix;

[0026] Performing at least one cyclic update on the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix through the cyclic update layer;

[0027] Determining the supporting points for matching the amino acid sequence according to the supporting point-residue pair representation matrix after the cyclic update;

[0028] The residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix after cyclic updating are processed by the structure prediction layer to obtain the chain structure corresponding to the amino acid sequence.

[0029] In a possible implementation, determining the supporting points for the amino acid sequence matching according to the cyclically updated supporting point-residue pair representation matrix includes:

[0030] Determine the residue closest to each of the support points and probability values ​​corresponding to a plurality of preset distance intervals according to the support point-residue pair representation matrix after cyclic update;

[0031] Filter the support points according to the probability value of the distance interval corresponding to the distance between each support point and the nearest residue;

[0032] The supporting points obtained after filtering are deduplicated to obtain the supporting points matching the amino acid sequence.

[0033] In a possible implementation, removing duplicates from the support points obtained after filtering to obtain the support points for the amino acid sequence match includes:

[0034] Determine the residues that match each support point obtained after filtering;

[0035] Determine that the supporting point whose distance among the supporting points corresponding to each of the residues is less than the first preset distance and whose distance corresponding probability value is the largest is the supporting point for matching the amino acid sequence.

[0036] In a possible implementation, the filtering of support points according to the chain structure corresponding to the amino acid sequence includes:

[0037] Determine the coordinate position of each target atom in the amino acid sequence according to the chain structure corresponding to the amino acid sequence;

[0038] Delete the support points whose distance from any target atom is less than the second preset distance.

[0039] In a possible implementation, determining the protein structure of the complex protein according to the chain structure corresponding to each of the amino acid sequences includes:

[0040] According to the supporting point position corresponding to each amino acid sequence, the chain structure is subjected to position adjustment processing;

[0041] The protein structure of the complex protein is obtained by adjusting the splicing position of each chain structure.

[0042] In a possible implementation, the training process of the chain structure prediction model includes:

[0043] determining a plurality of sample amino acid sequences having corresponding annotated chain structures and sample density plots;

[0044] Each of the sample amino acid sequences is randomly cropped, and model training is performed based on the cropped sample amino acid sequences, the corresponding sample density map, and the annotated chain structure to obtain a chain structure prediction model.

[0045] In a possible implementation, randomly cropping each of the sample amino acid sequences, and performing model training according to the cropped sample amino acid sequences and the corresponding sample density map and the annotated chain structure to obtain a chain structure prediction model includes:

[0046] Determining a first trimming interval for each of the sample amino acid sequences;

[0047] trimming the sample density map corresponding to each of the sample amino acid sequences according to a second trimming interval including the first trimming interval;

[0048] Model training is performed based on the cropped sample amino acid sequence, the cropped sample density map and the annotated chain structure to obtain a chain structure prediction model.

[0049] In one possible implementation, the loss in the chain structure prediction model training process includes a point segmentation loss determined according to the support point pair representation matrix after cyclic updating, and the point segmentation loss is used to characterize whether two support points that actually belong to the same sequence belong to the same sequence in the prediction result.

[0050] In a possible implementation manner, the process of determining the point segmentation loss includes:

[0051] Marking the true label of the support point pair representation matrix corresponding to each of the sample density graphs according to a preset rule, wherein the true label is used to indicate whether the corresponding support point pairs belong to the same sequence;

[0052] Input the support point pair representation matrix after cyclic update into the linear layer to obtain the predicted label of each support point pair belonging to the same sequence;

[0053] The cross entropy loss is calculated according to the true label and the predicted label corresponding to each sample density map to obtain the point segmentation loss.

[0054] According to a second aspect of the present disclosure, a complex protein structure determination device is provided, the device comprising:

[0055] An information determination module, used to determine the electron density map of the complex protein and a plurality of amino acid sequences constituting the complex protein;

[0056] A support point extraction module, used to extract support points in the electron density map, wherein the support points are pixel points located on the main chain of the complex protein in the electron density map;

[0057] A chain structure determination module, used to predict the chain structure corresponding to each of the amino acid sequences in sequence according to a preset prediction order;

[0058] A support point filtering module, used for filtering support points according to the chain structure corresponding to the amino acid sequence after determining the chain structure corresponding to the amino acid sequence each time;

[0059] The protein structure determination module is used to determine the protein structure of the complex protein according to the chain structure corresponding to each amino acid sequence in response to the predicted chain structure corresponding to all the amino acid sequences.

[0060] In a possible implementation, the support point extraction module is further used to:

[0061] The probability prediction model obtained through training predicts the confidence corresponding to each pixel point in the electron density map, wherein the confidence degree is used to characterize the probability that the corresponding pixel point belongs to the main chain of the complex protein;

[0062] A plurality of supporting points are obtained by screening according to the confidence level corresponding to each pixel point.

[0063] In a possible implementation, the support point extraction module is further used to:

[0064] Obtaining a first preset number of pixel points with the highest confidence values ​​as candidate points;

[0065] Calculate the distance between each candidate point and other candidate points multiple times in an iterative manner;

[0066] The candidate points with low confidence between the two candidate points with the smallest distance are removed until a second preset number of candidate points are screened out, the iterative process is stopped, and the current candidate point is determined as a supporting point.

[0067] In a possible implementation, the prediction order is to predict the supporting points corresponding to each amino acid sequence in order from long to short.

[0068] In a possible implementation, the chain structure determination module is further used to:

[0069] According to a preset prediction order, the support point pair representation matrix is ​​determined multiple times according to the current support points in an iterative manner;

[0070] Determine the support point-residue pair representation matrix corresponding to the current amino acid sequence;

[0071] The corresponding chain structure is predicted based on the current amino acid sequence and the corresponding support point pair representation matrix and support point-residue pair representation matrix.

[0072] In a possible implementation, the chain structure determination module is further used to:

[0073] The current amino acid sequence, the support point representation matrix corresponding to the amino acid sequence and the support point-residue pair representation matrix are input into the trained chain structure prediction model to predict the corresponding chain structure.

[0074] In a possible implementation, the chain structure determination module is further used to:

[0075] Inputting the amino acid sequence, the support point pair representation matrix and the support point-residue pair representation matrix into the chain structure prediction model, and processing the amino acid sequence through the pre-processing layer to obtain a residue pair representation matrix;

[0076] Performing at least one cyclic update on the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix through the cyclic update layer;

[0077] Determining the supporting points for matching the amino acid sequence according to the supporting point-residue pair representation matrix after the cyclic update;

[0078] The residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix after cyclic updating are processed by the structure prediction layer to obtain the chain structure corresponding to the amino acid sequence.

[0079] In a possible implementation, the chain structure determination module is further used to:

[0080] Determine the residue closest to each of the support points and probability values ​​corresponding to a plurality of preset distance intervals according to the support point-residue pair representation matrix after cyclic update;

[0081] Filter the support points according to the probability value of the distance interval corresponding to the distance between each support point and the nearest residue;

[0082] The supporting points obtained after filtering are deduplicated to obtain the supporting points matching the amino acid sequence.

[0083] In a possible implementation, the chain structure determination module is further used to:

[0084] Determine the residues that match each support point obtained after filtering;

[0085] Determine that the supporting point whose distance among the supporting points corresponding to each of the residues is less than the first preset distance and whose distance corresponding probability value is the largest is the supporting point for matching the amino acid sequence.

[0086] In a possible implementation, the support point filtering module is further used to:

[0087] Determine the coordinate position of each target atom in the amino acid sequence according to the chain structure corresponding to the amino acid sequence;

[0088] Delete the support points whose distance from any target atom is less than the second preset distance.

[0089] In a possible implementation, the protein structure determination module is further used to:

[0090] According to the supporting point position corresponding to each amino acid sequence, the chain structure is subjected to position adjustment processing;

[0091] The protein structure of the complex protein is obtained by adjusting the splicing position of each chain structure.

[0092] In a possible implementation, the training process of the chain structure prediction model includes:

[0093] determining a plurality of sample amino acid sequences having corresponding annotated chain structures and sample density plots;

[0094] Each of the sample amino acid sequences is randomly cropped, and model training is performed based on the cropped sample amino acid sequences, the corresponding sample density map, and the annotated chain structure to obtain a chain structure prediction model.

[0095] In a possible implementation, randomly cropping each of the sample amino acid sequences, and performing model training according to the cropped sample amino acid sequences and the corresponding sample density map and the annotated chain structure to obtain a chain structure prediction model includes:

[0096] Determining a first trimming interval for each of the sample amino acid sequences;

[0097] trimming the sample density map corresponding to each of the sample amino acid sequences according to a second trimming interval including the first trimming interval;

[0098] Model training is performed based on the cropped sample amino acid sequence, the cropped sample density map and the annotated chain structure to obtain a chain structure prediction model.

[0099] In one possible implementation, the loss in the chain structure prediction model training process includes a point segmentation loss determined according to the support point pair representation matrix after cyclic updating, and the point segmentation loss is used to characterize whether two support points that actually belong to the same sequence belong to the same sequence in the prediction result.

[0100] In a possible implementation manner, the process of determining the point segmentation loss includes:

[0101] Marking the true label of the support point pair representation matrix corresponding to each of the sample density graphs according to a preset rule, wherein the true label is used to indicate whether the corresponding support point pairs belong to the same sequence;

[0102] Input the support point pair representation matrix after cyclic update into the linear layer to obtain the predicted label of each support point pair belonging to the same sequence;

[0103] The cross entropy loss is calculated according to the true label and the predicted label corresponding to each sample density map to obtain the point segmentation loss.

[0104] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0105] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, wherein the computer program instructions implement the above method when executed by a processor.

[0106] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0107] In an embodiment of the present disclosure, the electron density map of the complex protein and the multiple amino acid sequences constituting the complex protein are determined, the supporting points located on the main chain in the electron density map are extracted, the supporting points corresponding to each amino acid sequence are matched in sequence according to a preset prediction order, and the corresponding chain structure is predicted based on the amino acid sequence and the corresponding supporting points. After each determination of the chain structure corresponding to the amino acid sequence, the supporting points are filtered according to the chain structure corresponding to the amino acid sequence. In the case where the chain structures corresponding to all amino acid sequences are predicted, the protein structure of the complex protein is determined according to the chain structure corresponding to each amino acid sequence. The present disclosure can improve the accuracy of the prediction results by predicting the protein folding structure of the complex protein based on the electron density map by accurately determining the chain structure of each amino acid sequence in order, and then jointly predicting the protein folding structure of the complex protein based on all chain structures.

[0108] Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

[0110] Figure 1 A flow chart showing a method for determining a complex protein structure according to an embodiment of the present disclosure;

[0111] Figure 2 A schematic diagram showing a process of predicting the chain structure of each amino acid sequence according to an embodiment of the present disclosure;

[0112] Figure 3 A schematic diagram showing a complex protein structure according to an embodiment of the present disclosure;

[0113] Figure 4 A schematic diagram showing a complex protein structure determination device according to an embodiment of the present disclosure;

[0114] Figure 5 A schematic diagram showing an electronic device according to an embodiment of the present disclosure;

[0115] Figure 6 A schematic diagram showing another electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0116] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0117] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0118] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

[0119] The complex protein structure determination method of the embodiment of the present disclosure can be performed by an electronic device such as a terminal device or a server. Among them, the terminal device can be any fixed or mobile terminal such as a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The server can be a single server or a server cluster composed of multiple servers. Any electronic device can implement the complex protein structure determination method of the embodiment of the present disclosure by a processor calling a computer-readable instruction stored in a memory.

[0120] Figure 1 A flow chart of a method for determining a complex protein structure according to an embodiment of the present disclosure is shown. Figure 1 As shown, the method for determining the complex protein structure of the embodiment of the present disclosure may include the following steps S10-S50.

[0121] Step S10: determining the electron density map of the complex protein and the multiple amino acid sequences constituting the complex protein.

[0122] In a possible implementation, an electronic device determines an electron density map and multiple amino acid sequences corresponding to a complex protein whose folding mode needs to be predicted. The complex protein includes multiple chains, and each chain corresponds to an amino acid sequence. Each amino acid sequence may include at least two amino acids arranged in order, which are used to characterize the amino acids included in a chain in the complex protein, and the arrangement order of each amino acid. The electron density map can be a three-dimensional point cloud image, in which the three-dimensional point cloud included is used to characterize the overall three-dimensional spatial structure of the complex protein. Optionally, the electron density map in the embodiment of the present disclosure can be determined by any image acquisition device, for example, it can be obtained by performing image acquisition of the complex protein by a cryo-electron microscope. That is, the electron density map can be directly obtained by controlling the cryo-electron microscope to acquire images of the complex protein by an electronic device, or by controlling the cryo-electron microscope to acquire images of the complex protein by other devices and then sending them to the electronic device.

[0123] Step S20: extracting support points in the electron density map.

[0124] In a possible implementation, after determining the electron density map corresponding to the complex protein, the electronic device extracts multiple support points in the three-dimensional point cloud in the electron density map. Among them, the support points are pixel points located on the main chain of the complex protein in the electron density map, that is, each extracted support point can be used to jointly characterize the contour information composed of multiple main chains in the complex protein. In order to increase the robustness and generalization ability of the embodiments of the present disclosure, the number of support points can be pre-set. Optionally, the number of support points can be determined based on the total length of each amino acid sequence in the complex protein, that is, the total number of amino acids included in the complex protein.

[0125] Optionally, the electronic device can first extract three-dimensional points that may belong to the protein main chain from the electron density map, and then further select at least one supporting point from the three-dimensional points that may belong to the protein main chain. That is, the method of determining the supporting points in the electron density map can be to predict the confidence corresponding to each pixel in the electron density map through the probability prediction model obtained through training, and the confidence is used to characterize the probability that the corresponding pixel belongs to the main chain of the complex protein. Then, multiple supporting points are obtained by screening according to the confidence corresponding to each pixel. That is, the electronic device can input the electron density map into the probability prediction model, and the probability prediction model predicts the probability that each pixel belongs to the protein main chain, and outputs the confidence map corresponding to the electron density map. Each pixel value in the confidence map is a numerical value between 0-1, which is used to characterize the probability that the pixel at the corresponding position in the electron density map belongs to the protein main chain of the complex protein. Among them, the probability prediction model can be an image segmentation model such as a Unet model.

[0126] Further, after the electronic device first determines the confidence that each pixel in the electron density map belongs to the main chain of the protein, at least one supporting point is screened from multiple pixel points according to the confidence of the pixel and the position coordinates. Wherein, the screening process of the supporting point may include two steps, the first step is to screen only according to the confidence to obtain multiple candidate points, the second step is to screen the supporting point from multiple candidate points according to the confidence and the pixel coordinates, and the second step screening process may be an iterative cycle screening process, and the better candidate points are retained each time. Optionally, the electronic device may first obtain the first preset number of pixel points with the highest confidence value as candidate points, and then calculate the distance between each candidate point and other candidate points in an iterative manner for multiple times, and remove the candidate points with low confidence between the two candidate points with the smallest distance, until the second preset number of candidate points is screened, the iterative process is stopped and the current candidate point is determined as a supporting point. Wherein, the first preset number and the second preset number can be preset, the second preset number is the number of supporting points to be determined, and can be determined according to the total length of the amino acid sequence, and the first preset number can be an integer greater than the second preset number, for example, it can be an integer multiple of the second preset number.

[0127] Exemplarily, the electronic device can set the first preset number to 3N and the second preset number to N. After determining the confidence of each pixel in the electron density map, the 3N pixels with the largest corresponding confidence are obtained as candidate points, and each candidate point has a corresponding coordinate (x, y, z). After obtaining the candidate points, the distance between the two candidate points is calculated according to the coordinates corresponding to each candidate point, and then the candidate point with low confidence is deleted from the two candidate points with the shortest distance. Further, the two candidate points with the shortest distance among the remaining candidate points are re-determined, and the candidate points with low confidence are deleted until the number of remaining candidate points is the second preset number. Among them, the distance between each two candidate points can be a Euclidean distance.

[0128] Step S30: predicting the chain structure corresponding to each of the amino acid sequences in sequence according to a preset prediction order.

[0129] In a possible implementation, after extracting the supporting points in the electron density map, the electronic device sorts the amino acid sequences included in the complex protein according to a preset prediction order, so as to match the supporting points corresponding to each amino acid sequence in sequence according to the arrangement order, and determine the chain structure corresponding to the amino acid sequence according to the amino acid sequence and the corresponding supporting points, and the chain structure is the spatial folding structure of the main chain corresponding to the corresponding amino acid sequence in the complex protein. Among them, the preset sequencing order can be to predict the chain structure and the corresponding supporting points of each amino acid sequence in sequence according to the length from long to short. Optionally, the chain structure corresponding to each amino acid sequence can be determined according to the supporting point pair representation matrix and the supporting point-residue pair representation matrix. That is, the electronic device can determine the supporting point pair representation matrix according to the current supporting points and the current amino acid sequence in an iterative manner for multiple times according to the preset prediction order, and determine the supporting point-residue pair representation matrix corresponding to the current amino acid sequence, and predict the corresponding chain structure. The chain structure prediction process can be completed based on the chain structure prediction model obtained through training, that is, the current amino acid sequence, the supporting point representation matrix corresponding to the current amino acid sequence, and the supporting point-residue pair representation matrix can be input into the trained chain structure prediction model to predict the corresponding chain structure and match the supporting points corresponding to the amino acid sequence.

[0130] Optionally, the support point pair representation matrix is ​​used to characterize the positional relationship between every two support points. In each iteration process, the support point pair representation matrix can be a matrix initialized by the positional relationship, which is used to determine the corresponding chain structure after multiple iterations. Among them, the positional relationship includes relative position features and distance features, that is, the electronic device can first determine the ordered support point pairs composed of every two support points, and determine the corresponding direction feature vector according to the positions of the two support points in each ordered support point pair. Then determine the corresponding distance feature vector according to the distance between the two support points in each ordered support point pair. Determine the support point pair representation matrix according to the distance feature vector and the direction feature vector corresponding to each ordered support point pair, and each row and column of the support point pair representation matrix represents a corresponding support point, and each element represents the positional relationship between the corresponding support point in the row and the corresponding support point in the column. The ordered support point pair is composed of two support points in order, that is, the same two support points can determine two support point pairs. For example, support point 1 and support point 2 can form two ordered support point pairs (support point 1, support point 2) and (support point 2, support point 1).

[0131] Further, the support point-residue pair representation matrix corresponding to the amino acid sequence is used to characterize the attention relationship between each support point and each amino acid in the amino acid sequence, and the attention relationship can characterize the relative spatial position between the support point and the amino acid. Optionally, the support point-residue pair representation matrix can be determined by random initialization, and the support point-residue pair representation matrix is ​​an initialized matrix used to determine the corresponding chain structure after multiple iterations of the loop. That is, the electronic device can determine a second matrix in which each row represents an amino acid and each column represents a support point. In each coordinate position in the second matrix, a vector with the same number of channels as the support point pair representation matrix is ​​randomly generated to obtain a support point-residue pair representation matrix. Among them, the number of channels is the dimension of the value in each coordinate in the matrix. For example, when the support point pair representation matrix is ​​(m, m, c) and the number of amino acids in the amino acid sequence is n, a second matrix of size m×n can be determined, and a c-dimensional vector is generated as a coordinate value at each coordinate position therein to obtain a support point-residue pair representation matrix (m, n, c).

[0132] In a possible implementation, the chain structure prediction model of the embodiment of the present disclosure includes a pre-processing layer, a cyclic update layer and a structure prediction layer. Each time the electronic device predicts the chain structure corresponding to the current amino acid sequence, the amino acid sequence, the support point pair representation matrix and the support point-residue pair representation matrix can be input into the chain structure prediction model, and the amino acid sequence is processed by the pre-processing layer to obtain the residue pair representation matrix. Then, the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix are cyclically updated at least once by the cyclic update layer. The supporting points matching the amino acid sequence are determined according to the support point-residue pair representation matrix after the cyclic update, and the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix after the cyclic update are processed by the structure prediction layer to obtain the chain structure corresponding to the amino acid sequence. Among them, the residue pair representation matrix is ​​used to characterize the spatial position relationship between every two amino acids in the amino acid sequence corresponding to the target protein.

[0133] Optionally, the pre-processing layer can also be the same as determining a multiple sequence comparison representation matrix based on the amino acid sequence, so as to perform at least one cyclic update on the multiple sequence comparison representation matrix, the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix through the cyclic update layer. Among them, the multiple sequence comparison representation matrix is ​​used to characterize the covariation information of every two amino acids in the amino acid sequence of other known folding modes similar to the target protein amino acid sequence, that is, the multiple sequence comparison representation matrix can be determined based on the common evolution information of the species, and is used to characterize the amino acid positions that tend to change together in similar amino acid sequences. In the case where the folding mode prediction model is the alphafold2 model, the cyclic update layer is the Evoformer layer therein, which is used to update and correct the input multiple representation matrices multiple times in a cyclic iteration manner to obtain an accurate feature matrix after correction. The cyclic update layer can update each matrix for a preset number of times respectively, and use the result obtained after each update as the input for the next update.

[0134] For each amino acid sequence input into the chain structure prediction model, the cyclic update layer can perform multiple corrections on the support point-residue pair representation matrix based on the multiple sequence comparison representation matrix and the support point pair representation matrix during each cyclic update process to obtain an accurate support point-residue pair representation matrix. The electronic device can determine the support points of the amino acid sequence match according to the support point-residue pair representation matrix determined after the cyclic update. Among them, the support point of each amino acid sequence match can be determined by the distance between the support point position and the predicted residue position. Exemplarily, the electronic device can determine the residue closest to each support point and the probability value corresponding to the preset multiple distance intervals according to the support point-residue pair representation matrix after the cyclic update. Then filter the support points according to the probability value of the distance interval corresponding to the distance between each support point and the nearest distance residue, and remove the duplicates of the support points obtained after filtering to obtain the support points of the amino acid sequence match. The way to remove the duplicates of the filtered support points can be to determine the residues matched by each support point obtained after filtering, and then determine that the support point corresponding to each residue has a distance less than the first preset distance, and the support point with the largest distance corresponding probability value is the support point of the amino acid sequence match.

[0135] Optionally, the support point-residue pair representation matrix output by the chain structure prediction model can also be used to give the predicted probability of each support point and residue corresponding to each preset distance interval based on the support point-residue pair matrix output by it. The prediction can be performed by introducing a linear layer, and the predicted probability is used to characterize the probability value that the predicted distance between the support point and the predicted nearest residue is less than the preset distance. After determining the mixed updated support point-residue pair representation matrix, the electronic device can calculate the probability value and the minimum of several target distance intervals (e.g., 0-0.5, 0.5-1, 1-2, 2-3) in each preset distance interval, and filter the support points according to the probability value and the probability value to retain the support points with probability values ​​greater than the probability value and the probability value. In the case where there are more than one support points corresponding to the nearest distance of a residue, the support point with the largest probability value is retained as the support point corresponding to the residue.

[0136] Figure 2 A schematic diagram of a process for predicting the chain structure of each amino acid sequence according to an embodiment of the present disclosure is shown. Figure 2 As shown, when the electronic device predicts the chain structure corresponding to each amino acid sequence 20 in the complex protein, it first extracts the support points according to the electron density map 21 of the complex protein and determines the support point pair representation matrix 22 and the support point-residue pair representation matrix 23. Further, the amino acid sequence 20 is processed according to the pre-processing layer 24 in the model to obtain the residue pair representation matrix 25, and the support point pair representation matrix 22, the support point-residue pair representation matrix 23 and the residue pair representation matrix 25 are cyclically updated together by the cyclic update layer 26, and then the structure prediction layer 27 performs structure prediction according to the cyclically updated support point-residue pair representation matrix 23 to obtain the chain structure corresponding to the input amino acid sequence 20 and obtain the prediction of the support point-residue correspondence relationship. Optionally, the pre-processing layer 24 can also extract the multi-sequence comparison representation matrix, and input it into the cyclic update layer together with the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix for cyclic update to further improve the accuracy of the final output chain structure.

[0137] In a possible implementation, the training process of the chain structure prediction model may include determining a plurality of sample amino acid sequences with corresponding annotated chain structures and sample density maps, and then randomly cropping each sample amino acid sequence, and performing model training based on the cropped sample amino acid sequence, the corresponding sample density map, and the annotated chain structure to obtain a chain structure prediction model. Among them, the electronic device may randomly crop only the sample amino acid sequence, or crop the sample amino acid sequence and the sample density map at the same time. That is, the electronic device may determine the first cropping interval of each sample amino acid sequence, and then crop the sample density map corresponding to each sample amino acid sequence according to the second cropping interval including the first cropping interval, and perform model training based on the cropped sample amino acid sequence, the cropped sample density map, and the annotated chain structure to obtain a chain structure prediction model. Optionally, the ratio of the first cropping interval to the second cropping interval may be preset, and the second cropping interval may also cover intervals on other chains. At the same time, in order to adapt to the situation where the homologous multimer includes multiple completely identical chains during the inference process, the first cropping interval and the second cropping interval selected during the training process have a certain probability of forcing the selection of the same residue sequence on different copies on the homologous chain, where the first cropping interval is a copy and the second cropping interval is a region composed of multiple copies.

[0138] Furthermore, in order to improve the prediction effect of the chain structure prediction model, the point segmentation loss determined according to the support point pair representation matrix after cyclic update can be introduced in the training process to characterize whether two support points that actually belong to the same sequence belong to the same sequence in the prediction result. That is, the chain structure prediction model can also introduce an additional task to determine whether any two support points belong to the same sequence. This task takes the updated support point pair representation matrix as input, and outputs a binary classification result for each point pair through a linear layer to determine whether they belong to the same sequence. During the training process, the true label of the support point pair representation matrix corresponding to each sample density map is marked according to the preset rules. The true label is used to characterize whether the corresponding support point pair belongs to the same sequence. The support point pair representation matrix after cyclic update is then input into the linear layer to obtain the predicted label that each support point pair belongs to the same sequence. The cross entropy loss is calculated based on the true label and predicted label corresponding to each sample density map to obtain the point segmentation loss.

[0139] Exemplarily, during the training process of the structure prediction model, a real label of 1 or 0 can be added to all support point pairs. The label can be added by first obtaining the main chain atom position closest to the support point and recording the distance from the support point to the main chain atom. When the distance is greater than or equal to the preset value, the point is determined to be a noise point and marked as -1. After all support points are marked, the real label is determined according to the marking result. Among them, when the two support point marks are the same and not -1, the real label of the support point pair is determined to be 1, and when the two support point marks are different, or the same but the value is -1, the real label of the support point pair is determined to be 0. After marking the predicted labels of the support point pair representation matrix, the electronic device can calculate the cross entropy loss according to the real label and predicted label corresponding to each sample density map to obtain the point segmentation loss.

[0140] Step S40: After the chain structure corresponding to the amino acid sequence is determined each time, support point filtering is performed according to the chain structure corresponding to the amino acid sequence.

[0141] In a possible implementation, after the electronic device determines the chain structure according to the current amino acid sequence each time, in order to avoid the determined chain structure from interfering with the prediction process of other subsequent amino acid sequences, it is necessary to filter the support points according to the chain structure corresponding to the current amino acid sequence, so as to predict the chain structure of the next amino acid sequence through the support points obtained after filtering. Among them, the electronic device can filter the support points by determining the coordinate position of each target atom according to the chain structure corresponding to the amino acid sequence, and then delete the support points whose distance from any target atom is less than the second preset distance. The target atom can be a carbon alpha atom, and the second preset distance can be 5 angstroms, that is, the electronic device can determine the coordinate positions of all carbon alpha atoms in the chain structure of the current amino acid sequence after determining the chain structure, and remove the support points whose distance from any carbon alpha atom is less than 5 angstroms.

[0142] Optionally, before determining the coordinates of the target atom position in the amino acid sequence, the chain structure needs to be rotated and translated according to the correspondence between the support points and the residues, so that the sum of the distances between the corresponding support points and the residues is minimized. After adjusting the spatial position of the amino acid sequence corresponding to the chain structure, the coordinate position of the target atom is calculated to improve the accuracy of the support point filtering results.

[0143] Step S50: In response to predicting the chain structures corresponding to all the amino acid sequences, determining the protein structure of the complex protein according to the chain structure corresponding to each of the amino acid sequences.

[0144] In a possible implementation, after predicting the chain structure of all amino acid sequences, the electronic device jointly determines the protein structure of the complex protein as a whole according to the chain structure corresponding to each amino acid sequence. Optionally, the electronic device can first adjust the chain structure according to the support point position corresponding to each amino acid sequence, and then splice each chain structure after the position adjustment to obtain the protein structure of the complex protein. Among them, the position adjustment of the chain structure includes translation and rotation. The electronic device can determine the position of each residue in the amino acid sequence, and the support point position corresponding to each residue, and then calculate a rotation and translation matrix based on the support point position and the residue position through the SVDSuperimposer library of BioPython, and obtain the chain structure aligned with the corresponding area of ​​the electron density map after the chain structure is processed by the rotation and translation matrix. Further, the electronic device adjusts the chain structure of all amino acid sequences in the complex protein and splices them to obtain the protein structure of the complex protein.

[0145] Figure 3 A schematic diagram of a complex protein structure according to an embodiment of the present disclosure is shown. Figure 3 As shown, the complex protein structure determined by the complex protein structure determination method according to the embodiment of the present disclosure can fit well with the real structure, that is, the prediction result has high accuracy.

[0146] Based on the above technical features, the complex protein structure determination method of the disclosed embodiment can predict the complex protein by accurately predicting the chain structure of each amino acid sequence in order based on the electron density map, and then predicting the protein folding structure of the complex protein based on all the chain structures. At the same time, the method of erasing some supporting points after each prediction of the chain structure reduces the interference in the prediction process and improves the accuracy of the prediction results.

[0147] Figure 4 A schematic diagram of a complex protein structure determination device according to an embodiment of the present disclosure is shown. Figure 4 As shown, the complex protein structure determination device of the embodiment of the present disclosure may include:

[0148] An information determination module 40, for determining an electron density map of a complex protein and a plurality of amino acid sequences constituting the complex protein;

[0149] A support point extraction module 41 is used to extract support points in the electron density map, where the support points are pixel points located on the main chain of the complex protein in the electron density map;

[0150] A chain structure determination module 42, used to predict the chain structure corresponding to each of the amino acid sequences in sequence according to a preset prediction order;

[0151] A support point filtering module 43, configured to perform support point filtering according to the chain structure corresponding to the amino acid sequence after determining the chain structure corresponding to the amino acid sequence each time;

[0152] The protein structure determination module 44 is used to determine the protein structure of the complex protein according to the chain structure corresponding to each amino acid sequence in response to the predicted chain structures corresponding to all the amino acid sequences.

[0153] In a possible implementation, the support point extraction module 41 is further used to:

[0154] The probability prediction model obtained through training predicts the confidence corresponding to each pixel point in the electron density map, wherein the confidence degree is used to characterize the probability that the corresponding pixel point belongs to the main chain of the complex protein;

[0155] A plurality of supporting points are obtained by screening according to the confidence level corresponding to each pixel point.

[0156] In a possible implementation, the support point extraction module 41 is further used to:

[0157] Obtaining a first preset number of pixel points with the highest confidence values ​​as candidate points;

[0158] Calculate the distance between each candidate point and other candidate points multiple times in an iterative manner;

[0159] The candidate points with low confidence between the two candidate points with the smallest distance are removed until a second preset number of candidate points are screened out, the iterative process is stopped, and the current candidate point is determined as a supporting point.

[0160] In a possible implementation, the prediction order is to predict the supporting points corresponding to each amino acid sequence in order from long to short.

[0161] In a possible implementation, the chain structure determination module 42 is further configured to:

[0162] According to a preset prediction order, the support point pair representation matrix is ​​determined multiple times according to the current support points in an iterative manner;

[0163] Determine the support point-residue pair representation matrix corresponding to the current amino acid sequence;

[0164] The corresponding chain structure is predicted based on the current amino acid sequence and the corresponding support point pair representation matrix and support point-residue pair representation matrix.

[0165] In a possible implementation, the chain structure determination module 42 is further configured to:

[0166] The current amino acid sequence, the support point representation matrix corresponding to the amino acid sequence and the support point-residue pair representation matrix are input into the trained chain structure prediction model to predict the corresponding chain structure.

[0167] In a possible implementation, the chain structure determination module 42 is further configured to:

[0168] Inputting the amino acid sequence, the support point pair representation matrix and the support point-residue pair representation matrix into the chain structure prediction model, and processing the amino acid sequence through the pre-processing layer to obtain a residue pair representation matrix;

[0169] Performing at least one cyclic update on the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix through the cyclic update layer;

[0170] Determining the supporting points for matching the amino acid sequence according to the supporting point-residue pair representation matrix after the cyclic update;

[0171] The residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix after cyclic updating are processed by the structure prediction layer to obtain the chain structure corresponding to the amino acid sequence.

[0172] In a possible implementation, the chain structure determination module 42 is further configured to:

[0173] Determine the residue closest to each of the support points and probability values ​​corresponding to a plurality of preset distance intervals according to the support point-residue pair representation matrix after cyclic update;

[0174] Filter the support points according to the probability value of the distance interval corresponding to the distance between each support point and the nearest residue;

[0175] The supporting points obtained after filtering are deduplicated to obtain the supporting points matching the amino acid sequence.

[0176] In a possible implementation, the chain structure determination module 42 is further configured to:

[0177] Determine the residues that match each support point obtained after filtering;

[0178] Determine that the supporting point whose distance among the supporting points corresponding to each of the residues is less than the first preset distance and whose distance corresponding probability value is the largest is the supporting point for matching the amino acid sequence.

[0179] In a possible implementation, the support point filtering module 43 is further used to:

[0180] Determine the coordinate position of each target atom in the amino acid sequence according to the chain structure corresponding to the amino acid sequence;

[0181] Delete the support points whose distance from any target atom is less than the second preset distance.

[0182] In a possible implementation, the protein structure determination module 44 is further used to:

[0183] According to the supporting point position corresponding to each amino acid sequence, the chain structure is subjected to position adjustment processing;

[0184] The protein structure of the complex protein is obtained by adjusting the splicing position of each chain structure.

[0185] In a possible implementation, the training process of the chain structure prediction model includes:

[0186] determining a plurality of sample amino acid sequences having corresponding annotated chain structures and sample density plots;

[0187] Each of the sample amino acid sequences is randomly cropped, and model training is performed based on the cropped sample amino acid sequences, the corresponding sample density map, and the annotated chain structure to obtain a chain structure prediction model.

[0188] In a possible implementation, randomly cropping each of the sample amino acid sequences, and performing model training according to the cropped sample amino acid sequences and the corresponding sample density map and the annotated chain structure to obtain a chain structure prediction model includes:

[0189] Determining a first trimming interval for each of the sample amino acid sequences;

[0190] trimming the sample density map corresponding to each of the sample amino acid sequences according to a second trimming interval including the first trimming interval;

[0191] Model training is performed based on the cropped sample amino acid sequence, the cropped sample density map and the annotated chain structure to obtain a chain structure prediction model.

[0192] In one possible implementation, the loss in the chain structure prediction model training process includes a point segmentation loss determined according to the support point pair representation matrix after cyclic updating, and the point segmentation loss is used to characterize whether two support points that actually belong to the same sequence belong to the same sequence in the prediction result.

[0193] In a possible implementation manner, the process of determining the point segmentation loss includes:

[0194] Marking the true label of the support point pair representation matrix corresponding to each of the sample density graphs according to a preset rule, wherein the true label is used to indicate whether the corresponding support point pairs belong to the same sequence;

[0195] Input the support point pair representation matrix after cyclic update into the linear layer to obtain the predicted label of each support point pair belonging to the same sequence;

[0196] The cross entropy loss is calculated according to the true label and the predicted label corresponding to each sample density map to obtain the point segmentation loss.

[0197] In some embodiments, the functions or modules included in the device provided by the embodiments of the present disclosure can be used to execute the method described in the above method embodiments. The specific implementation can refer to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0198] The embodiment of the present disclosure also provides a computer-readable storage medium on which computer program instructions are stored, and the computer program instructions implement the above method when executed by a processor. The computer-readable storage medium can be a volatile or non-volatile computer-readable storage medium.

[0199] An embodiment of the present disclosure further proposes an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to implement the above method when executing the instructions stored in the memory.

[0200] The embodiments of the present disclosure also provide a computer program product, including a computer-readable code, or a non-volatile computer-readable storage medium carrying the computer-readable code. When the computer-readable code runs in a processor of an electronic device, the processor in the electronic device executes the above method.

[0201] Figure 5 A schematic diagram of an electronic device 800 according to an embodiment of the present disclosure is shown. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.

[0202] Reference Figure 5 , the electronic device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .

[0203] The processing component 802 generally controls the overall operation of the electronic device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the above-mentioned method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.

[0204] The memory 804 is configured to store various types of data to support operations on the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0205] The power supply component 806 provides power to the various components of the electronic device 800. The power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the electronic device 800.

[0206] The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the electronic device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

[0207] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the electronic device 800 is in an operating mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.

[0208] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.

[0209] The sensor assembly 814 includes one or more sensors for providing various aspects of status assessment for the electronic device 800. For example, the sensor assembly 814 can detect the open / closed state of the electronic device 800, the relative positioning of the components, such as the display and keypad of the electronic device 800, and the sensor assembly 814 can also detect the position change of the electronic device 800 or a component of the electronic device 800, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration / deceleration of the electronic device 800, and the temperature change of the electronic device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

[0210] The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.

[0211] In an exemplary embodiment, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above methods.

[0212] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 804 including computer program instructions, which can be executed by a processor 820 of an electronic device 800 to perform the above method.

[0213] Figure 6 1 is a schematic diagram showing another electronic device 1900 according to an embodiment of the present disclosure. For example, the electronic device 1900 may be provided as a server or a terminal device. Figure 6 , the electronic device 1900 includes a processing component 1922, which further includes one or more processors, and a memory resource represented by a memory 1932 for storing instructions executable by the processing component 1922, such as an application. The application stored in the memory 1932 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1922 is configured to execute instructions to perform the above method.

[0214] The electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to a network, and an input / output (I / O) interface 1958. The electronic device 1900 may operate based on an operating system stored in the memory 1932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™ or the like.

[0215] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to perform the above method.

[0216] The present disclosure may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present disclosure.

[0217] A computer-readable storage medium may be a tangible device that can hold and store instructions used by an instruction execution device. A computer-readable storage medium may be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media (a non-exhaustive list) include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which instructions are stored, and any suitable combination of the foregoing. As used herein, a computer-readable storage medium is not to be interpreted as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through a wire.

[0218] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0219] The computer program instructions for performing the operation of the present disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed completely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or completely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be customized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present disclosure.

[0220] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0221] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0222] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0223] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.

[0224] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A method for determining the structure of a complex protein, characterized in that: The method comprises: Determining the electron density map of the complex protein and the multiple amino acid sequences constituting the complex protein; Extracting supporting points in the electron density map, wherein the supporting points are pixel points located on the main chain of the complex protein in the electron density map; Predicting the chain structure corresponding to each of the amino acid sequences in sequence according to a preset prediction order; After determining the chain structure corresponding to the amino acid sequence each time, filtering the supporting points according to the chain structure corresponding to the amino acid sequence; In response to predicting the chain structures corresponding to all the amino acid sequences, the protein structure of the complex protein is determined according to the chain structure corresponding to each of the amino acid sequences.

2. The method according to claim 1, characterized in that The extracting the supporting points in the electron density map comprises: The probability prediction model obtained through training predicts the confidence corresponding to each pixel point in the electron density map, wherein the confidence degree is used to characterize the probability that the corresponding pixel point belongs to the main chain of the complex protein; A plurality of supporting points are obtained by screening according to the confidence level corresponding to each pixel point.

3. The method according to claim 2, characterized in that The step of screening and obtaining a plurality of support points according to the confidence level corresponding to each pixel point includes: Obtaining a first preset number of pixel points with the highest confidence values ​​as candidate points; Calculate the distance between each candidate point and other candidate points multiple times in an iterative manner; The candidate points with low confidence between the two candidate points with the smallest distance are removed until a second preset number of candidate points are screened out, the iterative process is stopped, and the current candidate point is determined as a supporting point.

4. The method according to any one of claims 1 to 3, characterized in that: The prediction order is to predict the supporting points corresponding to each amino acid sequence in order from long to short.

5. The method according to any one of claims 1 to 4, characterized in that: The predicting of the chain structure corresponding to each of the amino acid sequences comprises: According to a preset prediction order, the support point pair representation matrix is ​​determined multiple times according to the current support points in an iterative manner; Determine the support point-residue pair representation matrix corresponding to the current amino acid sequence; The corresponding chain structure is predicted based on the current amino acid sequence and the corresponding support point pair representation matrix and support point-residue pair representation matrix.

6. The method according to claim 5, characterized in that The predicting of the corresponding chain structure according to the current amino acid sequence and the corresponding support point pair representation matrix and support point-residue pair representation matrix includes: The current amino acid sequence, the support point representation matrix corresponding to the amino acid sequence and the support point-residue pair representation matrix are input into the trained chain structure prediction model to predict the corresponding chain structure.

7. The method according to claim 6, characterized in that The step of inputting the current amino acid sequence, the support point representation matrix and the support point-residue pair representation matrix corresponding to the amino acid sequence into the trained chain structure prediction model to predict the corresponding chain structure comprises: Inputting the amino acid sequence, the support point pair representation matrix and the support point-residue pair representation matrix into the chain structure prediction model, and processing the amino acid sequence through the pre-processing layer to obtain a residue pair representation matrix; Performing at least one cyclic update on the residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix through the cyclic update layer; Determining the supporting points for matching the amino acid sequence according to the supporting point-residue pair representation matrix after the cyclic update; The residue pair representation matrix, the support point pair representation matrix and the support point-residue pair representation matrix after cyclic updating are processed by the structure prediction layer to obtain the chain structure corresponding to the amino acid sequence.

8. The method according to claim 7, characterized in that The step of determining the supporting points for the amino acid sequence matching according to the supporting point-residue pair representation matrix after the cyclic update includes: Determine the residue closest to each of the support points and probability values ​​corresponding to a plurality of preset distance intervals according to the support point-residue pair representation matrix after cyclic update; Filter the support points according to the probability value of the distance interval corresponding to the distance between each support point and the nearest residue; The supporting points obtained after filtering are deduplicated to obtain the supporting points matching the amino acid sequence.

9. The method according to claim 8, characterized in that The step of removing duplicates from the support points obtained after filtering to obtain the support points matching the amino acid sequence includes: Determine the residues that match each support point obtained after filtering; Determine that the supporting point whose distance among the supporting points corresponding to each of the residues is less than the first preset distance and whose distance corresponding probability value is the largest is the supporting point for matching the amino acid sequence.

10. The method according to any one of claims 1 to 9, characterized in that: The filtering of supporting points according to the chain structure corresponding to the amino acid sequence includes: Determine the coordinate position of each target atom in the amino acid sequence according to the chain structure corresponding to the amino acid sequence; Delete the support points whose distance from any target atom is less than the second preset distance.

11. The method according to any one of claims 1 to 10, characterized in that: Determining the protein structure of the complex protein according to the chain structure corresponding to each of the amino acid sequences comprises: According to the supporting point position corresponding to each amino acid sequence, the chain structure is subjected to position adjustment processing; The protein structure of the complex protein is obtained by adjusting the splicing position of each chain structure.

12. The method according to any one of claims 6 to 9, characterized in that: The training process of the chain structure prediction model includes: determining a plurality of sample amino acid sequences having corresponding annotated chain structures and sample density plots; Each of the sample amino acid sequences is randomly cropped, and model training is performed based on the cropped sample amino acid sequences, the corresponding sample density map, and the annotated chain structure to obtain a chain structure prediction model.

13. The method according to claim 12, characterized in that The step of randomly clipping each of the sample amino acid sequences and performing model training according to the clipped sample amino acid sequences and the corresponding sample density map and the annotated chain structure to obtain a chain structure prediction model includes: Determining a first trimming interval for each of the sample amino acid sequences; trimming the sample density map corresponding to each of the sample amino acid sequences according to a second trimming interval including the first trimming interval; Model training is performed based on the cropped sample amino acid sequence, the cropped sample density map and the annotated chain structure to obtain a chain structure prediction model.

14. The method according to claim 12 or 13, characterized in that The loss in the chain structure prediction model training process includes a point segmentation loss determined according to the support point pair representation matrix after cyclic updating, and the point segmentation loss is used to characterize whether two support points that actually belong to the same sequence belong to the same sequence in the prediction result.

15. The method according to claim 14, characterized in that The process of determining the point segmentation loss includes: Marking the true label of the support point pair representation matrix corresponding to each of the sample density graphs according to a preset rule, wherein the true label is used to indicate whether the corresponding support point pairs belong to the same sequence; Input the support point pair representation matrix after cyclic update into the linear layer to obtain the predicted label of each support point pair belonging to the same sequence; The cross entropy loss is calculated according to the true label and the predicted label corresponding to each sample density map to obtain the point segmentation loss.

16. A complex protein structure determination device, characterized in that: The device comprises: An information determination module, used to determine the electron density map of the complex protein and a plurality of amino acid sequences constituting the complex protein; A support point extraction module, used to extract support points in the electron density map, wherein the support points are pixel points located on the main chain of the complex protein in the electron density map; A chain structure determination module, used to predict the chain structure corresponding to each of the amino acid sequences in sequence according to a preset prediction order; A support point filtering module, used for filtering support points according to the chain structure corresponding to the amino acid sequence after determining the chain structure corresponding to the amino acid sequence each time; The protein structure determination module is used to determine the protein structure of the complex protein according to the chain structure corresponding to each amino acid sequence in response to the predicted chain structure corresponding to all the amino acid sequences.

17. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; Wherein, the processor is configured to implement the method described in any one of claims 1 to 15 when executing the instructions stored in the memory.

18. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 15 is implemented.

Images