Deep learning algorithm for predicting monomer and compound structure of cyclic peptide and linear peptide containing non-standard amino acid

Abstract

The invention discloses a deep learning algorithm for predicting a monomer and compound structure of cyclic peptide and linear peptide containing non-standard amino acid. The deep learning algorithm specifically comprises the following steps: constructing a training data set containing head and tail cyclopeptide monomers and compounds, disulfide bond cyclopeptide monomers and compounds, and linear peptide monomers and compounds of non-standard amino acid; training and optimizing the peptide structure prediction model containing the basic protein knowledge; constructing a test data set of a linear peptide monomer containing non-standard amino acid, a linear peptide compound, a head-tail cyclopeptide monomer and a disulfide bond cyclopeptide monomer; and testing on the test data set by using the trained peptide structure prediction model. Compared with an existing peptide structure prediction model, high-precision structure prediction of monomers and compounds of cyclic peptide and linear peptide containing non-standard amino acid is proposed for the first time, structural information or pocket information of any target protein is not needed, and only one-dimensional sequences of the protein and side chain groups of the non-standard amino acid need to be input, so that the non-standard amino acid containing cyclic peptide and linear peptide is obtained. And the three-dimensional structure of the polypeptide can be accurately predicted.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer-aided drug design and development, and specifically relates to a deep learning algorithm for predicting the monomer and complex structures of cyclic peptides and linear peptides containing non-standard amino acids. Technical Background

[0002] In recent years, peptides have been continuously explored as a potential drug. Compared with small molecule drugs, peptides have unique physicochemical properties, including larger size and more flexible skeletons, which give them intrinsic advantages as inhibitors or activators of protein-protein interactions (PPIs). Compared with linear peptides, cyclic peptides have advantages in stability and membrane permeability. Considering the significant potential of peptides in the therapeutic market, it is very important to understand the structural information of peptides before further designing new peptides. In recent years, with the rise of deep learning, biomolecular modeling has made significant progress. AlphaFold2 (AF2) and RoseTTAFold have achieved unprecedented accuracy in protein structure prediction. Studies have shown that even if the training set of AF2 excludes peptides with a length of less than 16, it still learns the folding mechanism of proteins in tens of millions of data sets, and the essence of peptide binding is the extension of folding, so it can still make relatively accurate predictions for peptide-protein complexes. Based on this cognition, AfCycDesign adds loop constraints to the AF2 network to embed cyclic information for structural prediction of cyclic peptides. HighFold then further comprehensively considered the structural information of the cyclic head and tail and the disulfide bond bridge, and further improved the accuracy of disulfide bond prediction on the basis of ensuring the head and tail cyclization.

[0003] Despite these findings, there are still limitations of idealism-realism in actual drug development. The most important point is that the above models can only predict the structure of peptides containing natural amino acids. Although such peptides have the potential to be used as therapeutic agents, they are rarely used in practice, mainly because of their low stability to proteolytic enzymes and low bioavailability. The pharmacokinetic characteristics of peptides can be significantly improved by introducing non-natural amino acids. However, to date, there is no model specifically designed to predict the structure of peptides containing non-standard amino acids. Considering the current demand for structural prediction of peptides containing non-standard amino acids, a high-precision peptide structure calculation method that does not require any three-dimensional structural information is urgently needed. Summary of the invention

[0004] In response to the above problems, the purpose of the present invention is to provide a deep learning algorithm for predicting the monomer and complex structures of cyclic peptides and linear peptides containing non-standard amino acids, so as to help researchers quickly understand the peptide structures containing non-standard amino acids and promote the research progress of peptide drugs.

[0005] The specific technical solutions are as follows:

[0006] A deep learning algorithm for predicting monomer and complex structures of cyclic peptides and linear peptides containing non-standard amino acids, comprising the following steps:

[0007] Step 1, obtain a training data set containing head-to-tail cyclic peptide monomers and complexes, disulfide bond cyclic peptide monomers and complexes, linear peptide monomers containing non-standard amino acids and complexes from the RCSB Protein Data Bank database;

[0008] Step 2: Further process the obtained training data set and organize it into a data format that can be used by the model: a txt file containing the characteristic information of the peptide chains and their atoms in all data samples for training the RHT module; a YAML file for training peptide structure prediction;

[0009] Step 3: The model adds cyclization information by defining the relative position matrix of the model feature construction module, and introduces a Transformer-based RHT module in the model feature construction to enhance the model's ability to parse different levels of details of amino acids;

[0010] Step 4: Use the training data set prepared in step 1 to pre-train the RHT module, and merge the trained RHT module with the main model;

[0011] Step 5: Input the file prepared in step 2 into the model for training. After obtaining the encoder output of the RHT module and the predicted three-dimensional coordinates of the atom, calculate the loss function value and perform gradient backpropagation. Iterate the above process until the model converges.

[0012] Step 6, obtaining a test data set of head-tail cyclic peptide monomers, disulfide bond cyclic peptide monomers, linear peptide monomers and linear peptide complexes containing non-standard amino acids from the PDB database;

[0013] Step 7: Repeat the steps in step 2 to configure the YAML file;

[0014] Step 8: Input the YAML file into the model to obtain the atomic three-dimensional coordinates of the peptide monomer or complex and save it in PDB format; after aligning the predicted structure with the real structure, calculate the RMSD. For the complex, calculate the DockQ value and F nat The value is used to evaluate the performance of the model in protein-peptide docking.

[0015] Furthermore, except for the monomer structure of the head-to-tail cyclic peptide, the rest of the data samples in the training data set in step one are obtained from the literature, and then PDB data are downloaded in batches from the PDB database according to PDB_ID. For the monomer structure of the head-to-tail cyclic peptide, the sequence of the head-to-tail cyclic peptide is extracted and input into the HighFold model to predict its three-dimensional structure. The predicted three-dimensional structure is used as a supplement to the head-to-tail cyclic peptide data as part of the training data set.

[0016] Furthermore, the txt file in step 2 includes residue sequence, atom name, atomic three-dimensional coordinates, the number of atoms of each residue, peptide chain identifier, atom identifier, atom index, the number of non-standard amino acids, peptide chain length, the number of atoms of each residue, and complex identification. The construction process of the txt file is: extracting the residue sequence, atom name, atomic three-dimensional coordinates, the number of atoms of each residue, peptide chain identifier, atom identifier, atom index, the number of non-standard amino acids, peptide chain length, the number of atoms of each residue, and complex identification from the PDB file obtained in step 1.

[0017] Furthermore, the YAML file in step 2 contains the file path of the input sequence, the SDF file path of the side chain small molecule group required by the non-standard amino acid, the covalent binding position information of the small molecule and the peptide chain, and whether the head-to-tail loop is required. For the linear peptide monomer and its complex file containing non-standard amino acids, the non-standard amino acids are split into standard amino acids and side chain small molecule residues; the FASTA file of the one-dimensional amino acid sequence is placed in the proteininputs position of the configuration file, and the side chain small molecule residues are placed in the sminputs position of the configuration file.

[0018] Furthermore, the cyclization information in step three is realized by modifying the relative position matrix. The relative position matrix of NCPepFold is divided into two parts. One part is the relative position matrix of standard amino acid residues, which is NX×NX in size, where N is the length of the peptide chain and X is the number of non-standard amino acids contained in the peptide chain. The relative positions of the N-terminus and C-terminus of the peptide are set to 1, and the relative positions of other residues are changed accordingly, thereby realizing the ring constraint. The basic idea is:

[0019]

[0020] Among them, c ij represents the relative position of the ith amino acid and the jth amino acid in the relative position matrix of the cyclic peptide, d ij is the relative position of the ith amino acid and the jth amino acid in the relative position matrix of the linear peptide, L Sis the number of residues in the peptide chain, including non-standard amino acids, |·| represents the absolute value sign, signal(x) represents the direction of relative position, signal(x)=1 represents that amino acid i is in the positive direction relative to amino acid j, signal(x)=-1 represents that amino acid i is in the negative direction relative to amino acid j, and signal(x)=0 represents that the relative position between amino acid i and amino acid i itself is 0.

[0021] Furthermore, the RHT module in step 3 extracts the encoder output and merges it with the feature embedding of the two relative position matrices to give the model different levels of information and enhance the model's understanding of peptide information. The dimension of the encoder output is Where N AT is the total number of atoms in the peptide chain, and the dimensions of the two relative position matrix feature embeddings are Where N RA It is the sum of the number of standard amino acids and the total number of non-standard amino acids. In order to match the two, a converter is used to summarize the information of the standard amino acid part in the encoder output from atoms to amino acid level. The atomic representation of the non-standard amino acid part remains unchanged, and then the paired features between residues and atoms are calculated by Einstein summation. Finally, the amino acid paired features and atom paired features are merged by diagonal splicing of blocks to obtain features with the same dimension as the feature embedding of the two relative position matrices.

[0022] Furthermore, the loss function when training the RHT module in step 4 is the cross entropy loss term, and the specific calculation formula is as follows:

[0023]

[0024] Where N is the number of samples, R is the number of categories, and y ji represents the true label of the jth sample in the i-th category, represents the predicted probability of the jth sample in the i-th category.

[0025] Furthermore, the loss function during training in step 5 is the frame alignment error, which is used for the structural loss of the atomic three-dimensional coordinate prediction module. The specific calculation formula is as follows:

[0026]

[0027] Among them, N frames Indicates the number of frames, N atoms represents the number of atoms, p represents the real structural coordinates of the atoms, represents the predicted structural coordinates of atoms.

[0028] Furthermore, the data samples in Step 6 are all obtained from the literature, and then the PDB data is downloaded in batches from the PDB database according to the PDB_ID; the YAML file in Step 7 contains the file path of the input sequence, the file path of the side chain small molecule group SDF file required for non-standard amino acids, the covalent binding position information of the small molecule and the peptide chain, and the information on whether to form a loop at the beginning and end.

[0029] Furthermore, the RMSD in Step 8 is used to measure the average distance deviation of the corresponding atoms between the native structure and the predicted structure. The RMSD includes the peptide chain RMSD and the non-standard residue RMSD, which respectively represent the deviation of the whole peptide chain and the accuracy of the local structure of non-standard amino acids. The calculation formula of RMSD is as follows:

[0030]

[0031] where p 1 is the true structure of the peptide chain, p 2 is the predicted structure of the peptide chain, i is the serial number of the C α atom, x i , y i, z i respectively represent the three-dimensional coordinates of the structure. The chain RMSD is to first extract the C α atoms shared by the native structure and the predicted structure. For monomer structures, the Kabsch algorithm is used to align the coordinates of the whole peptide chain, and the RMSD of the C α atoms of the whole peptide chain is calculated. For complex structures, the Kabsch algorithm is used to align the coordinates of the protein part, and the RMSD of the C α atoms of the whole peptide chain; the non-standard residue RMSD is to align the heavy atoms on the peptide chain within the non-standard residue and calculate the all-atom RMSD of the non-standard residue. If there are multiple non-standard residues on a peptide chain, the RMSD of each non-standard residue is calculated and then the average value is taken; the DockQ in Step 8 is used to evaluate the docking quality of the peptide-protein complex. The DockQ is a comprehensive score of three indicators, namely F nat , LRMS and iRMS. The DockQ divides the structure quality of the complex into four grades. DockQ < 0.23 is unacceptable quality, 0.23 < DockQ < 0.49 is acceptable quality, 0.49 < DockQ < 0.8 is medium quality, and DockQ > 0.8 is high-quality prediction. The calculation formula of DockQ is as follows:

[0032] DockQ(F nat , LRMS, iRMS, d 1 , d 2 ) = (F nat + RMS scaled(LRMS,d 1 )+RMS scaled (iRMS,d 2 )) / 3, where F nat It is defined as the number of heavy atoms in a crystal complex. For protein-peptide residue pairs, LRMS is the local root mean square error, iRMS is the global root mean square error, d 1 , d 2 is a scaling factor used to adjust the weight of the local RMSD,

[0033] F nat It is used to indicate the success rate of grounding. For peptide-protein complexes, F nat Complexes with a value of ≥0.5 were considered to be of high quality. nat The calculation formula is as follows:

[0034]

[0035] Among them, S N S represents the set of amino acid pairs whose distance between the ligand sequence and the receptor sequence in the real structure is less than the threshold θ. M It represents the set of amino acid pairs whose distance between the ligand sequence and the receptor sequence in the predicted structure is less than the threshold θ, and ∩ represents the intersection.

[0036] The beneficial effects of the present invention are as follows: on the one hand, by introducing cyclization constraints, the model's understanding of the cyclic structure is completed, the drawback that the original technology cannot be used for cyclic peptide structure prediction is solved, and high-precision prediction of cyclic peptide structures containing non-standard amino acids is achieved; on the other hand, a RHT module based on the Transformer model is introduced to expand the residue atomization range to the entire polypeptide chain, thereby achieving a comprehensive characterization of peptides of different granularities at the residue and atomic levels, and effectively integrating the inherent hierarchical relationship of the peptide chain. This integrated method significantly improves the model's ability to parse multi-level details of amino acids, especially non-standard amino acids, and further improves the prediction accuracy of the model; the structure prediction algorithm of monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids involved in the present invention will provide valuable insights into the influence of non-standard amino acids on the local structure of polypeptides, which is of particular importance for the discovery and design of polypeptide drugs. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Flowchart of the algorithm for structure prediction of monomers and complexes of cyclic and linear peptides containing non-standard amino acids.

[0038] Figure 2 Flowchart for model training based on transfer learning. DETAILED DESCRIPTION

[0039] The present invention will be described in detail below in conjunction with the accompanying drawings, and the technical solutions in the embodiments of the present invention will be clearly and completely described. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Example 1

[0041] like Figure 1 As shown, a deep learning algorithm for predicting monomer and complex structures of cyclic peptides and linear peptides containing non-standard amino acids comprises the following steps:

[0042] Step 1, obtain a training data set containing head-to-tail cyclic peptide monomers and complexes, disulfide bond cyclic peptide monomers and complexes, and linear peptide monomers and complexes containing non-standard amino acids from the PDB database. Except for the head-to-tail cyclic peptide monomer structure, the rest of the data samples in the training data set are obtained from the literature. Then, the PDB data is downloaded in batches from the PDB database according to the PDB_ID. For the monomer structure of the head-to-tail cyclic peptide, the sequence of the head-to-tail cyclic peptide is extracted and input into the HighFold model to predict its three-dimensional structure. The predicted three-dimensional structure is used as a supplement to the head-to-tail cyclic peptide data as part of the training data set.

[0043] Step 2: further process the obtained training data set and organize it into a data format that can be used by the model: a txt file containing the characteristic information of the peptide chains and their atoms in all data samples for training the RHT module; a YAML file for training peptide structure prediction; the txt file includes residue sequence, atom name, atomic three-dimensional coordinates, the number of atoms per residue, peptide chain identifier, atom identifier, atom index, number of non-standard amino acids, peptide chain length, number of atoms per residue, and complex identifier. The construction process of the txt file is as follows: extract the residue sequence, atom name, atomic three-dimensional coordinates, number of atoms per residue, peptide chain identifier, atom identifier, atom index, number of non-standard amino acids, peptide chain length, number of atoms per residue, and complex identifier from the PDB file obtained in step 1; the YAML file contains the file path of the input sequence, the SDF file path of the side chain small molecule group required for the non-standard amino acid, the covalent binding position information of the small molecule and the peptide chain, and whether the head-to-tail ring is required. For the linear peptide monomer and its complex file containing non-standard amino acids, the non-standard amino acids are split into standard amino acids and side chain small molecule residues; put the FASTA file of the amino acid one-dimensional sequence into the protein The inputs position puts the side chain small molecule residues into the sminputs position of the configuration file.

[0044] Step three, the model adds cyclization information by defining the relative position matrix of the model feature construction module, and introduces a Transformer-based RHT module in the model feature construction to enhance the model's ability to parse different levels of details of amino acids. The cyclization information in step three is achieved by modifying the relative position matrix. The relative position matrix of NCPepFold is divided into two parts. One part is the relative position matrix of standard amino acid residues, with a size of NX×NX, where N is the peptide chain length and X is the number of non-standard amino acids contained in the peptide chain. The relative positions of the N-terminus and C-terminus of the peptide are set to 1, and the relative positions of other residues change accordingly, thereby achieving ring constraints. The basic idea is:

[0045]

[0046] Among them, c ij represents the relative position of the ith amino acid and the jth amino acid in the relative position matrix of the cyclic peptide, d ij is the relative position of the ith amino acid and the jth amino acid in the relative position matrix of the linear peptide, L s is the number of residues in the peptide chain, including non-standard amino acids, |·| represents the absolute value sign, signal(x) represents the direction of relative position, signal(x)=1 represents that amino acid i is in the positive direction relative to amino acid j, signal(x)=-1 represents that amino acid i is in the negative direction relative to amino acid j, and signal(x)=0 represents that the relative position between amino acid i and amino acid i itself is 0.

[0047] The RHT module in step 3 extracts the encoder output and merges it with the feature embedding of the two relative position matrices to give the model different levels of information and enhance the model's understanding of peptide information. The dimension of the encoder output is Where N AT is the total number of atoms in the peptide chain, and the dimensions of the two relative position matrix feature embeddings are Where N RA It is the sum of the number of standard amino acids and the total number of non-standard amino acids. In order to match the two, a converter is used to summarize the information of the standard amino acid part in the encoder output from atoms to amino acid level. The atomic representation of the non-standard amino acid part remains unchanged, and then the paired features between residues and atoms are calculated by Einstein summation. Finally, the amino acid paired features and atom paired features are merged by diagonal splicing of blocks to obtain features with the same dimension as the feature embedding of the two relative position matrices.

[0048] Step 4: Use the training data set prepared in step 1 to pre-train the RHT module, and merge the trained RHT module with the main model. The loss function when training the RHT module in step 4 is the cross entropy loss term. The specific calculation formula is as follows:

[0049]

[0050] Where N is the number of samples, R is the number of categories, and y ji represents the true label of the jth sample in the i-th category, represents the predicted probability of the jth sample in the i-th category.

[0051] Step 5: Input the file prepared in step 2 into the model for training. After obtaining the encoder output of the RHT module and the predicted 3D coordinates of the atom, calculate the loss function value and perform gradient backpropagation. Iterate the above process until the model converges. The loss function during training in step 5 is the frame alignment error, which is used for the structural loss of the atomic 3D coordinate prediction module. The specific calculation formula is as follows:

[0052]

[0053] Among them, N frames Indicates the number of frames, N atoms represents the number of atoms, p represents the real structural coordinates of the atoms, represents the predicted structural coordinates of atoms.

[0054] Step six, obtain the test data set of linear peptide monomers, linear peptide complexes, head-to-tail cyclic peptide monomers, and disulfide bond cyclic peptide monomers containing non-standard amino acids from the PDB database. The data samples in step six are all obtained from the literature, and then the PDB data are downloaded in batches from the PDB database according to PDB_ID.

[0055] Step 7: Repeat the steps in step 2 to configure the YAML file. The YAML file in step 7 includes the file path of the input sequence, the SDF file path of the side chain small molecule group required by the non-standard amino acid, the covalent binding position information of the small molecule and the peptide chain, and whether the head-to-tail loop is required;

[0056] Step 8: Input the YAML file into the model to obtain the atomic three-dimensional coordinates of the peptide monomer or complex and save it in PDB format; after aligning the predicted structure with the real structure, calculate the RMSD. For the complex, calculate the DockQ value and F natValues are used to evaluate the performance of the model in protein-peptide docking. The RMSD in Step 8 is used to measure the average distance deviation of corresponding atoms between the native structure and the predicted structure. RMSD includes peptide chain RMSD and non-standard residue RMSD, which respectively represent the deviation of the overall peptide chain and the accuracy of the local structure of non-standard amino acids. The RMSD calculation formula is as follows:

[0057]

[0058] where p 1 is the true structure of the peptide chain, p 2 is the predicted structure of the peptide chain, i is the serial number of the C α atom, x i , y i, z i respectively represent the three-dimensional coordinates of the structure. The chain RMSD is to first extract the C α atoms shared by the native structure and the predicted structure. For monomer structures, the Kabsch algorithm is used to align the coordinates of the entire peptide chain, and the RMSD of the C α atoms of the entire peptide chain is calculated. For complex structures, the Kabsch algorithm is used to align the coordinates of the protein part, and the RMSD of the C α atoms of the entire peptide chain; the non-standard residue RMSD is to align the heavy atoms on the peptide chain within the non-standard residue and calculate the all-atom RMSD of the non-standard residue. If there are multiple non-standard residues on a peptide chain, the RMSD of each non-standard residue is calculated and then the average value is taken; DockQ in Step 8 is used to evaluate the docking quality of the peptide-protein complex. DockQ is a comprehensive score of three indicators, namely F nat 、LRMS and iRMS. DockQ divides the structural quality of the complex into four grades. DockQ < 0.23 represents unacceptable quality, 0.23 < DockQ < 0.49 represents acceptable quality, 0.49 < DockQ < 0.8 represents medium quality, and DockQ > 0.8 represents high-quality prediction. The DockQ calculation formula is as follows:

[0059] DockQ(F nat , LRMS, iRMS, d 1 , d 2 ) = (F nat + RMS scaled (LRMS, d 1 ) + RMS scaled (iRMS, d 2 )) / 3 where, F nat is defined as the heavy atoms in the crystal complex in For protein-peptide residue pairs, LRMS is the local root mean square error, iRMS is the global root mean square error, d 1 , d 2 is a scaling factor used to adjust the weight of the local RMSD,

[0060] F nat It is used to indicate the success rate of grounding. For peptide-protein complexes, F nat Complexes with a value of ≥0.5 were considered to be of high quality. nat The calculation formula is as follows:

[0061]

[0062] Among them, S N S represents the set of amino acid pairs whose distance between the ligand sequence and the receptor sequence in the real structure is less than the threshold θ. M It represents the set of amino acid pairs whose distance between the ligand sequence and the receptor sequence in the predicted structure is less than the threshold value θ, and ∩ represents the intersection. The data in Tables 1 to 3 are calculated in the above manner.

[0063] Table 1 RMSD results of the head-to-tail ring peptide monomers

[0064] PDB ID Peptide RMSD Modified RMSD 2N8E 0.274 1.209 2MSQ 0.458 0.895 2M1P 1.239 1.207 1T9E 0.229 1.019 2M2G 2.397 1.463 4M6E 4.063 2.09 2J15 1.45 0.845 2KNN 2.108 1.741 2M2S 1.767 1.753 2M2X 2.418 3.402 Average 1.640 1.562 median 1.503 1.336

[0065] Table 2 RMSD results of disulfide bond-forming cyclic peptide monomers

[0066] PDB ID Peptide RMSD Modified RMSD PDB ID Peptide RMSD Modified RMSD 1P9G 2.104 2.301 2MFX 1.824 3.118 2CRD 1.203 1.313 2MG6 1.637 3.283 1BIG 1.252 1.49 3L09 0.474 1.218 1K64 1.74 1.132 4E86 0.763 0.907 2EW4 3.587 1.558 4E83 1.075 0.963 6MY3 1.684 1.484 3HJ 0.632 1.995 6MY2 1.399 1.126 3LO6 0.697 0.717 1OMC 1.032 1.864 7N21 1.513 1.751 6MY1 0.688 1.388 7N24 1.396 1.892 2M62 2.646 2.295 7N25 1.026 2.831 5UG3 2.116 2.714 7N20 1.728 2.76 1KFP 1.241 1.703 average value 1.455 1.818 median 1.396 1.751

[0067] The average RMSD of the peptide monomer is The median is The average value of disulfide bonded cyclopeptide monomers is The median is The modification RMSDs are slightly smaller than the peptide RMSDs. Among them, the modification RMSD of the cyclic peptide monomer is The modified RMSD of the disulfide bond-forming cyclic peptide monomer is The performance is improved by 81.9% and 21.1% respectively compared with the prediction results of RFAA. Therefore, these results confirm that NCPepFold can accurately predict structures containing modified residues.

[0068] Table 3 RMSD, DockQ, F of peptide monomers and complexes nat result

[0069]

[0070] Table 3 shows the remarkable closeness of the peptide monomer and complex to the experimental structure, where the average RMSD of the peptide monomer is The average modification RMSD is

[0071] NCPepFold shows impressive effectiveness in online peptide-protein docking. The average peptide RMSD of the complex is The average modification RMSD is Compared with the structural prediction of monomers, the complexity of the complex lies in that the wrong choice of binding sites will greatly affect the spatial configuration of the overall complex. Even if the structural reconstruction of the peptide chain itself is accurate, if the binding position deviation is large, the final complex structure still cannot be close to the experimental true value. NCPepFold shows the correctness of its capture of binding sites. DockQ is used to evaluate the accuracy of complex docking, and DockQ>0.8 indicates a high-quality prediction. The median DockQ value of the complex structure predicted by NCPepFold is 0.911, which also confirms the good performance of NCPepFold.

[0072] It should be noted that the peptide monomer FASTA and peptide complex FASTA are mainly from the PDB database, but the peptide monomer FASTA and peptide complex FASTA available in the present invention are not limited to the above data sources. The evaluation of the results of peptide monomers and peptide complexes requires the experimental structures of real peptide monomers and peptide complexes as references, which is an optional step rather than a necessary step.

[0073] The above description is only a preferred specific embodiment of the present invention, and does not limit the patent protection scope of the present invention. Any equivalent replacement or modification made by using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A deep learning algorithm for predicting monomer and complex structures of cyclic and linear peptides containing non-standard amino acids, characterized in that: The steps include: Step 1, obtain a training data set containing head-to-tail cyclic peptide monomers and complexes, disulfide bond cyclic peptide monomers and complexes, linear peptide monomers containing non-standard amino acids and complexes from the RCSB Protein Data Bank database; Step 2: Further process the obtained training data set and organize it into a data format that can be used by the model: a txt file containing the characteristic information of the peptide chains and their atoms in all data samples, which is used for training the RHT module; A YAML file for training peptide structure prediction; Step 3: The model adds cyclization information by defining the relative position matrix of the model feature construction module, and introduces a Transformer-based RHT module in the model feature construction to enhance the model's ability to parse different levels of details of amino acids; Step 4: Use the training data set prepared in step 1 to pre-train the RHT module, and merge the trained RHT module with the main model; Step 5: Input the file prepared in step 2 into the model for training. After obtaining the encoder output of the RHT module and the predicted three-dimensional coordinates of the atom, calculate the loss function value and perform gradient backpropagation. Iterate the above process until the model converges. Step 6, obtaining a test data set of head-tail cyclic peptide monomers, disulfide bond cyclic peptide monomers, linear peptide monomers and linear peptide complexes containing non-standard amino acids from the PDB database; Step 7: Repeat the steps in step 2 to configure the YAML file; Step 8: Input the YAML file into the model to obtain the atomic three-dimensional coordinates of the peptide monomer or complex and save it in PDB format; after aligning the predicted structure with the real structure, calculate the RMSD. For the complex, calculate the DockQ value and F nat The value is used to evaluate the performance of the model in protein-peptide docking.

2. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: Except for the monomer structure of the head-to-tail cyclic peptide, the rest of the data samples in the training data set in step 1 are obtained from the literature. Then, PDB data are downloaded in batches from the PDB database according to PDB_ID. For the monomer structure of the head-to-tail cyclic peptide, the sequence of the head-to-tail cyclic peptide is extracted and input into the HighFold model to predict its three-dimensional structure. The predicted three-dimensional structure is used as a supplement to the head-to-tail cyclic peptide data as part of the training data set.

3. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The txt file in step 2 includes residue sequence, atom name, atomic three-dimensional coordinates, the number of atoms per residue, peptide chain identifier, atom identifier, atom index, the number of non-standard amino acids, peptide chain length, the number of atoms per residue, and complex identifier. The construction process of the txt file is as follows: extract the residue sequence, atom name, atomic three-dimensional coordinates, the number of atoms per residue, peptide chain identifier, atom identifier, atom index, the number of non-standard amino acids, peptide chain length, the number of atoms per residue, and complex identifier from the PDB file obtained in step 1.

4. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The YAML file in step 2 contains the file path of the input sequence, the SDF file path of the side chain small molecule groups required by the non-standard amino acids, the covalent binding position information of the small molecule and the peptide chain, and whether the head-to-tail loop is required. For the linear peptide monomer and its complex files containing non-standard amino acids, the non-standard amino acids are split into standard amino acids and side chain small molecule residues; the FASTA file of the one-dimensional amino acid sequence is placed in the proteininputs position of the configuration file, and the side chain small molecule residues are placed in the sminputs position of the configuration file.

5. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The cyclization information in step three is achieved by modifying the relative position matrix. The relative position matrix of NCPepFold is divided into two parts. One part is the relative position matrix of standard amino acid residues, which is NX×NX in size, where N is the length of the peptide chain and X is the number of non-standard amino acids contained in the peptide chain. The relative positions of the N-terminus and C-terminus of the peptide are set to 1, and the relative positions of other residues are changed accordingly, thereby realizing the ring constraint. The basic idea is: Among them, c ij represents the relative position of the ith amino acid and the jth amino acid in the relative position matrix of the cyclic peptide, d ij is the relative position of the ith amino acid and the jth amino acid in the relative position matrix of the linear peptide, L s is the number of residues in the peptide chain, including non-standard amino acids, |·| represents the absolute value sign, signal(x) represents the direction of relative position, signal(x)=1 represents that amino acid i is in the positive direction relative to amino acid j, signal(x)=-1 represents that amino acid i is in the negative direction relative to amino acid j, and signal(x)=0 represents that the relative position between amino acid i and amino acid i itself is 0.

6. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The RHT module in step 3 extracts the encoder output and merges it with the feature embedding of the two relative position matrices to give the model different levels of information and enhance the model's understanding of peptide information. The dimension of the encoder output is Where N AT is the total number of atoms in the peptide chain, and the dimensions of the two relative position matrix feature embeddings are Where N RA It is the sum of the number of standard amino acids and the total number of non-standard amino acids. In order to match the two, a converter is used to summarize the information of the standard amino acid part in the encoder output from atoms to amino acid level. The atomic representation of the non-standard amino acid part remains unchanged, and then the paired features between residues and atoms are calculated by Einstein summation. Finally, the amino acid paired features and atom paired features are merged by diagonal splicing of blocks to obtain features with the same dimension as the feature embedding of the two relative position matrices.

7. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The loss function when training the RHT module in step 4 is the cross entropy loss term, and the specific calculation formula is as follows: Where N is the number of samples, R is the number of categories, and y ji represents the true label of the jth sample in the i-th category, represents the predicted probability of the jth sample in the i-th category.

8. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The loss function during training in step 5 is the frame alignment error, which is used for the structural loss of the atomic three-dimensional coordinate prediction module. The specific calculation formula is as follows: Among them, N frames Indicates the number of frames, N atoms represents the number of atoms, p represents the real structural coordinates of the atoms, represents the predicted structural coordinates of atoms.

9. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The data samples in step six are all obtained from the literature, and then the PDB data are downloaded in batches from the PDB database according to PDB_ID; the YAML file in step seven contains the file path of the input sequence, the SDF file path of the side chain small molecule group required for non-standard amino acids, the covalent binding position information of the small molecule and the peptide chain, and whether the head-to-tail ring is required.

10. A deep learning algorithm for predicting monomers and complexes of cyclic peptides and linear peptides containing non-standard amino acids as claimed in claim 1, characterized in that: The RMSD in step eight is used to measure the average distance deviation of corresponding atoms between the natural structure and the predicted structure. RMSD includes peptide chain RMSD and non-standard residue RMSD, which respectively represent the deviation of the entire peptide chain and the accuracy of the non-standard amino acid local structure. The RMSD calculation formula is as follows: where p1 is the true structure of the peptide chain, p2 is the predicted structure of the peptide chain, and i is the serial number of the C α atom, x i , y i, z i respectively represent the three-dimensional coordinates of the structure. The chain RMSD is to first extract the C α atoms shared by the native structure and the predicted structure. For monomeric structures, the Kabsch algorithm is used to align the coordinates of the entire peptide chain, and the RMSD of the C α atoms of the entire peptide chain is calculated. For complex structures, the Kabsch algorithm is used to align the coordinates of the protein part, and the RMSD of the C α atoms of the entire peptide chain; the non-standard residue RMSD is to align the heavy atoms on the peptide chain within the non-standard residue , and the all-atom RMSD of the non-standard residue is calculated. If there are multiple non-standard residues on a peptide chain, the RMSD of each non-standard residue is calculated and then averaged; DockQ in step eight is used to evaluate the docking quality of the peptide-protein complex. DockQ is a comprehensive score of three indicators, namely F nat , LRMS, and iRMS. DockQ divides the structural quality of the complex into four grades. DockQ < 0.23 represents unacceptable quality, 0.23 < DockQ < 0.49 represents acceptable quality, 0.49 < DockQ < 0.8 represents medium quality, and DockQ > 0.8 represents high-quality prediction. The calculation formula of DockQ is as follows: DockQ(F nat , LRMS, iRMS, d1, d2) = (F nat + RMS scaled (LRMS, d1) + RMS scaled (iRMS, d2)) / 3 where, F nat It is defined as the number of heavy atoms in a crystal complex. For protein-peptide residue pairs, LRMS is the local root mean square error, iRMS is the global root mean square error, d1 and d2 are scaling factors used to adjust the weight of the local RMSD. F nat It is used to indicate the success rate of grounding. For peptide-protein complexes, F nat Complexes with a value of ≥0.5 were considered to be of high quality. nat The calculation formula is as follows: Among them, S N S represents the set of amino acid pairs whose distance between the ligand sequence and the receptor sequence in the real structure is less than the threshold θ. M It represents the set of amino acid pairs whose distance between the ligand sequence and the receptor sequence in the predicted structure is less than the threshold θ, and ∩ represents the intersection.

Images