Optical quantum computer-based protein structure prediction method, system and apparatus

Abstract

The present invention relates to the technical field of protein structure prediction, and relates to an optical quantum computer-based protein structure prediction method, system and apparatus. The method comprises: 1) obtaining a multiple sequence alignment (MSA) matrix of a target sequence on the basis of target sequence alignment of a protein needing to be predicted; 2) encoding the obtained MSA matrix into a (0,1) matrix; 3) converting interaction of amino acids of the protein into an undirected graph model of (0,1) state nodes, and using a Boltzmann machine training mechanism and an optical quantum computer to perform training to obtain weight coefficients of edges connecting nodes in the undirected graph model; and 4) obtaining coefficients of interaction between different amino acids of the protein on the basis of the weight coefficients. The present invention improves the computation efficiency and solution result of protein structure prediction, and solves the problems in the prior art that it is difficult to perform computation for complex scheduling problems and accurate solutions cannot be obtained.

Description

Protein structure prediction methods, systems, and devices based on optical quantum computers Technical Field

[0001] This invention belongs to the field of protein structure prediction technology, and relates to a protein structure prediction method, system and device, and particularly to a protein structure prediction method, system and device based on optical quantum computers. Background Technology

[0002] Protein structure prediction is of paramount importance in life science research. In basic biology, proteins are among the most functionally important biomolecules, formed by the transcription of DNA sequences into mRNA, which is then translated into proteins to perform their primary functions. In biology, structure often determines the function of biomolecules. After mRNA is translated into the corresponding amino acid chains, it folds in space to form specific 3D structures. Therefore, obtaining precise spatial structures is fundamental to understanding protein function and developing drugs.

[0003] Current methods for obtaining protein structures include experimental methods and AI-based prediction methods. Experimental methods mainly include X-ray crystallography, nuclear magnetic resonance (NMR), and cryo-electron microscopy (Cryo-EM). By purifying and crystallizing protein samples, and then using the above experimental devices for structural analysis, this method can obtain accurate three-dimensional structures. However, it consumes a lot of human and financial resources, and the purification of some proteins is uncertain, making it difficult to obtain relatively pure protein samples. AI-based structure prediction methods, such as AlphaFold and RoseTTAFold, demonstrate higher efficiency than traditional experimental methods, obtaining the three-dimensional structure of proteins in a short time. However, they still have several shortcomings, mainly: 1. Insufficient sensitivity to mutations: These AI tools are not very sensitive to point mutations (single amino acid changes); 2. Difficulty in predicting dynamic regions: These AI tools target regions with fixed structures in their natural state. Therefore, current AI tools cannot accurately predict the structure of dynamic and disordered regions. The disordered structure of proteins plays an important role in protein function, such as phase separation processes. Abnormal protein phase separation processes are significantly correlated with the occurrence of Alzheimer's disease and cancer; 3. Inaccurate prediction of some rare proteins: These proteins have low similarity to currently known structural proteins, resulting in inaccurate predictions; 4. These AI tools still have limitations in simulating the physical interaction between drugs and proteins.

[0004] Therefore, in order to address the shortcomings of the existing technologies, it is necessary to develop a novel method, system, and device for protein structure prediction. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, this invention proposes a protein structure prediction method, system, and device based on optical quantum computers, which improves the computational efficiency and solution results of protein structure prediction and solves the problems of existing technologies being unable to compute complex scheduling problems and obtain accurate solutions.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A protein structure prediction method based on optical quantum computers, characterized by comprising the following steps:

[0008] 1) Obtain the multiple sequence alignment (MSA) matrix of the target sequence based on the alignment of the target sequence of the protein to be predicted;

[0009] 2) Encode the obtained multiple sequence alignment (MSA) matrix into a {0,1} matrix;

[0010] 3) The amino acid interactions of the protein are converted into an undirected graph model with {0,1} state nodes and trained using a Boltzmann machine training mechanism with an optical quantum computer to obtain the weight coefficients of the edges between nodes in the undirected graph model.

[0011] 4) Based on the weight coefficients of the edges connecting nodes in the undirected graph model, the interaction coefficients between different amino acids in the protein are obtained.

[0012] Preferably, step 1) specifically includes:

[0013] 11) Download the target sequence of the protein to be predicted from the protein sequence database;

[0014] 12) Using the downloaded target sequence as input, the multiple sequence alignment (MSA) matrix is ​​obtained by aligning the sequence with the multiple sequence alignment tool mmseqs2 and the Swiss-Prot protein database.

[0015] Preferably, step 2) specifically involves: performing one-hot encoding on the string-encoded multiple sequence alignment (MSA) matrix, with 21 choices for each amino acid position. For matrix alignment, empty spaces are represented by "-". An L*21 dimensional {0,1} matrix is ​​used to represent the information of one row in the multiple sequence alignment MSA matrix. The multiple sequence alignment MSA matrix is ​​transformed to obtain M L*21 dimensional {0,1} matrices, where L is the sequence length of the amino acid and M is the number of protein pairs obtained from the alignment.

[0016] Preferably, in step 3), when converting the amino acid interactions of a protein into an undirected graph model of {0,1} state nodes, firstly, the amino acids of the protein sequence to be predicted are used as vertices, and the interactions between amino acids are used as edges to establish a fully connected undirected graph; then, since each amino acid is a 21-dimensional one-hot vector, each amino acid is converted into 21 {0,1} state nodes, and the connection between any two {0,1} state nodes is used as edges to establish the undirected graph model of the {0,1} state nodes, wherein the state of each state node in the undirected graph model of the {0,1} state nodes is 0 or 1, and the weight is h. i The weight coefficient of the edge connecting any two state nodes is J. ij .

[0017] Preferably, in step 3), when using a Boltzmann machine training mechanism with a quantum computer to train the undirected graph model and obtain the weight coefficients of the edges connecting nodes, the training loss function C is:

[0018] In the formula, D kl Let N(θ) represent the KL divergence, and let N(θ) represent the negative log-conditional likelihood. DS The number of {0,1}} state nodes in the undirected graph model is represented by α, where α is a coefficient and α∈[0,1].

[0019] Preferably, step 4) specifically includes:

[0020] 41) Based on the weight coefficient J of the edges connecting nodes in the undirected graph model ij Establish the weight coefficient matrix J;

[0021] 42) Extract the submatrix J′ = [i*21:i*21+21,j*21:j*21+21] from the weight coefficient matrix J. This submatrix represents the coupling information between the 21*21 {0,1} state nodes corresponding to the i-th amino acid and the 21*21 {0,1} state nodes corresponding to the j-th amino acid.

[0022] 43) The submatrix is ​​mean-centered as follows to obtain the mean-centered matrix J″: J″=J′-μ0-μ1+μ

[0023] In the formula, μ0 is the column mean, μ1 is the row mean, and μ is the overall mean of the submatrix;

[0024] 44) The interaction coefficient S between the i-th amino acid and the j-th amino acid is calculated using the mean-centered matrix J″. ij S ij =||J″||2.

[0025] Furthermore, the present invention also provides a protein structure prediction system based on optical quantum computers, characterized in that it comprises:

[0026] The alignment module is used to obtain the multiple sequence alignment (MSA) matrix of the target sequence based on the target sequence of the protein to be predicted.

[0027] The encoding module is used to encode the obtained multiple sequence alignment (MSA) matrix into a {0,1} matrix.

[0028] The weight coefficient acquisition module is used to convert the amino acid interactions of the protein into an undirected graph model of {0,1} state nodes and train it using a Boltzmann machine training mechanism with a quantum computer to obtain the weight coefficients of the edges between nodes in the undirected graph model.

[0029] The interaction coefficient acquisition module is used to obtain the interaction coefficients between different amino acids of the protein based on the weight coefficients of the edges connecting the nodes in the undirected graph model.

[0030] Furthermore, the present invention also provides a protein structure prediction device based on an optical quantum computer, characterized in that it comprises:

[0031] One or more processors;

[0032] Memory, used to store one or more programs;

[0033] When the one or more programs are executed by the one or more processors, the one or more processors implement the protein structure prediction method based on optical quantum computers as described above.

[0034] Finally, the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements the steps of the protein structure prediction method based on optical quantum computers as described above.

[0035] Compared with the prior art, the protein structure prediction method, system, and apparatus based on optical quantum computers of the present invention have one or more of the following beneficial technical effects:

[0036] 1. This invention uses a computational method to predict protein structure from sequence. Compared with traditional experimental methods such as X-ray crystallography, nuclear magnetic resonance (NMR) and cryo-electron microscopy (Cryo-EM), it can obtain results in a short time, avoid the complicated steps of protein purification and expensive observation experimental equipment, and lower the threshold for protein structure analysis.

[0037] 2. Compared to AI-based methods such as AlphaFold and RoseTTAFold, this invention does not require a large amount of high-quality training data. Since some low-similarity proteins and highly complex proteins do not have known structures that can be trained, these AI-based methods cannot accurately predict the structures of these proteins. This invention does not rely on known structure data; it only needs to combine protein evolutionary information to train the relevant parameters.

[0038] 3. Currently, all technical methods, whether experimental or AI-based, only target segments with fixed structures for protein structure analysis, lacking the ability to analyze the structure of disordered segments. These disordered segments are crucial for proteins to perform their normal functions, promoting processes such as protein-protein phase separation, which are associated with various diseases such as Alzheimer's disease, diabetes, and cardiovascular disease. Therefore, the limitations of existing methods prevent the analysis of the structure of disordered segments, resulting in the lack of drugs targeting these segments. This invention can obtain the structural state of disordered segments by sampling their structural distribution through energy minimization.

[0039] 4. In the training process of Boltzmann machines, traditional methods based on contrastive divergence use other distributions to approximate the Boltzmann distribution. At the same time, due to the parameter complexity brought by fully connected components, the Boltzmann machine model has not been widely used. This invention uses an optical quantum computer to directly obtain the real Boltzmann distribution by performing energy sampling based on the quantum computer, and increases the number of nodes in the Boltzmann machine model to the scale of thousands, while traditional methods have difficulty converging with only twenty nodes. Attached Figure Description

[0040] Figure 1 is a flowchart of the protein structure prediction method based on optical quantum computer of the present invention.

[0041] Figure 2 shows a partial example of the MSA matrix obtained from the alignment.

[0042] Figure 3 shows a schematic diagram of the MSA matrix encoded as a {0,1} matrix.

[0043] Figure 4 is a schematic diagram of an exemplary undirected graph model of the present invention, which converts amino acid interactions into {0, 1} state nodes.

[0044] Figure 5 is a schematic diagram of the interaction coefficients between amino acids extracted from the weight coefficient matrix of the edges connecting nodes in the trained undirected graph model.

[0045] Figure 6 shows a schematic diagram of the protein structure prediction system based on optical quantum computer of the present invention. Detailed Implementation

[0046] Before detailing any embodiment of the invention, it should be understood that the invention, in its application, is not limited to the details of the construction and arrangement of the components set forth in the following description or illustrated in the following figures. The invention can have other embodiments and can be practiced or carried out in various ways. Furthermore, it should be understood that the wording and terminology used herein are for descriptive purposes and should not be considered limiting. The use of “comprising” or “having” and variations thereof is intended to cover the items set forth below and their equivalents, as well as any additional items. Unless otherwise specified or limited, the terms “installation,” “connection,” “support,” and “linkage,” and variations thereof are used broadly and cover both direct and indirect installation, connection, support, and linking. Moreover, “connection” and “linkage” are not limited to physical or mechanical connections or links.

[0047] Furthermore, in the disclosure of this invention, the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0048] During protein evolution, amino acid mutations can occur, disrupting the interactions between residues and leading to protein structural instability. If residues that interact with the mutated residues also mutate, forming new interactions and stabilizing the protein structure, this change is called protein co-evolution. Therefore, by utilizing the evolutionary information of proteins and comparing sequences of a large number of homologous proteins to find common evolutionary amino acid pairs, the amino acid residues that are close to each other in the protein's 3D structure can be deduced. Based on global amino acid interaction information, the spatial structure of the protein can be obtained.

[0049] Based on the above principles, this invention provides a protein structure prediction method, system, and device based on optical quantum computers. By binarizing protein sequence alignment data and introducing a Boltzmann machine training mechanism, a mathematical model suitable for quantum computing training is designed. Then, a quantum computer is used to sample and solve the problem, thereby improving the computational efficiency and solution results of protein structure prediction. This solves the problem that existing technologies are difficult to compute for complex scheduling problems and cannot obtain accurate solutions.

[0050] Figure 1 shows a flowchart of the protein structure prediction method based on optical quantum computers according to the present invention. As shown in Figure 1, the protein structure prediction method based on optical quantum computers according to the present invention includes the following steps:

[0051] 1. Obtain the multiple sequence alignment (MSA) matrix of the target sequence based on the alignment of the target sequence of the protein to be predicted.

[0052] In this invention, the target sequence of the protein to be predicted is first obtained. For example, a .fa format sequence file is downloaded from a protein sequence database such as SWISS-PROT. This file contains the amino acid sequence information of the protein to be predicted. As an example, the specific format of the full-length amino acid sequence file of the human FIND3A protein with the Fibronection_III domain is shown below.

[0053] >sp|Q9Y2H6|FND3A_HUMAN Fibronectin type-III domain-containing protein 3A OS=Homo sapiens OX=9606 GN=FNDC3A PE=1 SV=4

[0054] Next, using the protein sequences obtained above as input, multiple sequence alignment (MSA) matrices were obtained by comparing the sequences with the mmseqs2s tool and the Swiss-Prot protein database, as shown in Figure 2. In Figure 2, each row represents a protein sequence of one species. The sequence length of amino acids in Figure 2 is L, that is, it has L amino acids, and the number of aligned protein sequences is M, that is, there are M homologous proteins.

[0055] 2. Encode the obtained multiple sequence alignment (MSA) matrix into a {0,1} matrix.

[0056] The multiple sequence alignment (MSA) matrix is ​​a string-encoded matrix. In this invention, the string-encoded MSA matrix needs to be one-hot encoded. There are 21 choices for the amino acid at each position (i.e., 21 amino acid types are available). For matrix alignment, empty spaces are represented by "-". A single row of the MSA matrix is ​​represented by an L*21 dimensional {0,1} matrix. The MSA matrix is ​​transformed into M L*21 dimensional {0,1} matrices, where L is the amino acid sequence length and M is the number of aligned protein segments. For example, in Figure 3, the left half of the MSA matrix is ​​represented by an L*21 {0,1} matrix for each row, resulting in the right half of the one-hot matrix, i.e., the {0,1} matrix.

[0057] In the L*21 dimensional {0,1} matrix, one dimension in each row represents a type of amino acid. During the conversion, the type of amino acid corresponding to each row is determined. Therefore, the state of the dimension in that row corresponding to that type of amino acid is 1, and the remaining 20 dimensions are 0.

[0058] Third, the amino acid interactions of the protein are converted into an undirected graph model with {0,1} state nodes, and a Boltzmann machine training mechanism is used to train the model with an optical quantum computer to obtain the weight coefficients of the edges connecting the nodes in the undirected graph model.

[0059] For a stable protein, a thermodynamic statistical model can be used to assume that the protein's amino acid sequence follows a Boltzmann distribution:

[0060] In the formula, P represents the global probability of the amino acid distribution, and a1,…,a L Z represents the specific types of amino acids from position 1 to position L, and Z represents the partition function.

[0061] Where Z is the partition function; β is the reciprocal temperature; E i It refers to the energy of a system in a certain state, and in this invention, it refers to the energy corresponding to a certain amino acid distribution.

[0062] H(a1…a L This represents the sum of the energies of all amino acids under the Boltzmann distribution, including the energy of a single amino acid and the energy of the interaction between pairs of amino acids. Specifically, it can be written as:

[0063] In the formula, a i Let a represent the amino acid at position i. j H represents the amino acid at position j. i J represents the energy of the amino acid at position i. ij For amino acids to {a i ,a j}, that is, the interaction energy between the amino acid at position i and the amino acid at position j.

[0064] The objective of this invention is to learn the parameter θ = {h} from the MSA matrix obtained through alignment. i J ij}, such that arg maxp(x|θ)=arg max Πp(x i |θ)=argmax∑logp(x i |θ) (4)

[0065] In the formula, p(x|θ) represents the expression in the case of parameter θ={h i J ij The joint probability distribution of all occurrences of x under the given condition is the joint probability distribution of the occurrences of M data points; p(x) i |θ) represents the parameter θ={h i J ij}down xi The probability of occurrence, x in this invention i This represents the state distribution corresponding to a single amino acid data point; in this parameter θ = {h i J ij The following equation (1) has H(a1,...,a) L (To obtain the minimum value)

[0066] In order to train and obtain the above parameters θ={h} from the MSA matrix i J ij In this invention, a fully connected undirected graph model, namely an undirected graph model with {0,1} state nodes, is used to represent the interaction rules of protein amino acid residues, and the energy function of the whole equation (3) is mapped to the exp space.

[0067] Specifically, when converting the amino acid interactions of a protein into an undirected graph model of {0,1} state nodes, firstly, the amino acids of the protein sequence to be predicted are used as vertices, and the interactions between amino acids are used as edges to establish a fully connected undirected graph. Then, since each amino acid is a 21-dimensional one-hot vector, each amino acid is converted into 21 {0,1} state nodes. The connection between any two {0,1} state nodes is used as edges to establish the undirected graph model of the {0,1} state nodes. Each state node in the undirected graph model of the {0,1} state nodes has a state of 0 or 1 and a weight of h. i The weight coefficient of the edge connecting any two state nodes is J. ij .

[0068] For example, assuming a protein sequence is ABCD, meaning it has four amino acids, A, B, C, and D, it can be converted into a fully connected undirected graph, as shown in Figure 4. In the left graph of Figure 4, the four vertices correspond to the four amino acids of the protein sequence, and the six edges between the vertices represent the interactions between the amino acids. Each amino acid at a vertex is a 21-dimensional one-hot vector, which, after encoding, can be converted into the visible layer in the right graph of Figure 4 (note that, for simplicity, only some nodes in the visible layer are shown in the right graph of Figure 4). The value of each node in the visible layer of the right graph of Figure 4 is {0,1}. For example, node... like but A value of 1 indicates that the amino acid type of the i-th node is α; otherwise, it is 0. The interaction relationships of the four amino acids are ultimately transformed into an undirected graph model with 4*2¹ = 84 nodes in {0,1} states. The state of each node in the undirected graph model is either 0 or 1, corresponding to the one-hot matrix encoded by each amino acid in the MSA matrix. The final parameters θ = {h} are obtained by training from M such one-hot matrices.i J ij}. Among them, h i J represents the weight of the i-th node in an undirected graph model. ij The edge weight represents the interaction coefficient between the i-th node and the j-th node in the undirected graph model.

[0069] To solve for the parameter θ = {h i J ij This invention introduces a Boltzmann machine training mechanism, specifically training an undirected graph model with {0,1} state nodes as a Boltzmann machine model. During training, the nodes in this undirected graph model with {0,1} state nodes can be treated as visible layer nodes, with an additional small number of hidden layer nodes introduced, as shown in Figure 4 (right). Based on the characteristics of the converted one-hot encoded input data, the visible layer nodes are randomly divided into input and output nodes. For example, if there are 1000 nodes in the visible layer, 500 nodes can be used as visible layer input nodes, and the other 500 as visible layer output nodes, with approximately 20 additional hidden layer nodes introduced. Due to the increased parameter complexity caused by fully connected layers, the training effect of traditional Boltzmann machine models is not ideal. Furthermore, because the partition function is difficult to handle, other methods are often used to approximate the node state value probabilities to the Boltzmann distribution, resulting in unsatisfactory final network parameter training results.

[0070] Therefore, this invention proposes parameter training of the Boltzmann machine model based on an optical quantum computer. The training loss function C uses weighted KL divergence and negative log-likelihood, specifically:

[0071] In the formula, D kl Let N(θ) represent the KL divergence, and let N(θ) represent the negative log-conditional likelihood. DS This represents the number of visible layer nodes, that is, the number of {0,1} state nodes in the undirected graph model. α is a coefficient and α∈[0,1], and its specific value can be set as needed.

[0072] The specific training process is as follows: First, initialize the parameter θ = {h} i J ij In other words, the parameter h is given manually. i J ij And take the {0,1} matrix obtained by transforming the MSA matrix as input, using the initial parameter θ = {h i J ijThe Ising matrix obtained by encoding the Ising model using the following equation (6) is used to sample the states of the {0,1} nodes of the undirected graph model of the {0,1} state nodes on a photonic quantum computer, which have the minimum energy, and the probability of each state combination is obtained through the energy distribution. H=min(-∑h i x i +∑J ij x i x j (6)

[0073] In the formula, H represents the energy state corresponding to the amino acid combination; x i Let x represent the state of the amino acid at position i. j The state of the amino acid at position j is represented, with values ​​{0, 1}; h i J represents the energy of the amino acid at the i-th position; ij For amino acids to {a i ,a j}, that is, the interaction energy between the amino acid at position i and the amino acid at position j.

[0074] The Ising matrix is ​​obtained by encoding the Ising model above. It is then input into a quantum computer for sampling calculation, and the output shows the states of different amino acid combinations and the probabilities corresponding to different state combinations obtained through energy distribution. The KL divergence is calculated using the state combination distribution output by the Ising model in equation (6) and the input data distribution. The calculation formula is as follows:

[0075] Among them, D kl Let p(x) represent the KL divergence to be calculated, which is the KL divergence between the probability distribution of the original data of the {0,1} matrix and the probability distribution of the data output by the optical quantum computer through the Ising model. p(x) is the probability distribution of the original data of the {0,1} matrix, and q(x) is the probability distribution of the data output by the optical quantum computer through the Ising model, which is the state combination distribution.

[0076] The negative log-conditional likelihood N(θ) can be obtained as follows: Iteratively calculate the {0,1} matrix obtained from the M input MSA matrices. Using a quantum computer, infer the output {0,1} states and the probabilities of different state combinations from the input {0,1} states in the visible layer. The calculation method is as follows:

[0077] In the formula, N(θ) represents the expression for parameter θ = {h}. i J ij The negative log-conditional likelihood value under}, v I ,vO p(v) represents the state of the input node and the output node of the visible layer, respectively. O |v I ;θ) indicates that in parameter θ={h i J ij The undirected graphical model under input v I v appears O The conditional probability, v D This represents the state of the D-th input node in the visible layer.

[0078] The gradient is calculated from the sampled node state combinations and their corresponding probabilities. The specific calculation formula is as follows:

[0079] Where α is a coefficient and α∈[0,1], its specific value can be set as needed, and is the same as α in equation (5); Let represent the expected distribution of all node data (including visible layer nodes and hidden layer nodes) output by the optical quantum computer based on equation (6), which is the product of the output node data distribution and the probability of the state distribution; D represents the input dataset, i.e., M one-hot matrices; v represents a single input data, i.e., a single one-hot matrix data; q(v) represents the probability distribution of a single one-hot input data; N DS Indicates the number of visible layer nodes; The expected value of the visible layer node data distribution output by the optical quantum computer based on equation (6) is the product of the output visible layer node data distribution and the probability of the state distribution. Let represent the expected distribution of the visible layer input node data output by the optical quantum computer based on equation (6), which is the product of the output visible layer input node data distribution and the probability of the state distribution.

[0080] Based on the above calculations, the gradient update parameter θ = {h} i J ij}:

[0081] Where, θ new These are the updated parameters, where γ is the learning rate, which is an adjustable parameter.

[0082] Then, iterative training is performed using M input one-hot datasets until convergence, yielding the optimal parameters θ = {h}. i J ij}, thereby obtaining the weight coefficient J of the edges connecting nodes in the undirected graph model. ij .

[0083] IV. Based on the weight coefficients of the edges connecting nodes in the undirected graph model, the interaction coefficients between different amino acids in the protein are obtained.

[0084] The training method in step three yields the weight coefficients J of the edges connecting the 21*L state nodes in the undirected graph model. ij As shown in Figure 5, the weight coefficient J of the edges connecting the 21*L state nodes is finally determined. ij The interaction coefficient between L amino acids transferred to the cell.

[0085] In this invention, the specific conversion process is as follows:

[0086] To obtain the interaction coefficients between amino acids in the right figure of Figure 5, such as the interaction coefficient S between the i-th amino acid and the j-th amino acid. ij First, based on the weight coefficient J of the edges connecting nodes in the undirected graph model... ij Establish the weight coefficient matrix J shown in the left figure of Figure 5.

[0087] Secondly, from the weight coefficient matrix J of the edges connecting nodes in the undirected graph model in the left figure of Figure 5, a submatrix J′=[i*21:i*21+21,j*21:j*21+21] is extracted. This submatrix represents the coupling information between the 21*21 one-hot encoded nodes corresponding to the i-th amino acid and the 21*21 one-hot encoded nodes corresponding to the j-th amino acid.

[0088] Then, the submatrix J′ is mean-centered as follows: J″=J′-μ0-μ1+μ

[0089] In the formula, μ0 is the column mean, μ1 is the row mean, and μ is the overall mean of the submatrix.

[0090] Finally, the interaction coefficient between the i-th amino acid and the j-th amino acid is calculated using the mean-centered matrix J″. The specific calculation method is as follows: S ij =||J″||2.

[0091] This invention utilizes protein co-evolutionary information. By comparing sequences of numerous homologous proteins, it identifies common evolutionary amino acid pairs, allowing the inference of closely spaced amino acid residues in the protein's 3D structure. Based on global amino acid interaction information, the protein's spatial structure can be derived. Simultaneously, this invention encodes amino acid positions and types as {0,1} states and transforms them into a fully connected Boltzmann machine model (i.e., an undirected graphical model). By establishing the Ising model and training parameters using a quantum computer, the interaction relationships between amino acid residues are finally obtained. This invention can sample the structural states of disordered regions of proteins without relying on known protein structures, and simultaneously increases the trainable node scale of the Boltzmann machine model to thousands.

[0092] Furthermore, this invention also provides a protein structure prediction system based on an optical quantum computer. As shown in Figure 6, the protein structure prediction system based on an optical quantum computer of this invention includes:

[0093] 1. Comparison module.

[0094] The alignment module is used to obtain a multiple sequence alignment (MSA) matrix of the target sequence based on the target sequence of the protein to be predicted.

[0095] 2. Encoding module.

[0096] The editing module is used to encode the obtained multiple sequence alignment (MSA) matrix into a {0,1} matrix.

[0097] 3. Weight coefficient acquisition module.

[0098] The weight coefficient acquisition module is used to convert the amino acid interactions of the protein into an undirected graph model with {0,1} state nodes and train it using a Boltzmann machine training mechanism with a quantum computer to obtain the weight coefficients of the edges connecting the nodes in the undirected graph model.

[0099] 4. Interaction coefficient acquisition module.

[0100] The interaction coefficient acquisition module is used to obtain the interaction coefficients between different amino acids of the protein based on the weight coefficients of the edges connecting nodes in the undirected graph model.

[0101] Furthermore, the present invention also provides a protein structure prediction device based on an optical quantum computer, comprising: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the protein structure prediction method based on an optical quantum computer as described above.

[0102] Finally, the present invention also provides a computer storage medium storing a computer program, characterized in that the program, when executed by a processor, implements the steps of the protein structure prediction method based on optical quantum computers as described above.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Those skilled in the art can modify or make equivalent substitutions to the technical solutions of the present invention based on the concept of the present invention, without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A method for predicting protein structure based on optical quantum computer, characterized in that, Includes the following steps: 1) Obtain the multiple sequence alignment (MSA) matrix of the target sequence based on the alignment of the target sequence of the protein to be predicted; 2) Encode the obtained multi-sequence alignment MSA matrix into a {0,1} matrix; 3) The amino acid interactions of the protein are converted into an undirected graph model with {0,1} state nodes and trained using a Boltzmann machine training mechanism with an optical quantum computer to obtain the weight coefficients of the edges between nodes in the undirected graph model. 4) Based on the weight coefficients of the edges connecting nodes in the undirected graph model, the interaction coefficients between different amino acids in the protein are obtained.

2. The protein structure prediction method based on optical quantum computer according to claim 1, characterized in that, Step 1) specifically includes: 11) Download the target sequence of the protein to be predicted from the protein sequence database; 12) Using the downloaded target sequence as input, the multiple sequence alignment (MSA) matrix is ​​obtained by aligning the sequence with the multiple sequence alignment tool mmseqs2 and the Swiss-Prot protein database.

3. The method of claim 1, wherein the method is performed by a photonic quantum computer. Step 2) specifically involves: performing one-hot encoding on the string-encoded multiple sequence alignment (MSA) matrix. There are 21 choices for the amino acid at each position. For matrix alignment, empty spaces are represented by "-". An L*21 dimensional {0,1} matrix is ​​used to represent the information of one row in the multiple sequence alignment MSA matrix. The multiple sequence alignment MSA matrix is ​​transformed to obtain M L*21 dimensional {0,1} matrices, where L is the sequence length of the amino acid and M is the number of protein pairs obtained from the alignment.

4. The protein structure prediction method based on optical quantum computer according to claim 1, characterized in that, In step 3), when converting the amino acid interactions of a protein into an undirected graph model of {0,1} state nodes, firstly, the amino acids of the protein sequence to be predicted are used as vertices, and the interactions between amino acids are used as edges to establish a fully connected undirected graph. Then, since each amino acid is a 21-dimensional one-hot vector, each amino acid is converted into 21 {0,1} state nodes. The connection between any two {0,1} state nodes is used as edges to establish the undirected graph model of the {0,1} state nodes. Each state node in the undirected graph model of the {0,1} state nodes has a state of 0 or 1 and a weight of h. i The weight coefficient of the edge connecting any two state nodes is J. ij .

5. The protein structure prediction method based on optical quantum computer according to claim 4, characterized in that, In the step 3), the training is performed using a photonic quantum computer with a Boltzmann machine training mechanism to obtain the weight coefficients of the edges between the nodes in the undirected graph model, and the loss function C of the training is: where D kl denotes the KL divergence, N(0) denotes the negative log conditional likelihood, N DS denotes the number of {0,1} state nodes in the undirected graph model, a is a coefficient and a e [0,1].

6. The protein structure prediction method based on optical quantum computer according to claim 5, wherein, Step 4) specifically includes: 41) based on the weight coefficients J of the edges between the nodes in the undirected graph model ij , a matrix of weight coefficients J is established; 42) Extract the submatrix J′ = [i*21:i*21+21,j*21:j*21+21] from the weight coefficient matrix J. This submatrix represents the coupling information between the 21*21 {0,1} state nodes corresponding to the i-th amino acid and the 21*21 {0,1} state nodes corresponding to the j-th amino acid. 43) The submatrix is ​​mean-centered as follows to obtain the mean-centered matrix J″: J″=J′-μ0-μ1+μ In the formula, μ0 is the column mean, μ1 is the row mean, and μ is the overall mean of the submatrix; 44) The interaction coefficient S between the ith amino acid and the jth amino acid is calculated by obtaining the matrix J" of the mean centering of the obtained matrix J ij : S ij = ||J"||2.

7. A protein structure prediction system based on a photonic quantum computer, characterized by, include: The alignment module is used to obtain the multiple sequence alignment (MSA) matrix of the target sequence based on the target sequence of the protein to be predicted. The encoding module is used to encode the obtained multiple sequence alignment (MSA) matrix into a {0,1} matrix. The weight coefficient acquisition module is used to convert the amino acid interactions of the protein into an undirected graph model of {0,1} state nodes and train it using a Boltzmann machine training mechanism with a quantum computer to obtain the weight coefficients of the edges between nodes in the undirected graph model. The interaction coefficient acquisition module is used to obtain the interaction coefficients between different amino acids of the protein based on the weight coefficients of the edges connecting the nodes in the undirected graph model.

8. A protein structure prediction apparatus based on a photonic quantum computer, characterized by comprising: include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the protein structure prediction method based on optical quantum computers as described in any one of claims 1-6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The program, when executed by the processor, implements the steps of the method for predicting protein structure based on optical quantum computer according to any one of claims 1-6.

Images