A saxs data assisted two-stage multi-domain protein assembly method

Abstract

A two-stage, multi-domain protein assembly method assisted by SAXS data is disclosed. Within an evolutionary algorithm framework, the method involves: first, population initialization, followed by two-stage conformational mutation and crossover; second, calculating model similarity differences using SAXS experimental data to assist the DEMO energy function in selecting solutions, while maintaining conformational diversity during the selection process through a Monte Carlo probability reception criterion; and finally, optimizing the generated model through simulation to potentially include multiple states. This invention provides a two-stage, multi-domain protein assembly method assisted by SAXS data with high sampling efficiency and high predictive assembly accuracy.

Description

Technical Field

[0001] This invention relates to the fields of bioinformatics and computer applications, and in particular to a two-stage, multi-domain protein assembly method assisted by SAXS data. Background Technology

[0002] Proteins are the main carriers of life activities, supporting almost all functions of life, and most reactions within cells depend on proteins. The way proteins work and function depends on their unique three-dimensional structure, which is often referred to as "structure determines function." Protein structure prediction is a major research area in structural bioinformatics. In 2005, on the 125th anniversary of its founding, the journal *Science* listed "Can protein folding be predicted?" as one of the 125 most challenging scientific frontier questions of the 21st century. How protein molecular machines spontaneously assemble into specific functional structures is one of the most critical legacy problems in the complete picture of the central dogma of biology, and one of the major unsolved fundamental scientific problems in the life sciences. Furthermore, it is also one of the major engineering and technological problems of widespread concern in industries such as innovative drug development, vaccine design, precision diagnostics, and medicine. In 2018, Google first entered the field of protein structure prediction. Subsequently, Facebook, Microsoft, Amazon, as well as domestic and international high-tech companies such as Tencent, Huawei, Baidu, and ByteDance, have all challenged the problem of protein structure prediction in order to develop new drugs and compete for the technological high ground in intelligent drug design. In 2020, "AI protein structure prediction" was selected by CB Insights as one of the twelve game-changing industry sectors. Also in 2020, AlphaFold2, developed by Google's DeepMind team, won first place in the global protein structure prediction competition (CASP14). AlphaFold2 brought this cutting-edge basic research problem of protein structure prediction from the scientific realm into the public eye, making it a hot topic. This demonstrates that the deep integration of computer technology, information technology, and structural biology will effectively drive and accelerate new scientific discoveries and generate new economic growth points. However, multi-domain protein structure prediction and assembly methods still face many difficulties and challenges.

[0003] End-to-end prediction methods are directly based on deep learning models and utilize protein structure information in protein structure databases. After extensive training, they can directly infer the structure of the target protein. However, this end-to-end machine learning method for predicting the full-length structure of multi-domain proteins first has to address the following three issues: (1) The number of multi-domain protein structures in protein structure databases is relatively small compared to single-domain proteins, making it difficult to meet the comprehensive training requirements of deep learning methods, which directly affects the accuracy of its prediction model. (2) For large multi-domain proteins, this end-to-end prediction method requires high computational costs, making it difficult for general equipment to meet its training and inference requirements. (3) Multi-domain proteins often have multiple states, and end-to-end direct prediction methods can often only capture the structure of one state, resulting in poor diversity of results.

[0004] The inherent complexity of multi-domain protein prediction makes it a highly challenging research topic in the field of protein structure prediction. To enable the discovery of usable protein structure models within a vast sampling space using computers, it is essential to design efficient conformational space assembly optimization algorithms to transform this into a practically operable computational problem.

[0005] Since its introduction by Price and Storm in 1995, the Differential Evolutionary Algorithm (DE) has been widely used in the field of protein conformation space optimization due to its advantages such as simple structure, ease of implementation, strong robustness, and fast convergence speed. However, as the amino acid sequence increases, the degrees of freedom of the protein molecular system also increase, making it a challenging task to obtain the global optimal solution of large-scale protein conformation space using traditional swarm optimization algorithms.

[0006] Therefore, existing protein structure prediction methods are insufficient in terms of sampling efficiency and prediction accuracy, and need to be improved. Summary of the Invention

[0007] To overcome the problems of imbalance between the detection and enhancement stages and low prediction accuracy in existing protein structure prediction methods, this invention proposes a two-stage multi-domain protein assembly method assisted by SAXS data, which has high sampling efficiency and high prediction accuracy. This invention makes full use of existing protein structure biological experimental SAXS data and designs a two-stage enhanced sampling method to fully sample the conformation space. At the same time, it uses SAXS experimental data to assist the DEMO energy function to guide conformation selection.

[0008] The technical solution adopted by this invention to solve its technical problem is:

[0009] A two-stage multi-domain protein assembly method assisted by SAXS data, the method comprising the following steps: 1) Given the full-length sequence of the target multi-domain protein and small-angle X-ray scattering (SAXS) experimental data, including the protein scattering vector q obtained from the small-angle X-ray scattering experiment. exp Spectral Intensity I exp (q) and experimental error e exp (q);

[0010] 2) The target protein sequence is segmented into domains using the DomBpred domain segmentation server. Then, AlphaFold2 is used to model the domain structure based on the domain sequence to obtain a single-domain structure model of the multi-domain protein. The N-terminus and C-terminus of the single-domain structure model are linked together in the order of the sequence.

[0011] 3) Set parameters: first stage population size NP1, second stage population size NP2, first stage crossover factor CR1, second stage crossover factor CR2, first stage temperature factor β1, second stage temperature factor β2, current generation g, set iteration generation g = 0.

[0012] 4) Population Initialization: The movement of each domain model is represented by a rotation vector and a translation vector. The solution of the domain assembly is expressed as (x1, y1, z1, θ1, φ1, ω1, ..., x n ,y n ,z n ,θ n ,φ n ,ω n ), where x n ,y n ,z n Let θ represent the translation vector of the nth field. n ,φ n ,ω n Represents the rotation vector of the nth field;

[0013] 5) Randomly generate NP1 initial solutions C in the solution space S1. i , i = {1, 2, ..., NP1};

[0014] 6) Solution C in the population i Perform the following operations:

[0015] 6.1) Place C i Let C be the objective solution. target And randomly select three distinct solutions C from the population NP1. a C b and C c , From C respectively b C cRandomly select a rotation and translation matrix from a different domain and replace C respectively. a Mutated solution C is generated at the corresponding position. mutant ;

[0016] 6.2) Generate a random number pCR, where pCR∈(0,1). If pCR<CR1, then select from C. target Randomly select a translation and rotation vector from one domain and replace it in C. mutant Generate test solution C at the corresponding position trial Otherwise, just put C. mutant Let it be C trial ′;

[0017] 6.3) If C trial In the solution space S1, then C trial ' is denoted as C trial Otherwise, a solution is randomly generated in the solution space S1, denoted as C. trial ;

[0018] 6.4) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q);

[0019] 6.5) Calculate the value of C according to formula (1) target and C trial The difference in full-length model similarity resulting from rotation and translation, where M represents the number of scattering vectors. and Representing the spectral intensity of the SAXS experiment corresponding to the m-th scattering vector, and the intensity calculated by the FoXS server using C, respectively. target The spectral intensity of the full-length model after transformation and the C trial The spectral intensity of the full-length model after transformation;

[0020]

[0021] 6.6) If S≤0, then retain C. target ;

[0022] 6.7) If S > 0, calculate the energy after passing through C using the DEMO energy function. target C trial The energy E of the full-length model generated after the transformation DEMO (C trial E DEMO (Ctarget If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The conformation is received according to the Monte Carlo criterion;

[0023] 7) g = g + 1, iterate through step 6) until g is continuous. tolerance Once the best solution in the population no longer changes, proceed to step 8.

[0024] 8) Set up the difference solution space S2, and select the top NP1 / 2 excellent individuals from population NP1 to form a new population NP2. Then, for each individual C in population NP2... i Perform the following operations:

[0025] 8.1) The solution C in population NP2 i Set as target C target And randomly select two distinct solutions C from the population NP2. a and C b , From C respectively a C b Randomly select rotation and translation vectors from the same domain, and then calculate the difference between them to obtain C. a C b The rotation and translation difference vector at that position in the domain;

[0026] 8.2) If the vector is in the difference solution space S2, then in C target Add the difference vector to the corresponding position to generate the test solution C. trial Otherwise, a difference rotation and translation vector is randomly generated in the solution space and added to C. target The corresponding position forms the test solution C. trial ;

[0027] 8.3) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q);

[0028] 8.4) Calculate the value of C according to formula (1) target and C trialThe similarity difference of the full-length model caused by rotation and translation is such that if S≤0, then C is retained. target ;

[0029] 8.5) If S > 0, calculate C using the DEMO energy function. target C trial Energy E DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The solution is accepted using the Monte Carlo criterion;

[0030] 9) Iterate through step 8) until E is obtained in m consecutive generations. DEMO The lowest solution no longer changes, so the output is E. DEMO The lowest of the top n solutions is the final result.

[0031] Furthermore, in step 3), the first stage population size NP1 = 100, the second stage population size NP2 = NP1 / 2, the first stage crossover factor CR1 = 0.5, the second stage crossover factor CR2 = 0.2, the first stage temperature factor β1 = 10, and the second stage temperature factor β2 = 2.

[0032] Furthermore, in step 7), g tolerance =50.

[0033] In step 5), the upper limit of the solution space S1 is (100.0,100.0,100.0,2π,π,2π,…,100.0,100.0,100.0,2π,π,2π), and the lower limit of the solution space S1 is (-100.0,-100.0,-100.0,0,0,0,…,-100.0,-100.0,-100.0,0,0,0,0).

[0034] In step 8), the upper limit of the difference solution space S2 is set to (1.0,1.0,1.0,0.5,0.5,0.25,…,1.0,1.0,1.0,0.5,0.5,0.25), and the lower limit is set to (-1.0,-1.0,-1.0,0,0,0,…,-1.0,-1.0,-1.0,0,0,0).

[0035] In step 9), m is 100 and n is 5.

[0036] The technical concept of this invention is as follows: Within the framework of an evolutionary algorithm, firstly, the population is initialized, and conformations are mutated and crossovered in two stages. Secondly, the similarity difference of the model is calculated using SAXS experimental data to assist the DEMO energy function in selecting solutions, while maintaining conformational diversity during the selection process through the Monte Carlo probability acceptance criterion. This SAXS data-assisted two-stage multi-domain protein assembly method utilizes existing SAXS experimental data to assist in structure assembly prediction, which not only alleviates the problem of inaccurate energy functions but also reduces the computer hardware requirements for modeling large multi-domain proteins using a divide-and-conquer approach. Finally, the solutions generated through simulation optimization may contain multiple states, providing a reference for the study of multi-domain protein polymorphism. This invention provides a SAXS data-assisted two-stage multi-domain protein assembly method.

[0037] The beneficial effects of this invention are as follows: based on the historical information of population updates, the two-stage evolutionary algorithm can not only enhance the adaptability of the algorithm, but also improve the sampling efficiency and generate polymorphic models; at the same time, the use of SAXS experimental data to assist in conformation selection improves the optimization efficiency and alleviates the prediction error problem caused by the inaccuracy of the energy function, thereby improving the prediction accuracy. Attached Figure Description

[0038] Figure 1 This is a conformational distribution map obtained by sampling a 3-domain protein 1bf2A with a sequence length of 750 using a two-stage multi-domain protein assembly method assisted by SAXS data.

[0039] Figure 2 This is a schematic diagram of conformational updates when sampling a 3-domain protein 1bf2A with a sequence length of 750 using a two-stage multi-domain protein assembly method assisted by SAXS data.

[0040] Figure 3 The three-dimensional structure was predicted and assembled from a 750-byte 3-domain protein 1bf2A using a two-stage multi-domain protein assembly method assisted by SAXS data. Detailed Implementation

[0041] The present invention will now be further described with reference to the accompanying drawings.

[0042] Reference Figures 1-3 A two-stage, multi-domain protein assembly method assisted by SAXS data, comprising the following steps:

[0043] 1) Given the full-length sequence of the target multidomain protein and small-angle X-ray scattering (SAXS) experimental data, including the protein scattering vector q obtained from the small-angle X-ray scattering experiment. exp Spectral Intensity I exp (q) and experimental error e exp (q);

[0044] 2) The target protein sequence is segmented into domains using the DomBpred domain segmentation server. Then, AlphaFold2 is used to model the domain structure based on the domain sequence to obtain a single-domain structure model of the multi-domain protein. The N-terminus and C-terminus of the single-domain structure model are linked together in the order of the sequence.

[0045] 3) Set parameters: first stage population size NP1, second stage population size NP2, first stage crossover factor CR1, second stage crossover factor CR2, first stage temperature factor β1, second stage temperature factor β2, current generation g, set iteration generation g = 0.

[0046] 4) Population Initialization: The movement of each domain model is represented by a rotation vector and a translation vector. The solution of the domain assembly is expressed as (x1, y1, z1, θ1, φ1, ω1, ..., x n ,y n ,z n ,θ n ,φ n ,ω n ), where x n ,y n ,z n Let θ represent the translation vector of the nth field. n ,φ n ,ω n Represents the rotation vector of the nth field;

[0047] 5) Randomly generate NP1 initial solutions C in the solution space S1. i ,i={1,2,…,N…,NP1P1};

[0048] 6) Solution C in the population i Perform the following operations:

[0049] 6.1) Place C i Let C be the objective solution. target And randomly select three distinct solutions C from the population NP1. a C b and C c , From C respectively b C c Randomly select a rotation and translation matrix from a different domain and replace C respectively. a Mutated solution C is generated at the corresponding position. mutant ;

[0050] 6.2) Generate a random number pCR, where pCR∈(0,1). If pCR<CR1, then select from C. targetRandomly select a translation and rotation vector from one domain and replace it in C. mutant Generate test solution C at the corresponding position trial Otherwise, just put C. mutant Let it be C trial ′;

[0051] 6.3) If C trial In the solution space S1, then C trial ' is denoted as C trial Otherwise, a solution is randomly generated in the solution space S1, denoted as C. trial ;

[0052] 6.4) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q);

[0053] 6.5) Calculate the value of C according to formula (1) target and C trial The difference in full-length model similarity resulting from rotation and translation, where M represents the number of scattering vectors. and Representing the spectral intensity of the SAXS experiment corresponding to the m-th scattering vector, and the intensity calculated by the FoXS server using C, respectively. target The spectral intensity of the full-length model after transformation and the C trial The spectral intensity of the full-length model after transformation;

[0054]

[0055] 6.6) If S≤0, then retain C. target ;

[0056] 6.7) If S > 0, calculate the energy after passing through C using the DEMO energy function. target C trial The energy E of the full-length model generated after the transformation DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The conformation is received according to the Monte Carlo criterion;

[0057] 7) g = g + 1, iterate through step 6) until g is continuous. tolerance Once the best solution in the population no longer changes, proceed to step 8.

[0058] 8) Set up the difference solution space S2, and select the top NP1 / 2 excellent individuals from population NP1 to form a new population NP2. Then, for each individual C in population NP2... i Perform the following operations:

[0059] 8.1) The solution C in population NP2 i Set as target C target And randomly select two distinct solutions C from the population NP2. a and C b , From C respectively a C b Randomly select rotation and translation vectors from the same domain, and then calculate the difference between them to obtain C. a C b The rotation and translation difference vector at that position in the domain;

[0060] 8.2) If the vector is in the difference solution space S2, then in C target Add the difference vector to the corresponding position to generate the test solution C. trial Otherwise, a difference rotation and translation vector is randomly generated in the solution space and added to C. target The corresponding position forms the test solution C. trial ;

[0061] 8.3) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q);

[0062] 8.4) Calculate the value of C according to formula (1) target and C trial The full-length models generated by rotation and translation have poor similarity.

[0063] If S≤0, then keep C. target ;

[0064] 8.5) If S > 0, calculate C using the DEMO energy function. target C trialEnergy E DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The solution is accepted using the Monte Carlo criterion;

[0065] 9) Iterate through step 8) until 100 consecutive generations of E. DEMO The lowest solution no longer changes, so the output is E. DEMO The five lowest-ranking solutions are the final results.

[0066] Using the 750-byte 3-domain protein 1bf2A as an example, a two-stage multi-domain protein assembly method assisted by SAXS data includes the following steps:

[0067] 1) Given the full-length sequence of the target multidomain protein and small-angle X-ray scattering (SAXS) experimental data, including the protein scattering vector q obtained from the small-angle X-ray scattering experiment. exp Spectral Intensity I exp (q) and experimental error e exp (q);

[0068] 2) The target protein sequence is segmented into domains using the DomBpred domain segmentation server. Then, AlphaFold2 is used to model the domain structure based on the domain sequence to obtain a single-domain structure model of the multi-domain protein. The N-terminus and C-terminus of the single-domain structure model are linked together in the order of the sequence.

[0069] 3) Set parameters: first stage population size NP1 = 100, second stage population size NP2 = NP1 / 2, first stage crossover factor CR1 = 0.5, second stage crossover factor CR2 = 0.2, first stage temperature factor β1 = 10, second stage temperature factor β2 = 2, current generation g, set iteration generation g = 0;

[0070] 4) Population Initialization: The movement of each domain model is represented by a rotation vector and a translation vector. The solution of the domain assembly is expressed as (x1, y1, z1, θ1, φ1, ω1, ..., x n ,y n ,z n ,θ n ,φ n ,ω n ), where x n ,y n ,zn Let θ represent the translation vector of the nth field. n ,φ n ,ω n Represents the rotation vector of the nth field;

[0071] 5) Randomly generate NP1 initial solutions C in the solution space S1. i Let i = {1,2,…,N…,NP1P1}, where the upper bound of the solution space S1 is (100.0,100.0,100.0,2π,π,2π,…,100.0,100.0,100.0,2π,π,2π), and the lower bound of the solution space S1 is (-100.0,-100.0,-100.0,0,0,0,…,-100.0,-100.0,0,0,0,0).

[0072] 6) Solution C in the population i Perform the following operations:

[0073] 6.1) Place C i Let C be the objective solution. target And randomly select three distinct solutions C from the population NP1. a C b and C c , From C respectively b C c Randomly select a rotation and translation matrix from a different domain and replace C respectively. a Mutated solution C is generated at the corresponding position. mutant ;

[0074] 6.2) Generate a random number pCR, where pCR∈(0,1). If pCR<CR1, then select from C. target Randomly select a translation and rotation vector from one domain and replace it in C. mutant Generate test solution C at the corresponding position trial Otherwise, just put C. mutant Let it be C trial ′;

[0075] 6.3) If C trial In the solution space S1, then C trial ' is denoted as C trial Otherwise, a solution is randomly generated in the solution space S1, denoted as C. trial ;

[0076] 6.4) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trialSpectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q);

[0077] 6.5) Calculate the value of C according to formula (1) target and C trial The difference in full-length model similarity resulting from rotation and translation, where M represents the number of scattering vectors. and Representing the spectral intensity of the SAXS experiment corresponding to the m-th scattering vector, and the intensity calculated by the FoXS server using C, respectively. target The spectral intensity of the full-length model after transformation and the C trial The spectral intensity of the full-length model after transformation;

[0078]

[0079] 6.6) If S≤0, then retain C. target ;

[0080] 6.7) If S > 0, calculate the energy after passing through C using the DEMO energy function. target C trial The energy E of the full-length model generated after the transformation DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The conformation is received according to the Monte Carlo criterion;

[0081] 7) g = g + 1, iterate through step 6) until g is continuous. tolerance =If the best solution in the 50th generation population no longer changes, then proceed to step 8);

[0082] 8) Set the upper bound of the difference solution space S2 to (1.0,1.0,1.0,0.5,0.5,0.25,…,1.0,1.0,1.0,0.5,0.5,0.25) and the lower bound to (-1.0,-1.0,-1.0,0,0,0,…,-1.0,-1.0,-1.0,0,0,0), and select the top NP1 / 2 of the best individuals in population NP1 to form a new population NP2. Then, for each individual C in population NP2... i Perform the following operations:

[0083] 8.1) The solution C in population NP2 i Set as target C target And randomly select two distinct solutions C from the population NP2. a and C b , From C respectively a C b Randomly select rotation and translation vectors from the same domain, and then calculate the difference between them to obtain C. a C b The rotation and translation difference vector at that position in the domain;

[0084] 8.2) If the vector is in the difference solution space S2, then in C target Add the difference vector to the corresponding position to generate the test solution C. trial Otherwise, a difference rotation and translation vector is randomly generated in the solution space and added to C. target The corresponding position forms the test solution C. trial ;

[0085] 8.3) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q);

[0086] 8.4) Calculate the value of C according to formula (1) target and C trial The full-length models generated by rotation and translation have poor similarity.

[0087] If S≤0, then keep C. target ;

[0088] 8.5) If S > 0, calculate C using the DEMO energy function. target C trial Energy E DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The solution is accepted using the Monte Carlo criterion;

[0089] 9) Iterate through step 8) until 100 consecutive generations of E. DEMO The lowest solution no longer changes, so the output is E. DEMO The five lowest-ranking solutions are the final results.

[0090] Using the 750-byte 3-domain protein 1bf2A as an example, the near-native conformation of this protein was obtained using the above method. The average root mean square deviation between the obtained structure and the native structure was [value missing]. The minimum root mean square deviation is The predicted three-dimensional structure is as follows Figure 3 As shown.

[0091] The above describes the results of an example provided by the present invention. Obviously, the present invention is not only suitable for the above embodiments, but can also be implemented in various ways without departing from the basic spirit of the present invention and without exceeding the content involved in the substantive content of the present invention.

Claims

1. A two-stage, multi-domain protein assembly method assisted by SAXS data, characterized in that, The method includes the following steps: 1) Given the full-length sequence of the target multidomain protein and small-angle X-ray scattering (SAXS) experimental data, including the protein scattering vector q determined by the small-angle X-ray scattering experiment. exp Spectral Intensity I exp (q) and experimental error e exp (q); 2) The target protein sequence is segmented into domains using the DomBpred domain segmentation server. Then, AlphaFold2 is used to model the domain structure based on the domain sequence to obtain a single-domain structure model of the multi-domain protein. The N-terminus and C-terminus of the single-domain structure model are linked together in the order of the sequence. 3) Set parameters: first stage population size NP1, second stage population size NP2, first stage crossover factor CR1, second stage crossover factor CR2, first stage temperature factor β1, second stage temperature factor β2, current generation g, set iteration generation g = 0. 4) Population Initialization: The movement of each domain model is represented by a rotation vector and a translation vector. The solution of the domain assembly is expressed as (x1, y1, z1, θ1, φ1, ω1, ..., x n ,y n ,z n ,θ n ,φ n ,ω n ), where x n ,y n ,z n Let θ represent the translation vector of the nth field. n ,φ n ,ω n Represents the rotation vector of the nth field; 5) Randomly generate NP1 initial solutions C in the solution space S1. i , i = {1, 2, ..., NP1}; 6) Solution C in the population i Perform the following operations: 6.1) Place C i Let C be the objective solution. target And randomly select three distinct solutions C from the population NP1. a C b and C c , From C respectively b C c Randomly select a rotation and translation matrix from a different domain and replace C respectively. a Mutated solution C is generated at the corresponding position. mutant ; 6.2) Generate a random number pCR, where pCR∈(0,1). If pCR<CR1, then select from C. target Randomly select a translation and rotation vector from one domain and replace it in C. mutant Generate test solution C at the corresponding position trial ′, Otherwise, just put C mutant Let it be C trial ′; 6.3) If C trial In the solution space S1, then C trial ' is denoted as C trial Otherwise, a solution is randomly generated in the solution space S1, denoted as C. trial ; 6.4) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q); 6.5) Calculate the value of C according to formula (1) target and C trial The difference in full-length model similarity resulting from rotation and translation, where M represents the number of scattering vectors. and Representing the spectral intensity of the SAXS experiment corresponding to the m-th scattering vector, and the intensity calculated by the FoXS server using C, respectively. target The spectral intensity of the full-length model after transformation and the C trial The spectral intensity of the full-length model after transformation; 6.6) If S≤0, then retain C. target ; 6.7) If S > 0, calculate the energy after passing through C using the DEMO energy function. target C trial The energy E of the full-length model generated after the transformation DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), Then C trial Replace C target Otherwise, according to probability The conformation is received according to the Monte Carlo criterion; 7) g = g + 1, iterate through step 6) until g is continuous. tolerance Once the best solution in the population no longer changes, proceed to step 8. 8) Set up the difference solution space S2, and select the top NP1 / 2 excellent individuals from population NP1 to form a new population NP2. Then, for each individual C in population NP2... i Perform the following operations: 8.1) The solution C in population NP2 i Set as target C target And randomly select two distinct solutions C from the population NP2. a and C b , From C respectively a C b Randomly select rotation and translation vectors from the same domain, and then calculate the difference between them to obtain C. a C b The rotation and translation difference vector at that position in the domain; 8.2) If the vector is in the difference solution space S2, then in C target Add the difference vector to the corresponding position to generate the test solution C. trial Otherwise, a difference rotation and translation vector is randomly generated in the solution space and added to C. target The corresponding position forms the test solution C. trial ; 8.3) Calculate the values ​​of C using the FoXS server. target and C trial The scattering vector q of the full-length model generated after rotation and translation target q trial Spectral intensity I target (q), I trial (q), experimental error e target (q), e trial (q); 8.4) Calculate the value of C according to formula (1) target and C trial The full-length models generated by rotation and translation have poor similarity. If S≤0, then keep C. target ; 8.5) If S > 0, calculate C using the DEMO energy function. target C trial Energy E DEMO (C trial E DEMO (C target If E DEMO (C trial ) <E DEMO (C target ), then C trial Replace C target Otherwise, according to probability The solution is accepted using the Monte Carlo criterion; 9) Iterate through step 8) until E is obtained in m consecutive generations. DEMO The lowest solution no longer changes, so the output is E. DEMO The lowest of the top n solutions is the final result.

2. The SAXS data-assisted two-stage multi-domain protein assembly method as described in claim 1, characterized in that, In step 3), the population size in the first stage is NP1 = 100, the population size in the second stage is NP2 = NP1 / 2, the crossover factor in the first stage is CR1 = 0.5, the crossover factor in the second stage is CR2 = 0.2, the temperature factor in the first stage is β1 = 10, and the temperature factor in the second stage is β2 = 2.

3. A two-stage, multi-domain protein assembly method assisted by SAXS data as described in claim 1 or 2, characterized in that, In step 7), g tolerance =50.

4. A two-stage, multi-domain protein assembly method assisted by SAXS data as described in claim 1 or 2, characterized in that, In step 5), the upper limit of the solution space S1 is (100.0, 100.0, 100.0, 2π, π, 2π, ..., 100.0, 100.0, 100.0, 2π, π, 2π), and the lower limit of the solution space S1 is (-100.0, -100.0, -100.0, 0, 0, 0, ..., -100.0, -100.0, -100.

0. In step 8), the upper limit of the difference solution space S2 is set to (1.0,1.0,1.0,0.5,0.5,0.25,…,1.0,1.0,1.0,0.5,0.5,0.25), and the lower limit is (-1.0,-1.0,-1.0,0,0,0,…,-1.0,-1.0,-1.0,0,0,0).

5. A two-stage, multi-domain protein assembly method assisted by SAXS data as described in claim 1 or 2, characterized in that, In step 9), m is 100 and n is 5.

Images