Method for predicting protein solubility, neural network training method, device, equipment and medium

Abstract

The invention discloses a method for predicting protein solubility, a neural network training method, a neural network training device, equipment and a medium, and relates to the technical field of biological information, in particular to the technical field of biological computing. The method for predicting the protein solubility comprises the following steps: acquiring a target protein sequence comprising a plurality of amino acid residues; coding the target protein sequence by using a protein pre-training model to obtain respective coding feature vectors of a plurality of amino acid residues; determining respective replacement score vectors of the plurality of amino acid residues and contact probabilities of the plurality of amino acid residue pairs; processing the respective coding feature vectors of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs by using a first graph neural network; processing the respective replacement score vectors of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs by using a second graph neural network; and inputting results processed by the two graph neural networks into a prediction sub-network to obtain a prediction result of the solubility of the target protein sequence.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to Chinese patent application No. 202410878381.7 filed on July 2, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0002] The present disclosure relates to the field of bioinformatics, specifically artificial intelligence, biocomputing and other technical fields, and particularly to a method for predicting protein solubility, a neural network training method, an apparatus, a device and a medium. Background Art

[0003] Solubility is a key biophysical property of proteins, crucial for evaluating their effectiveness in biochemical engineering. Protein solubility depends on the interplay of various external physical conditions, such as pH and temperature, and intrinsic factors, such as the protein's amino acid composition and structure. Traditionally, protein solubility is often assessed using wet-phase assays, which are time-consuming and material-intensive. Therefore, a dry-phase assay is urgently needed to predict the solubility of potential proteins and reduce the costs associated with wet-phase assays. Summary of the Invention

[0004] It would be advantageous to provide a mechanism that alleviates, mitigates, or even eliminates one or more of the above-mentioned problems.

[0005] According to one aspect of the present disclosure, a method for predicting protein solubility using a neural network is provided, wherein the trained neural network includes a first graph neural network, a second graph neural network and a prediction subnetwork, the method including: obtaining a target protein sequence, the target protein sequence including multiple amino acid residues; encoding the target protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the multiple amino acid residues; obtaining a position-specific score matrix for the target protein sequence, the position-specific score matrix including position-specific score vectors for each of the multiple amino acid residues; determining the contact probability of multiple amino acid residue pairs in the target protein sequence; processing the encoding feature vectors for each of the multiple amino acid residues and the contact probabilities for the multiple amino acid residue pairs using the first graph neural network to obtain a first comprehensive feature vector for each of the multiple amino acid residues; processing the position-specific score vectors for each of the multiple amino acid residues and the contact probabilities for the multiple amino acid residue pairs using the second graph neural network to obtain a second comprehensive feature vector for each of the multiple amino acid residues; and inputting the first comprehensive feature vector and the second comprehensive feature vector for each of the multiple amino acid residues into the prediction subnetwork to obtain a prediction result for the solubility of the target protein sequence.

[0006] According to another aspect of the present disclosure, a neural network training method is provided, wherein the initial neural network includes a first initial graph neural network, a second initial graph neural network, and an initial prediction subnetwork, the method comprising: obtaining a sample protein sequence and the true solubility of the sample protein sequence, wherein the sample protein sequence includes multiple amino acid residues; encoding the sample protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the multiple amino acid residues; obtaining a position-specific score matrix for the sample protein sequence, wherein the position-specific score matrix includes position-specific score vectors for each of the multiple amino acid residues; determining contact probabilities for multiple pairs of amino acid residues in the sample protein sequence; and using the first initial graph neural network to train the protein sequence. The initial graph neural network processes the encoding feature vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a first comprehensive feature vector for each of the multiple amino acid residues; the second initial graph neural network processes the position-specific score vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a second comprehensive feature vector for each of the multiple amino acid residues; the first comprehensive feature vectors and the second comprehensive feature vectors of the multiple amino acid residues are input into the initial prediction subnetwork to obtain a prediction result of the protein solubility of the sample protein sequence; and based on the prediction result and the actual solubility, the parameters of the initial neural network are adjusted to obtain a trained neural network.

[0007] According to another aspect of the present disclosure, a device for predicting protein solubility using a neural network is provided, wherein the trained neural network includes a first graph neural network, a second graph neural network, and a prediction subnetwork, and the device includes: a first acquisition unit configured to acquire a target protein sequence, wherein the target protein sequence includes a plurality of amino acid residues; a first encoding unit configured to encode the target protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the plurality of amino acid residues; a second acquisition unit configured to acquire a position-specific score matrix for the target protein sequence, wherein the position-specific score matrix includes position-specific score vectors for each of the plurality of amino acid residues; a first determination unit configured to determine the target protein sequence. contact probabilities of multiple amino acid residue pairs in a sequence; a first processing unit, configured to use the first graph neural network to process the encoding feature vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain first comprehensive feature vectors of each of the multiple amino acid residues; a second processing unit, configured to use the second graph neural network to process the position-specific score vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain second comprehensive feature vectors of each of the multiple amino acid residues; and a first prediction unit, configured to input the first comprehensive feature vectors and the second comprehensive feature vectors of each of the multiple amino acid residues into the prediction subnetwork to obtain a prediction result of the solubility of the target protein sequence.

[0008] According to another aspect of the present disclosure, a device for training using a neural network is provided, wherein the initial neural network includes a first initial graph neural network, a second initial graph neural network, and an initial prediction subnetwork, and the device includes: a second acquisition unit, configured to acquire a sample protein sequence and the true solubility of the sample protein sequence, wherein the sample protein sequence includes multiple amino acid residues; a second encoding unit, configured to encode the sample protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the multiple amino acid residues; a third determination unit, configured to determine a replacement score vector for each of the multiple amino acid residues, wherein the replacement score vector includes a propensity score indicating that the corresponding amino acid residue is replaced with multiple preset amino acids at the position of the corresponding amino acid residue; and a fourth determination unit, configured to determine the connection between multiple amino acid residue pairs in the sample protein sequence. contact probability; a third processing unit is configured to use the first initial graph neural network to process the encoding feature vectors of each of the multiple amino acid residues and the contact probability of multiple amino acid residue pairs to obtain a first comprehensive feature vector for each of the multiple amino acid residues; a fourth processing unit is configured to use the second initial graph neural network to process the replacement score vectors of each of the multiple amino acid residues and the contact probability of multiple amino acid residue pairs to obtain a second comprehensive feature vector for each of the multiple amino acid residues; a second prediction unit is configured to input the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues into the initial prediction sub-network to obtain a prediction result of the protein solubility of the sample protein sequence; and a parameter adjustment unit is configured to adjust the parameters of the initial neural network based on the prediction result and the actual solubility to obtain a trained neural network.

[0009] According to another aspect of the present disclosure, a computer device is provided, comprising: at least one processor; and a memory on which a computer program is stored, wherein when the computer program is executed by the processor, the processor is caused to perform the above method.

[0010] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the processor is caused to perform the above method.

[0011] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, which, when executed by a processor, causes the processor to perform the above method.

[0012] According to another aspect of the present disclosure, a trained neural network obtained by the above neural network training method is provided.

[0013] According to one or more embodiments of the present disclosure, by combining the coding feature vector obtained by encoding the target protein sequence using a protein pre-training model and the contact probability of multiple amino acid residue pairs of the target protein sequence within the framework of a graph neural network, and combining the replacement score vectors of multiple amino acid residues of the target protein sequence and the contact probability of multiple amino acid residue pairs, the neural network can utilize richer protein information, thereby improving the diversity and comprehensiveness of feature representation, and can complete the prediction of protein solubility without obtaining or predicting the spatial structure information of the target protein sequence. By setting up two graph neural networks to process the above two combinations respectively, it is possible to fully explore the information related to protein solubility in the coding feature vectors and replacement score vectors obtained by different methods, and avoid interference between the two, thereby improving the predictive ability of the first comprehensive feature vector and the second comprehensive feature vector, and ultimately obtaining a more accurate solubility prediction result for the target protein sequence.

[0014] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings illustrate exemplary embodiments and constitute a part of the specification. Together with the description of the specification, they serve to explain exemplary implementation of the embodiments. The illustrated embodiments are for illustrative purposes only and do not limit the scope of the claims. Throughout the drawings, the same reference numerals designate similar, but not necessarily identical, elements.

[0016] Figure 1 A flowchart of a method for predicting protein solubility using a neural network according to an exemplary embodiment of the present disclosure is shown; Figure 2 A flowchart illustrating a process of processing encoding feature vectors of respective amino acid residues and contact probabilities of multiple amino acid residue pairs using a first graph neural network according to an exemplary embodiment of the present disclosure is shown; Figure 3 A flowchart illustrating a process of updating node features in first graph data using a first graph neural network according to an exemplary embodiment of the present disclosure is shown; Figure 4 A flowchart showing a process of processing the respective substitution score vectors of multiple amino acid residues and the contact probabilities of multiple amino acid residue pairs using a second graph neural network according to an exemplary embodiment of the present disclosure is shown; Figure 5A flowchart illustrating a process of updating node features in second graph data using a second graph neural network according to an exemplary embodiment of the present disclosure is shown; Figure 6 A flowchart of a process for processing first comprehensive feature vectors and second comprehensive features of respective amino acid residues using a prediction subnetwork according to an exemplary embodiment of the present disclosure is shown; Figure 7 A schematic diagram illustrating a neural network according to an exemplary embodiment of the present disclosure is shown; Figure 8 A flowchart of a neural network training method according to an exemplary embodiment of the present disclosure is shown; Figure 9 1 shows a structural block diagram of an apparatus for predicting protein solubility using a neural network according to an exemplary embodiment of the present disclosure; Figure 10 A structural block diagram of a neural network training apparatus according to an exemplary embodiment of the present disclosure is shown; and Figure 11 is a block diagram illustrating an exemplary computer device that can be used with the exemplary embodiments. DETAILED DESCRIPTION

[0017] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0018] In this disclosure, unless otherwise specified, the use of terms such as "first" and "second" to describe various elements is not intended to limit the positional relationship, temporal relationship, or importance relationship of these elements. Such terms are only used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of the element, while in some cases, based on the context of the description, they may also refer to different instances.

[0019] The terms used in the descriptions of the various examples described in this disclosure are for the purpose of describing specific examples only and are not intended to be limiting. Unless the context clearly indicates otherwise, if the number of elements is not specifically limited, the element may be one or more. In addition, the term "and / or" used in this disclosure encompasses any one and all possible combinations of the listed items.

[0020] Among the related technologies, the estimation and verification of protein solubility properties currently rely mainly on traditional laboratories, and there is a lack of more efficient biotechnology-based methods to predict protein solubility.

[0021] In order to solve the above problems, the present disclosure combines the coding feature vector obtained by encoding the target protein sequence using a protein pre-training model and the contact probability of multiple amino acid residue pairs of the target protein sequence under the framework of a graph neural network, and combines the replacement score vectors of multiple amino acid residues of the target protein sequence and the contact probability of multiple amino acid residue pairs, so that the neural network can utilize richer protein information, thereby improving the diversity and comprehensiveness of feature representation, and can complete the prediction of protein solubility without obtaining or predicting the spatial structure information of the target protein sequence. By setting up two graph neural networks to process the above two combinations respectively, it is possible to fully mine the information related to protein solubility in the coding feature vectors and replacement score vectors obtained by different methods, and avoid interference between the two, thereby improving the predictive ability of the first comprehensive feature vector and the second comprehensive feature vector, and ultimately obtaining a more accurate solubility prediction result for the target protein sequence.

[0022] The embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0023] According to one aspect of the present disclosure, a method for predicting protein solubility using a neural network is provided. Figure 1 FIG. 5 is a flow chart showing a method for predicting protein solubility using a neural network according to an exemplary embodiment of the present disclosure. Figure 1As shown, method 100 includes: step S101, obtaining a target protein sequence, wherein the target protein sequence includes multiple amino acid residues; step S102, encoding the target protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the multiple amino acid residues; step S103, determining a replacement score vector for each of the multiple amino acid residues, wherein the replacement score vector includes a propensity score indicating that the corresponding amino acid residue is replaced with multiple preset amino acids at the position of the corresponding amino acid residue; step S104, determining a contact probability for multiple amino acid residue pairs in the target protein sequence; step S105, processing the encoding feature vectors for each of the multiple amino acid residues and the contact probabilities for the multiple amino acid residue pairs using a first graph neural network to obtain a first comprehensive feature vector for each of the multiple amino acid residues; step S106, processing the replacement score vectors for each of the multiple amino acid residues and the contact probabilities for the multiple amino acid residue pairs using a second graph neural network to obtain a second comprehensive feature vector for each of the multiple amino acid residues; and step S107, inputting the first comprehensive feature vector and the second comprehensive feature vector for each of the multiple amino acid residues into a prediction subnetwork to obtain a prediction result for the solubility of the target protein sequence.

[0024] Therefore, by combining the encoding feature vector obtained by encoding the target protein sequence using the protein pre-training model and the contact probability of multiple amino acid residue pairs of the target protein sequence within the framework of the graph neural network, and combining the replacement score vectors of the multiple amino acid residues of the target protein sequence and the contact probability of multiple amino acid residue pairs, the neural network can utilize richer protein information, thereby improving the diversity and comprehensiveness of the feature representation, and can complete the prediction of protein solubility without obtaining or predicting the spatial structure information of the target protein sequence. By setting up two graph neural networks to process the above two combinations respectively, it is possible to fully explore the information related to protein solubility in the encoding feature vectors and replacement score vectors obtained by different methods, and avoid interference between the two, thereby improving the predictive ability of the first comprehensive feature vector and the second comprehensive feature vector, and ultimately obtaining a more accurate solubility prediction result for the target protein sequence.

[0025] In the present disclosure, although the various operations are depicted in the accompanying drawings as being performed in a specific order, this should not be understood as requiring that these operations must be performed in the specific order shown or in sequential order. For example, the order of step S102, step S103 and step S104 can be interchangeable or performed simultaneously. For example, the order of step S105 and step S106 can be interchangeable or performed simultaneously.

[0026] In some embodiments, in step S101, the sequence sources of the target protein include, but are not limited to: obtaining protein amino acid sequence information by mass spectrometry using peptide mass spectrometry and tandem mass spectrometry analysis, deducing the protein amino acid sequence by transcribing complementary deoxyribonucleic acid (cDNA) from messenger RNA (mRNA) and then performing deoxyribonucleic acid (DNA) sequencing, directly synthesizing the protein using synthetic biology, manually designing protein amino acid sequences, computer-designed protein amino acid sequences, NCBI database, EXProt database, UniProtKB database, and other common bioinformatics databases. The target protein sequence contains multiple amino acid residues, and these amino acid residues contain information related to protein properties, such as protein solubility information.

[0027] In some embodiments, in step S102, multiple amino acid residues included in the target protein sequence can be directly input into the protein pre-training model to obtain the encoding results output by the protein pre-training model, that is, the encoding feature vectors of each of the multiple amino acid residues.

[0028] A pre-trained model can be obtained by pre-learning language characteristics on a large amount of unlabeled data, which can then be fine-tuned on a small amount of labeled data. This training method can improve the performance of downstream tasks. The protein pre-trained model used in step S102 can be obtained by training on a large corpus of protein or nucleic acid sequences.

[0029] According to some embodiments, the protein pre-training model can be the ProtT5-XL model. Pre-trained on a large-scale protein sequence dataset, the ProtT5-XL model can capture sequence, structural, and functional information about proteins, providing rich and accurate feature representations. Using the ProtT5-XL model as a protein pre-training model significantly improves the quality of feature extraction, ultimately increasing the accuracy of protein solubility prediction.

[0030] In an exemplary embodiment, the ProtT5-XL model can be first trained on the BFD dataset and then fine-tuned using the UniRef50 dataset. The model contains a 24-layer attention structure, each layer consisting of 32 self-attention heads and 1024 neurons. The output of the last layer can be selected as the encoding feature vector of multiple amino acid residues. The length of each encoding feature vector is 1024, so the encoding feature vectors of multiple amino acid residues can form a size of L ×1024 matrix, where L is the number of amino acid residues in the target protein sequence.

[0031] In some embodiments, the protein pre-training model can also be ESM-1b, UniRef, ProteinBert, TAPE, ProtGPT2, ProtTXL, ProtBert, ProtXLNet, ProtAlbert, ProtElectra, ProtT5-XXL, Ankh or other models, which are not limited here.

[0032] It has been observed that the propensity score for amino acid residues to be replaced with other amino acids at specific positions is helpful for predicting protein solubility. By determining the substitution score vectors for each of the multiple amino acid residues in the target protein sequence (including the propensity score for each amino acid residue to be replaced with multiple preset amino acids at its position) and using a graph neural network to process the substitution score vectors and the base probabilities of multiple amino acid residue pairs, a feature vector with strong predictive power for protein solubility can be generated.

[0033] In some embodiments, step S103, determining the replacement score vectors of each of the multiple amino acid residues may include: searching a protein database for known protein sequences similar to the target protein sequence to obtain a multiple sequence alignment result; for each of the multiple amino acid residues, counting the target frequencies of multiple preset amino acids appearing at the position of the amino acid residue based on the multiple sequence alignment result; and determining, based on the target frequencies and the background frequencies of the multiple preset amino acids, a propensity score for the amino acid residue to be replaced with multiple preset amino acids at the position of the amino acid residue.

[0034] Background frequency is the probability of an amino acid occurring at all possible positions without any specific context or constraints. This is the global frequency of occurrence of each amino acid in a protein sequence and can be statistically determined from large-scale protein databases. The multiple preset amino acids can include natural amino acids as well as unnatural amino acids, such as synthetic amino acids, amino acid analogs with functions similar to natural amino acids, and amino acid mimetics. Natural amino acids may include, but are not limited to, 20 common natural amino acids, such as A (alanine), V (valine), L (leucine), I (isoleucine), F (phenylalanine), W (tryptophan), M (methionine), P (proline), G (glycine), T (threonine), S (serine), C (cysteine), N (aspartic acid), Q (glutamine), Y (tyrosine), K (lysine), R (arginine), H (histidine), D (aspartic acid), E (glutamic acid), B (aspartic acid or aspartic acid), Z (glutamine or glutamic acid), J (leucine or isoleucine), and may also include two less common pyrolysine and selenocysteine. The propensity score may be expressed, for example, as: S ij = log ( P ij / P j ),in S ij Indicates the i The position corresponding to the amino acid residue is replaced by amino acid j The score, P ij In the multiple sequence alignment results, i The amino acid residues appear at the position corresponding to j frequency, P j It is an amino acid j background frequency.

[0035] In an exemplary embodiment, the PSI-BLAST tool (e.g., version v2.4.0) can be used to search a protein database (e.g., UniRef90) with the target protein sequence as the query sequence, and a multiple sequence alignment result can be obtained after 3 iterations. The substitution score vectors of the multiple amino acid residues can form a vector of size L ×20 matrix, where L is the number of amino acid residues in the target protein sequence, and 20 standard amino acids are used as the multiple preset amino acids.

[0036] Contact probability is the likelihood that a pair of amino acid residues in a protein sequence will form a contact. In some embodiments, contact formation can be defined as the distance between atoms (e.g., Cβ atoms) of the two amino acid residues in the protein being less than a predetermined distance threshold (e.g., 8 Å).

[0037] It has been observed that by obtaining the contact probability of multiple amino acid residue pairs and combining the encoding feature vectors or replacement score vectors of multiple amino acid residues in the target protein sequence and then processing them by a graph neural network, it is possible to obtain a feature vector with strong predictive ability for protein solubility without obtaining or predicting the spatial structure information of the target protein sequence, thereby greatly reducing the cost of protein solubility prediction.

[0038] In some embodiments, step S104, determining the contact probabilities of multiple amino acid residue pairs in the target protein sequence, may include: determining multiple amino acid residue pairs in the target protein sequence; obtaining, for each of the multiple amino acid residue pairs, a high-dimensional feature vector between a first amino acid residue and a second amino acid residue included in the amino acid residue pair; and simultaneously inputting the high-dimensional feature vectors of the multiple amino acid residue pairs into a trained neural network for predicting contact probabilities to obtain the contact probabilities of the multiple amino acid residue pairs. In this manner, the neural network for predicting contact probabilities can consider information about other amino acids when predicting the contact probability between each amino acid residue pair, thereby improving the accuracy of the predicted contact probabilities.

[0039] According to some embodiments, the plurality of amino acid residue pairs may include amino acid residue pairs consisting of an optional first amino acid residue and a second amino acid residue in the target protein sequence. That is, the contact probability between any two amino acid residues in the target protein sequence can be determined in step S104. By determining the contact probability between all possible amino acid residue pairs, more comprehensive protein structure information can be provided, information omissions can be reduced, and the needs of complex proteins can be adapted. In addition, graph neural networks rely on the relationship between nodes (amino acid residues) and edges (amino acid residue pairs) for feature learning and prediction. By determining the contact probability of comprehensive amino acid residue pairs, more accurate edge topology and edge features can be provided, enabling the graph neural network to make more full use of this information, thereby improving the performance of the neural network and ultimately helping to improve the accuracy and reliability of protein solubility prediction.

[0040] In some embodiments, the high-dimensional feature vector between the first amino acid residue and the second amino acid residue (i.e., the high-dimensional feature vector of the amino acid residue pair) may include any combination of the following features: one-hot encoding of the first amino acid residue and the second amino acid residue, attention scores of the first amino acid residue and the second amino acid residue, and / or structural information of the target protein sequence. In an exemplary embodiment, the above-mentioned attention score may come from an attention map generated by a protein pre-training model based on the target protein sequence; the above-mentioned structural information may include the probability of 3-state secondary structure (SS3), 8-state secondary structure (SS8), solvent accessible surface area (ASA), hemispherical exposure (HSE), protein backbone dihedral angle ψ, , θ, τ and / or other protein structure information.

[0041] In some embodiments, the above features can be combined using concatenation, element-wise operations (eg, element-wise addition, subtraction, multiplication), or other methods to obtain a high-dimensional feature vector for the amino acid residue pair.

[0042] In some embodiments, the neural network used to predict contact probability may be a residual network (eg, ResNet). The neural network may be trained using labeled data describing whether amino acid residue pairs are in contact.

[0043] In some embodiments, in step S105, the encoding feature vectors of each of the multiple amino acid residues obtained in step S102 and the contact probabilities of the multiple amino acid residue pairs obtained in step S104 can be input into the first graph neural network to obtain a first comprehensive feature vector of each of the multiple amino acid residues output by the first graph neural network.

[0044] A graph neural network is a deep learning model based on graph-structured data, used to process structured data of nodes and edges. In some embodiments, the first graph neural network and the second graph neural network (described in detail below) can adopt a graph convolutional network (GCN), a graph attention network (GAT), a graph isomorphism network (GIN), a diffusion convolutional neural network (DCNN), or other graph neural networks.

[0045] According to some embodiments, the first and second graph neural networks can be based on GraphSAGE. GraphSAGE is a sampling-based graph neural network that effectively learns and represents node features by sampling and aggregating neighboring nodes. These features give GraphSAGE significant advantages in processing large-scale graph data, particularly when processing protein sequences to predict protein solubility, where GraphSAGE balances efficiency and accuracy.

[0046] Figure 2 FIG2 is a flowchart showing a process of processing the encoding feature vectors of multiple amino acid residues and the contact probabilities of multiple amino acid residue pairs using a first graph neural network according to an exemplary embodiment of the present disclosure. Process 200 can be used to implement step S105 in the above method 100. According to some embodiments, such as Figure 2 As shown, process 200 may include: step S201, using the encoding feature vectors of each of the multiple amino acid residues as node features, and using the contact probabilities of multiple amino acid residue pairs as edge features to construct first graph data; and step S202, using a first graph neural network to update the node features in the first graph data based on the node features and edge features in the first graph data to obtain a first comprehensive feature vector for each of the multiple amino acid residues.

[0047] Therefore, through the above method, the sequence information of the protein and the contact information of the amino acid residue pairs can be more effectively utilized under the framework of the graph neural network, thereby obtaining a first comprehensive feature vector with stronger predictive ability, thereby improving the accuracy and robustness of subsequent protein solubility predictions.

[0048] In step S201, multiple amino acid residues included in the target protein sequence can be determined as multiple nodes in first graph data, and a topological relationship of the first graph data can be determined based on the multiple amino acid residue pairs. In some embodiments, an edge can be set between two nodes corresponding to the first amino acid residue and the second amino acid residue included in each of the multiple amino acid residue pairs determined in step S104, thereby obtaining multiple edges in the first graph data. In some embodiments, the multiple amino acid residue pairs include amino acid residue pairs consisting of any first amino acid residue and second amino acid residue in the target protein sequence, and an edge can be set between any two nodes in the first graph data, that is, the first graph data is a complete graph.

[0049] In some embodiments, at least one amino acid residue pair with a contact relationship can be determined based on the contact probability of multiple amino acid residue pairs and a preset probability threshold, and the topological relationship of the first graph data can be determined based on the at least one amino acid residue pair. For example, an edge can be set between two nodes corresponding to the first amino acid residue and the second amino acid residue included in the amino acid residue pair with a contact probability greater than the preset probability threshold. In this way, it is possible to avoid establishing edges between nodes corresponding to amino acid residues without a contact relationship, thereby effectively reducing the number of edges, lowering the demand for computing and storage resources, and improving the reasoning efficiency of the first graph neural network.

[0050] According to some embodiments, the first graph neural network may include three first graph convolutional layers. By configuring the first graph neural network with three graph convolutional layers to process the encoded feature vectors and contact probabilities of multiple amino acid residues, optimal prediction performance can be achieved while maintaining reasonable computational resource consumption.

[0051] In some embodiments, in step S202, three first graph convolutional layers can be used to iteratively update the node features in the first graph data, and the updated features of the multiple amino acid residues output by the last first graph convolutional layer are obtained as the first comprehensive feature vector.

[0052] Figure 3 FIG3 is a flowchart showing a process of updating node features in first graph data using a first graph neural network according to an exemplary embodiment of the present disclosure. Process 300 can be used to implement step S202 in the above process 200. In some embodiments, as Figure 3 As shown, process 300 may include: step S301, using three first graph convolution layers to iteratively update the node features in the first graph data, and obtain the first intermediate feature vectors of each of the multiple amino acid residues respectively output by the three first graph convolution layers; and step S302, fusing the first intermediate feature vectors of each of the multiple amino acid residues respectively output by the three first graph convolution layers to obtain the first comprehensive feature vectors of each of the multiple amino acid residues.

[0053] Thus, through iterative updates of multi-layer graph convolutions, each convolution layer can collect information from a wider neighborhood, thereby aggregating more and more contextual information layer by layer. More specifically, the first layer mainly captures the local features of other amino acid residues that are close to the amino acid residue, the second layer begins to integrate a wider range of neighborhood information, and the third layer captures more global structural information. Ultimately, by fusing multiple layers of features, the neural network can comprehensively utilize the feature representations at each level to capture the complex relationships between amino acid residues and obtain a feature vector with stronger predictive capabilities.

[0054] In an exemplary embodiment, in step S301, each first graph convolution layer output size is L ×128 matrix, where L is the number of amino acid residues in the target protein sequence, and 128 is the hidden dimension of the first intermediate feature vector.

[0055] In some embodiments, in step S302, the three first intermediate feature vectors output by the three first graph convolutional layers of each amino acid residue can be fused by splicing, weighted summation, mean pooling, maximum pooling, attention mechanism, direct summation, processing using a multi-layer perceptron or other neural network, or any combination of the above means to obtain a first comprehensive feature vector of the amino acid residue.

[0056] According to some embodiments, the fusion in step S302 may include merging the first intermediate feature vectors of the multiple amino acid residues respectively output by the three first graph convolutional layers using a residual network structure.

[0057] Therefore, through the above method, the first intermediate feature vectors of different levels can be more effectively fused, and the expression ability of the first comprehensive feature vector obtained after fusion can be enhanced, thereby improving the performance of the neural network in protein solubility prediction.

[0058] In an exemplary embodiment, in step S302, for each amino acid residue, the three first intermediate feature vectors (length 128) of the amino acid residue outputted by the three first graph convolutional layers can be merged using a residual network structure to obtain a first comprehensive feature vector (length 384) of the amino acid residue. In other words, the first comprehensive feature vectors of the multiple amino acid residues can be constructed to form a network of size 384. L ×384 matrix.

[0059] Figure 4 FIG4 is a flowchart showing a process of processing the substitution score vectors of multiple amino acid residues and the contact probabilities of multiple amino acid residue pairs using a second graph neural network according to an exemplary embodiment of the present disclosure. Process 400 can be used to implement step S106 in the above method 100. According to some embodiments, such as Figure 4 As shown, process 400 may include: step S401, using the replacement score vectors of multiple amino acid residues as node features, and using the contact probabilities of multiple amino acid residue pairs as edge features to construct second graph data; and step S402, using the second graph neural network to update the node features in the second graph data based on the node features and edge features in the second graph data to obtain second comprehensive feature vectors for each of the multiple amino acid residues.

[0060] Therefore, through the above method, the propensity scores of amino acid residues in proteins being replaced by other amino acids and the contact information of amino acid residue pairs can be more effectively utilized under the framework of graph neural networks, thereby obtaining a second comprehensive feature vector with stronger predictive ability, thereby improving the accuracy and robustness of subsequent protein solubility predictions.

[0061] In some embodiments, the second graph data may have a similar or identical topological structure to the first graph data. In step S401, the plurality of amino acid residues included in the target protein sequence may be determined as a plurality of nodes in the second graph data, and the topological relationship of the second graph data may be determined based on the plurality of amino acid residue pairs. In some embodiments, edges may be set between two nodes corresponding to the first amino acid residue and the second amino acid residue included in each of the plurality of amino acid residue pairs determined in step S104, thereby obtaining a plurality of edges in the second graph data. In some embodiments, the plurality of amino acid residue pairs include amino acid residue pairs consisting of an optional first amino acid residue and a second amino acid residue in the target protein sequence, and edges may be set between any two nodes of the second graph data, that is, the second graph data is a complete graph.

[0062] In some embodiments, at least one amino acid residue pair with a contact relationship can be determined based on the contact probabilities of multiple amino acid residue pairs and a preset probability threshold, and the topological relationship of the second graph data can be determined based on the at least one amino acid residue pair. For example, an edge can be set between two nodes corresponding to the first and second amino acid residues included in the amino acid residue pair with a contact probability greater than the preset probability threshold. In this way, it is possible to avoid establishing edges between nodes corresponding to amino acid residues that do not have a contact relationship, thereby effectively reducing the number of edges, lowering the demand for computing and storage resources, and improving the reasoning efficiency of the second graph neural network.

[0063] According to some embodiments, the second graph neural network may include two second graph convolutional layers. By configuring the second graph neural network with two graph convolutional layers to process substitution score vectors and contact probabilities for multiple amino acid residues, optimal prediction performance can be achieved while maintaining reasonable computational resource consumption.

[0064] In some embodiments, in step S402, two second graph convolutional layers may be used to iteratively update node features in the second graph data, and updated features of multiple amino acid residues output by the last second graph convolutional layer may be obtained as a second comprehensive feature vector.

[0065] Figure 5FIG5 shows a flowchart of a process for updating node features in the second graph data using a second graph neural network according to an exemplary embodiment of the present disclosure. Process 500 can be used to implement step S402 in the above process 400. In some embodiments, as Figure 5 As shown, process 500 may include: step S501, using two second graph convolution layers to iteratively update the node features in the second graph data, and obtain the second intermediate feature vectors of each of the multiple amino acid residues respectively output by the two second graph convolution layers; and step S502, fusing the second intermediate feature vectors of each of the multiple amino acid residues respectively output by the two second graph convolution layers to obtain the second comprehensive feature vectors of each of the multiple amino acid residues.

[0066] Thus, through iterative updates of multi-layer graph convolution, each convolution layer can collect information from a wider neighborhood, thereby aggregating more and more contextual information layer by layer. More specifically, the first layer mainly captures the local features of other amino acid residues that are close to the amino acid residue, while the second layer begins to integrate information from a wider range of neighborhoods. Ultimately, by fusing multiple layers of features, the neural network can comprehensively utilize feature representations at all levels to capture the complex relationships between amino acid residues and obtain feature vectors with stronger predictive capabilities.

[0067] In an exemplary embodiment, in step S501, each second graph convolution layer output size is L ×32 matrix, where L is the number of amino acid residues in the target protein sequence, and 32 is the hidden dimension of the second intermediate feature vector.

[0068] In some embodiments, in step S502, the two second intermediate feature vectors output by the two second graph convolutional layers of each amino acid residue can be fused by splicing, weighted summation, mean pooling, maximum pooling, attention mechanism, direct summation, processing using a multi-layer perceptron or other neural network, or any combination of the above means to obtain a second comprehensive feature vector of the amino acid residue.

[0069] According to some embodiments, the fusion in step S502 may include merging the second intermediate feature vectors of the multiple amino acid residues respectively output by the two second graph convolutional layers using a residual network structure.

[0070] Therefore, through the above method, the second intermediate feature vectors of different levels can be more effectively fused, and the expression ability of the second comprehensive feature vector obtained after fusion can be enhanced, thereby improving the performance of the neural network in protein solubility prediction.

[0071] In an exemplary embodiment, in step S502, for each amino acid residue, the two second intermediate feature vectors (length 32) of the amino acid residue outputted by the two second graph convolutional layers can be merged using a residual network structure to obtain a second comprehensive feature vector (length 64) of the amino acid residue. In other words, the second comprehensive feature vectors of the amino acid residues can be constructed to form a vector of size 64. L ×64 matrix.

[0072] Figure 6 FIG. 6 is a flowchart illustrating a process of processing the first comprehensive feature vector and the second comprehensive feature of each of a plurality of amino acid residues using a prediction subnetwork according to an exemplary embodiment of the present disclosure. The process 600 can be used to implement step S107 in the above method 100, for example. According to some embodiments, such as Figure 6 As shown, process 600 may include: step S601, pooling the first comprehensive feature vectors of each of the multiple amino acid residues to obtain a first pooled feature vector of the target protein sequence; step S602, pooling the second comprehensive feature vectors of each of the multiple amino acid residues to obtain a second pooled feature vector of the target protein sequence; step S603, fusing the first pooled feature vector with the second pooled feature vector to obtain a target feature vector of the target protein sequence; and step S604, processing the target feature vector using a fully connected layer and an output layer to obtain a prediction result.

[0073] Therefore, through the above method, the comprehensive feature vector of multiple amino acid residues can be simplified into a feature vector of fixed length that represents the target protein sequence, making subsequent processing more efficient, and the first comprehensive feature vector and the second comprehensive feature vector representing different feature sources (encoding feature vector and replacement score vector) can be effectively fused, so that the neural network can make full use of this information to obtain more accurate protein solubility prediction results.

[0074] In some embodiments, in step S601, the first comprehensive feature vectors of the plurality of amino acid residues may be pooled along the direction of the amino acid residues to obtain the first pooled feature vector of the target protein sequence. In an exemplary embodiment, the matrix (size is ) formed by the first comprehensive feature vectors of the plurality of amino acid residues may be pooled. L × 128) and an average pooling operation is performed along the direction of the amino acid residues to obtain the first pooling feature vector with a size of 1 × 128.

[0075] In some embodiments, in step S602, the second comprehensive feature vectors of the plurality of amino acid residues may be pooled along the direction of the amino acid residues to obtain the second pooled feature vector of the target protein sequence. In an exemplary embodiment, the matrix (size is ) formed by the second comprehensive feature vectors of the plurality of amino acid residues may be pooled. L ×64) and an average pooling operation is performed along the amino acid residue direction to obtain a second pooling feature vector of size 1×64.

[0076] It is understandable that the present disclosure does not limit the execution order of step S601 and step S602. The execution order of these two steps can be interchangeable, or the two steps can be executed simultaneously.

[0077] In some embodiments, in step S603, the first pooled feature vector obtained in step S601 and step S602 can be fused with the second pooled feature vector by concatenation, weighted summation, mean pooling, maximum pooling, attention mechanism, direct summation, processing using a multi-layer perceptron or other neural network, or any combination of the above means to obtain a target feature vector of the target protein sequence.

[0078] According to some embodiments, the fusion in step S603 may include merging the first pooled feature vector and the second pooled feature vector using a residual network structure.

[0079] Therefore, through the above method, the first pooled feature vector and the second pooled feature vector based on different protein information sources can be more effectively fused, and the expressive ability of the target feature vector obtained after fusion can be enhanced, thereby improving the performance of the neural network in protein solubility prediction.

[0080] In an exemplary embodiment, in step S603, a residual network can be used to merge the first pooled feature vector of size 1×384 with the second pooled feature vector of size 1×64 to obtain a target feature vector of size 1×448 as a comprehensive representation characterizing the entire target protein sequence.

[0081] In some embodiments, in step S604, the target feature vector can be input into a prediction head comprising a fully connected layer and an output layer, to obtain a prediction result for the solubility of the target protein sequence output by the prediction head. It will be appreciated that the disclosed method can be used for both qualitative predictions to determine whether a target protein sequence is soluble, and quantitative predictions to determine the solubility of a target protein sequence. These different prediction methods can be implemented using different prediction heads (e.g., regression heads, classification heads).

[0082] In an exemplary embodiment, in step S604, the target feature vector obtained in step S603 can be input into two fully connected layers to obtain a feature vector of size 1×128, and the feature vector is sent to the output layer to obtain the final prediction result. Softmax can be used as the classification function of the output layer to predict the protein solubility probability. The protein solubility can be classified using a protein solubility probability value of 0.5 as a threshold. That is, when the protein solubility probability is higher than 0.5, the target protein sequence can be classified as a soluble protein and the output result can be marked as 1. Otherwise, it can be classified as an insoluble protein and the output result can be marked as 0.

[0083] In some embodiments, the neural network and protein pre-training model used in method 100 are both trained. The neural network can be trained using, for example, the neural network training method provided by the present disclosure (e.g., method 800 described below).

[0084] According to some embodiments, a trained neural network is obtained through end-to-end training using sample protein sequences and the actual solubility of the sample protein sequences. The encoded feature vectors (encoded using a pre-trained protein model) for each of the multiple sample amino acid residues corresponding to the sample protein sequence, the substitution score vectors for each of the multiple sample amino acid residues, and the contact probabilities for multiple sample amino acid residue pairs in the sample protein sequence can be input into a first initial graph neural network and a second initial graph neural network in an initial neural network to be trained. Sample prediction results output by the initial prediction subnetworks in the initial neural networks are then obtained. The initial neural networks are then end-to-end trained based on the sample prediction results and the actual solubility to obtain a trained neural network.

[0085] In some specific embodiments, the trained neural network is obtained by end-to-end training using a sample protein sequence and the actual solubility of the sample protein sequence, the training comprising: obtaining a sample protein sequence and the actual solubility of the sample protein sequence, the sample protein sequence comprising a plurality of amino acid residues; encoding the sample protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the plurality of amino acid residues; determining a replacement score vector for each of the plurality of amino acid residues, the replacement score vector comprising a propensity score indicating that the corresponding amino acid residue is replaced with a plurality of preset amino acids at the position of the corresponding amino acid residue; determining contact probabilities for a plurality of amino acid residue pairs in the sample protein sequence; and using the first initial The initial graph neural network processes the encoding feature vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a first comprehensive feature vector for each of the multiple amino acid residues; the second initial graph neural network processes the replacement score vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a second comprehensive feature vector for each of the multiple amino acid residues; the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues are input into the initial prediction subnetwork to obtain a prediction result of the protein solubility of the sample protein sequence; and based on the prediction result and the actual solubility, the parameters of the initial neural network are adjusted to obtain a trained neural network.

[0086] Figure 7 Schematic diagram of a neural network according to an exemplary embodiment of the present disclosure is shown. Figure 7 As shown, neural network 700 may include: a first graph convolutional network 720, a second graph convolutional network 730, and a prediction sub-network 740. The first graph convolutional network 720 is configured to receive encoding feature vectors 702 for each of a plurality of amino acid residues and contact probabilities 704 for a plurality of amino acid residue pairs, and output a first comprehensive feature vector for each of the plurality of amino acid residues. The second graph convolutional network 730 is configured to receive substitution score vectors 708 for each of the plurality of amino acid residues and contact probabilities 704 for a plurality of amino acid residue pairs, and output a second comprehensive feature vector for each of the plurality of amino acid residues. The prediction sub-network 740 is configured to receive the first comprehensive feature vector and the second comprehensive feature vector for each of the plurality of amino acid residues, and output a prediction result 712 of the solubility of the target protein sequence.

[0087] In some embodiments, first graph data 706 can be constructed based on the encoding feature vectors 702 of each of the multiple amino acid residues and the contact probabilities 704 of the multiple amino acid residue pairs. The first graph convolutional network 720 can include three first graph convolutional layers 722, configured to iteratively update node features in the first graph data and output first intermediate feature vectors for each of the multiple amino acid residues. The first graph convolutional network 720 can also include a first fusion module 724, configured to fuse the first intermediate feature vectors for each of the multiple amino acid residues output by the three first graph convolutional layers to obtain first comprehensive feature vectors for each of the multiple amino acid residues.

[0088] In some embodiments, second graph data 710 can be constructed based on the substitution score vectors 708 for each of the multiple amino acid residues and the contact probabilities 704 for the multiple amino acid residue pairs. The second graph convolutional network 730 can include two second graph convolutional layers 732 configured to iteratively update node features in the second graph data and output second intermediate feature vectors for each of the multiple amino acid residues. The second graph convolutional network 730 can also include a second fusion module 734 configured to fuse the second intermediate feature vectors for each of the multiple amino acid residues output by the two second graph convolutional layers to obtain second comprehensive feature vectors for each of the multiple amino acid residues.

[0089] In some embodiments, the prediction subnetwork 740 may include: a first pooling module 742, configured to pool the first comprehensive feature vectors of each of the multiple amino acid residues to obtain a first pooled feature vector of the target protein sequence; a second pooling module 744, configured to pool the second comprehensive feature vectors of each of the multiple amino acid residues to obtain a second pooled feature vector of the target protein sequence; a third fusion module 746, configured to fuse the first pooled feature vector with the second pooled feature vector to obtain a target feature vector of the target protein sequence; and a fully connected layer and output layer 748, configured to process the target feature vector to obtain a prediction result 712.

[0090] In this disclosure, neural network, neural network model, deep learning model, model or other similar terms can be used interchangeably and, without additional explanation, can be used to refer to the trained neural network used in method 100 or the initial neural network trained in method 800 to be introduced below.

[0091] According to another aspect of the present disclosure, a neural network training method is provided. Figure 8A flowchart of a neural network training method according to an exemplary embodiment of the present disclosure is shown. The method 800 comprises: step S801, obtaining a sample protein sequence and the actual solubility of the sample protein sequence, wherein the sample protein sequence comprises a plurality of amino acid residues; step S802, encoding the sample protein sequence using a protein pre-training model to obtain encoding feature vectors of the plurality of amino acid residues; step S803, determining a replacement score vector of the plurality of amino acid residues, wherein the replacement score vector comprises a tendency score indicating that the corresponding amino acid residue is replaced with a plurality of preset amino acids at the position of the corresponding amino acid residue; step S804, determining the contact probability of the plurality of amino acid residue pairs in the sample protein sequence; step S805, using a first initial graph neural network to perform a weighted average of the plurality of amino acid residues. The encoding feature vector and the contact probability of multiple amino acid residue pairs are processed to obtain a first comprehensive feature vector for each of the multiple amino acid residues; step S806, using a second initial graph neural network to process the replacement score vectors of each of the multiple amino acid residues and the contact probability of multiple amino acid residue pairs to obtain a second comprehensive feature vector for each of the multiple amino acid residues; step S807, inputting the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues into the initial prediction subnetwork to obtain a prediction result of the protein solubility of the sample protein sequence; and step S808, adjusting the parameters of the initial neural network based on the prediction result and the actual solubility to obtain a trained neural network.

[0092] Therefore, under the framework of the graph neural network, the encoding feature vector obtained by encoding the protein sequence using the protein pre-training model and the contact probability of multiple amino acid residue pairs of the protein sequence are combined and used, and the replacement score vectors of the multiple amino acid residues of the protein sequence and the contact probability of multiple amino acid residue pairs are combined and used, so that the neural network can utilize richer protein information, thereby improving the diversity and comprehensiveness of the feature representation, and can complete the prediction of protein solubility without obtaining or predicting the spatial structure information of the protein sequence. By setting up two graph neural networks to process the above two combinations respectively, it is possible to fully explore the information related to protein solubility in the encoding feature vectors and replacement score vectors obtained by different methods, and avoid interference between the two, thereby improving the predictive ability of the first comprehensive feature vector and the second comprehensive feature vector, and ultimately enabling the trained neural network to output more accurate solubility prediction results for the protein sequence.

[0093] It should be understood that the operations and effects of steps S801-S807 on the sample protein sequence in method 800 can be referenced above with respect to the description of steps S101-S107 in method 100. The implementation details of steps S101 and S107, as well as the processes for implementing steps S101-S107, are applicable to steps S801-S807 (simply replacing the target protein sequence with the sample protein sequence). In some embodiments, the neural network trained using method 800 can be used in the neural network for predicting protein solubility in method 100.

[0094] In addition, the order of step S802, step S803 and step S804 can be replaced with each other or performed simultaneously. For example, the order of step S805 and step S806 can be replaced with each other or performed simultaneously.

[0095] In some embodiments, in step S801, a sample protein sequence and its true solubility can be obtained from a predetermined training set. In an exemplary embodiment, the PSI: Biology data set provided by the NetSolP tool can be used as a training set. The data set has 11,226 original protein samples. After deleting samples with conflicting labels for the same protein sequence, the data set finally used to train the model contains 11,110 samples. The number of samples of soluble proteins and insoluble proteins in the data set is 7,402 and 3,708 respectively, and the classification labels of soluble proteins and insoluble proteins are 1 and 0 respectively. By training the initial neural network using the sample protein sequences and true solubility (classification labels) obtained from the above training set, the trained neural network can be equipped with the ability to output a classification result indicating whether the protein sequence is soluble.

[0096] In some embodiments, the solubility of sample protein sequences obtained by other methods can be used as the annotation data for the true solubility of the sample protein sequences. By using such sample data to train the initial neural network, the trained neural network can be equipped to output regression results describing the protein solubility.

[0097] In some embodiments, in step S802, the protein pre-training model may use the ProtT5-XL pre-training model, which encodes each of the multiple amino acid residues included in the sample protein sequence. Each amino acid residue is encoded to correspond to a vector of length 1024, i.e., a vector of length L The protein sequence will encode a protein of size L×1024 matrix. The ProtT5-XL model, pre-trained on a large-scale protein sequence dataset, captures protein sequence, structure, and function information, providing rich and accurate feature representations. Using the ProtT5-XL model as a protein pre-training model significantly improves the quality of feature extraction, ultimately increasing the accuracy of protein solubility prediction using trained neural networks.

[0098] In some embodiments, step S803, determining the replacement score vectors of each of the multiple amino acid residues may include: searching a protein database for known protein sequences similar to the sample protein sequence to obtain a multiple sequence alignment result; for each of the multiple amino acid residues, counting the target frequencies of multiple preset amino acids appearing at the position of the amino acid residue based on the multiple sequence alignment result; and determining, based on the target frequencies and the background frequencies of the multiple preset amino acids, a propensity score for the amino acid residue to be replaced with multiple preset amino acids at the position of the amino acid residue.

[0099] In an exemplary embodiment, the PSI-BLAST tool (e.g., version v2.4.0) can be used to search a protein database (e.g., UniRef90) using a sample protein sequence as a query sequence, and a multiple sequence alignment result can be obtained after three iterations. The substitution score vectors of the multiple amino acid residues can form a vector of size L ×20 matrix, where L is the number of amino acid residues in the sample protein sequence, and 20 standard amino acids are used as multiple preset amino acids.

[0100] Contact probability is the likelihood that a pair of amino acid residues in a protein sequence will form a contact. In some embodiments, contact formation can be defined as the distance between atoms (e.g., Cβ atoms) of the two amino acid residues in the protein being less than a predetermined distance threshold (e.g., 8 Å).

[0101] In some embodiments, step S804, determining the contact probabilities of multiple amino acid residue pairs in the sample protein sequence, may include: determining multiple amino acid residue pairs in the sample protein sequence; obtaining, for each of the multiple amino acid residue pairs, a high-dimensional feature vector between a first amino acid residue and a second amino acid residue included in the amino acid residue pair; and simultaneously inputting the high-dimensional feature vectors of the multiple amino acid residue pairs into a trained neural network for predicting contact probabilities to obtain the contact probabilities of the multiple amino acid residue pairs. In this manner, the neural network for predicting contact probabilities can consider information about other amino acids when predicting the contact probability between each amino acid residue pair, thereby improving the accuracy of the predicted contact probabilities.

[0102] In some embodiments, the plurality of amino acid residue pairs may include amino acid residue pairs consisting of an arbitrary first amino acid residue and a second amino acid residue in the sample protein sequence. That is, the contact probability between any two amino acid residues in the sample protein sequence can be determined in step S804. By determining the contact probability between all possible amino acid residue pairs, more comprehensive protein structure information can be provided, information omissions can be reduced, and the needs of complex proteins can be adapted. Furthermore, graph neural networks rely on the relationship between nodes (amino acid residues) and edges (amino acid residue pairs) for feature learning and prediction. By determining the contact probability of a comprehensive set of amino acid residue pairs, more accurate edge topology and edge features can be provided, enabling the graph neural network to more fully utilize this information, thereby improving the performance of the model and ultimately contributing to improved accuracy and reliability in protein solubility prediction using trained neural networks.

[0103] In some embodiments, the high-dimensional feature vector between the first amino acid residue and the second amino acid residue (i.e., the high-dimensional feature vector of the amino acid residue pair) may include any combination of the following features: one-hot encoding of the first amino acid residue and the second amino acid residue, attention scores of the first amino acid residue and the second amino acid residue, and / or structural information of the target protein sequence. In an exemplary embodiment, the above-mentioned attention score may come from an attention map generated by a protein pre-training model based on the target protein sequence; the above-mentioned structural information may include the probability of 3-state secondary structure (SS3), 8-state secondary structure (SS8), solvent accessible surface area (ASA), hemispherical exposure (HSE), protein backbone dihedral angle ψ, , θ, τ and / or other protein structure information.

[0104] In some embodiments, the above features can be combined using concatenation, element-wise operations (eg, element-wise addition, subtraction, multiplication), or other methods to obtain a high-dimensional feature vector for the amino acid residue pair.

[0105] In some embodiments, the neural network used to predict contact probability may be a residual network (eg, ResNet). The neural network may be trained using labeled data describing whether amino acid residue pairs are in contact.

[0106] In some embodiments, in step S805, the encoding feature vectors of each of the multiple amino acid residues obtained in step S802 and the contact probabilities of the multiple amino acid residue pairs obtained in step S804 can be input into the first initial graph neural network to obtain a first comprehensive feature vector of each of the multiple amino acid residues output by the first initial graph neural network.

[0107] According to some embodiments, step S805, using the first initial graph neural network to process the encoding feature vectors of each of the multiple amino acid residues and the contact probabilities of multiple amino acid residue pairs may include (not shown in the figure): step S8051, using the encoding feature vectors of each of the multiple amino acid residues as node features, and using the contact probabilities of multiple amino acid residue pairs as edge features to construct first graph data; and step S8052, using the first initial graph neural network to update the node features in the first graph data based on the node features and edge features in the first graph data to obtain a first comprehensive feature vector for each of the multiple amino acid residues.

[0108] Therefore, through the above method, the sequence information of the protein and the contact information of the amino acid residue pairs can be more effectively utilized within the framework of the graph neural network, thereby obtaining a first comprehensive feature vector with stronger predictive ability, thereby improving the accuracy and robustness of protein solubility prediction using the trained neural network.

[0109] In step S8051, the multiple amino acid residues included in the sample protein sequence can be determined as multiple nodes in the first graph data, and the topological relationship of the first graph data can be determined based on the multiple amino acid residue pairs. In some embodiments, an edge can be set between two nodes corresponding to the first amino acid residue and the second amino acid residue included in each of the multiple amino acid residue pairs determined in step S804, thereby obtaining multiple edges in the first graph data. In some embodiments, the multiple amino acid residue pairs include amino acid residue pairs consisting of any first amino acid residue and second amino acid residue in the sample protein sequence, and an edge can be set between any two nodes in the first graph data, that is, the first graph data is a complete graph.

[0110] In some embodiments, at least one amino acid residue pair with a contact relationship can be determined based on the contact probability of multiple amino acid residue pairs and a preset probability threshold, and the topological relationship of the first graph data can be determined based on the at least one amino acid residue pair. For example, an edge can be set between two nodes corresponding to the first amino acid residue and the second amino acid residue included in the amino acid residue pair with a contact probability greater than the preset probability threshold. In this way, it is possible to avoid establishing edges between nodes corresponding to amino acid residues without a contact relationship, thereby effectively reducing the number of edges, lowering the demand for computing and storage resources, and improving the reasoning efficiency of the first graph neural network.

[0111] According to some embodiments, the first initial graph neural network may include three first graph convolutional layers. By configuring the first graph neural network with three graph convolutional layers to process the encoded feature vectors and contact probabilities of multiple amino acid residues, optimal prediction performance can be achieved while maintaining reasonable computational resource consumption.

[0112] In some embodiments, in step S202, three first graph convolutional layers can be used to iteratively update the node features in the first graph data, and the updated features of the multiple amino acid residues output by the last first graph convolutional layer are obtained as the first comprehensive feature vector.

[0113] In some embodiments, in step S8052, three first graph convolutional layers can be used to iteratively update the node features in the first graph data, and the updated features of the multiple amino acid residues output by the last first graph convolutional layer are obtained as the first comprehensive feature vector.

[0114] In some embodiments, step S8052, using the first initial graph neural network, updating the node features in the first graph data based on the node features and edge features in the first graph data to obtain the first comprehensive feature vectors of each of the multiple amino acid residues may include (not shown in the figure): step S80521, using three first graph convolutional layers to iteratively update the node features in the first graph data, and obtaining the first intermediate feature vectors of each of the multiple amino acid residues respectively output by the three first graph convolutional layers; and step S80522, fusing the first intermediate feature vectors of each of the multiple amino acid residues respectively output by the three first graph convolutional layers to obtain the first comprehensive feature vectors of each of the multiple amino acid residues.

[0115] Thus, through iterative updates of multi-layer graph convolutions, each convolution layer can collect information from a wider neighborhood, thereby aggregating more and more contextual information layer by layer. More specifically, the first layer mainly captures the local features of other amino acid residues that are close to the amino acid residue, the second layer begins to integrate a wider range of neighborhood information, and the third layer captures more global structural information. Ultimately, by fusing multiple layers of features, the model can comprehensively utilize feature representations at all levels to capture the complex relationships between amino acid residues and obtain feature vectors with stronger predictive capabilities.

[0116] In an exemplary embodiment, in step S80521, each first graph convolution layer output size is L ×128 matrix, where L is the number of amino acid residues in the sample protein sequence, and 128 is the hidden dimension of the first intermediate feature vector.

[0117] In some embodiments, in step S80522, the three first intermediate feature vectors output by the three first graph convolutional layers of each amino acid residue can be fused by splicing, weighted summation, mean pooling, maximum pooling, attention mechanism, direct summation, processing using a multi-layer perceptron or other neural network, or any combination of the above means to obtain a first comprehensive feature vector of the amino acid residue.

[0118] According to some embodiments, the fusion in step S80522 may include merging the first intermediate feature vectors of the multiple amino acid residues respectively output by the three first graph convolutional layers using a residual network structure.

[0119] Therefore, through the above method, the first intermediate feature vectors of different levels can be more effectively fused, and the expression ability of the first comprehensive feature vector obtained after fusion can be enhanced, thereby improving the performance of the neural network in protein solubility prediction.

[0120] In an exemplary embodiment, in step S80522, for each amino acid residue, the three first intermediate feature vectors (length 128) of the amino acid residue outputted by the three first graph convolutional layers can be merged using a residual network structure to obtain the first comprehensive feature vector (length 384) of the amino acid residue. In other words, the first comprehensive feature vectors of the amino acid residues can be constructed with a size of L ×384 matrix.

[0121] According to some embodiments, step S806, using a second initial graph neural network to process the replacement score vectors of each of the multiple amino acid residues and the contact probabilities of multiple amino acid residue pairs to obtain second comprehensive feature vectors of each of the multiple amino acid residues may include (not shown in the figure): step S8061, using the replacement score vectors of each of the multiple amino acid residues as node features, and using the contact probabilities of multiple amino acid residue pairs as edge features to construct second graph data; and step S8062, using the second graph neural network to update the node features in the second graph data based on the node features and edge features in the second graph data to obtain second comprehensive feature vectors of each of the multiple amino acid residues.

[0122] Therefore, through the above method, the propensity scores of amino acid residues in proteins to be replaced by other amino acids and the contact information of amino acid residue pairs can be more effectively utilized under the framework of graph neural networks, thereby obtaining a second comprehensive feature vector with stronger predictive ability, thereby improving the accuracy and robustness of subsequent protein solubility prediction using trained neural networks.

[0123] In some embodiments, the second graph data may have a similar or identical topological structure to the first graph data. In step S8061, the plurality of amino acid residues included in the sample protein sequence may be determined as a plurality of nodes in the second graph data, and the topological relationship of the second graph data may be determined based on the plurality of amino acid residue pairs. In some embodiments, an edge may be set between two nodes corresponding to the first amino acid residue and the second amino acid residue included in each of the plurality of amino acid residue pairs determined in step S804, thereby obtaining a plurality of edges in the second graph data. In some embodiments, the plurality of amino acid residue pairs include amino acid residue pairs consisting of any first amino acid residue and second amino acid residue in the sample protein sequence, and an edge may be set between any two nodes in the second graph data, that is, the second graph data is a complete graph.

[0124] In some embodiments, at least one amino acid residue pair with a contact relationship can be determined based on the contact probabilities of multiple amino acid residue pairs and a preset probability threshold, and the topological relationship of the second graph data can be determined based on the at least one amino acid residue pair. For example, an edge can be set between two nodes corresponding to the first and second amino acid residues included in the amino acid residue pair with a contact probability greater than the preset probability threshold. In this way, it is possible to avoid establishing edges between nodes corresponding to amino acid residues that do not have a contact relationship, thereby effectively reducing the number of edges, lowering the demand for computing and storage resources, and improving the reasoning efficiency of the second graph neural network.

[0125] According to some embodiments, the second graph neural network may include two second graph convolutional layers. By configuring the second graph neural network with two graph convolutional layers to process substitution score vectors and contact probabilities for multiple amino acid residues, optimal prediction performance can be achieved while maintaining reasonable computational resource consumption.

[0126] In some embodiments, in step S8062, two second graph convolutional layers can be used to iteratively update the node features in the second graph data, and the updated features of the multiple amino acid residues output by the last second graph convolutional layer are obtained as the second comprehensive feature vector.

[0127] In some embodiments, step S8062, using the second graph neural network to update the node features in the second graph data based on the node features and edge features in the second graph data to obtain the second comprehensive feature vectors of each of the multiple amino acid residues may include (not shown in the figure): step S80621, using two second graph convolutional layers to iteratively update the node features in the second graph data, and obtain the second intermediate feature vectors of each of the multiple amino acid residues respectively output by the two second graph convolutional layers; and step S80622, fusing the second intermediate feature vectors of each of the multiple amino acid residues respectively output by the two second graph convolutional layers to obtain the second comprehensive feature vectors of each of the multiple amino acid residues.

[0128] Thus, through iterative updates of multi-layer graph convolutions, each convolution layer can collect information from a wider neighborhood, thereby aggregating more and more contextual information layer by layer. More specifically, the first layer mainly captures the local features of other amino acid residues that are close to the amino acid residue, while the second layer begins to integrate information from a wider range of neighborhoods. Ultimately, by fusing multiple layers of features, the model can comprehensively utilize feature representations at all levels to capture the complex relationships between amino acid residues and obtain feature vectors with stronger predictive capabilities.

[0129] In an exemplary embodiment, in step S80621, each second graph convolution layer output size is L ×32 matrix, where L is the number of amino acid residues in the sample protein sequence, and 32 is the hidden dimension of the second intermediate feature vector.

[0130] In some embodiments, in step S80622, the two second intermediate feature vectors output by the two second graph convolutional layers of each amino acid residue can be fused by splicing, weighted summation, mean pooling, maximum pooling, attention mechanism, direct summation, processing using a multi-layer perceptron or other neural network, or any combination of the above means to obtain a second comprehensive feature vector of the amino acid residue.

[0131] According to some embodiments, the fusion in step S80622 may include merging the second intermediate feature vectors of the multiple amino acid residues respectively output by the two second graph convolutional layers using a residual network structure.

[0132] Therefore, through the above method, the second intermediate feature vectors at different levels can be more effectively fused, and the expression ability of the second comprehensive feature vector obtained after fusion can be enhanced, thereby improving the performance of the trained neural network in protein solubility prediction.

[0133] In an exemplary embodiment, in step S80622, for each amino acid residue, the two second intermediate feature vectors (length 32) of the amino acid residue outputted by the two second graph convolutional layers can be merged using a residual network structure to obtain a second comprehensive feature vector (length 64) of the amino acid residue. In other words, the second comprehensive feature vectors of the amino acid residues can be constructed to form a vector of size 64. L ×64 matrix.

[0134] According to some embodiments, step S807, inputting the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues into the initial prediction sub-network to obtain a prediction result of the protein solubility of the sample protein sequence may include (not shown in the figure): step S8071, pooling the first comprehensive feature vectors of each of the multiple amino acid residues to obtain a first pooled feature vector of the sample protein sequence; step S8072, pooling the second comprehensive feature vectors of each of the multiple amino acid residues to obtain a second pooled feature vector of the sample protein sequence; step S8073, fusing the first pooled feature vector and the second pooled feature vector to obtain a target feature vector of the sample protein sequence; and step S8074, processing the target feature vector using a fully connected layer and an output layer to obtain a prediction result.

[0135] Therefore, through the above method, the comprehensive feature vector of multiple amino acid residues can be simplified into a feature vector of fixed length that characterizes the sample protein sequence, making subsequent processing more efficient, and the first comprehensive feature vector and the second comprehensive feature vector representing different feature sources (encoding feature vector and replacement score vector) can be effectively fused, ultimately enabling the trained neural network to make full use of this information to obtain more accurate protein solubility prediction results.

[0136] In some embodiments, in step S8071, the first comprehensive feature vectors of the plurality of amino acid residues may be pooled along the direction of the amino acid residues to obtain the first pooled feature vector of the sample protein sequence. In an exemplary embodiment, the matrix (size is L × 128) and an average pooling operation is performed along the direction of the amino acid residues to obtain the first pooling feature vector with a size of 1 × 128.

[0137] In some embodiments, in step S8072, the second comprehensive feature vectors of the plurality of amino acid residues may be pooled along the direction of the amino acid residues to obtain the second pooled feature vector of the sample protein sequence. In an exemplary embodiment, the matrix (size is ) formed by the second comprehensive feature vectors of the plurality of amino acid residues may be pooled. L×64) and an average pooling operation is performed along the amino acid residue direction to obtain a second pooling feature vector of size 1×64.

[0138] It is understandable that the present disclosure does not limit the execution order between step S8071 and step S8072. The execution order of these two steps can be replaced with each other, or the two steps can be executed simultaneously.

[0139] In some embodiments, in step S8073, the first pooled feature vector obtained in step S8071 and step S8072 can be fused with the second pooled feature vector by splicing, weighted summation, mean pooling, maximum pooling, attention mechanism, direct summation, processing using a multi-layer perceptron or other neural network, or any combination of the above means to obtain a target feature vector of the sample protein sequence.

[0140] According to some embodiments, the fusion in step S8073 may include merging the first pooled feature vector and the second pooled feature vector using a residual network structure.

[0141] Therefore, through the above method, the first pooled feature vector and the second pooled feature vector based on different protein information sources can be more effectively fused, thereby enhancing the expressive power of the target feature vector obtained after fusion, thereby improving the performance of the trained neural network in protein solubility prediction.

[0142] In an exemplary embodiment, in step S8073, a residual network can be used to merge the first pooled feature vector of size 1×384 with the second pooled feature vector of size 1×64 to obtain a target feature vector of size 1×448 as a comprehensive representation characterizing the entire sample protein sequence.

[0143] In some embodiments, in step S8074, the target feature vector can be input into a prediction head comprising a fully connected layer and an output layer to obtain a prediction result for the solubility of the sample protein sequence, output by the prediction head. It will be appreciated that the disclosed method can be used for both qualitative predictions of whether a sample protein sequence is soluble and quantitative predictions of the solubility of a sample protein sequence. The initial neural network can be trained by configuring different prediction heads (e.g., regression heads, classification heads) and using different true solubilities (e.g., a classification label indicating whether a sample protein sequence is soluble and a regression label describing the solubility of the sample protein sequence) to enable the trained neural network to implement the aforementioned different prediction methods.

[0144] In an exemplary embodiment, in step S8074, the target feature vector obtained in step S8073 can be input into two fully connected layers to obtain a feature vector of size 1×128, and the feature vector is sent to the output layer to obtain the final prediction result. Softmax can be used as the classification function of the output layer to predict the protein solubility probability. 0.5 can be used as the predicted protein solubility probability threshold to classify the protein. That is, when the protein solubility probability is higher than 0.5, the sample protein sequence can be classified as a soluble protein and the output result can be marked as 1. Otherwise, it can be classified as an insoluble protein and the output result can be marked as 0.

[0145] In some embodiments, in step S807, a loss value may be calculated based on the predicted results, the true solubility, and a predetermined loss function. The learnable parameters of each subnetwork in the initial neural network (the first initial graph convolutional network, the second initial graph convolutional network, and the initial prediction subnetwork) are then adjusted based on the loss value to obtain a trained neural network. In one exemplary embodiment, the loss function may use binary cross-entropy loss. During training, the neural network is adjusted and optimized based on the loss value until the performance of the neural network no longer significantly improves.

[0146] In an exemplary embodiment, the initial neural network can adopt any combination of the following multiple hyperparameters to obtain better training results and prediction performance: using the Adam optimizer; using the rectified linear unit (ReLU) as the activation function, using the L2 regularization term, and the coefficient can be set to 1e-03; the dropout rate can be set to 0.3; the learning rate can be set to 0.01; the method of aggregating neighbor node information in the first initial graph convolutional network and the second initial graph convolutional network can use the average pooling method; the batch size can be set to 512; the training cycle (epoch) can be set to 30, and all data will be trained according to the selected batch size in each training cycle.

[0147] In some embodiments, to test the generalization performance of a trained neural network, the following three independent test sets can be selected for model validation, and samples with a global protein sequence identity greater than 25% with the training set are removed from all test sets. The global sequence similarity alignment tool uses 32-bit USEARCH v11.0.667. The three independent test sets are: the NESG dataset, the S. cerevisiaes dataset, and the eSol dataset. These three independent test sets have the following characteristics: 1. NESG Dataset: The North East Structural Consortium (NESG) expressed 1,323 proteins in E. coli using a standardized production process and scored their solubility. This data was compared with the training set and screened, ultimately retaining 1,319 samples. Of these, 838 were soluble and 481 were insoluble.

[0148] 2. S. cerevisiaes dataset: Contains 109 protein samples expressed in Saccharomyces cerevisiae. Under identical external conditions, solubility was measured using the PURE cell-free expression system to minimize environmental impact. After comparison and filtering with the training set, 103 samples were retained for regression testing, with solubility values ​​being the input to the model. Classification was performed using the same threshold as for the eSol dataset, resulting in 72 samples for classification testing. These samples included 53 soluble proteins and 19 insoluble proteins.

[0149] 3. eSol dataset: The entire Escherichia coli protein was synthesized individually using an in vitro reconstructed translation system, and the aggregation tendency of the protein was analyzed, and 3173 proteins were successfully quantified. Among them, the solubility is the ratio of the protein content in the supernatant to the total protein content in a physical and chemical experiment called PURE, and thresholds of 30% and 70% are used to define the insoluble and soluble proteins. After comparing the eSol dataset with the training set and filtering, 2943 proteins were retained for the application of regression problems. According to the defined threshold, 2067 proteins were obtained for the application of classification problems, of which the number of samples of soluble proteins and insoluble proteins were 928 and 1139, respectively.

[0150] In order to better verify the accuracy of solubility data, the following performance indicators were selected to verify the solubility prediction effect of the neural network: True Positive (TP), False Positive (FP), True Negative (TN), False Negative (FN), Accuracy (ACC), Area Under Curve (AUC) of Receiver Operating Characteristic (ROC), Matthews correlation coefficient (MCC), precision, recall, F1 value and Pearson correlation coefficient (R). The specific definitions are as follows: Accuracy (ACC) is the most basic classification metric, representing the percentage of samples with correct predictions out of the total number of samples. ACC is calculated as ACC = (TP+TN) / (P+N), with a value ranging from 0 to 1. Generally, higher accuracy indicates better model performance.

[0151] The area under the receiver operating characteristic (ROC) curve, also known as AUC, is a popular parameter-independent metric used to describe binary classifiers. The AUC is the area under the receiver operating characteristic (ROC) curve and ranges from 0 to 1, with higher values ​​indicating better classifier performance.

[0152] The Matthews correlation coefficient (MCC) is a comprehensive performance metric for evaluating binary classification models, particularly useful for working with imbalanced datasets. MCC is calculated as MCC = (TP×TN-FP×FN) / (sqrt((TP+FP)×(TP+FN)×(TN+FP)×(TN+FN))) The MCC value ranges from -1 to +1, where +1 indicates a perfect prediction, 0 indicates a random prediction, and -1 indicates a complete discrepancy between the prediction and the actual observation.

[0153] Precision refers to the proportion of correctly predicted results among all predicted positive examples. The formula for calculating precision is: precision = TP / (TP+FP), where TP (False Negative) is the number of positive examples classified as positive, and FP (False Positive) is the number of negative examples classified as positive. Precision ranges from 0 to 1, with higher precision indicating fewer false positives (FP) generated by the model.

[0154] Recall refers to the ratio of correctly predicted positive examples to the total number of positive examples. Recall is calculated as: recall = TP / (TP + FN), where FN (False Negative) represents positive examples classified as negative. Recall values ​​range from 0 to 1, with higher recall indicating fewer instances of the model missing positive examples.

[0155] The F1 score is the harmonic mean of precision and recall, calculated as F1 = 2 × precision × recall / (precision + recall). The F1 score ranges from 0 to 1. A higher F1 score indicates that the model achieves a better balance between precision and recall.

[0156] Pearson's coefficient: In statistics, the Pearson correlation coefficient (also known as the Pearson product-moment correlation coefficient (PPMCC or PCCs)) measures the correlation (linear association) between two variables X and Y, with values ​​ranging from -1 to 1. The Pearson correlation coefficient ranges from -1 to 1. A value of 1 means that X and Y are perfectly described by a straight line equation, with all data points falling perfectly on the line and Y increasing as X increases. A value of -1 means that all data points fall perfectly on the line and Y decreases as X increases. A value of 0 means that there is no linear relationship between the two variables.

[0157] In order to better reflect the prediction accuracy of the method proposed in the present disclosure using a neural network to predict protein solubility, the following will provide comparative experimental results of the method proposed in the present disclosure with other protein solubility prediction methods. It should be noted that although the training method proposed in the present disclosure preferably uses the PSI:Biology dataset as the training set, in order to make a fair comparison, in the following experiment, NetSolP will be used as the five-fold cross-validation of the training set for performance comparison with other methods, and the test set will use the PSI:Biology dataset. In addition, the labels were balanced in the experiment, and each training and test dataset was guaranteed not to share sequences with a global sequence identity of >25%. The global sequence identity was determined using ggsearch36.

[0158] Furthermore, to better validate the accuracy of the neural network-based protein solubility prediction method proposed in this disclosure, two new methods were developed: Method A, based on Method 100 (Method 800) proposed in this disclosure, removes the second graph neural network. Specifically, only the first graph neural network is used to process the encoding feature vectors of multiple amino acid residues and the contact probabilities of multiple amino acid residue pairs, and only the first comprehensive feature vector is input into the prediction subnetwork for protein solubility prediction. Method B, based on Method A, replaces the encoding feature vectors of multiple amino acid residues processed by the first graph neural network with a fusion of the encoding feature vectors of multiple amino acid residues and the replacement score vector, while remaining unchanged.

[0159] The performance of different methods on this test set is shown in Table 1 below: Table 1 Performance of different methods on the PSI:Biology dataset

[0160] As shown in Table 1, the proposed method achieves significantly higher accuracy (ACC), classification ability (AUC), Matthews correlation coefficient (MCC), F1 score, and recall than other methods on the PSI:Biology dataset. Since SWI, SoluProt, NetSolP, and DeepSoluE all use models trained on the PSI:Biology dataset, they are not compared here.

[0161] To further evaluate the performance of the proposed method compared to other methods, we selected three independent test sets: NESG, S. cerevisiae, and eSol. It should be noted that for most models, the threshold for determining protein solubility was the value recommended by the model authors. Since CamSol is not a binary classification model, we used a threshold of 1; all other methods used a threshold of 0.5. The comparison results are as follows: Table 2 Performance of different methods on the independent test set NESG

[0162] The results show that the accuracy (ACC), classification ability (AUC), Matthews correlation coefficient (MCC), F1 score, and recall of the proposed method on the independent test set NESG are significantly higher than those of other methods. DeepSoluE's prediction results for the NESG dataset contain 7 NaN values.

[0163] Table 3 Performance of different methods on the independent test set S. cerevisiae

[0164] Results show that the proposed method achieves significantly higher accuracy (ACC), classification ability (AUC), F1 score, and Pearson correlation coefficient (R coefficient) than other methods on the independent test set S. cerevisiae. It is worth noting that the proteins in the S. cerevisiae dataset were expressed in Saccharomyces cerevisiae, while the other datasets, including the training set, were expressed in E. coli. This demonstrates that the proposed method exhibits excellent generalization performance across expression systems.

[0165] Table 4 Performance of different methods on the independent test set eSol

[0166] The results show that the proposed method achieves significantly higher accuracy (ACC), classification ability (AUC), Matthews correlation coefficient (MCC), F1 score, and Pearson correlation coefficient (R coefficient) than other methods on the independent test set eSol, demonstrating that the proposed method can address both classification and regression problems. The training sets for Protein-Sol, GraphSol, and HybridGCN are all derived from the eSol dataset and are not compared here.

[0167] The above comparison results can further prove that by combining the encoding feature vector obtained by encoding the target protein sequence using the protein pre-training model and the contact probability of multiple amino acid residue pairs of the target protein sequence within the framework of the graph neural network, and combining the replacement score vectors of multiple amino acid residues of the target protein sequence and the contact probability of multiple amino acid residue pairs, the neural network can utilize richer protein information, thereby improving the diversity and comprehensiveness of feature representation, and can complete the prediction of protein solubility without obtaining or predicting the spatial structure information of the target protein sequence. By setting up two graph neural networks to process the above two combinations respectively, it is possible to fully explore the information related to protein solubility in the encoding feature vectors and replacement score vectors obtained by different methods, and avoid interference between the two, thereby improving the predictive ability of the first comprehensive feature vector and the second comprehensive feature vector, and ultimately obtaining a more accurate solubility prediction result for the target protein sequence.

[0168] According to another aspect of the present disclosure, a device for predicting protein solubility using a neural network is provided. Figure 9 FIG. 5 shows a block diagram of a device for predicting protein solubility using a neural network according to an exemplary embodiment of the present disclosure. Figure 9As shown, the device 900 includes: a first acquisition unit 910, configured to acquire a target protein sequence, the target protein sequence including multiple amino acid residues; a first encoding unit 920, configured to encode the target protein sequence using a protein pre-training model to obtain encoding feature vectors of each of the multiple amino acid residues. The first determination unit 930 is configured to determine a replacement score vector for each of the multiple amino acid residues, the replacement score vector including a propensity score indicating that the corresponding amino acid residue is replaced with a plurality of preset amino acids at the position of the corresponding amino acid residue; the second determination unit 940 is configured to determine the contact probability of multiple amino acid residue pairs in the target protein sequence; the first processing unit 950 is configured to use a first graph neural network to process the encoding feature vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a first comprehensive feature vector for each of the multiple amino acid residues; the second processing unit 960 is configured to use a second graph neural network to process the replacement score vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a second comprehensive feature vector for each of the multiple amino acid residues; and the first prediction unit 970 is configured to input the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues into a prediction subnetwork to obtain a prediction result of the solubility of the target protein sequence.

[0169] It can be understood that the operations of units 910 to 970 in the apparatus 900 may refer to the above description of steps S101 to S107 in the method 100 and are not described in detail here.

[0170] According to another aspect of the present disclosure, a neural network training device is provided. Figure 10 FIG. 1 shows a structural block diagram of a neural network training device according to an exemplary embodiment of the present disclosure. Figure 10As shown, the apparatus 1000 includes: a second acquisition unit 1010, configured to obtain a sample protein sequence and a real solubility of the sample protein sequence, wherein the sample protein sequence includes a plurality of amino acid residues; a second encoding unit 1020, configured to encode the sample protein sequence using a protein pre-training model to obtain encoding feature vectors of each of the plurality of amino acid residues; a third determination unit 1030, configured to determine a replacement score vector of each of the plurality of amino acid residues, the replacement score vector including a propensity score indicating that the corresponding amino acid residue is replaced with a plurality of preset amino acids at the position of the corresponding amino acid residue; a fourth determination unit 1040, configured to determine the contact probability of a plurality of amino acid residue pairs in the sample protein sequence; a third processing unit 1050, configured to use a first initial graph neural network to generate a plurality of amino acid residues; The network processes the encoding feature vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a first comprehensive feature vector for each of the multiple amino acid residues; the fourth processing unit 1060 is configured to use the second initial graph neural network to process the replacement score vectors of each of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain a second comprehensive feature vector for each of the multiple amino acid residues; the second prediction unit 1007 is configured to input the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues into the initial prediction sub-network to obtain a prediction result of the protein solubility of the sample protein sequence; and the parameter adjustment unit 1080 is configured to adjust the parameters of the initial neural network based on the prediction result and the actual solubility to obtain a trained neural network.

[0171] It can be understood that the operations of units 1010 to 1080 in the apparatus 1000 may refer to the above description of steps S801 to S808 in the method 800 and are not described in detail here.

[0172] According to one aspect of the present disclosure, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory. The processor is configured to execute the computer program to implement the steps of any one of the method embodiments described above.

[0173] According to one aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any method embodiment described above are implemented.

[0174] According to one aspect of the present disclosure, a computer program product is provided, which includes a computer program. When the computer program is executed by a processor, the steps of any one of the method embodiments described above are implemented.

[0175] In the following, combined Figure 11Illustrative examples of such a computer device, non-transitory computer-readable storage medium, and computer program product are described.

[0176] Figure 11 The figure shows an example configuration of a computer device 1100 that can be used to implement the methods described herein. The apparatus 900 and the apparatus 1000 described above can also be fully or at least partially implemented by the computer device 1100 or a similar device or system.

[0177] Computer device 1100 can be a variety of different types of devices. Examples of computer device 1100 include, but are not limited to, a desktop computer, a server computer, a laptop or netbook computer, a mobile device (e.g., a tablet computer, a cellular or other wireless phone (e.g., a smartphone), a notepad computer, a mobile station), a wearable device (e.g., eyeglasses, a watch), an entertainment device (e.g., an entertainment appliance, a set-top box communicatively coupled to a display device, a game console), a television or other display device, an automobile computer, and the like.

[0178] Computer device 1100 may include at least one processor 1102, memory 1104, communication interface(s) 1106, a display device 1108, other input / output (I / O) devices 1110, and one or more mass storage devices 1112, all capable of communicating with one another, such as via a system bus 1114 or other appropriate connection.

[0179] The processor 1102 may be a single processing unit or multiple processing units, all of which may include a single or multiple computing units or multiple cores. The processor 1102 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuits, and / or any device that manipulates signals based on operational instructions. Among other capabilities, the processor 1102 may be configured to retrieve and execute computer-readable instructions stored in the memory 1104, mass storage device 1112, or other computer-readable media, such as program code for an operating system 1116, program code for application programs 1118, program code for other programs 1120, and the like.

[0180] Memory 1104 and mass storage device 1112 are examples of computer-readable storage media for storing instructions that are executed by processor 1102 to implement the various functions described above. For example, memory 1104 may generally include both volatile memory and non-volatile memory (e.g., RAM, ROM, etc.). Furthermore, mass storage device 1112 may generally include a hard drive, a solid-state drive, removable media, including external and removable drives, memory cards, flash memory, floppy disks, optical disks (e.g., CDs, DVDs), storage arrays, network-attached storage, storage area networks, and the like. Memory 1104 and mass storage device 1112 may be collectively referred to herein as memory or computer-readable storage media and may be non-transitory media capable of storing computer-readable, processor-executable program instructions as computer program code, which may be executed by processor 1102 as a specific machine configured to implement the operations and functions described in the examples herein.

[0181] A number of programs may be stored on mass storage device 1112. These programs include an operating system 1116, one or more application programs 1118, other programs 1120, and program data 1122, and may be loaded into memory 1104 for execution.

[0182] Although Figure 11 1100, but modules 1116, 1118, 1120, and 1122, or portions thereof, may be implemented using any form of computer-readable media accessible by the computer device 1100. As used herein, "computer-readable media" includes at least two types of computer-readable media, namely, computer-readable storage media and communication media.

[0183] Computer-readable storage media includes volatile and nonvolatile, removable and non-removable media implemented by any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer-readable storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disks (DVDs), or other optical storage devices, magnetic cassettes, magnetic tape, magnetic disk storage devices, or other magnetic storage devices, or any other non-transmission media that can be used to store information for access by a computer device. In contrast, communication media can embody computer-readable instructions, data structures, program modules, or other data in a modulated data signal, such as a carrier wave or other transport mechanism. Computer-readable storage media, as defined herein, does not include communication media.

[0184] One or more communication interfaces 1106 are used to exchange data with other devices, such as through a network, a direct connection, etc. Such communication interfaces can be one or more of the following: any type of network interface (e.g., a network interface card (NIC)), a wired or wireless (such as an IEEE 802.11 wireless LAN (WLAN)) interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth TM The communication interface 1106 may include a wireless network interface, a near field communication (NFC) interface, and the like. The communication interface 1106 may facilitate communication within a variety of network and protocol types, including wired networks (e.g., LAN, cable, etc.) and wireless networks (e.g., WLAN, cellular, satellite, etc.), the Internet, and the like. The communication interface 1106 may also provide for communication with external storage devices (not shown) such as storage arrays, network attached storage, storage area networks, and the like.

[0185] In some examples, a display device 1108 such as a monitor may be included for displaying information and images to the user. Other I / O devices 1110 may be devices that receive various inputs from the user and provide various outputs to the user, and may include a touch input device, a gesture input device, a camera, a keyboard, a remote control, a mouse, a printer, an audio input / output device, and the like.

[0186] The techniques described herein can be supported by these various configurations of computer device 1100 and are not limited to the specific examples of the techniques described herein. For example, the functionality can also be implemented in whole or in part in a "cloud" using a distributed system. The cloud includes and / or represents a platform for resources. The platform abstracts the underlying functionality of the cloud's hardware (e.g., servers) and software resources. Resources can include applications and / or data that can be used when performing computing processing on servers remote from computer device 1100. Resources can also include services provided over the Internet and / or through subscriber networks such as cellular or Wi-Fi networks. The platform can abstract resources and functionality to connect computer device 1100 with other computer devices. Therefore, the implementation of the functionality described herein can be distributed throughout the cloud. For example, functionality can be implemented partially on computer device 1100 and partially through the platform that abstracts the functionality of the cloud.

[0187] Although the present disclosure has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description are to be considered illustrative and exemplary and not restrictive; the present disclosure is not limited to the disclosed embodiments. Variations to the disclosed embodiments will be understood and effected by those skilled in the art in practicing the claimed subject matter by studying the drawings, the disclosure and the appended claims. In the claims, the word "comprising" does not exclude other elements or steps that are not listed, the indefinite article "a" or "an" does not exclude a plurality, the term "plurality" means two or more, and the term "based on" should be interpreted as "based at least in part on". The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

Claims

1. A method for predicting protein solubility using a neural network, wherein: The trained neural network includes a first graph neural network, a second graph neural network, and a prediction subnetwork, and the method includes: obtaining a target protein sequence, wherein the target protein sequence comprises a plurality of amino acid residues; Encoding the target protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the multiple amino acid residues; determining a substitution score vector for each of the plurality of amino acid residues, the substitution score vector comprising a propensity score indicating that the corresponding amino acid residue is substituted with a plurality of predetermined amino acids at the position of the corresponding amino acid residue; determining contact probabilities of multiple amino acid residue pairs in the target protein sequence; Processing the encoding feature vectors of each of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs using the first graph neural network to obtain a first comprehensive feature vector for each of the plurality of amino acid residues; Using the second graph neural network, processing the substitution score vectors of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain second comprehensive feature vectors for the multiple amino acid residues; and The first comprehensive feature vector and the second comprehensive feature vector of each of the plurality of amino acid residues are input into the prediction subnetwork to obtain a prediction result of the solubility of the target protein sequence.

2. The method according to claim 1, wherein Processing the encoding feature vectors of each of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs using the first graph neural network to obtain a first comprehensive feature vector for each of the plurality of amino acid residues includes: Using the encoding feature vectors of each of the plurality of amino acid residues as node features, and using the contact probabilities of the plurality of amino acid residue pairs as edge features, to construct first graph data; and Utilizing the first graph neural network, the node features in the first graph data are updated based on the node features and edge features in the first graph data to obtain a first comprehensive feature vector for each of the multiple amino acid residues.

3. The method according to claim 2, wherein The first graph neural network includes three first graph convolutional layers. The first graph neural network is used to update the node features in the first graph data based on the node features and edge features in the first graph data to obtain the first comprehensive feature vectors of the plurality of amino acid residues. Iteratively updating node features in the first graph data using the three first graph convolutional layers, and obtaining first intermediate feature vectors of the plurality of amino acid residues respectively output by the three first graph convolutional layers; and The first intermediate feature vectors of the multiple amino acid residues respectively output by the three first graph convolutional layers are fused to obtain the first comprehensive feature vectors of the multiple amino acid residues respectively.

4. The method according to claim 3, wherein: The fusion includes merging the first intermediate feature vectors of the multiple amino acid residues respectively output by the three first graph convolutional layers using a residual network structure.

5. The method according to claim 1, wherein Using the second graph neural network to process the substitution score vectors of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain the second comprehensive feature vectors of the multiple amino acid residues includes: Using the respective substitution score vectors of the plurality of amino acid residues as node features, and using the contact probabilities of the plurality of amino acid residue pairs as edge features, to construct second graph data; and The second graph neural network is used to update the node features in the second graph data based on the node features and edge features in the second graph data to obtain a second comprehensive feature vector for each of the multiple amino acid residues.

6. The method according to claim 5, wherein The second graph neural network includes two second graph convolutional layers. The second graph neural network is used to update the node features in the second graph data based on the node features and edge features in the second graph data to obtain a second comprehensive feature vector for each of the multiple amino acid residues, including: Iteratively updating node features in the second graph data using the two second graph convolutional layers, and obtaining second intermediate feature vectors of the plurality of amino acid residues respectively output by the two second graph convolutional layers; and The second intermediate feature vectors of the multiple amino acid residues respectively output by the two second graph convolutional layers are fused to obtain the second comprehensive feature vectors of the multiple amino acid residues.

7. The method according to claim 6, wherein: The fusion includes merging the second intermediate feature vectors of the multiple amino acid residues respectively output by the two second graph convolutional layers using a residual network structure.

8. The method according to claim 1, wherein Inputting the first comprehensive feature vector and the second comprehensive feature vector of each of the plurality of amino acid residues into the prediction subnetwork to obtain a prediction result of the solubility of the target protein sequence includes: Pooling the first comprehensive feature vectors of the plurality of amino acid residues to obtain a first pooled feature vector of the target protein sequence; Pooling the second comprehensive feature vectors of the plurality of amino acid residues to obtain a second pooled feature vector of the target protein sequence; fusing the first pooled feature vector and the second pooled feature vector to obtain a target feature vector of the target protein sequence; and The target feature vector is processed using a fully connected layer and an output layer to obtain the prediction result.

9. The method according to claim 8, wherein The fusion includes merging the first pooled feature vector and the second pooled feature vector using a residual network structure.

10. The method according to any one of claims 1 to 9, wherein The protein pre-training model is the ProtT5-XL model.

11. The method according to any one of claims 1 to 9, wherein The first graph neural network and the second graph neural network are based on GraphSAGE.

12. The method according to any one of claims 1 to 9, wherein The plurality of amino acid residue pairs include amino acid residue pairs consisting of a first amino acid residue and a second amino acid residue optionally selected from the target protein sequence.

13. The method according to any one of claims 1 to 9, wherein: The trained neural network is obtained by end-to-end training using sample protein sequences and the actual solubility of the sample protein sequences.

14. A neural network training method, wherein: The initial neural network includes a first initial graph neural network, a second initial graph neural network, and an initial prediction subnetwork. The method includes: obtaining a sample protein sequence and actual solubility of the sample protein sequence, wherein the sample protein sequence comprises a plurality of amino acid residues; Encoding the sample protein sequence using a protein pre-training model to obtain encoding feature vectors of each of the multiple amino acid residues; determining a substitution score vector for each of the plurality of amino acid residues, the substitution score vector comprising a propensity score indicating that the corresponding amino acid residue is substituted with a plurality of predetermined amino acids at the position of the corresponding amino acid residue; determining contact probabilities of multiple amino acid residue pairs in the sample protein sequence; Using the first initial graph neural network, processing the encoding feature vectors of each of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs to obtain a first comprehensive feature vector for each of the plurality of amino acid residues; Using the second initial graph neural network, processing the replacement score vectors of the multiple amino acid residues and the contact probabilities of the multiple amino acid residue pairs to obtain second comprehensive feature vectors for the multiple amino acid residues; Inputting the first comprehensive feature vector and the second comprehensive feature vector of each of the plurality of amino acid residues into the initial prediction subnetwork to obtain a prediction result of the protein solubility of the sample protein sequence; and Based on the prediction results and the true solubility, the parameters of the initial neural network are adjusted to obtain a trained neural network.

15. A device for predicting protein solubility using a neural network, wherein: The trained neural network includes a first graph neural network, a second graph neural network, and a prediction subnetwork, and the apparatus includes: A first acquisition unit is configured to acquire a target protein sequence, where the target protein sequence includes a plurality of amino acid residues; a first encoding unit configured to encode the target protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the plurality of amino acid residues; a first determining unit configured to determine a substitution score vector for each of the plurality of amino acid residues, the substitution score vector comprising a propensity score indicating that the corresponding amino acid residue is substituted with a plurality of preset amino acids at the position of the corresponding amino acid residue; a second determining unit, configured to determine contact probabilities of a plurality of amino acid residue pairs in the target protein sequence; a first processing unit configured to process the encoding feature vectors of each of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs using the first graph neural network to obtain a first comprehensive feature vector for each of the plurality of amino acid residues; a second processing unit configured to process the respective substitution score vectors of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs using the second graph neural network to obtain respective second comprehensive feature vectors of the plurality of amino acid residues; and The first prediction unit is configured to input the first comprehensive feature vector and the second comprehensive feature vector of each of the multiple amino acid residues into the prediction sub-network to obtain a prediction result of the solubility of the target protein sequence.

16. A neural network training device, wherein: The initial neural network includes a first initial graph neural network, a second initial graph neural network, and an initial prediction subnetwork. The apparatus includes: a second acquiring unit configured to acquire a sample protein sequence and a real solubility of the sample protein sequence, wherein the sample protein sequence includes a plurality of amino acid residues; a second encoding unit configured to encode the sample protein sequence using a protein pre-training model to obtain encoding feature vectors for each of the plurality of amino acid residues; a third determining unit configured to determine a substitution score vector for each of the plurality of amino acid residues, the substitution score vector comprising a propensity score indicating that the corresponding amino acid residue is substituted with a plurality of preset amino acids at the position of the corresponding amino acid residue; a fourth determining unit, configured to determine contact probabilities of a plurality of amino acid residue pairs in the sample protein sequence; a third processing unit configured to process the encoding feature vectors of each of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs using the first initial graph neural network to obtain a first comprehensive feature vector for each of the plurality of amino acid residues; a fourth processing unit configured to process the replacement score vectors of the plurality of amino acid residues and the contact probabilities of the plurality of amino acid residue pairs using the second initial graph neural network to obtain second comprehensive feature vectors of the plurality of amino acid residues; a second prediction unit configured to input the first comprehensive feature vector and the second comprehensive feature vector of each of the plurality of amino acid residues into the initial prediction sub-network to obtain a prediction result of the protein solubility of the sample protein sequence; and A parameter adjustment unit is configured to adjust the parameters of the initial neural network based on the prediction result and the actual solubility to obtain a trained neural network.

17. A computer device comprising: at least one processor; as well as a memory having a computer program stored thereon, When the computer program is executed by the processor, the processor is caused to perform the method according to any one of claims 1 to 14.

18. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 14.

19. A computer program product comprising a computer program, which, when executed by a processor, causes the processor to perform the method according to any one of claims 1 to 14.

20. A trained neural network obtained by the neural network training method according to claim 14.

Images