Liquid-liquid phase separation protein prediction method and system based on multivariate characteristics

Abstract

The invention discloses a multivariate feature-based liquid-liquid phase separation protein prediction method, which comprises the following steps of: extracting sequence features, secondary structure features and spatial structure features during prediction, and constructing a residue contact diagram according to a spatial structure; a graph attention network is adopted to extract a spatial structure graph constructed according to a residue contact graph and sequence embedding to excavate deep structure features, and prediction performance is effectively improved based on the characteristics closely related to protein phase separation behaviors and structures thereof; besides, according to the scheme, a three-layer stacked graph attention network is adopted to extract structural features of the protein, physical and chemical features extracted from the sequence are combined, the liquid-liquid phase separation protein is predicted through a stacked integrated learning model, and the model prediction performance can be further improved.

Description

Technical Field

[0001] The invention belongs to the field of computer bioinformatics, and in particular relates to a liquid-liquid phase separation protein prediction method and system based on multivariate features. Background Art

[0002] Liquid-liquid phase separation (LLPS) refers to the interaction between certain biomacromolecules (such as proteins and RNA) in cells to form phase-separated droplets with different components and properties. These droplets are similar to the separation state of oil droplets in water, forming a unique substructure in cells. The identification of liquid-liquid phase separation proteins is of great significance in the study and treatment of diseases. Existing methods for identifying liquid-liquid phase separation proteins include traditional experimental methods (fluorescence microscopy, atomic force microscopy, and nuclear magnetic resonance, etc.), first-generation prediction tools (PLAAC, LARKS, CatGRANULE, PScore, etc.), and second-generation prediction tools (FuzDrop, etc.).

[0003] Traditional experimental methods identify key proteins associated with LLPS by directly observing phase separation behavior. However, although experimental methods can provide precise qualitative and quantitative information, the high cost and low throughput of experimental operations have obvious limitations in large-scale data screening.

[0004] The first generation of prediction tools used computational methods, most of which were not originally developed specifically for predicting phase separation tendencies. They were later shown to be able to predict phase-separating proteins, but the features used by these methods were simple and could not capture the complex drivers of phase separation. The prediction accuracy was low and they were only suitable for preliminary screening.

[0005] The second generation of prediction tools are based on a large amount of experimental data and complex physical and chemical characteristics, combined with machine learning algorithms, and can perform large-scale predictions of LLPS proteins. However, these methods are limited to limited sequence features, and the accuracy and stability of the predictions still need to be improved. Summary of the invention

[0006] The purpose of the present invention is to propose a liquid-liquid phase separation protein prediction method and system based on multivariate features in view of the problems existing in the prior art.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions: A method for predicting liquid-liquid phase separation proteins based on multivariate features, the method comprising: Prepare positive and negative sample datasets; Obtain sequence data and structural data of each positive and negative sample; Extract sequence features based on sequence data; Generate sequence embeddings based on sequence data; A residue contact graph is constructed based on the structural data, a spatial structure graph is obtained based on the residue contact graph and sequence embedding, and spatial structural features of the protein are extracted based on the spatial structure graph through a three-layer stacked graph attention network; Extraction of secondary structural features based on structural data; Splicing the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; The concatenated feature set is used to train an integrated learner consisting of a base learner and a meta learner, wherein the base learner takes the concatenated features as input and outputs secondary features, and the meta learner takes the secondary features as input and outputs a prediction result of protein liquid-liquid phase separation behavior.

[0008] In the above-mentioned multivariate feature-based liquid-liquid phase separation protein prediction method, proteins with liquid-liquid phase separation behavior and without modification are obtained from the liquid-liquid phase separation database as positive sample data sets; Proteins that have never been reported to have liquid-liquid phase separation behavior in various organisms are used as negative sample datasets.

[0009] In the above-mentioned liquid-liquid phase separation protein prediction method based on multivariate features, the sequence data of all proteins are obtained from the UniProt database according to the UniProt ID of the protein in the data set, and the structural data of all proteins are obtained from the AlphaFold database.

[0010] In the above-mentioned liquid-liquid phase separation protein prediction method based on multivariate features, extracting sequence features based on sequence data specifically includes: The scores of intrinsic disordered regions and low-complexity regions were calculated using the ESpritz and SEG algorithms, respectively; The particle formation propensity score of the protein was calculated by the catGRANULE algorithm; The prion-like propensity score of the protein was calculated using the PLAAC algorithm; The Localcider toolkit was used to calculate the length, proportion of charged residues, net charge per residue, Kappa value, Omega value, polyproline II helical propensity, average hydrophobicity, isoelectric point, proportion of residues that contribute to chain extension, and proportion of residues that promote disordered regions for each protein sequence; The frequency of π-π interactions was calculated using PScore; Protein solubility was obtained using the Protein-sol software package.

[0011] In the above-mentioned liquid-liquid phase separation protein prediction method based on multivariate features, the extraction of secondary structure features based on structural data specifically includes directly obtaining secondary structure features such as α-helix, β-sheet, β-turn, etc. of the protein through the DSSP software package.

[0012] In the above-mentioned liquid-liquid phase separation protein prediction method based on multivariate features, a residue contact map is constructed based on structural data, a spatial structure map is obtained based on the residue contact map and sequence embedding, and the spatial structural features of the protein are extracted based on the spatial structure map through a three-layer stacked graph attention network, specifically including: The residue contact map is constructed based on the structural data. The contact map of a protein with a length of L is represented by an L-order square matrix C = {c ij}, i,j=1,2,…,L: (1) If the Euclidean distance between the alpha carbon atoms of two residues i and j is less than 8 , then the two residues are defined as contacting each other, which is represented by 1, otherwise it is 0; The spatial structure graph of a protein is defined as G = (V, E), where V represents a set of nodes, each node corresponds to a residue of a protein, the initial node feature is a sequence embedding obtained using the SeqVec model with sequence data as input, and E is an adjacency matrix. Derived from residue contact map; That is, the spatial structure graph consists of two parts: the sequence embedding and the adjacency matrix (i.e., the residue contact graph); The three-layer stacked graph attention network takes the spatial structure graph with initial node features as input, and updates the node features to output spatial structure features through the following formula: (2) Among them, K represents the number of heads of the multi-head attention mechanism, is a learnable linear transformation matrix, is the first-order neighbor set of node i, || represents the feature concatenation operation, is the activation function, is the normalized attention coefficient calculated by the kth attention mechanism.

[0013] First, the fixed parameters of the graph attention network are trained separately, and then the trained graph attention network is used to extract spatial structure features for subsequent training of base learners and meta-learners.

[0014] In the above-mentioned liquid-liquid phase separation protein prediction method based on multivariate features, the base learners include random forest, XGBoost and LightGBM, which respectively perform probability prediction based on the input spatial structure graph and output probability values ​​respectively. , probability value , probability value , concatenating the output results of the three base learners to obtain the secondary features: in , Respectively represent the predicted probability of the i-th sample in the three base learners, and N is the total number of samples; The meta-learner includes logistic regression.

[0015] In the above-mentioned liquid-liquid phase separation protein prediction method based on multivariate features, the method also includes using oversampling technology to increase the number of positive samples to balance the number of positive and negative samples.

[0016] A method for predicting liquid-liquid phase separation proteins based on multivariate features, the method comprising: Obtain sequence data and structural data of the protein to be predicted; Extract sequence features based on sequence data; Generate sequence embeddings based on sequence data; constructing a residue contact graph based on the structural data, obtaining a spatial structure graph based on the residue contact graph and sequence embedding, and extracting spatial structural features of the protein based on the spatial structure graph using a trained three-layer stacked graph attention network; Extraction of secondary structural features based on structural data; Splicing the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; Taking the splicing feature set as input, a trained ensemble learner is used to output the prediction results of the liquid-liquid phase separation behavior of the protein to be predicted.

[0017] A liquid-liquid phase separation protein prediction system based on multivariate features, comprising a first-type feature extraction module, a second-type feature extraction module, a feature splicing module and a prediction module; The first type of feature extraction module is used to extract sequence features based on sequence data; The second type of feature extraction module is used to extract secondary structure features based on structural data, construct a residue contact map based on the structural data, obtain a spatial structure map based on the residue contact map and the sequence embedding generated based on the sequence data, and extract the spatial structure features of the protein based on the spatial structure map through a graph attention module; The feature splicing module is used to splice the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; The prediction module is used to output the prediction result of protein liquid-liquid phase separation behavior based on the splicing feature set.

[0018] The advantages of the present invention are: This scheme extracts sequence features, secondary structure features and spatial structure features at the same time. In addition, a residue contact graph is constructed based on the spatial structure, and a graph attention network is used to extract the spatial structure graph constructed based on the residue contact graph and sequence embedding to mine deep structural features. Based on the fact that protein phase separation behavior is closely related to its structure, the prediction performance can be effectively improved. This scheme uses a three-layer stacked graph attention network to extract the structural features of proteins, and combines them with the physicochemical features extracted from the sequence to predict liquid-liquid phase separation proteins through a stacked ensemble learning model, which can further improve the model prediction performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The flowchart of the method for implementing the liquid-liquid phase separation protein prediction method based on multivariate features of the present invention; Figure 2 The model framework diagram of the liquid-liquid phase separation protein prediction method based on multivariate features of the present invention; Figure 3 It is a box plot comparing some features between positive and negative data sets in the liquid-liquid phase separation protein prediction method based on multivariate features of the present invention; Figure 4 The contribution of manually extracted features; Figure 5 A prediction flow chart of the method for predicting liquid-liquid phase separation proteins based on multivariate features of the present invention; Figure 6 ROC curve diagram of the liquid-liquid phase separation protein prediction method based on multivariate features of the present invention and other methods in the prior art; Figure 7 A system block diagram of the liquid-liquid phase separation protein prediction system based on multivariate features of the present invention.

[0020] Reference numerals: first-category feature extraction module 01 ; second-category feature extraction module 02 ; image attention module 021 ; feature splicing module 03 ; prediction module 04 . DETAILED DESCRIPTION

[0021] This protocol provides a method for predicting proteins in liquid-liquid phase separation based on multivariate features, e.g. Figure 1 and Figure 2 As shown, the specific steps are as follows: The first step is to construct a data set: screen and obtain experimentally verified or calculated high-confidence natural proteins with liquid-liquid phase separation behavior and no modification as positive data sets from databases such as PhaSepDB, LLPSDB, PhaSePro, and DrLLPS; Proteins that have never been reported to have liquid-liquid phase separation behavior were collected from 10 representative species, including human, Caenorhabditis elegans, Drosophila melanogaster, Escherichia coli, Arabidopsis thaliana, mouse, rat, Saccharomyces cerevisiae, fission yeast, and Xenopus laevis, as negative data sets; Proteins whose structures could not be found in known databases were removed, and finally the CD-HIT tool was used to remove redundancy to obtain the final data set.

[0022] The second step is to obtain sequence data and structural data: According to the UniProt ID (an identifier used to uniquely identify proteins in the UniProt database) of the protein in the dataset, the FASTA format sequence data of all proteins in the dataset are obtained from the UniProt (Universal Protein Resource) database, and the PDB structural data files of all proteins are obtained from the AlphaFold database.

[0023] The third step is sequence feature extraction: based on the sequence data, the ESpritz and SEG algorithms are used to calculate the scores of intrinsic disordered regions (IDRs) and low complexity regions (LCRs), respectively, to reflect the characteristics of protein sequences in terms of flexibility, dynamics, and repetitive sequences; The granule-formationpropensity of the protein was calculated using the catGRANULE algorithm. The prion-like propensity score (LLR) of the protein was calculated using the PLAAC algorithm; The Localcider toolkit is used to calculate the length (Length), the proportion of charged residues (FCR), the net charge per residue (NCPR), the Kappa value (a measure of the distribution pattern of charged residues in the sequence), the Omega value (reflecting the distribution pattern between charged residues and proline residues and other amino acid residues), the polyproline II helix tendency (PPIIPropensity), the mean hydropathicity (Mean Hydropathy), the isoelectric point (Isoelectric Point), the proportion of residues that contribute to chain extension (Fraction Expanding), and the proportion of residues that promote disordered regions (DisorderPromotingFraction) of each protein sequence; Use PScore to calculate the frequency of π-π interactions (Pi interaction); Protein solubility (Percent-Sol) was obtained using the Protein-sol software package.

[0024] Some feature comparisons between positive and negative datasets are as follows Figure 3As shown, Figure 3 It reflects the distribution difference of the values ​​of each feature between the positive and negative data sets. From the figure, we can see that these feature distribution differences are obvious, which is helpful for the classification prediction of positive and negative classes.

[0025] The fourth step is structural feature extraction: Based on the structural data file PDB, the secondary structural features of the protein such as α-helix (Alpha_Helix), β-fold (Beta_Sheet), and β-turn (Beta_Turn) are directly obtained through the DSSP software package.

[0026] In addition, the residue contact map is further constructed based on the structural data file PDB. The contact map of a protein with a length of L is represented by an L-order square matrix C={c ij}, i,j=1,2,…,L.

[0027] That is, if the Euclidean distance between the alpha carbon atoms of two residues i and j of a protein is less than 8 , then the two residues are defined as being in contact with each other, represented by 1, otherwise 0.

[0028] Secondly, the spatial structure graph of the protein is defined as G = (V, E), where V represents a set of nodes, each node corresponds to a residue of the protein, and the adjacency matrix Derived from the residue contact map.

[0029] The structural feature extraction module consisting of a three-layer stacked graph attention network GAT is used to extract protein structural features of fixed length through global maximum pooling operations. GAT introduces an attention mechanism when aggregating neighbor node features, which can assign different weight coefficients according to the degree of influence of neighbor nodes on the target node, thereby more effectively capturing the connection between residues that are far away but have spatial dependencies. The details are as follows: The initial input is the node feature , the feature dimension of each node is , this scheme models protein amino acids as vectors of fixed dimension d=1024 through SeqVec embedding. 3 After layer GAT processing, new node features are generated The node feature update formula is: Among them, K represents the number of heads of the multi-head attention mechanism, is a learnable linear transformation matrix, is the first-order neighbor set of node i, || represents the feature concatenation operation, is the activation function, is the normalized attention coefficient calculated by the k-th attention mechanism and is defined as follows: j is defined in the feature update formula , that is, the elements in the first-order neighbor set of node i.

[0030] In this scheme, the number of heads of the multi-head attention mechanism is set to K = 3 in the first two layers of GAT, and to 1 in the last layer of GAT. The node features of the last layer are subjected to the global maximum pooling operation, and a feature vector with a fixed length of 128 is output.

[0031] In the fifth step, the obtained sequence features and structural features are spliced, and then SMOTE oversampling is used to increase the number of positive samples and balance the number of positive and negative samples. The contribution of the manually extracted features - sequence features and secondary structure features is as follows Figure 4 shown.

[0032] In the sixth step, the concatenated feature samples are input into the ensemble learning model for training and verification. This solution constructs a two-level model framework combining base learners and meta-learners. First, Random Forest, XGBoost and LightGBM are used as base learners. These three models are all decision tree-based algorithms with strong nonlinear modeling capabilities and robustness. Among them: Random forest achieves classification and regression of input features through the integration of multiple decision trees, which can effectively reduce the risk of overfitting of a single model. Its core idea is to build multiple decision trees for the input feature X and output the prediction results by majority voting or average value. : Where T is the number of decision trees, is the prediction result of the tth tree.

[0033] XGBoost is an algorithm based on the gradient boosting framework, which uses a weighted iterative optimization strategy to optimize the loss function. The first and second order derivatives of are approximated and iteratively optimized, and the prediction results It is expressed as: in is the learning rate is the incremental prediction model of the t-th tree.

[0034] LightGBM is an efficient implementation based on histogram decision trees. It can quickly process large-scale high-dimensional data by grouping data and accelerating gradient updates. Its optimization process is similar to XGBoost, but it further improves the split point selection strategy to improve efficiency. By training the input feature samples X separately, these three base learners generate the prediction probability value corresponding to each sample , , .

[0035] Then, the output results of the above base learners are used as secondary features to construct the training data for the input meta-learner. Specifically, the prediction probabilities output by the base learners are concatenated to form a new secondary feature matrix Z: in , They represent the predicted probabilities of the i-th sample in the three base learners, and N is the total number of samples.

[0036] Logistic Regression is used as a meta-learner to further model and optimize the secondary features. Logistic regression is defined as: in are the parameters of logistic regression, optimized by minimizing the cross entropy loss function: Where N is the number of samples, and are the true label and predicted probability of the i-th sample respectively.

[0037] In this process, in order to ensure the generalization performance of the model, the K-fold cross-validation technique is used to verify the base learner and meta-learner. The data set is divided into K subsets, one of which is used as the validation set, and the remaining subsets are used as the training set. The model is trained and validated in a cycle, and the final validation results are combined to evaluate the stability and performance of the model. Finally, the trained integrated model is applied to unseen test feature samples. Generate initial prediction results through the base learner , , , and then the meta-learner optimizes and integrates the secondary feature matrix Z to obtain the final prediction result .

[0038] like Figure 5 As shown, the test process is as follows, using the test feature sample as the protein to be predicted: Obtain sequence data and structural data of the protein to be predicted; Extract sequence features based on sequence data. The specific method is the same as the training process. Generate sequence embeddings based on sequence data; constructing a residue contact graph based on the structural data, obtaining a spatial structure graph based on the residue contact graph and sequence embedding, and extracting spatial structural features of the protein based on the spatial structure graph using a trained three-layer stacked graph attention network; Extraction of secondary structural features based on structural data; The specific method of extracting spatial structure features and secondary structure features is the same as the training process.

[0039] Splicing the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; Taking the splicing feature set as input, the ensemble learner trained as above is used to output the prediction result of the liquid-liquid phase separation behavior of the protein to be predicted. .

[0040] In order to verify the performance and effect of this solution, this solution method is compared with the following existing methods: PLAAC, for predicting the prion-like propensity of proteins, evaluates their ability to form liquid-liquid phase separation by calculating the composition and distribution of specific amino acids in the protein sequence; CatGRANUL, which is used to predict the particle-forming tendency of proteins by analyzing the physicochemical properties and structural features of protein sequences to evaluate their ability to form liquid-liquid phase separation; PScore, used to predict the liquid-liquid phase separation tendency of proteins, evaluates their ability to form liquid-liquid phase separation by calculating the frequency of π-π interactions in protein sequences; FuzDrop combines multiple sequence and structural features to evaluate the ability of proteins to form liquid-liquid phase separation through a deep neural network model.

[0041] The area under the ROC curve (AUROC) was used as the performance evaluation indicator of the model, and the ROC curve was drawn. The results are as follows Figure 6 As shown, it can be seen that the ROC curve of this method forms a shape close to a right angle. Compared with several other methods, it has the best curve shape. The model has the lowest false positive rate and the highest true positive rate, showing better model performance.

[0042] Furthermore, if Figure 7 As shown, the present solution also provides a liquid-liquid phase separation protein prediction system based on multivariate features, including a first-type feature extraction module 01, a second-type feature extraction module 02, a feature splicing module 03 and a prediction module 04; The first type of feature extraction module 01 is used to extract sequence features based on sequence data. The extraction method is the same as the above model training process and will not be described here; The second type of feature extraction module 02 is used to extract secondary structure features based on structural data, construct a residue contact map based on the structural data, obtain a spatial structure map based on the residue contact map and the sequence embedding generated based on the sequence data, and extract the spatial structure features of the protein based on the spatial structure map through the graph attention module 021. The graph attention module 021 is used to implement a three-layer stacked graph attention network, and the spatial structure features are extracted through the graph attention network.

[0043] Feature splicing module 03, used for splicing sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; The prediction module 04 is used to implement the above-mentioned integrated learner and output the prediction result of protein liquid-liquid phase separation behavior based on the splicing feature set.

[0044] This scheme uses a three-layer stacked graph attention network to extract the structural features of proteins, and combines them with the physicochemical features extracted from the sequence to predict liquid-liquid phase separation proteins through a stacked ensemble learning model, which effectively improves the system's prediction accuracy of whether the protein phase separation occurs or not. Compared with the existing various protein liquid-liquid phase separation prediction methods, it has higher prediction performance.

[0045] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0046] It should be noted that there is no absolute order relationship between the steps in the above specific implementation method when they are put into use. It only means that this embodiment is the order given above. Those skilled in the art can change the order of the steps without deviating from the spirit of the present invention or exceeding the scope defined by the attached claims.

Claims

1. A method for predicting liquid-liquid phase separation proteins based on multivariate features, characterized in that: The method includes: Prepare positive and negative sample datasets; Obtain sequence data and structural data of each positive and negative sample; Extract sequence features based on sequence data; Generate sequence embeddings based on sequence data; A residue contact graph is constructed based on the structural data, a spatial structure graph is obtained based on the residue contact graph and sequence embedding, and spatial structural features of the protein are extracted based on the spatial structure graph through a three-layer stacked graph attention network; Extraction of secondary structural features based on structural data; Splicing the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; The concatenated feature set is used to train an integrated learner consisting of a base learner and a meta learner, wherein the base learner takes the concatenated features as input and outputs secondary features, and the meta learner takes the secondary features as input and outputs a prediction result of protein liquid-liquid phase separation behavior.

2. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: Obtain proteins with liquid-liquid phase separation behavior and without modification from the liquid-liquid phase separation database as positive sample data sets; Proteins that have never been reported to have liquid-liquid phase separation behavior in various organisms are used as negative sample datasets.

3. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: The sequence data of all proteins were obtained from the UniProt database according to the UniProt ID of the proteins in the dataset, and the structural data of all proteins were obtained from the AlphaFold database.

4. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: Extracting sequence features based on sequence data specifically includes: The scores of intrinsic disordered regions and low-complexity regions were calculated using the ESpritz and SEG algorithms, respectively; The particle formation propensity score of the protein was calculated by the catGRANULE algorithm; The prion-like propensity score of the protein was calculated using the PLAAC algorithm; The Localcider toolkit was used to calculate the length, proportion of charged residues, net charge per residue, Kappa value, Omega value, polyproline II helical propensity, average hydrophobicity, isoelectric point, proportion of residues that contribute to chain extension, and proportion of residues that promote disordered regions for each protein sequence; The frequency of π-π interactions was calculated using PScore; Protein solubility was obtained using the Protein-sol software package.

5. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: Extracting secondary structural features based on structural data specifically includes directly obtaining secondary structural features of proteins including α-helix, β-sheet, and β-turn through the DSSP software package.

6. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: A residue contact graph is constructed based on the structural data, a spatial structure graph is obtained based on the residue contact graph and sequence embedding, and the spatial structural features of the protein are extracted based on the spatial structure graph through a three-layer stacked graph attention network, specifically including: The residue contact map is constructed based on the structural data. The contact map of a protein with a length of L is represented by an L-order square matrix C = {c ij }, i,j=1,2,…,L: （1） If the Euclidean distance between the alpha carbon atoms of two residues i and j is less than 8 , then the two residues are defined as contacting each other, which is represented by 1, otherwise it is 0; The spatial structure graph of the protein is defined as G = (V, E), where V represents a node set, each node corresponds to a residue of the protein, and the sequence embedding obtained by using the SeqVec model with sequence data as input is the initial node feature —, and E is an adjacency matrix. Derived from residue contact map; The three-layer stacked graph attention network takes the spatial structure graph with initial node features as input, and updates the node features to output spatial structure features through the following formula: （2） Among them, K represents the number of heads of the multi-head attention mechanism, is a learnable linear transformation matrix, is the first-order neighbor set of node i, || represents the feature concatenation operation, is the activation function, is the normalized attention coefficient calculated by the kth attention mechanism.

7. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: The base learners include random forest, XGBoost and LightGBM. Random forest, XGBoost and LightGBM respectively make probability predictions based on the input spatial structure graph and output probability values ​​respectively. , probability value , probability value , concatenating the output results of the three base learners to obtain the secondary features: in , Respectively represent the predicted probability of the i-th sample in the three base learners, and N is the total number of samples; The meta-learner includes logistic regression.

8. The method for predicting liquid-liquid phase separation proteins based on multivariate features according to claim 1, characterized in that: The method also includes using oversampling technology to increase the number of positive samples to balance the number of positive and negative samples.

9. A method for predicting liquid-liquid phase separation proteins based on multivariate features, characterized in that: The method includes: Obtain sequence data and structural data of the protein to be predicted; Extract sequence features based on sequence data; Generate sequence embeddings based on sequence data; constructing a residue contact graph based on the structural data, obtaining a spatial structure graph based on the residue contact graph and sequence embedding, and extracting spatial structural features of the protein based on the spatial structure graph using a trained three-layer stacked graph attention network; Extraction of secondary structural features based on structural data; Splicing the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; Taking the splicing feature set as input, a trained ensemble learner is used to output the prediction results of the liquid-liquid phase separation behavior of the protein to be predicted.

10. A liquid-liquid phase separation protein prediction system based on multivariate features, characterized in that: It includes a first-class feature extraction module, a second-class feature extraction module, a feature splicing module and a prediction module; The first type of feature extraction module is used to extract sequence features based on sequence data; The second type of feature extraction module is used to extract secondary structure features based on structural data, construct a residue contact map based on the structural data, obtain a spatial structure map based on the residue contact map and the sequence embedding generated based on the sequence data, and extract the spatial structure features of the protein based on the spatial structure map through the graph attention module; the feature splicing module is used to splice the sequence features, spatial structure features and secondary structure features to obtain a splicing feature set; The prediction module is used to output the prediction result of protein liquid-liquid phase separation behavior based on the splicing feature set.

Images