Method for predicting rna structure based on multiscale molecular dynamics simulation

Abstract

The present application relates to a method for RNA structure prediction based on multi-scale molecular dynamics simulation, and a model for representing one nucleotide by three particles is constructed, which includes the linkage relationship mode of each particle in the coarse-grained model, the fitting method of the parameters of each particle in the coarse-grained model, and the fitting method of the mapping relationship between the coarse-grained model and the all-atom model. The present application provides a method for RNA structure prediction based on all-atom and coarse-grained iterative simulation. The simulation efficiency of the model is higher than that of the all-atom model, and the simulation accuracy of the all-atom model is ensured to a certain extent.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of molecular dynamics simulation, and particularly relates to a method for structure prediction based on multi-scale RNA molecular dynamics simulation. BACKGROUND

[0002] RNA structure prediction is one of the current hot topics in life science, which almost exists in any life process, such as enzyme catalysis, signal transduction, immune response, and gene central dogma. Research on the dynamic process and mechanism of RNA molecule folding has important significance for understanding pathogenesis and drug research and design.

[0003] Existing experimental methods can analyze RNA molecular structure through cryo-electron microscopy or X-ray diffraction, and the resolution reaches the atomic scale, but the structure is static and is not convenient for studying the dynamic process. The spatial resolution of existing ultrafast time resolution experimental methods cannot reach the atomic scale. Traditional computer simulation methods can achieve the required spatial and temporal resolution, but the simulation efficiency is too low to meet the research needs. However, the accuracy and resolution of the calculation model can be appropriately reduced to improve the calculation efficiency. The calculation model with reduced resolution is called a coarse-grained model.

[0004] The coarse-grained model simplifies multiple atoms in a molecule into a virtual atom, and then connects these simplified virtual atoms according to the connection mode of the molecule to form a simplified molecular model. The coarse-grained model gives the virtual atom the same physical and chemical properties as the real atom, such as bond relationship, atomic size, and charge amount. The coarse-grained model greatly reduces the degrees of freedom of the molecular system, thereby reducing the calculation amount of the molecular simulation. A high-quality coarse-grained model can improve the calculation efficiency while ensuring a certain calculation accuracy, and can quickly and stably simulate the molecular motion process. SUMMARY

[0005] The application provides a method for structure prediction based on multi-scale RNA molecular dynamics simulation. The application aims to establish a method for structure prediction based on multi-scale RNA molecular dynamics simulation, and provides a method for fitting a coarse-grained force field with an all-atom dynamics trajectory based on an all-atom force field. The two scales of all-atom and coarse-grained are iteratively simulated, the all-atom force field is mainly responsible for conformation correction, and the coarse-grained force field is responsible for efficient conformation sampling, so as to adjust the relationship between simulation accuracy and efficiency.

[0006] The technical scheme adopted by the application to achieve the above-mentioned purpose is as follows: the method for RNA structure prediction based on multi-scale molecular dynamics simulation comprises the following steps:

[0007] (1) using RNA sequence to predict secondary structure, identifying paired base region, constructing a-form double-stranded RNA, taking a-form double-helical RNA fragment as target for open RNA all-atom model, and performing stretch dynamics simulation in segments to obtain RNA all-atom model trajectory;

[0008] (2) constructing a coarse-grained model, each nucleotide being represented by 3 coarse-grained particles;

[0009] (3) fitting van der Waals parameters of the coarse-grained particles in the coarse-grained model and positions of the coarse-grained particles;

[0010] (4) obtaining mapping relationship between the RNA all-atom model and the coarse-grained model according to the positions of the coarse-grained particles determined in step (3);

[0011] (5) establishing bond relationship of the coarse-grained particles in the coarse-grained model;

[0012] (6) fitting bonding parameters of the coarse-grained particles in the coarse-grained model according to the RNA all-atom model trajectory and the bond relationship of the coarse-grained particles;

[0013] (7) performing molecular dynamics simulation of the coarse-grained model; for the conformation generated by the molecular dynamics simulation, finding the center of the largest class as the representative conformation through clustering analysis, and converting the coarse-grained model into an all-atom model through stretch dynamics.

[0014] In step (1), the following is specifically performed:

[0015] The target RNA sequence is subjected to secondary structure prediction to identify double-helical regions through software SPOT-RNA or RNAfold prediction method;

[0016] For the identified double-helical regions, a-form RNA double helix is constructed using 3DNA software;

[0017] Taking the a-form double-helical RNA fragment as target for the open RNA all-atom model, stretch dynamics simulation is performed on each double-helical fragment in turn under vacuum and water-free condition, and after each fragment is stretched, the constraint of base pairing of the fragment is added in the force field; until all predicted double-helical regions appear.

[0018] In step (2), the positions of the coarse-grained particles in the coarse-grained model are obtained through the following steps:

[0019] For each nucleotide, the initial positions of the three coarse-grained particles are determined through clustering analysis.

[0020] Step (3) is specifically performed as follows:

[0021] 1) Generate a plurality of probes within the range of the periphery of the RNA all-atom model, each probe is subjected to the van der Waals potential of the RNA all-atom model;

[0022] 2) Firstly, the van der Waals parameters of the coarse-grained particles are obtained by fitting the coarse-grained particle positions as known quantities;

[0023] 3) Then, the particle positions describing the non-bonding potential of the periphery of the molecule are obtained by fitting the van der Waals parameters as known quantities and the positions of the coarse-grained particles as unknown quantities;

[0024] 4) Steps 2) to 3) are repeatedly performed until the van der Waals parameters and positions of the coarse-grained particles are converged.

[0025] In step (5), the hydrogen bonds of base pairing are added to the force field file as chemical bonds to increase the constraint strength of the bond angle and dihedral angle of the coarse-grained particles in the coarse-grained force field and ensure the stability of the secondary structure.

[0026] Step (6) is as follows:

[0027] According to the motion trajectory of the RNA all-atom model and the bonding and non-bonding mapping relationship between the all-atom model and the coarse-grained model, the position of the coarse-grained particle and the corresponding coarse-grained trajectory are determined.

[0028] The bonding parameters of all bonds, bond angles and dihedral angles in the bonding relationship of the coarse-grained particles are obtained by inverse Boltzmann fitting.

[0029] An RNA structure prediction device based on multi-scale molecular dynamics simulation, comprising a memory and a processor; the memory is used for storing a computer program; the processor is used for realizing the RNA structure prediction method based on multi-scale molecular dynamics simulation when the computer program is executed.

[0030] A computer readable storage medium, the storage medium stores a computer program, when the computer program is executed by a processor, a RNA structure prediction method based on multi-scale molecular dynamics simulation is realized.

[0031] The present application has the following beneficial effects and advantages:

[0032] 1. The conventional modeling method usually adopts fixed mapping relationship of coarse-grained particles and all atoms - the same type of base mapping relationship is the same. The present application attempts to establish a mapping relationship for each base. Compared with the fixed mapping mode, it can have better description effect.

[0033] 2. Due to the small time step and limited conformational changes inherent in all-atom molecular dynamics simulation, it is impossible to achieve RNA folding prediction. Coarse-grained (CG) model is fast in calculation but low in accuracy. The comprehensive advantages of atomic model and CG model will become a new method to promote RNA structure prediction and will have an impact on this field.

[0034] 3. The present application is a physical-based modeling method, which has the unique advantage of not relying on experimental data and has a wider range of applications. The coarse-grained force field parameters are fitted by all-atom dynamics simulation trajectory. And RNA folding is predicted by performing all-atom and coarse-grained iterative simulation. All-atom simulation is mainly responsible for fine-tuning RNA structure, and coarse-grained simulation is mainly responsible for efficient sampling.

[0035] 4. Compared with artificial intelligence and knowledge-based methods, this multi-scale modeling strategy can handle modified RNA. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 is a schematic diagram of a multi-scale iterative process;

[0037] Figure 2 is a flowchart of a method for constructing a coarse-grained model of arbitrary resolution;

[0038] Figure 3 is a tRNA secondary structure prediction result;

[0039] Figure 4 is a tRNA natural crystal conformation prediction structure and tRNA tertiary structure prediction result. DETAILED DESCRIPTION

[0040] The present application will be further described in detail below in combination with the drawings and examples.

[0041] The present application proposes a method based on multi-scale molecular dynamics simulation, which constructs a model representing one nucleotide with three particles. This includes the bonding relationship mode of each particle in the coarse-grained model; the fitting method of the particle parameters in the coarse-grained model; and the fitting method of the mapping relationship between the coarse-grained model and the all-atom model. The present application provides a user with an RNA structure prediction method based on all-atom and coarse-grained iterative simulation. This model is more efficient than the all-atom model in simulation efficiency, and to some extent, it guarantees the simulation accuracy of the all-atom model.

[0042] As shown in Figure 1 , Figure 2 A method for structure prediction based on multi-scale RNA molecular dynamics simulation, the specific steps are as follows:

[0043] (1) Predict the secondary structure of the RNA primary sequence, get several base-paired double helix regions, model these double helix regions as a-form RNA double helix.

[0044] (2) Take the open all-atom model conformation of the RNA molecule as the initial structure, use gromacs+plumed to sequentially target each standard a-form double helix segment, complete the all-atom stretching dynamics simulation using rmsd as the distance measure. Once the current double helix stretching is complete, add the relevant base pair constraint information to the force field until all double helix regions appear.

[0045] (2) Coarse-grained resolution: represent each nucleotide by 3 coarse-grained particles.

[0046] This coarse-grained model is composed of one type of particles: each nucleotide is divided into one segment, and 3 coarse-grained particles are reasonably placed in the spatial range of each segment through a clustering algorithm, i.e., the initial coarse-grained particle position. The position of the coarse-grained particle will be optimized.

[0047] (3) Iteratively fit the van der Waals parameters and positions of the coarse-grained particles until convergence.

[0048] A series of probes are generated within a certain range outside the all-atom model of the molecule, and each probe will be subject to the van der Waals potential of the all-atom model of the molecule. This potential should be equal to the potential that the probe receives from the coarse-grained model of the molecule. The van der Waals parameters of the particles can be theoretically solved by numerical fitting methods. However, one term in the van der Waals potential is inversely proportional to the 12th power of the distance between particles, and the effective range of the van der Waals potential is close and extremely sensitive to distance. Therefore, during fitting, it can be approximately considered that the probe only generates a van der Waals potential with the nearest particle (or its replacement part), and an inverse function transformation is performed on the van der Waals potential during fitting to convert the nonlinear fitting into linear fitting, thereby reducing the complexity and error of fitting. This fitting method allows the coarse-grained model to most closely reproduce the van der Waals potential of the all-atom model outside the molecule.

[0049] However, the initial particle position is not necessarily the most suitable for describing the non-bonding potential of the molecule's periphery. Taking the above obtained van der Waals parameters as known quantities and the coarse-grained particle positions as unknown quantities, the particle positions that are more suitable for describing the non-bonding potential of the molecule's periphery can also be obtained through numerical fitting methods. Then, iteratively fit the van der Waals parameters and positions of the particles until convergence.

[0050] (4) Fit the electrostatic parameters of the coarse-grained particles.

[0051] A series of probes are generated within a certain range around the all-atomic model of the molecule. Each probe is subjected to the electrostatic potential of the all-atomic model of the molecule. This potential energy should be equal to the potential energy of the probe subjected to the coarse-grained model of the molecule. The electrostatic parameters of the particles can be solved by numerical fitting.

[0052] (5) Find the mapping relationship between the all-atom model and the coarse-grained model.

[0053] The optimized particle positions are uncertain; therefore, it is necessary to solve the mapping relationship between the all-atom model and the corresponding coarse-grained model particles in order to convert any all-atom model conformation into the corresponding coarse-grained model. The positions of all atoms in the all-atom model and the positions of all particles in the corresponding coarse-grained model are known quantities, and the mapping relationship can be solved using numerical fitting methods.

[0054] (6) Establish the bonding relationship between particles.

[0055] Each coarse-grained particle is bonded only to other coarse-grained particles that are connected (through all-atomic connections).

[0056] (7) Fit the bonding parameters of the particles.

[0057] First, based on the full atom trajectory and the mapping relationship between the full atom model and the coarse-grained model, the position of the particle and its corresponding coarse-grained trajectory are determined. Then, the bonding parameters such as all bonds, bond angles, and dihedral angles are obtained by fitting using the "anti-Boltzmann method".

[0058] (8) Molecular dynamics simulation of the coarse-grained model was performed. The dynamics software used was grmacs, and the sampling method used was multiple copy exchange. After the dynamics simulation was completed, the largest cluster center was found through cluster analysis, and then the coarse-grained model was transformed into a full-atom model through virtual site technology and stretching dynamics.

[0059] Example 1

[0060] This implementation example illustrates the establishment of a coarse-grained tRNA model and explores the optimal resolution for this molecule. The steps are as follows:

[0061] Molecular stretching dynamics simulation using an all-atom model.

[0062] (a) Download the tRNA crystal structure of E. coli 2zm5 from the pdb website (https: / / www.rcsb.org / ) and extract the sequence seq from the crystal structure file.

[0063] Its sequence is as follows:

[0064] GCCCGGAUAGCUCAGUCGGUAGAGCAGGGGAUUGAAAAUCCCCGUGUCCUUGGUUCGAUUCCGAGUCCGGGCAC

[0065] To keep consistent with the crystal structure, we did not directly use the sequence from the pdb website, because the crystal structure might have missing nucleotides. We used spot-rna to predict the secondary structure of the known sequence, and identified the paired base regions. The secondary structure prediction results are as follows Figure 3 We can see that the tRNA has four double helix regions. For the four predicted double helix regions, we used 3dna (http: / / web.x3dna.org / mutation_file / upload) to construct standard a-form double-stranded RNA fragments.

[0066] (b) Using gromacs+plumed, we used the standard a-form double-stranded RNA fragments constructed in the previous step as targets for the open RNA all-atom model, and performed stretching dynamics simulation in a vacuum and water-free state. After each fragment stretching is completed, the base pairing constraint of the fragment is added to the force field. Until all predicted double helix regions appear.

[0067] Add the constraint information of the predicted non-wc base pairs to the force field, then add water and 0.15 mol / L Nacl to simulate for 100 nanoseconds. Because the predicted non-wc base pairs may have errors, the constraint cannot be too strict, and the constraint distance is set to 1 nm.

[0068] (2) Define the resolution of the coarse-grained model

[0069] The resolution of the coarse-grained particles is defined as 3, and 3 coarse-grained particles are placed in each residue. The particles are placed in the most suitable spatial position within each residue by clustering algorithm.

[0070] (3) Iterative fitting of van der Waals parameters and positions of coarse-grained particles

[0071] Place probes with an interval of 0.1 nm within the range of 0.2-0.4 nm from the molecule, fix the position in the first step and fit the van der Waals parameters in the second step, and iterate 12 steps until convergence.

[0072] (4) Fitting of coarse-grained particle electrostatic parameters

[0073] Place probes with an interval of 0.25 nm within the range of 0.4-1.5 nm from the molecule, and fit the electrostatic parameters according to the coarse-grained particle positions obtained in step (3).

[0074] (5) Fitting of the coarse-grained mapping relationship between the all-atom model and the coarse-grained model

[0075] According to all atom model all atom position and step (3) obtained coarse-grained particle position, fitting coarse-grained mapping relationship.

[0076] According to step (2) and step (5) obtained coarse-grained mapping relationship, the trajectory of the all atom model is converted into the corresponding coarse-grained model trajectory.

[0077] (6) Establish the bond relationship of coarse-grained particles

[0078] In order to stabilize the secondary and tertiary structure of nucleic acid, hydrogen bond, π-π interaction in the trajectory of the all atom model also need to be considered, and connected into a bond corresponding to the coarse-grained model particles. From the bond relationship, the bond angle, dihedral angle is derived.

[0079] (7) Fitting of particle bonding parameters

[0080] According to step (5) obtained coarse-grained model trajectory, by "anti-Boltzmann method" fitting to obtain all the bond, bond angle, dihedral angle of step (6) bonding parameters.

[0081] (8) Realize the coarse-grained model molecular dynamics simulation

[0082] According to step (5), a certain all atom model conformation is converted into a coarse-grained model conformation, and the conformation and all bond relationships and model parameters are input into the dynamics simulation software, so as to realize the dynamics simulation of the coarse-grained model of the molecule. The dynamics software uses grmacs, and the sampling method uses multiple copy exchange. After the end of the dynamics simulation, the largest class center is found by cluster analysis, and then the coarse-grained model is converted into an all atom model by virtual site technology and stretching dynamics. As shown in Figure 4 The yellow represents the natural crystal conformation of tRNA, and the green represents the predicted structure.

Claims

1. A method for predicting RNA structure based on multi-scale molecular dynamics simulations, characterized in that, Includes the following steps: (1) Secondary structure prediction was performed using RNA sequences to identify paired base regions and construct α-shaped double-stranded RNA. The open RNA whole-atom model was targeted with α-shaped double-helix RNA fragments and the stretching dynamics simulation was performed in segments to obtain the motion trajectory of the RNA whole-atom model. (2) Construct a coarse-grained model, where each nucleotide is represented by 3 coarse-grained particles; (3) Fit the van der Waals parameters of the coarse-grained particles and the location of the coarse-grained particles in the coarse-grained model; (4) Based on the location of the coarse-grained particles determined in step (3), obtain the mapping relationship between the RNA full-atom model and the coarse-grained model; (5) Establish the bonding relationship of coarse-grained particles in the coarse-grained model; (6) Based on the trajectory of the RNA whole atom model and the bonding relationship of the coarse-grained particles, fit the bonding parameters of the coarse-grained particles in the coarse-grained model; (7) Perform molecular dynamics simulation of the coarse-grained model; for the conformations generated by the molecular dynamics simulation, find the center of the largest class as the representative conformation through cluster analysis, and transform the coarse-grained model into an all-atom model through stretching dynamics.

2. The RNA structure prediction method based on multi-scale molecular dynamics simulation according to claim 1, characterized in that, In step (1), the specific details are as follows: The secondary structure of the target RNA sequence is predicted and double-helix regions are identified using software SPOT-RNA or RNAfold prediction methods. For the identified double-helix regions, α-shaped RNA double helices were constructed using 3DNA software; Using the α-shaped double helix RNA fragment as the target in the open RNA whole-atom model, stretching dynamics simulations were performed on each double helix fragment sequentially under vacuum and anhydrous conditions. After stretching of each fragment, base pairing constraints for that fragment were added to the force field until all predicted double helix regions appeared.

3. The RNA structure prediction method based on multi-scale molecular dynamics simulation according to claim 1, characterized in that, In step (2), the positions of the coarse-grained particles in the coarse-grained model are obtained through the following steps: For each nucleotide, cluster analysis was used to determine the initial positions of the three coarse-grained particles.

4. The RNA structure prediction method based on multi-scale molecular dynamics simulation according to claim 1, characterized in that, Step (3) is as follows: 1) Generate multiple probes within a defined area surrounding the RNA whole-atom model, with each probe subject to the van der Waals potential of the RNA whole-atom model relative to it; 2) First, the position of the coarse-grained particles is considered as a known quantity, and the van der Waals parameters of the coarse-grained particles are obtained by fitting. 3) Then, the van der Waals parameters are treated as known quantities, and the positions of the coarse-grained particles are treated as unknown quantities. The particle positions describing the non-bonded potential of the molecule's periphery are obtained by fitting. 4) Repeat steps 2)-3) until the van der Waals parameters and positions of the coarse-grained particles converge.

5. The RNA structure prediction method based on multi-scale molecular dynamics simulation according to claim 1, characterized in that, In step (5), hydrogen bonds of base pairing are added as chemical bonds to the force field file to increase the constraint strength of bond angles and dihedral angles of coarse-grained particles in the coarse-grained force field, thereby ensuring the stability of the secondary structure.

6. The RNA structure prediction method based on multi-scale molecular dynamics simulation according to claim 1, characterized in that, Step (6) is as follows: Based on the trajectory of the RNA whole-atom model and the bonding and non-bonding mapping relationship between the whole-atom model and the coarse-grained model, the location of the coarse-grained particles and their corresponding coarse-grained trajectories are determined. Then, by fitting using the inverse Boltzmann method, the bonding relationship of the coarse-grained particles is obtained, which includes all bonding parameters such as bonds, bond angles, and dihedral angles.

7. An RNA structure prediction device based on multi-scale molecular dynamics simulation, characterized in that, It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the RNA structure prediction method based on multi-scale molecular dynamics simulation as described in any one of claims 1-6 when the computer program is executed.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the RNA structure prediction method based on multi-scale molecular dynamics simulation as described in any one of claims 1-6.

Images