Method and system for predicting protein necessity under different cell lines

Abstract

The invention provides a method and a system for predicting protein necessity under different cell lines. The method comprises the following steps: collecting protein sequence data; exon information is collected, and the exon information is used for assisting in predicting the protein necessity under different cell lines; the method comprises the following steps: setting a DeepExon network structure, wherein the DeepExon network structure comprises a coding module, a CNN, a multi-head self-attention module, a Bi-LSTM and a prediction module; in the coding module, the amino acid position weight and the dipeptide binding probability weight are fused into coding operation; extracting local features from the original sequence and exon information by adopting a CNN (Convolutional Neural Network); the obtained sequence information features and exon features are spliced and then sent to a multi-head attention module; the LSTM captures and stores long-term information from the splicing features by introducing memory cells and a gating mechanism; and predicting protein necessity. According to the method, the technical problem that the prediction accuracy and efficiency of essential proteins are relatively low due to the fact that the specificity of the cell line is not fully considered and information mining is insufficient is solved.

Description

Technical Field

[0001] The present invention relates to the field of gene feature recognition, and in particular to a method and system for predicting protein essentiality in different cell lines. Background Art

[0002] Genes are the key to any organism, and essential genes are essential for cell survival. The expression of the same gene is different in different cell environments. For different cell types, because they perform different biological functions, the essentiality of the same gene will be very different. Essential proteins are the products of essential genes, so the essentiality of proteins will also change with changes in the cell environment. Therefore, if the essentiality of proteins can be predicted more accurately under different cell environments and conditions, it is particularly important for precision medicine and personalized treatment. Since the essentiality of proteins will be affected by environmental factors, the essentiality of the same protein will be different in different environments. The features obtained by predicting through sequences alone will be very limited.

[0003] Existing methods for determining essential genes include: single gene knockout or multiple gene knockout, mutagenesis screening, and RNA interference screening. There are also many ways to identify essential proteins, including transposon mutation, single gene knockout, RNA interference, and CRISPR gene editing technology. However, both the determination of essential genes and the identification of essential proteins require more manpower, material resources, and financial resources. Therefore, it is necessary to develop a more effective, faster, and more accurate computational method to predict essential proteins.

[0004] At present, the computational methods can be roughly divided into two categories. One is based on network centrality measurement. The centrality-lethality rule was first proposed by Jeong et al., and then the centrality betweenness centrality (BC), closeness centrality (CC), eigenvector centrality (EC), local average centrality (LAC) and maximum neighborhood component (MNC) were used to identify essential proteins. The other is based on deep learning methods. With the development of high-throughput sequencing technology, many related studies have emerged. Kuang et al. (2021) developed a machine learning model that combines gradient boosting trees, support vector machines and multilayer perceptrons (MLP) to predict essential genes. Zeng et al. (2021) developed an integrated deep learning model by integrating multiple gradient boosting decision tree (GBDT) base classifiers to achieve accurate prediction. Li et al. (2021) developed an integrated deep learning model EP-EDL, which uses convolutional neural networks (CNN) to predict essential human proteins from evolutionary information. The deep learning model DeepCellEss proposed by Li et al. (2023) can predict the essentiality of proteins based on different cell line environments.

[0005] Although there are many methods to predict the essentiality of proteins, they still have certain limitations. Before the DeepCellEss model proposed by Li (2023), most computational methods did not consider cell line specificity, which made it impossible to accurately identify essential proteins in different cell lines. After the DeepCellEss model was proposed, although it can accurately identify essential proteins in different cell lines, it is difficult to mine more relevant information to assist in identifying essential proteins based on sequence information.

[0006] The existing invention patent application document "A method for identifying biologically essential proteins" with publication number CN114242168A, the existing method includes: supplementing the absent data to improve robustness. Then, the PPI network topology structure, Pearson correlation coefficient, and homology correlation coefficient are constructed to reduce the convergence speed of the deep neural network. Finally, the best correlation relationship between the three characteristics of node degree, Pearson correlation coefficient, and homology correlation coefficient is found through the deep neural network. However, the aforementioned existing scheme does not take into account cell specificity. From the various existing data, it can be seen that the necessity of protein is closely related to the cell environment. The activity level of the same protein gene in different cell lines is different, and its necessity also varies greatly.

[0007] Secondly, the existing protein essentiality prediction technologies in different cell lines all rely solely on sequence information for prediction. Sequence information alone is not sufficient to obtain better prediction results and should be supplemented with other feature information for prediction.

[0008] Finally, traditional embedding codes are composed of word embedding and position embedding. Although traditional embedding codes can better connect contextual relationships during training, they lack the introduction of sequence information.

[0009] In summary, the existing technologies have technical problems such as insufficient consideration of cell line specificity and insufficient information mining, which leads to low prediction accuracy and efficiency for essential proteins. Summary of the invention

[0010] The technical problem to be solved by the present invention is: how to solve the technical problem that the prior art does not fully consider cell line specificity and information mining is not sufficient, resulting in low prediction accuracy and efficiency for essential proteins.

[0011] The present invention adopts the following technical solution to solve the above technical problem: A method for predicting the essentiality of proteins in different cell lines comprises:

[0012] S1. Collect protein sequence data;

[0013] S2. Collect exon information based on protein sequence data, obtain a dataset of labeled exon positions through annotation operations, and use exon information to assist in predicting the essentiality of proteins in different cell lines;

[0014] S3. Setting a DeepExon network structure, wherein the DeepExon network structure includes: an encoding module, a CNN network, a multi-head self-attention module, and a prediction module; wherein, in the encoding operation of the encoding module, amino acid position weights and dipeptide binding probability weights are integrated; using a CNN network, extracting local features from protein data and exon information; performing a splicing operation on sequence information features and exon features in the local features to obtain splicing features; using Bi-LSTM, capturing and storing long-term information from the splicing features;

[0015] S4. Using the prediction module, the protein essentiality prediction results are obtained based on the long-term information prediction.

[0016] The present invention collects the exon positions of proteins and combines features such as embedded coding to predict the essentiality of proteins in different cell lines, and can make full use of auxiliary features to help the model better predict the essentiality of proteins in different environments.

[0017] In a more specific technical solution, in S1, the data set uses CD-HIT and CD-HIT-est to remove redundant sequences at the protein level and nucleotide level to obtain an applicable sequence data set, wherein the sequence data set includes: cell lines, binary essential labels of proteins, and sequence information.

[0018] In a more specific technical scheme, in S2, the exon information corresponding to each protein sequence data is found on NCBI according to the consensus coding provided in the applicable dataset;

[0019] Based on the exon positions in the exon information and combined with the applicable sequence dataset, a dataset of annotated exon positions was generated.

[0020] According to the original data, the present invention collects the corresponding exon information, adopts embedded coding, integrates sequence position information, amino acid position information and other features for prediction. At the same time, the data set used by the present invention can predict the essentiality of proteins for different cell lines. The present invention proposes to use exon position information as an additional feature. Since the functional domain of the protein directly encoded by the exon determines the specific function of the protein in different cells and biological environments, the introduction of exon position information can enable the neural network to better analyze the relationship between the sequence and the exon position. In addition, the present invention has conducted a large number of experiments on DeepExon. Compared with the DeepCellEss model, the present invention has improved the AUC by 2.3% and the AUPRC by 2.1%, which proves that exons play an important role in the prediction of essential proteins in different cell lines. In addition, the present invention also conducted feature ablation experiments to prove that the feature method we proposed has outstanding contributions.

[0021] In a more specific technical solution, in the DeepExon network structure in S3, the encoding modules include: adaptive embedding encoding, one_hot encoding;

[0022] In the encoding operation, the non-exon part is shielded, and the exon part is encoded using one-hot encoding to obtain the amino acid fixed feature matrix for highlighting the exon information;

[0023] For protein sequence data, adaptive embedding coding is used, including learnable embedding layer and position embedding. Position embedding coding is used to obtain the amino acid sequence in protein sequence data, so as to understand the sequence information and perform model enhancement and tuning operations.

[0024] In adaptive embedded coding, the position weights of amino acids and the probability of dipeptide binding are incorporated to understand sequence information. For the position weights of amino acids, for amino acid X, the number of occurrences of amino acid X at each position in the applicable sequence data set is traversed, and then the total number of occurrences of amino acids at the position is compared with the number of occurrences of amino acid X at the position to obtain the position weight of amino acid X at the position;

[0025] CNN is used to extract local features from protein sequence data and exon information. CNN has powerful feature extraction capabilities and performs well in many biological information predictions. However, CNN has obvious limitations because it is only suitable for extracting local information of sequences.

[0026] For the sequence information extraction part, in the CNN network, a dual-channel CNN is combined to use different convolution kernels to extract sequence information from the protein sequence data and perform element-by-element multiplication.

[0027] The present invention proposes an improved embedding coding method, which adopts two weights to solve the problem of missing sequence information in embedding coding and can also accelerate the convergence speed.

[0028] The present invention proposes the DeepExon model, which is a deep learning framework improved from DeepCellEss, for the prediction of cell line-specific essential proteins. The present invention collects 16,408 protein exon location information and combines other features such as sequence information and amino acid location information to assist prediction. After a large number of experimental comparisons, the DeepExon model proposed in the present invention has been significantly improved compared to the DeepCellEss model.

[0029] In a more specific technical solution, in the adaptive embedding coding, for the amino acid at the pth position in the protein sequence data, the position embedding is represented as d k The vector PE(p) of , using the following logic, expresses the i-th element of the vector PE(p):

[0030]

[0031] Where 2i and 2i+1 represent odd and even dimensions respectively.

[0032] In a more specific technical solution, the position weight of amino acid X is determined using the following logic:

[0033]

[0034] Where, L X represents the weight distribution of amino acid X, X i represents the number of occurrences of amino acid X at the i-th position, N represents the sequence length, L i Represents the total number of occurrences of the amino acid at the ith position.

[0035] In a more specific technical solution, the dipeptide binding probability is used to represent the binding probability of two amino acids according to the following logic:

[0036]

[0037] Wherein, (X, Y) represents the probability of binding of amino acid X to amino acid Y, A1 represents the total number of bindings of amino acid X to amino acid Y, and A2 represents the total number of bindings of amino acid X to the remaining amino acids.

[0038] The present invention improves the traditional embedded coding and proposes to use exon information to assist in predicting the essentiality of proteins in different cell lines. For traditional embedded coding, the present invention proposes to integrate amino acid position weights and dipeptide binding probability weights into the coding. The present invention has conducted different ablation experiments to verify the effectiveness of our improved embedded coding. Secondly, we proposed to use exon information as another part of the feature to assist in predicting essentiality. The present invention has conducted different experiments to verify that exons have excellent effects on predicting protein essentiality. In addition, the present invention has also been independently trained on 323 cell line benchmark data sets to verify that the model of the present invention can better predict the essentiality of proteins in different cell lines.

[0039] In a more specific technical solution, the multi-head attention mechanism in the multi-head attention module includes: no less than 2 self-attentions:

[0040]

[0041] Where Q, K, and V represent query, key, and value respectively, and W Q , W K , W V represents the weight matrix, is the dot product scaling factor.

[0042] In a more specific technical solution, in S3, Bi-LSTM includes: forward LSTM and backward LSTM; wherein, the forward LSTM and backward LSTM are expressed using the following logic:

[0043]

[0044] In a more specific technical solution, a system for predicting protein essentiality in different cell lines includes:

[0045] A sequence acquisition module to collect protein sequence data;

[0046] An exon information collection module is used to collect exon information based on protein sequence data, obtain a labeled exon position data set through labeling operations, and use the exon information to assist in predicting the essentiality of proteins in different cell lines. The exon information collection module is connected to the sequence acquisition module;

[0047] The DeepExon network feature processing module is used to set up and use the DeepExon network structure, wherein the DeepExon network structure includes: an encoding module, a CNN network, a multi-head self-attention module, and a prediction module; wherein, in the encoding operation of the encoding module, amino acid position weights and dipeptide binding probability weights are incorporated; a CNN network is used to extract local features from protein data and exon information; sequence information features in local features are spliced ​​with exon features to obtain splicing features; Bi-LSTM is used to capture and store long-term information from splicing features, and the DeepExon network feature processing module is connected with the exon information collection module and the sequence acquisition module;

[0048] The essentiality prediction module is used to utilize the prediction module to obtain the protein essentiality prediction result according to the long-term information prediction. The essentiality prediction module is connected with the DeepExon network feature processing module.

[0049] Compared with the prior art, the present invention has the following advantages:

[0050] The present invention collects the exon positions of proteins and combines features such as embedded coding to predict the essentiality of proteins in different cell lines, and can make full use of auxiliary features to help the model better predict the essentiality of proteins in different environments.

[0051] Based on the original data, the present invention collects the corresponding exon information, adopts embedded coding, and integrates sequence position information, amino acid position information and other features for prediction. In addition, the present invention conducts a large number of experiments on DeepExon. Compared with the DeepCellEss model, the present invention improves AUC by 2.3% and AUPRC by 2.1%, proving that exons play an important role in the prediction of essential proteins in different cell lines. In addition, the present invention also conducts feature ablation experiments to prove that the feature method we proposed has outstanding contributions.

[0052] The present invention proposes the DeepExon model, which is a deep learning framework improved from DeepCellEss, for the prediction of cell line-specific essential proteins. The present invention collects 16,408 protein exon location information and combines other features such as sequence information and amino acid location information to assist prediction. After a large number of experimental comparisons, the DeepExon model proposed in the present invention has been significantly improved compared to the DeepCellEss model.

[0053] The present invention improves the traditional embedded coding and proposes to use exon information to assist in predicting the essentiality of proteins in different cell lines. For traditional embedded coding, the present invention proposes to integrate amino acid position weights and dipeptide binding probability weights into the coding. The present invention has conducted different ablation experiments to verify the effectiveness of our improved embedded coding. Secondly, we proposed to use exon information as another part of the feature to assist in predicting essentiality. The present invention has conducted different experiments to verify that exons have excellent effects on predicting protein essentiality. In addition, the present invention has also been independently trained on 323 cell line benchmark data sets to verify that the model of the present invention can better predict the essentiality of proteins in different cell lines.

[0054] The present invention solves the technical problems existing in the prior art that cell line specificity is not fully considered and information mining is not sufficient, resulting in low prediction accuracy and efficiency for essential proteins. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a schematic diagram of the basic steps of a method for predicting protein essentiality in different cell lines according to Example 1 of the present invention;

[0056] Figure 2 This is a schematic diagram of the DeepExon model structure of Example 1 of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Example 1

[0059] like Figure 1 As shown, the present invention provides a method for predicting the essentiality of proteins in different cell lines, comprising the following basic steps:

[0060] S1. Collect protein sequence data;

[0061] The dataset used in this example is the protein essentiality data of large-scale cell lines provided by Li (2023). These data were identified by a large number of systematic genome-level CRISPR-Cas9 exit screenings, including different binary basic scores for 17,485 human protein-coding genes in 328 different human cell lines, specifically, 1 means important and 0 means unimportant. And the dataset uses CD-HIT and CD-HIT-est to remove redundant sequences at the protein level and nucleotide level. The final dataset contains binary essentiality tags and sequence information for 328 cell lines and 16,408 proteins.

[0062] S2. Collect exon information and use it to help predict the essentiality of proteins in different cell lines;

[0063] In this embodiment, exons, as part of eukaryotic genes, are fragments of genes retained after mRNA shearing, and can be expressed as proteins during protein biosynthesis. Since exons are directly involved in protein coding, exon information plays a key role in predicting and understanding protein essentiality. In this embodiment, according to the consensus coding provided in the original data set, the exon information corresponding to each protein sequence is found on NCBI, and then a new data set is generated based on the exon position and combined with the original data set to mark the exon position.

[0064] S3. Set up a DeepExon network structure, wherein the DeepExon network structure includes: an encoding module, CNN, a multi-head self-attention module, a Bi-LSTM and a prediction module; in the encoding module, the amino acid position weights and the dipeptide binding probability weights are integrated into the encoding operation; a CNN network is used to extract local features from the original sequence and exon information; the acquired sequence information features are spliced ​​with the exon features and sent to the multi-head attention module; LSTM captures and stores long-term information from the splicing features by introducing memory cells and gating mechanisms; wherein the Bi-LSTM includes: a forward LSTM and a backward LSTM; using the prediction module, a protein essentiality prediction result is obtained based on the long-term information prediction.

[0065] like Figure 2 As shown, in this embodiment, the DeepExon network structure combines adaptive embedding learning and one_hot encoding. In the encoding process of this embodiment, in order to highlight the exon information, the non-exon part is shielded, and in order to ensure the uniqueness of each amino acid, the exon part is encoded with one_hot, which is a fixed feature matrix for each amino acid.

[0066] In this embodiment, we use embedded coding for sequence information. Traditional embedded coding consists of a learnable embedding layer and position embedding, where position coding can provide the amino acid sequence of the protein sequence. Position coding can help the model understand the sequence information and enhance the performance and flexibility of the model.

[0067] In this example, for the amino acid at the pth position in the sequence, the position embedding is represented as a d k The i-th element of the vector PE(p) can be expressed as:

[0068]

[0069] Among them, 2i and 2i+1 represent odd dimensions and even dimensions respectively. Although the introduction of position embedding allows the model to capture the order information of amino acids in the sequence, it lacks other information about amino acids. For this reason, this embodiment incorporates the position weights of amino acids and the probability of dipeptide binding in the embedded coding so that the model can better understand the sequence information. For the position weights of amino acids, the principle is that for a certain amino acid X, the number of occurrences of amino acid X at each position in the data set is traversed, and then the ratio of the total number of amino acid occurrences at the position to the number of occurrences of the amino acid at the position is used to obtain the weight of the amino acid at the position. The formula for calculating the position weight of the amino acid is as follows:

[0070]

[0071] Among them, L X represents the weight distribution of amino acid X, X i represents the number of occurrences of amino acid X at the i-th position, N represents the sequence length, L i Represents the total number of occurrences of the amino acid at the ith position.

[0072] In this embodiment, the dipeptide binding probability represents the binding probability of two amino acids, which is expressed as:

[0073]

[0074] Where (X, Y) represents the probability of binding of amino acid X to amino acid Y, A1 represents the total number of binding of amino acid X to amino acid Y, that is, the total number of dipeptide XY, and A2 represents the total number of binding of amino acid X to the remaining amino acids, that is, the total number of binding to amino acid X to form a dipeptide.

[0075] In this embodiment, the DeepExon network structure also includes: CNN, multi-head self-attention, Bi-LSTM and prediction module;

[0076] In the CNN module of this embodiment, the CNN network is used to extract local features from the original sequence and exon information. CNN has a strong feature extraction capability and has a strong performance in many biological information predictions. However, the limitations of CNN are obvious because CNN is only suitable for extracting local information of the sequence;

[0077] In this embodiment, for extracting sequence information, a dual-channel CNN combination is used, and different convolution kernels are used to extract features from sequence information, and then element-by-element multiplication is performed. Using different convolution kernels allows CNN to capture features of different scales, and multi-scale features can better understand the structure and pattern of data; using element-by-element multiplication is to implement a gating mechanism similar to that of gating. When performing element-by-element multiplication, if one of the values ​​is large, the corresponding value of the other will also be amplified. On the contrary, if one is very small, the corresponding value of the other will be suppressed. Such a mechanism can more flexibly adjust the strength of the feature and help the model better handle high-dimensional and complex nonlinear data.

[0078] In the multi-head self-attention module of this embodiment, the acquired sequence information features are spliced ​​with the exon features and then sent to the multi-head attention module. The multi-head attention mechanism consists of multiple self-attentions, and its mathematical formula is as follows:

[0079]

[0080] Among them, Q, K, and V represent query, key, and value respectively, which are obtained by pairing X and weight matrix W. Q , W K , W V The linear transformation is obtained. is the dot product scaling factor.

[0081] In the Bi-LSTM module of this embodiment, LSTM is a special recurrent neural network (RNN) designed specifically for processing long-term dependency problems in sequence data. LSTM effectively captures and stores long-term information by introducing memory cells and gating mechanisms (input gate, forget gate, output gate). Bi-LSTM is composed of two LSTM layers, a forward LSTM and a backward LSTM. The two LSTM layers process different directions of the input sequence respectively. The following is the mathematical expression of the two LSTM layers:

[0082] For the forward LSTM:

[0083]

[0084]

[0085] The backward LSTM is the same as the forward one, but the input sequence is reversed.

[0086] Example 2

[0087] In this embodiment, since the number of samples of essential proteins and non-essential proteins in the data set is extremely unbalanced, the traditional hierarchical segmentation strategy will result in an unbalanced test set. Therefore, 20% of the essential proteins and an equal amount of non-essential proteins were randomly selected as independent test sets, and the remaining samples were used as training sets, and the model was trained using five-fold cross validation. In order to solve the imbalance problem of training data during the training process, the binary entropy loss function of the benchmark paper was used. All training processes used Nvidia GeForce RTX 4090GPU.

[0088] The model indicators in this embodiment continue to use the same evaluation indicators as the benchmark paper: the area under the characteristic curve (AUPROC) and the area under the precision-recall curve (AUPRC). The values ​​of AUROC and AUPRC are both between 0 and 1, and the larger the value, the better the model performance. It is worth noting that AUPRC is more sensitive to positive samples, and for protein essentiality prediction under different cell lines, the number of samples of non-essential and essential amino acids in the data set is inherently unbalanced, so the higher the value of AUPRC, the better the performance of the model when processing unbalanced data sets.

[0089] In this embodiment, the experimental results are discussed, among which the prediction performance on large-scale data sets of different cell lines. The ultimate point of the research purpose in this embodiment is to predict the essentiality of proteins in different cell lines, so we use the same parameter settings to independently train DeepExon on 323 cell line benchmark data sets. Although DeepExon performs differently on 328 cell lines, from the experimental results, the indicators obtained for more than 90% of the data sets are better than the corresponding indicators of the benchmark papers. Secondly, it can be observed that on the EFO-21 cell line, the improvement of the present invention is the largest, and the AUROC and AUPRC are improved by 4.2% and 4.8% respectively compared with the benchmark paper. Looking at the data sets under all cell lines, DeepExon can well predict the essential proteins under different cell lines. In summary, DeepExon can better complete the prediction of protein essentiality, and its performance is better than the DeepCellEss model of the benchmark paper.

[0090] In the comparison process of this embodiment with the benchmark paper, the performance comparison of the model and the benchmark paper model is mainly carried out. First, the benchmark paper DeepCellEss obtained an AUROC of 0.782 and an AUPRC of 0.795 under the HCT-116 cell line, while the AUROC obtained by our DeepExon model under the HCT-116 cell line was 0.805 and AUPRC was 0.816, that is, the AUROC and AUPRC were improved by 2.3% and 2.1% respectively. It can be observed that the performance of DeepExon is better than DeepCellEss. First of all, it is not enough to obtain features by sequence alone. Introducing the position information of exons as features into the model can greatly improve this point; secondly, we note that the benchmark paper uses one-hot encoding, but the vector after one-hot encoding will be very sparse, which will cause great computational costs, and other features such as sequence continuity cannot be perceived by one-hot encoding, and the embedded encoding used by the present invention that combines amino acid position weights and dipeptide binding probability weights can handle the above problems well. Moreover, the embedding layer in the embedded coding can be adjusted and optimized according to the training conditions, which makes the embedded coding essentially superior to the one-hot coding method. The improved embedded coding of the present invention can incorporate more sequence features, which can enhance the understanding of protein sequences and thus improve the prediction performance of the model.

[0091] In this embodiment, an embedding coding ablation experiment is conducted. In order to prove that our improved embedding coding has made a great contribution, we conducted an embedding coding ablation experiment. First, our improved embedding coding combines the amino acid position information weight and the dipeptide fusion probability weight. In order to prove the effectiveness of this method, we will remove the features and conduct experiments one by one. According to the experimental results, we found that the performance indicators after removing the amino acid position information weight and the dipeptide fusion weight are shown in Table 1 below.

[0092] Table 1. Performance indicators for removing amino acid position information weights and dipeptide fusion weights

[0093]

[0094] As can be seen from Table 1 above, after removing the two parts of weights, AUROC and AUPRC are significantly reduced. In this embodiment, the same model as the original paper was experimented, but traditional embedding coding was used instead of one-hot coding, and improved embedding coding was used for the experiment. The experimental results are shown in Table 2 below:

[0095] Table 2 Experimental results of improved embedded coding

[0096]

[0097]

[0098] As can be seen from Table 2 above, traditional embedding coding is better than one-hot coding. Embedding coding can learn and reflect the similarities between categories. Especially in protein research, embedding coding can capture the connection between amino acids and the hidden information between sequences. Secondly, the incorporation of amino acid position weight features can enable the model to better capture the unique effects and functions between sequences, especially amino acids at different positions, and the introduction of dipeptide binding probability weights can help the model better understand the strength of amino acid interactions.

[0099] In this embodiment, for exon features, since exons are directly involved in and synthesize proteins, the introduction of exon features is very effective for protein essentiality prediction. To prove this, we conducted an exon feature ablation experiment, the purpose of which is to confirm the contribution of exons. We conducted the following experiments:

[0100] 1) We fixed the network structure of sequence information and stripped away the exon feature extraction network.

[0101] 2) We masked the exon information in the sequence and only retained the non-exon information.

[0102] Table 3. Results of exon feature ablation experiments

[0103]

[0104] From the experimental results in Table 3 above, the present invention can find that the model performance of the network with exon characteristics is the best. Exons directly encode the functional domain of proteins, and the functional domain determines the specific function of proteins in different cells and biological environments. By analyzing the exon coding region, we can better understand the functional requirements of proteins in different cell types.

[0105] In this embodiment, a comparison is made with the existing methods. Specifically, DeepExon is compared with three other protein essentiality prediction methods, namely EP-GBDT, EP-EDL, Pheg, and DeepCellEss. It can be seen that the DeepExon model adopted by the present invention is superior to other methods in terms of AUROC and AUPRC. The comparative nucleotide sequence features proposed in DeepCellEss, the present invention can find that for this reason we continue to compare DeepCellEss-nc based on nucleotide sequence features, it can be seen that DeepExon is still better than the model based on nucleotide sequence. See Table 4 below for details:

[0106] Table 4. Comparison with existing methods

[0107]

[0108]

[0109] In summary, the present invention collects the exon positions of proteins and combines features such as embedded coding to predict the essentiality of proteins in different cell lines, and can make full use of auxiliary features to help the model better predict the essentiality of proteins in different environments.

[0110] Based on the original data, the present invention collects the corresponding exon information, adopts embedded coding, and integrates sequence position information, amino acid position information and other features for prediction. In addition, the present invention conducts a large number of experiments on DeepExon. Compared with the DeepCellEss model, the present invention improves AUC by 2.3% and AUPRC by 2.1%, proving that exons play an important role in the prediction of essential proteins in different cell lines. In addition, the present invention also conducts feature ablation experiments to prove that the feature method we proposed has outstanding contributions.

[0111] The present invention proposes the DeepExon model, which is a deep learning framework improved from DeepCellEss, for the prediction of cell line-specific essential proteins. The present invention collects 16,408 protein exon location information and combines other features such as sequence information and amino acid location information to assist prediction. After a large number of experimental comparisons, the DeepExon model proposed in the present invention has been significantly improved compared to the DeepCellEss model.

[0112] The present invention improves the traditional embedded coding and proposes to use exon information to assist in predicting the essentiality of proteins in different cell lines. For traditional embedded coding, the present invention proposes to integrate amino acid position weights and dipeptide binding probability weights into the coding. The present invention has conducted different ablation experiments to verify the effectiveness of our improved embedded coding. Secondly, we proposed to use exon information as another part of the feature to assist in predicting essentiality. The present invention has conducted different experiments to verify that exons have excellent effects on predicting protein essentiality. In addition, the present invention has also been independently trained on 323 cell line benchmark data sets to verify that the model of the present invention can better predict the essentiality of proteins in different cell lines.

[0113] The present invention solves the technical problems existing in the prior art that cell line specificity is not fully considered and information mining is not sufficient, resulting in low prediction accuracy and efficiency for essential proteins.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting protein essentiality in different cell lines, characterized in that: The method comprises: S1. Collect protein sequence data; S2. Collecting exon information based on the protein sequence data, obtaining a labeled exon position data set through a labeling operation, and using the exon information to assist in predicting the essentiality of proteins in different cell lines; S3, setting a DeepExon network structure, wherein the DeepExon network structure includes: an encoding module, a CNN network, a multi-head self-attention module and a prediction module; wherein, in the encoding operation of the encoding module, amino acid position weights and dipeptide binding probability weights are integrated; using the CNN network, extracting local features from the protein data and the exon information; performing a splicing operation on the sequence information features and the exon features in the local features to obtain splicing features; using Bi-LSTM, capturing and storing long-term information from the splicing features; S4. Utilizing a prediction module, a protein essentiality prediction result is obtained according to the long-term information prediction.

2. A method for predicting protein essentiality in different cell lines according to claim 1, characterized in that: In S1, the data set uses CD-HIT and CD-HIT-est to remove redundant sequences at the protein level and the nucleotide level to obtain an applicable sequence data set, wherein the sequence data set includes: cell lines, binary essential labels of proteins, and sequence information.

3. A method for predicting protein essentiality in different cell lines according to claim 1, characterized in that: In S2, according to the consensus code provided in the applicable dataset, the exon information corresponding to each of the protein sequence data is found on NCBI; The annotated exon position dataset is generated based on the exon positions in the exon information and combined with the applicable sequence dataset.

4. A method for predicting protein essentiality in different cell lines according to claim 1, characterized in that: In the DeepExon network structure in S3, the encoding module includes: adaptive embedding encoding and one_hot encoding; In the encoding operation, the non-exon part is shielded, and the one-hot encoding is used for the exon part to obtain an amino acid fixed feature matrix for highlighting the exon information; For the protein sequence data, the adaptive embedding coding is used, including: a learnable embedding layer and position embedding, using the position embedding coding to obtain the amino acid sequence in the protein sequence data, thereby understanding the sequence information and performing model enhancement and tuning operations; In the adaptive embedded coding, the position weight of the amino acid and the dipeptide binding probability are incorporated for understanding the sequence information. For the position weight of the amino acid, for the amino acid X, the number of occurrences of the amino acid X at each position in the applicable sequence data set is traversed, and then the total number of occurrences of the amino acid at the position is compared with the number of occurrences of the amino acid X at the position to obtain the position weight of the amino acid X at the position; The local features are extracted from the protein sequence data and the exon information using the CNN network. CNN has a strong feature extraction capability and has a strong performance in many biological information predictions. However, the limitations of CNN are obvious because CNN is only suitable for extracting local information of the sequence; For extracting sequence information, in the CNN network, a dual-channel CNN is combined to use different convolution kernels to extract sequence information from the protein sequence data and perform element-by-element multiplication.

5. A method for predicting protein essentiality in different cell lines according to claim 4, characterized in that: In the adaptive embedding coding, for the amino acid at the pth position in the protein sequence data, the position embedding is represented as d k The vector PE(p) of , using the following logic, expresses the i-th element of the vector PE(p): Where 2i and 2i+1 represent odd and even dimensions respectively.

6. A method for predicting protein essentiality in different cell lines according to claim 4, characterized in that: The position weight of the amino acid X is determined using the following logic: Where, L X represents the weight distribution of amino acid X, X i represents the number of occurrences of amino acid X at the i-th position, N represents the sequence length, L i Represents the total number of occurrences of the amino acid at the ith position.

7. A method for predicting protein essentiality in different cell lines according to claim 4, characterized in that: Using the dipeptide binding probability, the binding probability of the two amino acids is expressed according to the following logic: Wherein, (X, Y) represents the probability of binding of amino acid X to amino acid Y, A1 represents the total number of bindings of amino acid X to amino acid Y, and A2 represents the total number of bindings of amino acid X to the remaining amino acids.

8. A method for predicting protein essentiality in different cell lines according to claim 4, characterized in that: The multi-head attention mechanism in the multi-head attention module includes: no less than 2 self-attentions: Where Q, K, and V represent query, key, and value respectively, and W Q , W K , W V represents the weight matrix, is the dot product scaling factor.

9. The method for predicting protein essentiality in different cell lines according to claim 1, characterized in that: In S3, the Bi-LSTM includes: a forward LSTM and a backward LSTM; wherein the forward LSTM and the backward LSTM are expressed using the following logic:

10. A system for predicting protein essentiality in different cell lines, characterized in that: The system comprises: A sequence acquisition module to collect protein sequence data; An exon information collection module, used to collect exon information according to the protein sequence data, obtain a labeled exon position data set through a labeling operation, and use the exon information to assist in predicting the essentiality of proteins in different cell lines. The exon information collection module is connected to the sequence acquisition module; A DeepExon network feature processing module is used to set up and use the DeepExon network structure, wherein the DeepExon network structure includes: an encoding module, a CNN network, a multi-head self-attention module and a prediction module; wherein, in the encoding operation of the encoding module, amino acid position weights and dipeptide binding probability weights are incorporated; the CNN network is used to extract local features from the protein data and the exon information; the sequence information features in the local features are spliced ​​with the exon features to obtain splicing features; Bi-LSTM is used to capture and store long-term information from the splicing features, and the DeepExon network feature processing module is connected to the exon information collection module and the sequence acquisition module; The essentiality prediction module is used to utilize the prediction module to obtain the protein essentiality prediction result according to the long-term information prediction, and the essentiality prediction module is connected to the DeepExon network feature processing module.

Images