Method, device and readable medium for predicting functional khib sites based on protein sequences

Abstract

The application discloses a kind of based on protein sequence functional Khib site prediction method, device and readable medium, comprising: obtaining the 2-hydroxyisobutyryl group data to be predicted and extracting from it to the HSP (10, 10) data to be predicted, the HSP (10, 10) data to be predicted is carried out feature coding, obtain several sequence features and several structure features, the HSP (10, 10) data to be predicted and several sequence features and several structure features are input functional Khib site prediction model, extract multidimensional feature and obtain the score vector that several sequence features, several structure features and multidimensional feature corresponding score constitute, score vector is input trained first classification module, obtain final score, whether the HSP (10, 10) data to be predicted is located in functional Khib site vicinity according to final score is determined, to realize more accurate and efficient functional Khib site prediction.

Description

Technical Field

[0001] This invention relates to the field of data processing, and specifically to a method, apparatus, and readable medium for predicting functional Khib sites based on protein sequences. Background Technology

[0002] Khib (2-hydroxyisobutyryl) sites play crucial roles in biological processes, including gene transcription, chromatin function regulation, and metabolism. With the rapid development of mass spectrometry, the efficiency of identifying Khib sites has increased significantly. However, methods for detecting functional Khib sites remain time-consuming and technically expensive, exhibiting limitations. Furthermore, only 16 functional Khib sites have been experimentally identified to date. Therefore, effective methods are needed for the predictive sequencing of functional Khib sites. This would not only greatly enhance our understanding of post-translational modifications (PTMs) of this novel protein but also guide future experimental research. Summary of the Invention

[0003] The purpose of this application is to provide a method, apparatus, and readable medium for predicting functional Khib sites based on protein sequences, addressing the aforementioned technical problems.

[0004] In a first aspect, the present invention provides a method for predicting functional Khib sites based on protein sequences, comprising the following steps:

[0005] 2-hydroxyisobutyryl group data were obtained and HSP(10,10) data were extracted from them. HSP(10,10) data are Khib peptide data with 10 residues upstream and 10 residues downstream. HSP(10,10) data near Khib sites and HSP(10,10) data near other lysine sites were labeled to construct the first training data. HSP(10,10) data near functional Khib sites and HSP(10,10) data near other Khib sites were labeled to construct the second training data.

[0006] A Khib site prediction model and a first classification module are constructed. The Khib site prediction model includes a protein sequence processing module and a second classification module. The Khib site prediction model and the first classification module are trained using the first training data to obtain a pre-trained Khib site prediction model and a trained first classification module. The pre-trained Khib site prediction model is used as the source model, and a functional Khib site prediction model is trained using a few-shot transfer learning method based on the second training data.

[0007] The 2-hydroxyisobutyryl group data to be predicted is obtained, and the HSP(10,10) data to be predicted is extracted from it. The HSP(10,10) data to be predicted is feature-encoded to obtain several sequence features and several structural features. The HSP(10,10) data to be predicted, along with several sequence features and several structural features, are input into the functional Khib site prediction model to extract multi-dimensional features and obtain a score vector composed of the scores corresponding to several sequence features, several structural features, and multi-dimensional features. The score vector is input into the trained first classification module to obtain the final score. Based on the final score, it is determined whether the HSP(10,10) data to be predicted is located near the functional Khib site.

[0008] Preferably, in the Khib site prediction model and the functional Khib site prediction model, the HSP(10,10) data to be predicted is input into the protein sequence processing module to obtain multi-dimensional features. Several sequence features, several structural features, and multi-dimensional features are input into the second classification module to obtain their corresponding scores. The scores output by the Khib site prediction model are used to predict whether the HSP(10,10) data is located near a Khib site. The scores output by the functional Khib site prediction model are used to predict whether the HSP(10,10) data is located near a functional Khib site.

[0009] Preferably, the protein sequence processing module includes a token and position embedding layer, an autoregressive encoder layer, a pooling layer and a first dropout layer connected in sequence. The output features of the autoregressive encoder layer are then repeatedly input into the autoregressive encoder layer.

[0010] Preferably, both the first classification module and the second classification module are deep neural networks, including a first dense layer, a first ReLU function layer, a second Dropout layer, a second dense layer, a second ReLU function layer, a third Dropout layer, and an output layer connected in sequence. The output layer consists of two neurons with sigmoid functions, which output scores.

[0011] Preferably, several sequence features include GPS, PseAAC, CKSAAPs, OBC, AAindex, PSSM, and ACFs; several structural features include ASA, SS, and BTAs; and these, together with the multidimensional features output by the protein sequence processing module, constitute 11 features, with the score vector being an 11-dimensional vector V = (D1, D2, D3, ..., D...). 11 ); where D1, D2, D3, ..., D 11 The scores obtained by the second classification module correspond to the 11 features respectively.

[0012] Preferably, the multi-dimensional feature is a 128-dimensional vector output by the first Dropout layer.

[0013] Secondly, the present invention provides a functional Khib site prediction device based on protein sequence, comprising:

[0014] The data acquisition module is configured to acquire 2-hydroxyisobutyryl group data and extract HSP(10,10) data from it. HSP(10,10) data are Khib peptide data with 10 residues upstream and 10 residues downstream. The HSP(10,10) data near the Khib site and the HSP(10,10) data near other lysine sites are labeled to construct the first training data. The HSP(10,10) data near the functional Khib site and the HSP(10,10) data near other Khib sites are labeled to construct the second training data.

[0015] The model building module is configured to build a Khib site prediction model and a first classification module. The Khib site prediction model includes a protein sequence processing module and a second classification module. The Khib site prediction model and the first classification module are trained using the first training data to obtain a pre-trained Khib site prediction model and a trained first classification module. The pre-trained Khib site prediction model is used as the source model, and a functional Khib site prediction model is trained using a few-shot transfer learning method based on the second training data.

[0016] The prediction module is configured to acquire the 2-hydroxyisobutyryl group data to be predicted and extract the HSP(10,10) data to be predicted from it. It performs feature encoding on the HSP(10,10) data to be predicted to obtain several sequence features and several structural features. It inputs the HSP(10,10) data to be predicted, as well as several sequence features and several structural features, into the functional Khib site prediction model to extract multi-dimensional features and obtain a score vector composed of the scores of several sequence features, several structural features, and multi-dimensional features. It inputs the score vector into the trained first classification module to obtain the final score. Based on the final score, it determines whether the HSP(10,10) data to be predicted is located near the functional Khib site.

[0017] Thirdly, the present invention provides an electronic device including one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any implementation of the first aspect.

[0018] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any of the implementations of the first aspect.

[0019] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the method as described in any of the implementations in the first aspect.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] (1) The functional Khib site prediction method based on protein sequence proposed in this invention adopts machine learning and combines a two-step training method to integrate multi-feature hybrid learning into a protein language model (PLM) framework to improve prediction accuracy.

[0022] (2) The functional Khib site prediction method based on protein sequence proposed in this invention uses a few-shot transfer learning strategy to help the model effectively generalize with limited labeled data and improve its prediction performance.

[0023] (3) The functional Khib site prediction method based on protein sequence proposed in this invention uses the class_weight function, which effectively solves the significant imbalance problem of the benchmark dataset and produces more reliable and robust predictions, thus achieving more accurate and efficient functional Khib site prediction. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a schematic flowchart illustrating the functional Khib site prediction method based on protein sequence, as an embodiment of this application.

[0026] Figure 2 This is a flowchart illustrating a method for predicting functional Khib sites based on protein sequences, as described in an embodiment of this application.

[0027] Figure 3 This is a schematic diagram of a protein sequence-based functional Khib site prediction device according to an embodiment of this application.

[0028] Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0030] Figure 1 This application illustrates an embodiment of a method for predicting functional Khib sites based on protein sequences, comprising the following steps:

[0031] S1. Obtain 2-hydroxyisobutyryl group data and extract HSP(10,10) data from it. HSP(10,10) data are Khib peptide data with 10 residues upstream and 10 residues downstream. HSP(10,10) data near Khib sites and HSP(10,10) data near other lysine sites are labeled to construct the first training data. HSP(10,10) data near functional Khib sites and HSP(10,10) data near other Khib sites are labeled to construct the second training data.

[0032] Specifically, 24,299 experimentally identified Khib (2-hydroxyisobutyryl) sites were collected from PubMed, CPLM databases, and experimental data from this application, and 16 experimentally validated functional Khib sites were obtained from the literature. HSP(10,10) was defined as a Khib peptide with 10 residues upstream and 10 residues downstream, and HSP(10,10) data was used for training. The collected HSP(10,10) data near the 24,299 Khib sites were used as positive pre-training data, and HSP(10,10) data near other lysine sites were used as negative data to train the Khib site prediction model and the first classification module. Similarly, the HSP(10,10) data near the 16 known functional Khib sites were used as positive data, and the HSPs near the remaining Khib sites were used as negative data to fine-tune the pre-trained Khib site prediction model, in order to develop a functional Khib site prediction model for predicting functional Khib sites. The character "*" is used to supplement lysine residues near the ends of protein sequences in HSP(10,10) data. Labels 0 and 1 are used to represent negative and positive data, respectively. Embodiments of this application employ a two-step training process to avoid the risk of overfitting due to insufficient data on functional Khib sites.

[0033] S2, construct a Khib site prediction model and a first classification module. The Khib site prediction model includes a protein sequence processing module and a second classification module. The Khib site prediction model and the first classification module are trained using the first training data to obtain a pre-trained Khib site prediction model and a trained first classification module. The pre-trained Khib site prediction model is used as the source model, and a functional Khib site prediction model is trained using a few-shot transfer learning method based on the second training data.

[0034] In a specific embodiment, the protein sequence processing module includes a token and position embedding layer, an autoregressive encoder layer, a pooling layer and a first dropout layer connected in sequence. The output features of the autoregressive encoder layer are cyclically input into the autoregressive encoder layer for repeated multiple times.

[0035] In a specific embodiment, both the first classification module and the second classification module are deep neural networks, including a first dense layer, a first ReLU function layer, a second Dropout layer, a second dense layer, a second ReLU function layer, a third Dropout layer, and an output layer connected in sequence. The output layer consists of two neurons with sigmoid functions, which output scores.

[0036] Specifically, embodiments of this application design a Transformer-like protein sequence processing module to process protein sequences. This module includes a token and position embedding layer, an autoregressive encoder layer, a pooling layer, and a first Dropout layer, but no decoder layer. The first Dropout layer is then connected to a second classification module, which is a four-layer deep neural network (DNN). The DNN includes a first dense layer, a first ReLU function layer, a second Dropout layer, a second dense layer, a second ReLU function layer, a third Dropout layer, and an output layer, connected sequentially. The structure of the first classification module is identical to that of the second classification module, forming a framework based on a protein language model (PLM).

[0037] refer to Figure 2 The embodiments of this application encode seven types of sequence features based on HSP(10,10) data, namely GPS, PseAAC, CKSAAPs, OBC, AAindex, PSSM, and ACFs, and three types of structural features, namely ASA, SS, and BTAs. The seven types of sequence features and the three types of structural features are described below.

[0038] 1. PseAAC

[0039] The frequencies of 21 amino acids in the HSP(10,10) peptide were directly calculated using PseAAC. The frequencies F of 21 different pseudo-amino acid types in HSP(10,10) were collected in alphabetical order (A, C, D, ..., Y, *), and each peptide i was converted into a 21-dimensional vector.

[0040] V i =(F A ,F C ,F D ,…,F Y ,F * ) 21 .

[0041] 2. CKSAAP

[0042] CKSAAP represents the composition of amino acid pairs separated by k other residues within an HSP(10,10) peptide and is used to illustrate short-range interactions between amino acid residues within an HSP(10,10). For example, if a pair of amino acids (AAs) appears 1 time in the peptide, then the composition of a pair of AAs in the vector is equal to 1(Q). AA =l). Consider all k = 0, 1, ..., k max The amino acid pairs, and each HSP(10,10) i Encoded as a 21*21*(k) max+1 A numerical vector of 3D, specifically as follows:

[0043]

[0044] Evaluate different k max Choose the value, k max =0 was used as the baseline dataset for the pre-trained Khib site prediction model and the functional Khib site prediction model. Finally, a 21*21 dimensional vector was used as the CKSAAP feature.

[0045] 3. OBC

[0046] The 21 amino acids are ordered alphabetically, and each amino acid is converted into a binary vector. For the j-th amino acid, the j-th position is set to 1, and the remaining positions are set to 0. For example, "Alanine" is represented as [1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0], and "Cysteine" is represented as [0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]. For HSP(10, 10), a 21*20 dimensional vector is generated from its neighboring residues.

[0047] 4. AAindex

[0048] AAindex is a database containing 566 physicochemical properties collected from various databases, reflecting the diverse physicochemical and biological properties of amino acids. To reflect the diversity of physicochemical properties, each amino acid is assigned a corresponding numerical value. For a given physicochemical property h, each HSP(10, 10) is assigned a value. i It is converted into a 21-dimensional numerical vector:

[0049] v i =h1,h2,h3,…,h 21 ;

[0050] H p The AAindex physicochemical property represents the p-th residue (p = 1, 2, ..., 21). A penalized logistic regression method was used to independently train the computational model using 566 physicochemical properties for each dataset. AUC values ​​were calculated using 5-fold cross-validation. The top 10 most informative physicochemical parameters in each dataset, based on the highest AUC scores, were selected. Therefore, each HSP(10, 10) containing 21 residues... i It is converted into a 21×10 dimensional numerical vector:

[0051] V i =v1,v2,v3,…,v 10 .

[0052] 5. ACF

[0053] ACF represents sequence order information surrounding the PTM site. The HSP(10,10) data were encoded using the top 10 most informative physicochemical features selected from the AAindex database. For a given physicochemical property h, each HSP(10,10)... i It is converted into a 20-dimensional numerical vector:

[0054] v i =h1,h2,h3,…,h 21 ;

[0055] Among them, h p The AAindex physicochemical properties of the p-th residue (p = 1, 2, ..., 21) are represented, and the central Khib site is also encoded. Then, ACF is defined as:

[0056]

[0057] Where L is the length of HSP(10,10), which is equal to 21, and m=1 represents the correlation between two adjacent amino acid residues. Then, for each HSP(10,10)...i This generates a 20*10 dimensional vector:

[0058] V i =[r1,r2,r3,…,r 20 ]1,[r1,r2,r3,…,r 20 ]2,…,[r1,r2,r3,…,r 20 ] i ,…,[r1,r2,r3,…,r 20 ] 10 ;

[0059] Where i represents the i-th AAindex physicochemical property.

[0060] 6. GPS

[0061] GPS is a phosphorylation site prediction method that predicts kinase-specific phosphorylation sites in proteins by assuming that short peptides with similar biological characteristics share similar features. In this application, a positive dataset is specified and an HSP(10, 10) is used. p The average similarity score (S) between them was modified from the original GPS 3.0 algorithm:

[0062]

[0063] L is the length of HSP(10,10), i.e., L=21, N is the number of positive sample peptides, and T st Indicates positive sample T s The amino acid at position t (s=1,2,3,…,N), W t Here, is the weight value at position t, and M is an amino acid substitution matrix, such as the BLOSUM62 matrix used in this application. Initially, the weight of each position in HSP(10, 10) is 1. For each dataset, 5-fold cross-validation was performed after training the model using penalized logistic regression. W with the highest AUC value was selected. t The vector serves as the optimal weight vector. Subsequently, the provided HSP(10, 10) is defined. P The average similarity score (PS) between amino acid a in the middle dataset and residue b in the positive dataset. ab ), as shown below:

[0064]

[0065] Among them, C t This represents the number of amino acid pairs ab at position t. For 21 types of amino acids, there are a total of [21*(21+1)] / 2 = 231 unique S. ab Fractions, where S ab=S ba This GPS matrix quantifies the degree of location-weighted similarity between a given HSP(10, 10) and all Khib sites in the positive dataset. Therefore, HSP(10, 10) can be represented as a 231-dimensional vector, as shown below:

[0066] V = (S AA ,S AC ,S AD ,…,S ** ) 231 .

[0067] 7. PSSM

[0068] The position-specific scoring matrix (PSSM) feature extraction technique was used to evaluate the evolutionary information of Khib sites. Using the position-specific iterative BLAST (PSI-BLAST) tool in the BLAST software package, all HSP(10,10) peptides in each dataset were aligned with Swiss-Prot protein sequences from the UniProt database. A unique PSSM was developed for each HSP(10,10), providing probability values ​​(Pi, i = 1, 2, 3, ..., 21) for 20 common amino acids distributed at 21 positions. Subsequently, each HSP(10,10) was... i Encoded as a 420-dimensional numerical vector, as shown below:

[0069]

[0070] 8. ASA

[0071] ASA is an estimated area of ​​amino acid residues exposed in the solvent. As mentioned earlier, the potential ASA value A for each amino acid present in the peptide is calculated using the SPIDER2 tool. p (p = 1, 2, 3, ..., 21). Then, for each HSP(10, 10)... i It is represented as a 21-dimensional numerical vector, as shown below:

[0072] V i =(A1,A2,…,A) 21 )

[0073] 9. SS

[0074] The probability fraction Y of each amino acid in the α-helix, β-sheet, or coil structure was calculated using the SPIDER2 tool. p (p = 1, 2, 3, ..., 21). Each HSP(10, 10) i It is converted into a 21*3 dimension vector through the following steps:

[0075] V i =(Y1,Y2,…,Y) 21 ) α-helix (Y1,Y2,…,Y 21 ) β-strand (Y1,Y2,…,Y 21 ) coil

[0076] 10. BTA

[0077] SPIDER2 was used to compute each HSP(10, 10). i The four parameters: Cα i -Cα i+1 (τ), Cαi-1-Cαi-Cαi+1(θ), and the main chain twist angles φ and Ψ. Each HSP(10,10) i It is converted into a 4×21 dimensional vector using the following method:

[0078]

[0079] In addition, each HSP(10,10) data point is processed by the protein sequence processing module in the Khib site prediction model to generate a 128-dimensional vector as a multi-dimensional feature:

[0080] V i =(C1,C2,…,C 127 C 128 ) 128 .

[0081] In the first and second classification modules, the first and second ReLU function layers introduce nonlinear characteristics into the deep neural network and solve the gradient vanishing problem, as detailed below:

[0082]

[0083] Two neurons with sigmoid functions are used in the output layer to calculate the final score for a given HSP(10,10) peptide y, as shown below:

[0084]

[0085] For each HSP(10,10) dataset, the 11 features are independently scored by the second classification module, generating an 11-dimensional score vector V = (D1,D2,D3,…,D…) containing these 11 scores. 11 ).

[0086] To combine information from different features, the score vector V is input into the first classification module to obtain the final score.

[0087] The protein sequence and its corresponding label are input into the protein sequence processing module. The multi-dimensional features output from the first dropout layer are used as the 11th feature. Each of the seven sequence features, three structural features, and one multi-dimensional feature is encoded as a one-dimensional vector for training and prediction in the first and second classification modules. The multi-dimensional features output from the protein sequence processing module, along with the seven sequence features and three structural features, are input into the second classification module to generate their corresponding scores. The scores for each feature are then input into the first classification module to obtain the final score. This final score is compared with the initial label, and the parameters of the Khib site prediction model and the first classification module are optimized using the error backpropagation algorithm. During training, five-fold cross-validation is used to ensure the robustness and generalization ability of the model.

[0088] Furthermore, a few-shot transfer learning strategy was used to fine-tune the pre-trained Khib site prediction model. The transfer learning strategy was applied by using the pre-trained Khib site prediction model (pFKhib-pre) as the source model and employing secondary training data to train a functional Khib site prediction model (pFKhib) for predicting different expressed Khib sites. Sixteen reported and experimentally validated functional Khib sites were collected as positive data, and other Khib sites lacking functional experimental evidence were collected as negative data. During the transfer learning process, the Keras function `class_weight = {0:1., 1:10.}` was used to increase the weights of positive and negative data to balance them. After few-shot transfer learning, the functional Khib site prediction model was obtained.

[0089] S3. Obtain the 2-hydroxyisobutyryl group data to be predicted and extract the HSP(10,10) data to be predicted from it. Encode the HSP(10,10) data to be predicted to obtain several sequence features and several structural features. Input the HSP(10,10) data to be predicted, along with the several sequence features and several structural features, into the functional Khib site prediction model. Extract multi-dimensional features and obtain a score vector composed of the scores corresponding to the several sequence features, several structural features, and multi-dimensional features. Input the score vector into the trained first classification module to obtain the final score. Determine whether the HSP(10,10) data to be predicted is located near the functional Khib site based on the final score.

[0090] In a specific embodiment, in the Khib site prediction model and the functional Khib site prediction model, the HSP(10,10) data to be predicted is input into the protein sequence processing module to obtain multi-dimensional features. Several sequence features, several structural features, and multi-dimensional features are input into the second classification module to obtain their corresponding scores. The score output by the Khib site prediction model is used to predict whether the HSP(10,10) data is located near a Khib site. The score output by the functional Khib site prediction model is used to predict whether the HSP(10,10) data is located near a functional Khib site.

[0091] In a specific embodiment, several sequence features include GPS, PseAAC, CKSAAPs, OBC, AAindex, PSSM, and ACFs; several structural features include ASA, SS, and BTAs; and together with the multidimensional features output by the protein sequence processing module, they constitute 11 features, with a score vector of 11 dimensions V = (D1, D2, D3, ..., D...). 11 ); where D1, D2, D3, ..., D 11 The scores obtained by the second classification module correspond to the 11 features respectively.

[0092] In a specific embodiment, the multi-dimensional feature is a 128-dimensional vector output by the first Dropout layer.

[0093] Specifically, in the transfer learning process, the pre-trained Khib site prediction model is used as the source model. The source model is then fine-tuned using the second training data to obtain the target model, namely the functional Khib site prediction model. This functional Khib site prediction model demonstrates excellent performance in predicting functional Khib sites and can identify the most likely functional Khib sites in the 2-hydroxyisobutyryl group data, providing guidance for future experimental research.

[0094] Further reference Figure 3 As an implementation of the methods shown in the above figures, this application provides an embodiment of a functional Khib site prediction device based on protein sequences. This device embodiment is similar to... Figure 1 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.

[0095] This application provides a device for predicting functional Khib sites based on protein sequences, comprising:

[0096] Data acquisition module 1 is configured to acquire 2-hydroxyisobutyryl group data and extract HSP(10,10) data from it. HSP(10,10) data are Khib peptide data with 10 residues upstream and 10 residues downstream. HSP(10,10) data near Khib sites and HSP(10,10) data near other lysine sites are labeled to construct the first training data. HSP(10,10) data near functional Khib sites and HSP(10,10) data near other Khib sites are labeled to construct the second training data.

[0097] Model building module 2 is configured to build a Khib site prediction model and a first classification module. The Khib site prediction model includes a protein sequence processing module and a second classification module. The Khib site prediction model and the first classification module are trained using the first training data to obtain a pre-trained Khib site prediction model and a trained first classification module. The pre-trained Khib site prediction model is used as the source model, and a functional Khib site prediction model is trained using a few-shot transfer learning method based on the second training data.

[0098] Prediction module 3 is configured to acquire the 2-hydroxyisobutyryl group data to be predicted and extract the HSP(10,10) data to be predicted from it. It performs feature encoding on the HSP(10,10) data to be predicted to obtain several sequence features and several structural features. It inputs the HSP(10,10) data to be predicted, as well as several sequence features and several structural features, into the functional Khib site prediction model to extract multi-dimensional features and obtain a score vector composed of the scores of several sequence features, several structural features, and multi-dimensional features. It inputs the score vector into the trained first classification module to obtain the final score. Based on the final score, it determines whether the HSP(10,10) data to be predicted is located near the functional Khib site.

[0099] Figure 4 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. For example... Figure 4 As shown, the electronic device in this embodiment includes a processor 401 and a memory 402; wherein the memory 402 is used to store computer execution instructions; and the processor 401 is used to execute the computer execution instructions stored in the memory to implement the various steps performed by the electronic device in the above embodiment. For details, please refer to the relevant descriptions in the foregoing method embodiments.

[0100] Alternatively, the memory 402 can be either standalone or integrated with the processor 401.

[0101] When the memory 402 is set up independently, the electronic device also includes a bus 403 for connecting the memory 402 and the processor 401.

[0102] This invention also provides a computer program product, including a computer program that, when executed by a processor 401, implements the above-described method.

[0103] In the embodiments provided by this invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0104] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0105] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0106] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor 401 to execute some steps of the methods of the various embodiments of this application.

[0107] It should be understood that the processor 401 described above can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor, or the processor 401 can be any conventional processor 401. The steps of the method disclosed in this invention can be directly manifested as the hardware processor 401 executing the steps, or as a combination of hardware and software modules within the processor 401 executing the steps.

[0108] The memory 402 may include high-speed RAM memory, and may also include non-volatile memory NVM, such as at least one disk storage device, and may also be a USB flash drive, portable hard drive, read-only memory, disk or optical disc, etc.

[0109] Bus 403 can be an Industry Standard Architecture (ISA), a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Bus 403 can be divided into address bus, data bus, control bus, etc. For ease of illustration, the bus 403 in the accompanying drawings of this application is not limited to only one bus 403 or one type of bus 403.

[0110] The aforementioned storage medium can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium accessible to general-purpose or special-purpose computers.

[0111] An exemplary storage medium is coupled to processor 401, enabling processor 401 to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of processor 401. Processor 401 and storage medium can reside in application-specific integrated circuits (ASICs). Alternatively, processor 401 and storage medium can exist as discrete components in an electronic device or host device.

[0112] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting functional Khib sites based on protein sequences, characterized in that, Includes the following steps: 2-hydroxyisobutyryl group data were obtained and HSP(10,10) data were extracted from them. The HSP(10,10) data are Khib peptide data with 10 residues upstream and 10 residues downstream. The HSP(10,10) data near the Khib site and the HSP(10,10) data near other lysine sites were labeled to construct the first training data. The HSP(10,10) data near the functional Khib site and the HSP(10,10) data near other Khib sites were labeled to construct the second training data. A Khib site prediction model and a first classification module are constructed. The Khib site prediction model includes a protein sequence processing module and a second classification module. The Khib site prediction model and the first classification module are trained using the first training data to obtain a pre-trained Khib site prediction model and a trained first classification module. The pre-trained Khib site prediction model is used as the source model, and a functional Khib site prediction model is trained using a few-shot transfer learning method based on the second training data. Acquire the 2-hydroxyisobutyryl group data to be predicted and extract the HSP(10,10) data to be predicted from it. Encode the HSP(10,10) data to be predicted using features to obtain several sequence features and several structural features. Input the HSP(10,10) data to be predicted, along with the sequence features and structural features, into the functional Khib site prediction model to extract multi-dimensional features and obtain a score vector composed of the scores corresponding to the sequence features, structural features, and multi-dimensional features. Input the score vector into the trained first classification module to obtain the final score. Determine the HSP(10,10) to be predicted based on the final score. The Khib site prediction model determines whether the HSP(10,10) data is located near a functional Khib site. In the Khib site prediction model and the functional Khib site prediction model, the HSP(10,10) data to be predicted is input into the protein sequence processing module to obtain multi-dimensional features. The sequence features, structural features, and multi-dimensional features are then input into the second classification module to obtain their corresponding scores. The scores output by the Khib site prediction model are used to predict whether the HSP(10,10) data is located near a Khib site. The scores output by the functional Khib site prediction model are used to predict whether the HSP(10,10) data is located near a functional Khib site.

2. The method for predicting functional Khib sites based on protein sequences according to claim 1, characterized in that, The protein sequence processing module includes a token and position embedding layer, an autoregressive encoder layer, a pooling layer, and a first dropout layer connected in sequence. The output features of the autoregressive encoder layer are cyclically input into the autoregressive encoder layer for repeated processing multiple times.

3. The method for predicting functional Khib sites based on protein sequences according to claim 1, characterized in that, Both the first classification module and the second classification module are deep neural networks, including a first dense layer, a first ReLU function layer, a second Dropout layer, a second dense layer, a second ReLU function layer, a third Dropout layer, and an output layer connected in sequence. The output layer consists of two neurons with sigmoid functions, which output scores.

4. The method for predicting functional Khib sites based on protein sequences according to claim 3, characterized in that, The sequence features include GPS, PseAAC, CKSAAPs, OBC, AAindex, PSSM, and ACFs; the structural features include ASA, SS, and BTAs; and together with the multidimensional features output by the protein sequence processing module, they constitute 11 features, where the score vector is an 11-dimensional vector V = (D1, D2, D3, …, D…). 11 ); where D1, D2, D3, …, D 11 The scores obtained by the second classification module correspond to the 11 features respectively.

5. The method for predicting functional Khib sites based on protein sequences according to claim 2, characterized in that, The multi-dimensional feature is a 128-dimensional vector output by the first Dropout layer.

6. A device for predicting functional Khib sites based on protein sequences, characterized in that, include: The data acquisition module is configured to acquire 2-hydroxyisobutyryl group data and extract HSP(10,10) data from it. The HSP(10,10) data is Khib peptide data with 10 residues upstream and 10 residues downstream. The HSP(10,10) data near the Khib site and the HSP(10,10) data near other lysine sites are labeled to construct the first training data. The HSP(10,10) data near the functional Khib site and the HSP(10,10) data near other Khib sites are labeled to construct the second training data. The model building module is configured to build a Khib site prediction model and a first classification module. The Khib site prediction model includes a protein sequence processing module and a second classification module. The Khib site prediction model and the first classification module are trained using the first training data to obtain a pre-trained Khib site prediction model and a trained first classification module. The pre-trained Khib site prediction model is used as the source model, and a functional Khib site prediction model is trained based on the second training data using a few-shot transfer learning approach. The prediction module is configured to acquire the 2-hydroxyisobutyryl group data to be predicted and extract the HSP(10,10) data to be predicted from it. It performs feature encoding on the HSP(10,10) data to be predicted, obtaining several sequence features and several structural features. The HSP(10,10) data to be predicted, along with the sequence and structural features, are input into the functional Khib site prediction model to extract multi-dimensional features and obtain a score vector composed of the scores corresponding to the sequence, structural, and multi-dimensional features. This score vector is input into the trained first classification module to obtain the final score. Based on the final score, the predicted HSP(10,10) site is determined. Whether the HSP(10,10) data is located near a functional Khib site; in the Khib site prediction model and the functional Khib site prediction model, the HSP(10,10) data to be predicted is input into the protein sequence processing module to obtain multi-dimensional features. The several sequence features, several structural features and multi-dimensional features are respectively input into the second classification module to obtain their corresponding scores; the score output by the Khib site prediction model is used to predict whether the HSP(10,10) data is located near a Khib site; the score output by the functional Khib site prediction model is used to predict whether the HSP(10,10) data is located near a functional Khib site.

7. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-5.

Images