RNA-protein interaction prediction methods, devices, media, and electronic equipment

Abstract

A method, apparatus, medium, and electronic device for predicting RNA-protein interactions are disclosed, relating to the field of artificial intelligence technology. The method includes: acquiring an RNA-protein pair to be predicted (S210); extracting features from the RNA-protein pair to obtain sequence features (S220); vectorizing the RNA-protein pair to obtain RNA sequence representation vectors and protein sequence representation vectors (S230); based on the sequence features, RNA sequence representation vectors, and protein sequence representation vectors of the RNA-protein pair, using multiple interaction prediction models to obtain multiple interaction prediction values ​​for the RNA-protein pair (S240); and determining the interaction between the RNA and protein based on the multiple interaction prediction values ​​(S250).

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and more specifically, to an RNA-protein interaction prediction method, an RNA-protein interaction prediction device, a computer-readable storage medium, and an electronic device. Background Technology

[0002] Noncoding RNAs (ncRNAs) participate in many complex cellular processes, playing crucial roles in life processes such as alternative splicing, chromatin modification, and epigenetics, and are closely linked to many diseases. Studies have shown that most ncRNAs exert their regulatory functions through interactions with proteins. Therefore, researching the interactions between ncRNAs and proteins is of great significance for elucidating the molecular mechanisms of ncRNA action in human diseases and life activities, and has become one of the important approaches to analyzing the functions of ncRNAs and proteins.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] This disclosure provides a method for predicting RNA-protein interactions, an apparatus for predicting RNA-protein interactions, a computer-readable storage medium, and an electronic device.

[0005] This disclosure provides a method for predicting RNA-protein interactions, including:

[0006] Obtain the RNA-protein pairs to be predicted;

[0007] Feature extraction is performed on the RNA-protein pair to be predicted to obtain the sequence features of the RNA-protein pair to be predicted;

[0008] Vectorize the RNA-protein pair to be predicted to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted;

[0009] Based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, multiple interaction prediction models are used to obtain multiple interaction prediction values ​​of the RNA-protein pair to be predicted.

[0010] The interaction between the RNA and the protein is determined based on the multiple interaction prediction values.

[0011] In one exemplary embodiment of this disclosure, the step of extracting features from the RNA-protein pair to be predicted to obtain the sequence features of the RNA-protein pair to be predicted includes:

[0012] Obtain the original sequence feature set;

[0013] The sequence features of the RNA-protein pair to be predicted are determined based on the original sequence feature set.

[0014] In one exemplary embodiment of this disclosure, determining the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set includes:

[0015] The RNA and protein sequences in the RNA-protein pair to be predicted are converted into k-mer subsequences, respectively.

[0016] Each k-mer subsequence is searched in the original sequence feature set, and the sequence features of the RNA-protein pair to be predicted are obtained based on the search results.

[0017] In one exemplary embodiment of this disclosure, determining the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set includes:

[0018] The RNA sequence and protein sequence in the RNA-protein pair to be predicted are converted into k-mer sub-sequences, which include RNA k-mer sub-sequences and protein k-mer sub-sequences.

[0019] By combining RNA k-mer subsequences and protein k-mer subsequences, various RNA-protein k-mer subsequence pairs can be obtained;

[0020] Search for each RNA-protein k-mer subsequence pair in the original sequence feature set, and obtain the sequence features of the RNA-protein pair to be predicted based on the search results.

[0021] In one exemplary embodiment of this disclosure, determining the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set includes:

[0022] The RNA sequence and protein sequence in the RNA-protein pair to be predicted are converted into k-mer sub-sequences, which include RNA k-mer sub-sequences and protein k-mer sub-sequences.

[0023] The first sequence feature is obtained by searching for each k-mer subsequence in the original sequence feature set;

[0024] By combining the RNA k-mer subsequences and protein k-mer subsequences, various RNA-protein k-mer subsequence pairs can be obtained;

[0025] The second sequence feature is obtained by searching for each RNA-protein k-mer subsequence pair in the original sequence feature set;

[0026] The sequence features of the RNA-protein pair to be predicted consist of the first sequence features and the second sequence features.

[0027] In one exemplary embodiment of this disclosure, obtaining the original sequence feature set includes:

[0028] Obtain the original dataset;

[0029] Feature extraction is performed on each RNA-protein pair in the original dataset to obtain the original sequence feature set.

[0030] In one exemplary embodiment of this disclosure, the step of extracting features from each RNA-protein pair in the original dataset to obtain the original sequence feature set includes:

[0031] The basic units of RNA and protein are arranged and combined to obtain k-mer subsequences;

[0032] The frequency of occurrence of each k-mer subsequence in the original dataset is counted, and the variance of each k-mer subsequence is calculated based on the frequency of occurrence.

[0033] The original sequence feature set is determined based on the variance of each k-mer sequence.

[0034] In one exemplary embodiment of this disclosure, the step of counting the frequency of each k-mer subsequence in the original dataset and calculating the variance of each k-mer subsequence based on the frequency of occurrence includes:

[0035] Calculate the frequency of occurrence of each k-mer subsequence in the original dataset;

[0036] The frequency of occurrence of each k-mer subsequence in the original dataset is calculated based on the frequency of occurrence.

[0037] The original dataset is traversed, and each k-mer subsequence is marked as appearing in each RNA-protein pair.

[0038] The variance of each k-mer subsequence was calculated based on its frequency of occurrence in the original dataset and its label value in each RNA-protein pair.

[0039] In one exemplary embodiment of this disclosure, calculating the variance of each k-mer subsequence based on its frequency of occurrence in the original dataset and its label value in each RNA-protein pair includes:

[0040] according to:

[0041]

[0042] Calculate the variance Var of the i-th k-mer subsequence. i .in, Freq represents the label value of the i-th k-mer subsequence in the n-th RNA-protein pair. i Let represent the frequency of the i-th k-mer subsequence in the original dataset, and N be the total number of RNA-protein pairs in the original dataset.

[0043] In one exemplary embodiment of this disclosure, determining the original sequence feature set based on the variance of each k-mer sequence includes:

[0044] Based on the variance of each k-mer subsequence, k-mer subsequences that meet preset conditions are determined, and the original sequence feature set is composed of the k-mer subsequences that meet the preset conditions.

[0045] In one exemplary embodiment of this disclosure, the step of extracting features from each RNA-protein pair in the original dataset to obtain the original sequence feature set further includes:

[0046] The RNA sequence and protein sequence in each RNA-protein pair are converted into k-mer sub-sequences, and the k-mer sub-sequences form a first candidate set, which includes RNA k-mer sub-sequences and protein k-mer sub-sequences.

[0047] The frequency of each k-mer subsequence in the first candidate itemset is counted in the original dataset, and frequent itemsets are formed from the k-mer subsequences that meet the preset frequency threshold.

[0048] The RNA k-mer subsequences and protein k-mer subsequences in the frequent itemset are cross-combined to form a second candidate itemset.

[0049] The frequency of occurrence of each k-mer subsequence pair in the second candidate set in the original dataset is counted to obtain the support of each k-mer subsequence pair.

[0050] The original sequence feature set is composed of k-mer subsequence pairs whose support meets the preset conditions.

[0051] In one exemplary embodiment of this disclosure, the vectorization of the RNA-protein pair to be predicted, to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, includes:

[0052] The RNA sequence and protein sequence in the RNA-protein pair to be predicted are converted into k-mer sub-sequences, which include M RNA k-mer sub-sequences and N protein k-mer sub-sequences.

[0053] Each RNA k-mer subsequence is vectorized to obtain M RNA k-mer vectors;

[0054] The RNA sequence representation vector is obtained by concatenating the M RNA k-mer vectors.

[0055] Each protein k-mer sequence is vectorized to obtain N protein k-mer vectors;

[0056] The protein sequence representation vector is obtained by concatenating the N protein k-mer vectors.

[0057] In one exemplary embodiment of this disclosure, the step of obtaining multiple interaction prediction values ​​for the RNA-protein pair to be predicted using multiple interaction prediction models based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, includes:

[0058] The sequence features of the RNA-protein pair to be predicted are input into at least one first interaction prediction model to obtain at least one first interaction prediction value.

[0059] The RNA sequence representation vector and protein sequence representation vector of the RNA-protein pair to be predicted are input into at least one second interaction prediction model to obtain at least one second interaction prediction value.

[0060] In one exemplary embodiment of this disclosure, the step of obtaining multiple interaction prediction values ​​for the RNA-protein pair to be predicted using multiple interaction prediction models based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, includes:

[0061] The sequence features of the RNA-protein pair to be predicted are input into at least one conventional machine learning model to obtain at least one first interaction prediction value;

[0062] The RNA sequence representation vector and protein sequence representation vector of the RNA-protein pair to be predicted are input into at least one deep learning model to obtain at least one second interaction prediction value.

[0063] In one exemplary embodiment of this disclosure, each deep learning model includes at least two sub-deep learning models; the step of inputting the RNA sequence representation vector and protein sequence representation vector of the RNA-protein pair to be predicted into at least one deep learning model to obtain at least one second interaction prediction value includes:

[0064] The RNA sequence representation vector in the RNA-protein pair to be predicted is input into the first sub-deep learning model to obtain the first sequence feature;

[0065] The protein sequence representation vector in the RNA-protein pair to be predicted is input into the second sub-deep learning model to obtain the second sequence feature;

[0066] The first sequence feature and the second sequence feature are fused together, and the second interaction prediction value is obtained based on the fused features.

[0067] In one exemplary embodiment of this disclosure, the conventional machine learning model includes at least one of a logistic regression model, a support vector machine model, and a decision tree model, and the deep learning model includes at least one of a convolutional neural network model and a recurrent neural network model.

[0068] In one exemplary embodiment of this disclosure, determining the interaction between the RNA and the protein based on the plurality of interaction prediction values ​​includes:

[0069] The multiple interaction prediction values ​​are weighted and summed.

[0070] The interaction between the RNA and the protein was determined based on the calculation results.

[0071] In one exemplary embodiment of this disclosure, determining the interaction between the RNA and the protein based on the calculation results includes:

[0072] If the calculation result is greater than the preset interaction prediction threshold, it is determined that there is an interaction between the RNA and the protein;

[0073] If the calculation result is less than or equal to a preset interaction prediction threshold, it is determined that there is no interaction between the RNA and the protein.

[0074] In one exemplary embodiment of this disclosure, the method further includes:

[0075] The multiple interaction prediction models are jointly trained.

[0076] In one exemplary embodiment of this disclosure, the joint training of the plurality of interaction prediction models includes:

[0077] The joint training of the multiple interaction prediction models includes:

[0078] Obtain a training dataset, which includes positive RNA-protein pairs and negative RNA-protein pairs;

[0079] The multiple interaction prediction models are used to obtain multiple interaction prediction values ​​for each RNA-protein pair in the training dataset;

[0080] The joint prediction value for each RNA-protein pair in the training dataset is obtained based on the multiple interaction prediction values;

[0081] The loss function is used to calculate the joint predicted value and label value of each RNA-protein pair in the training dataset to obtain the corresponding loss value;

[0082] The model parameters of the multiple interaction prediction models are adjusted based on the loss value.

[0083] In one exemplary embodiment of this disclosure, obtaining the joint prediction value for each RNA-protein pair in the training dataset based on the plurality of interaction prediction values ​​includes:

[0084] The weighted summation of multiple interaction predictions for each RNA-protein pair in the training dataset yields the joint prediction value for each RNA-protein pair.

[0085] In one exemplary embodiment of this disclosure, the step of weighted summing of multiple interaction prediction values ​​for each RNA-protein pair in the training dataset to obtain a joint prediction value for each RNA-protein pair includes:

[0086] according to:

[0087] y out =α*y1+β*y2+γ*y3

[0088] The joint prediction value y for each RNA-protein pair in the training dataset was calculated. outWhere y1 is the output value of the traditional machine learning model, y2 is the output value of the convolutional neural network model, y3 is the output value of the recurrent neural network model, and α, β, and γ are the weight parameters of the traditional machine learning model, the convolutional neural network model, and the recurrent neural network model, respectively.

[0089] In one exemplary embodiment of this disclosure, adjusting the model parameters of the plurality of interaction prediction models according to the loss value includes:

[0090] Based on the loss value, the model parameters of the multiple interaction prediction models are iteratively updated. When the iteration termination condition is met, the training of the model parameters of the multiple interaction prediction models is completed, so as to use the trained multiple interaction prediction models to predict the interaction of the RNA-protein pair to be predicted.

[0091] In one exemplary embodiment of this disclosure, the method further includes:

[0092] Output the predicted results of the interaction between the RNA and the protein.

[0093] This disclosure provides an RNA-protein interaction prediction device, comprising:

[0094] The data acquisition module is used to acquire the RNA-protein pairs to be predicted;

[0095] The feature extraction module is used to extract features from the RNA-protein pair to be predicted, and obtain the sequence features of the RNA-protein pair to be predicted.

[0096] The data vectorization module is used to vectorize the RNA-protein pair to be predicted, and obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted;

[0097] The interaction prediction module is used to obtain multiple interaction prediction values ​​for the RNA-protein pair to be predicted based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, using multiple interaction prediction models respectively.

[0098] An interaction determination module is used to determine the interaction between the RNA and the protein based on the plurality of interaction prediction values.

[0099] This disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any one of the above descriptions.

[0100] This disclosure provides an electronic device, including: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method described in any of the preceding methods by executing the executable instructions.

[0101] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0102] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0103] Figure 1 A schematic diagram of an exemplary system architecture for predicting RNA-protein interactions, to which embodiments of the present disclosure can be applied, is shown;

[0104] Figure 2 A flowchart illustrating an RNA-protein interaction prediction method according to an embodiment of the present disclosure is shown schematically.

[0105] Figure 3 A flowchart illustrating the determination of sequence characteristics of an RNA-protein pair to be predicted according to an embodiment of the present disclosure is shown schematically.

[0106] Figure 4 A flowchart illustrating the acquisition of an original sequence feature set according to an embodiment of the present disclosure is shown schematically;

[0107] Figure 5 A flowchart illustrating the acquisition of an original sequence feature set according to another embodiment of the present disclosure is shown schematically;

[0108] Figure 6 A flowchart illustrating the training of multiple interaction prediction models according to one embodiment of the present disclosure is shown schematically;

[0109] Figure 7 A block diagram of an RNA-protein interaction prediction device according to an embodiment of the present disclosure is shown schematically;

[0110] Figure 8 A schematic diagram of the structure of a computer system suitable for implementing the embodiments of the present disclosure is shown. Detailed Implementation

[0111] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0112] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0113] Figure 1 A schematic diagram of a system architecture for an exemplary application environment in which an RNA-protein interaction prediction method and apparatus according to embodiments of the present disclosure can be applied is shown.

[0114] like Figure 1 As shown, system architecture 100 may include one or more of terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables. Terminal devices 101, 102, and 103 may be various electronic devices, including but not limited to desktop computers, laptops, smartphones, and tablets. It should be understood that... Figure 1The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be used. For example, server 105 can be a single server, a server cluster consisting of multiple servers, a cloud computing platform, or a virtualization center. Specifically, server 105 can be used to perform the following: acquiring RNA-protein pairs to be predicted; extracting features from the RNA-protein pairs to obtain their sequence features; vectorizing the RNA-protein pairs to obtain RNA sequence representation vectors and protein sequence representation vectors; based on the sequence features, RNA sequence representation vectors, and protein sequence representation vectors of the RNA-protein pairs, using multiple interaction prediction models to obtain multiple interaction prediction values ​​for the RNA-protein pairs; and determining the interaction between the RNA and protein based on the multiple interaction prediction values.

[0115] The RNA-protein interaction prediction method provided in this embodiment is generally executed by server 105. Correspondingly, the RNA-protein interaction prediction device is generally installed in server 105. The server can send the prediction results of the RNA-protein pair interactions to be predicted to the terminal device, and the terminal device can display them to the user. However, it is readily understood by those skilled in the art that the RNA-protein interaction prediction method provided in this embodiment can also be executed by one or more of terminal devices 101, 102, and 103. Correspondingly, the RNA-protein interaction prediction device can also be installed in terminal devices 101, 102, and 103. For example, after execution by the terminal device, the prediction results can be directly displayed on the terminal device's screen, or the prediction results can be provided to the user through voice broadcast. This exemplary embodiment does not impose any special limitations on this.

[0116] The technical solutions of the embodiments of this disclosure are described in detail below:

[0117] Currently, experimental methods can be used to study noncoding RNA-protein interactions (ncRPI). Traditional experimental methods can obtain valuable data through experiments to construct ncRNA-protein interaction networks, but they are expensive and time-consuming.

[0118] This exemplary embodiment provides a method for predicting RNA-protein interactions. This method can be applied to the server 105 described above, or to one or more of the terminal devices 101, 102, and 103 described above; no special limitation is made in this exemplary embodiment. (See reference...) Figure 2 As shown, the RNA-protein interaction prediction method may include the following steps S210 to S250:

[0119] Step S210. Obtain the RNA-protein pair to be predicted;

[0120] Step S220. Extract features from the RNA-protein pair to be predicted to obtain the sequence features of the RNA-protein pair to be predicted;

[0121] Step S230. Vectorize the RNA-protein pair to be predicted to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted;

[0122] Step S240. Based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, multiple interaction prediction models are used to obtain multiple interaction prediction values ​​of the RNA-protein pair to be predicted.

[0123] Step S250. Determine the interaction between the RNA and the protein based on the plurality of interaction prediction values.

[0124] In the RNA-protein interaction prediction method provided in the exemplary embodiments of this disclosure, an RNA-protein pair to be predicted is obtained; features are extracted from the RNA-protein pair to obtain its sequence features; the RNA-protein pair is vectorized to obtain its RNA sequence representation vector and protein sequence representation vector; based on the sequence features, RNA sequence representation vector, and protein sequence representation vector of the RNA-protein pair, multiple interaction prediction models are used to obtain multiple interaction prediction values ​​for the RNA-protein pair; and the interaction between the RNA and protein is determined based on the multiple interaction prediction values. On the one hand, by performing feature extraction and vectorization on the RNA-protein pair, the connection between RNA and protein sequences can be fully explored, so as to accurately predict the interaction between RNA and protein; on the other hand, effectively combining the characteristics of multiple interaction prediction models can further improve the accuracy of predicting RNA-protein interactions.

[0125] The steps described above in this example implementation will now be explained in more detail.

[0126] In step S210, the RNA-protein pair to be predicted is obtained.

[0127] In this example implementation, at least one RNA-protein pair to be predicted can be obtained. The interaction between the RNA and protein in each RNA-protein pair to be predicted is unknown. For example, a user can input the RNA-protein pair to be predicted via a terminal device. For instance, the user can input the RNA-protein pair to be predicted manually or via voice; this example does not specifically limit this. For instance, one RNA and one protein can be input, and the input order is not limited. For instance, RNA and protein can be input into different text boxes or into the same text box. For instance, after inputting, clicking the "Start Prediction" button will begin executing the prediction steps provided in some embodiments of this application.

[0128] The interaction between RNA and protein refers to the function of a protein manifested through its interactions with other proteins and RNA. For example, the interaction between protein and RNA plays a crucial role in protein synthesis. Simultaneously, the performance of many RNA functions depends on its interaction with proteins. Interactions can be regulatory, guiding, etc., and are not limited here. For instance, in the presence of an interaction, RNA can guide protein synthesis, or RNA can regulate protein function. The interaction between RNA and protein can also refer to their ability to regulate each other's life cycle and function through physical interactions. For example, RNA-coding sequences can guide protein synthesis, and correspondingly, proteins can regulate RNA expression and function.

[0129] After obtaining the RNA-protein pairs to be predicted, multiple interaction prediction models can be used to predict the interactions of each input RNA-protein pair, and the presence or absence of interaction between each pair can be determined based on the prediction results. Simultaneously, the prediction results of the RNA-protein pair interactions can be output to a terminal device for user viewing. For example, the prediction results can be displayed directly on the terminal device's screen, or they can be provided to the user via voice announcement; this example does not specify any particular limitation.

[0130] In other examples, at least one RNA sequence to be predicted can be obtained, and multiple interaction prediction models can be used to search a database for protein sequences that interact with each input RNA sequence. For instance, after a user inputs the RNA sequence to be predicted via a terminal device, they can select at least one protein sequence from the database. Multiple RNA-protein pairs are formed by combining the RNA sequence to be predicted with each protein sequence. Then, multiple interaction prediction models can predict the interaction of each RNA-protein pair, and the protein sequences that can interact with the RNA sequence to be predicted are output based on the prediction results. Preferably, several types of protein sequences can be pre-stored in a database for easy retrieval when predicting RNA-protein pair interactions. For example, protein sequences can be stored in a Redis database or a MySQL database, allowing for real-time querying and selection of the protein sequence to be predicted. Redis is a key-value storage system; when stored in a Redis database, it can include key-value pairs formed by sequence identifiers (such as sequence numbers) and corresponding protein sequences, where the key is the sequence identifier and the value is the corresponding protein sequence. Redis, as a high-efficiency caching technology, supports read / write frequencies exceeding 100KB / s, offering advantages in data reading and storage speed. MySQL, on the other hand, is a relational database management system. Relational databases store data across different tables, rather than storing all data uniformly, increasing storage speed and flexibility. They also offer stability in data storage and can prevent data loss.

[0131] It is understandable that several types of RNA sequences can be pre-stored in a database for easy retrieval when predicting RNA-protein pair interactions. Therefore, at least one protein sequence to be predicted can be obtained, and multiple interaction prediction models can be used to search the database for RNA sequences that interact with each input protein sequence. Similarly, after a user inputs a protein sequence via a terminal device, they can select at least one RNA sequence from the database, forming multiple RNA-protein pairs with the protein sequence to be predicted and various RNA sequences. Multiple interaction prediction models can then predict the interaction of each RNA-protein pair, and based on the prediction results, output the RNA sequences that can interact with the protein sequence to be predicted. This disclosure does not specifically limit this approach.

[0132] In step S220, feature extraction is performed on the RNA-protein pair to be predicted to obtain the sequence features of the RNA-protein pair to be predicted.

[0133] In one example implementation, the method involves obtaining at least one RNA-protein pair to be predicted and predicting their interactions. Before predicting the interactions of each RNA-protein pair using multiple interaction prediction models, it is necessary to obtain the input features for each interaction prediction model. For example, feature extraction can be performed on the RNA-protein pair to be predicted, that is, extracting features from the RNA and protein sequences of the pair sequentially to obtain corresponding RNA sequence features and protein sequence features. These features constitute the sequence features of the RNA-protein pair to be predicted, which can then be used as the input to the interaction prediction model. Alternatively, the RNA-protein pair can be vectorized, that is, the RNA and protein sequences are vectorized separately to obtain corresponding RNA sequence representation vectors and protein representation vectors, which are then used as the input to the interaction prediction model. The RNA-protein pair to be predicted can also be simultaneously subjected to feature extraction and vectorization to obtain the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein representation vector in the RNA-protein pair to be predicted. The sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein representation vector in the RNA-protein pair to be predicted can all be used as inputs to the interaction prediction model. This disclosure does not make any specific limitations on this.

[0134] refer to Figure 3 As shown, features can be extracted from the RNA-protein pairs to be predicted according to steps S310 and S320.

[0135] In step S310, the original sequence feature set is obtained.

[0136] In one example implementation, a raw dataset can be obtained, and features can be extracted from each RNA-protein pair in the raw dataset to obtain a raw sequence feature set. For example, the RPI1807 dataset can be used as the raw dataset, which may contain 3243 RNA-protein pairs, including 1807 positive pairs and 1436 negative pairs. Positive pairs indicate that there is an interaction between the RNA and protein in the RNA-protein pair, while negative pairs indicate that there is no interaction between the RNA and protein in the RNA-protein pair. It is understood that other examples may also use the RPI2241 dataset, RPI369 dataset, etc., as the raw dataset for experiments, and this disclosure does not specifically limit this.

[0137] After obtaining the original dataset, refer to Figure 4 As shown, features can be extracted from RNA-protein pairs in the original dataset according to steps S410 to S430 to obtain the original sequence feature set.

[0138] Step S410. Arrange and combine the basic units of RNA and protein respectively to obtain k-mer subsequences.

[0139] For example, bases are the basic units of RNA. An RNA sequence can include four bases: adenine (A), uracil (U), guanine (G), and cytosine (C). All k-mer subsequences of the RNA sequence can be obtained by arranging and combining these four bases. For example, amino acids are the basic units of proteins. A protein sequence can include 20 amino acids, which are sequentially encoded as A, G, V, I, L, F, P, Y, M, T, S, H, N, Q, W, R, K, D, E, and C. For instance, based on their physicochemical properties, the 20 amino acids can be divided into seven categories: {A, G, V}, {I, L, F, P}, {Y, M, T, S}, {H, N, Q, W}, {R, K}, {D, E}, and {C}. Each category of amino acids can then be re-encoded, for example, sequentially as 1, 2, 3, 4, 5, 6, and 7. For example, the protein sequence ALQDVG can be converted to 124611. Then, the k-mer subsequences of the amino acid sequence can be obtained by arranging and combining the seven classes of amino acids. In other examples, the 20 amino acids can be classified according to their composition, or the k-mer subsequences of the amino acid sequence can be obtained directly by arranging and combining the 20 amino acids without classification. This disclosure does not specifically limit this approach.

[0140] Here, a k-mer subsequence refers to a k-connector consisting of k bases or k classes of amino acids as a group. Correspondingly, in the exemplary embodiments of this disclosure, the k-mer subsequence may include RNA k-mer subsequences and protein k-mer subsequences. For example, a k-mer subsequence may refer to an RNA k-mer subsequence obtained by arranging and combining 4 types of bases, where for a certain value of k, 4... k A k-mer subsequence. A k-mer subsequence can also refer to a protein k-mer subsequence obtained by arranging and combining seven classes of amino acids. For a given k value, seven k-mer subsequences can be obtained. k The k-mer subsequence. It's understandable that dividing the 20 amino acids into 7 categories is merely illustrative and can be omitted. Similarly, the 4 bases in an RNA sequence can be classified according to actual needs.

[0141] In this exemplary embodiment, the value of k can be one or more, and the specific value of k can be adjusted according to the actual situation, without limitation. In one example, k can be described using values ​​of 3 and 4. The 3-mer subsequence of the RNA sequence has a total of 4... 3 =64 types, there are 4 types of 4-mer subsequences 4 = 256 types. There are 7 types of 3-mer subsequences in a protein sequence. 3 = 343 types, with a total of 7 4-mer subsequences 4 = 2401 possibilities. For example, AAA and AUC are two 3-mer subsequences of RNA sequences, and AAAA and AAAU are two 4-mer subsequences of RNA sequences. 111 and 112 are two 3-mer subsequences of protein sequences, and 1111 and 1122 are two 4-mer subsequences of protein sequences. In other examples, k can also be only 3 or only 4, and this disclosure does not specifically limit this.

[0142] Step S420. Calculate the frequency of occurrence of each k-mer subsequence in the original dataset, and calculate the variance of each k-mer subsequence based on the frequency of occurrence.

[0143] In one example implementation, step S410 can yield all 3-mer and 4-mer subsequences of the RNA and protein sequences, including 64 3-mer and 256 4-mer subsequences for the RNA sequence and 343 3-mer and 2401 4-mer subsequences for the protein sequence. The frequency of each 3-mer or 4-mer subsequence in the original dataset can be calculated, and the variance of each 3-mer or 4-mer subsequence can be calculated based on its frequency. Before calculating the frequency of each 3-mer or 4-mer subsequence in the original dataset, the RNA and protein sequences of each RNA-protein pair in the original dataset need to be converted into 3-mer and 4-mer subsequences. For example, for the RNA sequence "AGAUGG", the 3-mer subsequences of this sequence may include "AGA", "GAU", "AUG", and "UGG", and the 4-mer subsequences of this sequence may include "AGAU", "GAUG", and "AUGG". That is, the corresponding 3-mer or 4-mer subsequences can be obtained by reading the RNA sequence in a forward overlap manner. Similarly, the corresponding 3-mer or 4-mer subsequences can also be obtained by reading the RNA sequence in a reverse overlap manner. For example, the 3-mer subsequences of this sequence may also include "GGU", "GUA", "UAG", and "AGA", and the 4-mer subsequences of this sequence may also include "GGUA", "GUAG", and "UAGA". In some embodiments, the corresponding 3-mer or 4-mer subsequences can also be obtained by reading the RNA sequence in a non-overlapping manner. For example, the 3-mer subsequences of this sequence may also include "AGA" and "UGG". This disclosure does not specifically limit this approach.

[0144] For example, the frequency of each 3-mer and / or 4-mer subsequence of RNA and protein sequences can be counted in the original dataset, and the occurrence frequency of each 3-mer and / or 4-mer subsequence in the original dataset can be calculated based on the occurrence frequency. For instance, the occurrence frequency of a certain 3-mer subsequence in the original dataset can be calculated as the ratio between the occurrence frequency of a certain 3-mer subsequence in the original dataset and the total number of RNA-protein pairs in the original dataset. By traversing the original dataset, each 3-mer and / or 4-mer subsequence can be labeled as appearing in each RNA-protein pair. The variance of each k-mer subsequence can be calculated based on the occurrence frequency of each 3-mer and / or 4-mer subsequence in the original dataset and the label value in each RNA-protein pair.

[0145] For example, for the i-th k-mer subsequence, this subsequence can be a 3-mer subsequence of an RNA or protein sequence, or a 4-mer subsequence of an RNA or protein sequence. We can first count the frequency of this subsequence in the RPI1807 dataset. For example, we can iterate through N RNA-protein pairs (N=3243) in the RPI1807 dataset. If the subsequence appears in the current RNA-protein pair, the frequency is incremented by 1; if it does not appear in the current RNA-protein pair, the frequency remains unchanged. Then we determine whether the subsequence appears in the next RNA-protein pair. The frequency of the i-th k-mer subsequence in the RPI1807 dataset is denoted as num. i Therefore, it can be determined based on the frequency of occurrence, num. i The frequency Freq of this subsequence in the RPI1807 dataset was calculated. i .Right now:

[0146]

[0147] After determining the frequency of the i-th k-mer subsequence in the RPI1807 dataset, we can iterate through the RPI1807 dataset to check the occurrence of this subsequence in each RNA-protein pair and mark the occurrences as follows: In other words, if the subsequence appears in the nth RNA-protein pair, the tag value is... If it does not appear in the nth RNA-protein pair, then the tag value is...

[0148] The frequency of this subsequence in the RPI1807 dataset was statistically determined using the value Freq. i and the label value in the nth RNA-protein pair Then, it can be based on:

[0149]

[0150] The variance Var of the i-th k-mer subsequence in the RPI1807 dataset is calculated by summing the squares of the differences between the label value of the i-th k-mer subsequence in each RNA-protein pair and the frequency of occurrence of that k-mer subsequence in the RPI1807 dataset. i .in, Freq represents the label value of the i-th k-mer subsequence in the n-th RNA-protein pair. i Let represent the frequency of the i-th k-mer subsequence in the RPI1807 dataset, and N be the total number of RNA-protein pairs in the RPI1807 dataset.

[0151] Step S430. Determine the original sequence feature set based on the variance of each k-mer sequence.

[0152] After calculating the variance of each k-mer subsequence, k-mer subsequences that meet preset conditions can be determined based on the magnitude of their variances, and an original sequence feature set can be formed from these k-mer subsequences. For example, all 3-mer and 4-mer subsequences of RNA sequences and all 3-mer and 4-mer subsequences of protein sequences can be sorted according to their variances, such as in descending order. The top-ranked k-mer subsequences can then be selected to form the original sequence feature set. For instance, the top 560 k-mer subsequences can be selected, and this set can be used to form the original sequence feature set. This set may include the 3-mer subsequences of the first 60 RNA sequences, the 4-mer subsequences of the first 200 RNA sequences, the 3-mer subsequences of the first 200 protein sequences, and the 4-mer subsequences of the first 100 protein sequences. It is understood that the number of selected k-mer subsequences is merely illustrative; any number of k-mer subsequences can be selected according to actual needs. In other examples, a variance threshold can be preset to filter out k-mer subsequences with variance greater than the threshold, and these filtered k-mer subsequences can be used to form the original sequence feature set. For example, when the preset variance threshold is 3, k-mer subsequences with variance greater than 3 can be selected to form the original sequence feature set. It should be noted that when selecting features, features with larger variances can be preferred. Larger variances indicate greater data variability for that feature, meaning that the feature can better distinguish samples, thereby improving the classification and prediction capabilities of the interaction prediction model.

[0153] In another example implementation, the average number of occurrences of each k-mer subsequence in each RNA-protein pair can be calculated, and the variance of each k-mer subsequence can be calculated based on the average number of occurrences. Then, the original sequence feature set can be determined based on the variance of each k-mer subsequence.

[0154] For example, by traversing the original dataset, the frequency of each 3-mer or 4-mer subsequence in each RNA-protein pair can be determined. Statistical analysis of the frequency of each 3-mer or 4-mer subsequence in each RNA-protein pair yields the total frequency of that subsequence in the original dataset. Based on this total frequency, the average frequency of each 3-mer or 4-mer subsequence in each RNA-protein pair can be calculated. Finally, the variance of each subsequence can be calculated based on the average frequency of each 3-mer or 4-mer subsequence in each RNA-protein pair and its frequency in each RNA-protein pair.

[0155] For example, for the i-th k-mer subsequence, this subsequence can be a 3-mer subsequence of an RNA or protein sequence, or it can be a 4-mer subsequence of an RNA or protein sequence. We can first count the total number of occurrences of this subsequence in the RPI1807 dataset. For example, we can iterate through the n RNA-protein pairs (n = 3243) in the RPI1807 dataset and count the occurrences of this subsequence in each RNA-protein pair as x1, x2, ..., x... n x1,x2,…,x n The total number of occurrences of the subsequence in the RPI1807 dataset is obtained by superimposing the results, denoted as num. i Therefore, it can be determined based on the total number of occurrences, num. i The average number of times the subsequence appeared in each RNA-protein pair was calculated, m. i According to:

[0156]

[0157] The average occurrence count of the i-th k-mer subsequence in each RNA-protein pair is calculated. The variance of this subsequence can be calculated using the average occurrence count and the total occurrence count of the i-th k-mer subsequence in each RNA-protein pair, i.e., based on:

[0158]

[0159] Calculate the variance s of the i-th k-mer subsequence 2 Where n is the number of RNA-protein pairs in the RPI1807 dataset, and m i x is the average number of times this subsequence appears in each RNA-protein pair. nx1 represents the number of times the subsequence appears in the nth RNA-protein pair. Similarly, x2 represents the number of times the subsequence appears in the first RNA-protein pair, and x3 represents the number of times the subsequence appears in the second RNA-protein pair.

[0160] After calculating the variance of each k-mer subsequence, for example, all k-mer subsequences can be sorted according to their variance, such as in descending order, and the top-ranked k-mer subsequences can be selected to form the original sequence feature set. In other examples, a variance threshold can be preset to filter out k-mer subsequences with variances greater than the threshold, and the filtered k-mer subsequences can be used to form the original sequence feature set.

[0161] In the exemplary embodiments of this disclosure, k-mer features of each RNA-protein pair in the original dataset can be extracted, and the original sequence feature set is composed of the extracted k-mer features of the RNA sequence and the protein sequence. Taking the k-mer features of the RNA sequence as an example, the k-mer features can contain information about the monomeric components of the RNA sequence (i.e., the individual bases contained therein) and sequence order information. Therefore, using k-mer features can better characterize an RNA sequence; that is, an RNA sequence can be more accurately identified based on k-mer features, and different RNA sequences can also be distinguished using k-mer features. To further explore the relationship between RNA and protein sequences, frequent itemset features of each RNA-protein pair in the original dataset can also be extracted, and the original sequence feature set is composed of the extracted frequent itemset features. Frequent itemset features can combine the k-mer features of the RNA sequence and the protein sequence. Therefore, using frequent itemset features can better distinguish between RNA-protein pairs that interact and those that do not. It is also possible to extract k-mer features and frequent itemset features simultaneously, and combine them to form the original sequence feature set. By combining the characteristics of k-mer features and frequent itemset features, the interaction between RNA and protein in unknown RNA-protein pairs can be predicted more accurately. This disclosure does not impose specific limitations on this. Frequent itemset features refer to k-mer subsequence pairs formed by combining RNA k-mer subsequences and protein k-mer subsequences with a certain support in the original dataset. Support refers to the proportion of transactions containing both A and B out of all transactions. For example, the subsequence pair (AAU,137) represents a 3-mer subsequence pair composed of an RNA 3-mer subsequence AAU and a protein 3-mer subsequence 137. The support of this subsequence pair is the proportion of RNA-protein pairs containing both subsequences AAU and 137 out of all RNA-protein pairs in the original dataset.

[0162] refer to Figure 5 As shown, frequent itemset features of each RNA-protein pair in the original dataset can be extracted according to steps S510 to S550 to obtain the original sequence feature set.

[0163] Step S510. Convert the RNA sequence and protein sequence in each RNA-protein pair into k-mer sub-sequences respectively, and form a first candidate set of k-mer sub-sequences, wherein the k-mer sub-sequences include RNA k-mer sub-sequences and protein k-mer sub-sequences.

[0164] In one example implementation, the RNA sequence and protein sequence of each RNA-protein pair in the RPI1807 dataset can be first converted into 3-mer and 4-mer subsequences, respectively. By traversing the RPI1807 dataset, all RNA 3-mer, RNA 4-mer, protein 3-mer, and protein 4-mer subsequences in the dataset can be identified, and a first candidate selection set C1 is formed from all the 3-mer and 4-mer subsequences in the dataset.

[0165] Step S520. Calculate the frequency of each k-mer subsequence in the first candidate itemset in the original dataset, and form a frequent itemset by k-mer subsequences that meet the preset frequency threshold.

[0166] We can count the frequency of each k-mer subsequence in the first candidate set C1 in the original dataset. For example, for the j-th k-mer subsequence, this subsequence can be a 3-mer subsequence of an RNA or protein sequence, or a 4-mer subsequence of an RNA or protein sequence. We can first count the frequency of this subsequence in the RPI1807 dataset. For example, we can iterate through N RNA-protein pairs in the RPI1807 dataset; if the subsequence appears in the current RNA-protein pair, the frequency is incremented by 1; if it does not appear in the current RNA-protein pair, the frequency remains unchanged. The counted frequency of the j-th k-mer subsequence in the RPI1807 dataset is denoted as num. j Therefore, it can be determined based on the frequency of occurrence, num. j The frequency Freq of this subsequence in the RPI1807 dataset was calculated. j .Right now:

[0167]

[0168] Similarly, the frequency of each 3-mer or 4-mer subsequence in the first candidate itemset C1 in the RPI1807 dataset can be calculated. Then, all 3-mer and 4-mer subsequences can be filtered according to a preset frequency threshold. For example, RNA 3-mer subsequences with a frequency greater than the first threshold, RNA 4-mer subsequences with a frequency greater than the second threshold, protein 3-mer subsequences with a frequency greater than the third threshold, and protein 4-mer subsequences with a frequency greater than the fourth threshold can be selected to form a frequent itemset L1. The first, second, third, and fourth thresholds can be the same or different; this disclosure does not impose specific limitations on this. In other examples, the frequencies of 3-mer and 4-mer subsequences of RNA sequences and protein sequences can be sorted in descending order, and the frequent itemset L1 can be formed from the highest-ranking subsequences; this disclosure does not impose specific limitations on this either.

[0169] Step S530. Cross-combine the RNA k-mer subsequences and protein k-mer subsequences in the frequent itemset to form a second candidate itemset.

[0170] The RNA 3-mer and protein 3-mer sequences, as well as the RNA 4-mer and protein 4-mer sequences in the frequent itemset L1, can be cross-combined pairwise to obtain multiple 3-mer and 4-mer sequence pairs. These multiple sequence pairs are then used to form the second candidate itemset C2. For example, if the frequent itemset includes [AAU, AUC, 137, 123, AAUU, AGUC, 1737, 1234], "AAU" and "137" can be combined to obtain the 3-mer sequence pair "AAU_137", and "AAU" and "123" can be combined to obtain the 3-mer sequence pair "AAU_123". Similarly, by cross-combining RNA 3-mer subsequences and protein 3-mer subsequences, we can obtain 3-mer subsequence pairs “AUC_137” and “AUC_123”, and by cross-combining RNA 4-mer subsequences and protein 4-mer subsequences, we can obtain 4-mer subsequence pairs “AAUU_1737”, “AAUU_1234”, “AGUC_1737”, and “AGUC_1234”.

[0171] Step S540. Calculate the frequency of occurrence of each k-mer subsequence pair in the original dataset from the second candidate set to obtain the support of each k-mer subsequence pair.

[0172] The frequency of each subsequence pair in the second candidate set C2 can be counted in the RPI1807 dataset. For example, for the f-th k-mer subsequence pair, this subsequence pair can be either a 3-mer or a 4-mer subsequence pair. The frequency of this subsequence pair in the RPI1807 dataset can be counted first. For example, iteratively, N RNA-protein pairs in the RPI1807 dataset can be used; if the subsequence pair appears in the current RNA-protein pair, its frequency is incremented by 1; otherwise, the frequency remains unchanged. The counted frequency of the f-th k-mer subsequence pair in the RPI1807 dataset is denoted as num. f Therefore, it can be determined based on the frequency of occurrence, num. f The frequency of occurrence of the subsequence pair in the RPI1807 dataset is calculated, which gives the support of the subsequence pair. f .Right now:

[0173]

[0174] Similarly, the support of each subsequence pair in the second candidate set C2 can be calculated.

[0175] Step S550. The original sequence feature set is composed of k-mer subsequence pairs whose support satisfies the preset conditions.

[0176] After calculating the support of each type of sub-sequence pair, sub-sequence pairs that meet preset conditions can be determined based on their support levels, and these sub-sequence pairs can be used to form the original sequence feature set. For example, a preset support threshold can be used to filter out sub-sequence pairs with support greater than the threshold, and these filtered sub-sequence pairs can be used to form the original sequence feature set. For instance, if there are 370 sub-sequence pairs with support greater than the threshold, these 370 sub-sequence pairs are 370 frequent itemset features, and the original sequence feature set can be formed from these 370 frequent itemset features. In other examples, all sub-sequence pairs in the second candidate itemset C2 can be sorted in descending order based on their support levels, and the top-ranked sub-sequence pairs can be selected to form the original sequence feature set; this disclosure does not specifically limit this approach.

[0177] In another example implementation, the RNA sequence and protein sequence in each RNA-protein pair can be converted into k-mer sub-sequences to obtain k-mer sub-sequence pairs. The frequency of each k-mer sub-sequence pair in the original dataset is counted, and the k-mer sub-sequence pairs that meet the preset frequency conditions are used as frequent itemset features and form the original sequence feature set.

[0178] For example, the RNA and protein sequences of all positive RNA-protein pairs in the RPI1807 dataset can be converted into positive 3-mer and 4-mer subsequences, respectively. Similarly, the RNA and protein sequences of all negative RNA-protein pairs in the dataset can be converted into negative 3-mer and 4-mer subsequences, respectively. By traversing the RPI1807 dataset, all positive and negative RNA 3-mer, RNA 4-mer, protein 3-mer, and protein 4-mer subsequences in the dataset can be identified. The RNA 3-mer and protein 3-mer sequences, and RNA 4-mer and protein 4-mer sequences in the dataset can be cross-combined pairwise to obtain various 3-mer and 4-mer subsequence pairs. For example, positive RNA 3-mer sequences and positive protein 3-mer sequences can be cross-combined to obtain positive 3-mer subsequence pairs. Negative RNA 3-mer sequences and negative protein 3-mer sequences can be cross-combined to obtain negative 3-mer sequence pairs. Similarly, positive RNA 4-mer sequences and positive protein 4-mer sequences can be cross-combined to obtain positive 4-mer sequence pairs. Finally, negative RNA 4-mer sequences and negative protein 4-mer sequences can be cross-combined to obtain negative 4-mer sequence pairs.

[0179] The frequency of each seed sequence pair in the dataset can be counted. For example, for any positive 3-mer subsequence pair, it can be calculated based on:

[0180]

[0181] The frequency Freq of the positive example 3-mer subsequence pair in the dataset is calculated. Here, num is the number of times the positive example 3-mer subsequence pair appears in the dataset, and NUM is the total number of times all positive example 3-mer subsequence pairs appear in the dataset.

[0182] After calculating the frequency of each k-mer subsequence pair in the original dataset, for example, all 3-mer and 4-mer subsequence pairs can be sorted according to their frequency, such as in descending order. The top-ranked k-mer subsequence pairs can then be selected to form frequent itemsets. For instance, sorting all positive 3-mer subsequence pairs in descending order yields the first m pairs to form frequent itemset A1. Sort all positive 4-mer subsequence pairs in descending order yields the first n pairs to form frequent itemset A2. Sort all negative 3-mer subsequence pairs in descending order yields the first p pairs to form frequent itemset A3. Sort all negative 4-mer subsequence pairs in descending order yields the first q pairs to form frequent itemset A4. These four frequent itemsets A1, A2, A3, and A4 then form the original sequence feature set. In other examples, a frequency threshold can be preset to filter out k-mer subsequence pairs that have a frequency greater than the threshold, and the filtered k-mer subsequence pairs can be used as frequent itemset features to form the original sequence feature set. This disclosure does not make specific limitations on this.

[0183] In the exemplary embodiments of this disclosure, frequent itemset features can combine the kmer features of RNA sequences and the kmer features of protein sequences. Using frequent itemset features can better distinguish between RNA-protein pairs that interact and those that do not. Therefore, when an original sequence feature set is composed of frequent itemset features, and features are extracted from the RNA-protein pair to be predicted based on this original sequence feature set, the presence or absence of interaction between the RNA-protein pair to be predicted can be determined more accurately based on the extracted sequence features of the RNA-protein pair to be predicted.

[0184] Step S320. Determine the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set.

[0185] In one example implementation, the RNA and protein sequences in the RNA-protein pair to be predicted can be converted into k-mer subsequences to obtain the original sequence feature set. Then, each k-mer subsequence can be searched within the original sequence feature set, and the sequence features of the RNA-protein pair to be predicted can be obtained based on the search results. The sequence features of the RNA-protein pair to be predicted can refer to a complete sequence feature composed of both RNA and protein sequence features.

[0186] For example, the original sequence feature set can consist of 560 k-mer subsequences, such as [CCC, ..., AGU, CCCC, ..., CUGG, 777, ..., 373, 7774, ..., 7571]. The RNA-protein pair to be predicted is converted into 3-mer and 4-mer subsequences, resulting in RNA 3-mer, RNA 4-mer, protein 3-mer, and protein 4-mer subsequences. Based on the original sequence feature set, feature calculations can be performed on the RNA and protein sequences in the RNA-protein pair to be predicted to obtain the sequence features of that pair.

[0187] Specifically, for the original sequence feature set [CCC, ..., AGU, CCCC, ..., CUGG, 777, ..., 373, 7774, ..., 7571] comprising 560 feature dimensions, each feature dimension corresponds to a k-mer subsequence. For example, subsequence CCC is a feature of the first feature dimension, and subsequence 7571 is a feature of the 560th feature dimension. All 3-mer and 4-mer subsequences of the RNA-protein pair to be predicted can be searched in the original sequence feature set, and the presence of a feature in each feature dimension can be determined based on the search results. If it exists, the feature value in that feature dimension is 1; otherwise, it is 0. For example, if the RNA sequence in the RNA-protein pair to be predicted is AAAACCCGGG, it can be seen that feature CCC in the first feature dimension of the original sequence feature set is also a 3-mer subsequence of that RNA-protein pair. Therefore, it can be determined that feature CCC in the first feature dimension of the original sequence feature set exists, and the corresponding feature value can be recorded as 1. For example, if the feature AGU in the original sequence feature set does not exist in the RNA sequence, then the feature value in the feature dimension corresponding to the feature AGU in the original sequence feature set can be recorded as 0. Finally, a 560-dimensional feature value vector [1, 0, ..., 0, 1, ...] can be calculated, which is the sequence feature of the RNA-protein pair to be predicted. It can be understood that each feature value in this feature value vector corresponds one-to-one with the feature value of each feature dimension in the original sequence feature set.

[0188] In this example implementation, k-mer features of each RNA-protein pair in the original dataset are extracted, forming an original sequence feature set composed of the extracted RNA sequence k-mer features and the protein sequence k-mer features. Based on this original sequence feature set, feature extraction is performed on the RNA-protein pair to be predicted to obtain the sequence features of that pair. Taking the RNA sequence k-mer features as an example, the k-mer features can contain information about the monomeric components of the RNA sequence (i.e., the individual bases contained) and sequence order information. Therefore, using k-mer features can better characterize an RNA sequence; that is, an RNA sequence can be more accurately determined based on k-mer features, and different RNA sequences can also be distinguished using k-mer features.

[0189] In another example implementation, after obtaining the original sequence feature set, the RNA and protein sequences in the RNA-protein pair to be predicted can be converted into k-mer subsequences, and the RNA k-mer subsequences and protein k-mer subsequences can be cross-combined to obtain multiple RNA-protein k-mer subsequence pairs. For example, after obtaining the RNA 3-mer, protein 3-mer, RNA 4-mer, and protein 4-mer subsequences of the RNA-protein pair to be predicted, the RNA 3-mer and protein 3-mer subsequences, and the RNA 4-mer and protein 4-mer subsequences can be cross-combined pairwise to obtain multiple 3-mer and 4-mer subsequence pairs. Each RNA-protein k-mer subsequence pair can be searched in the original sequence feature set, and the sequence characteristics of the RNA-protein pair can be obtained based on the search results.

[0190] For example, the original sequence feature set can consist of 370 frequent itemset features, such as [CUG_122, AAU_122, ..., CUUU_1312, UCUG_1312, ...]. The RNA-protein pair to be predicted is converted into 3-mer and 4-mer subsequences, resulting in RNA 3-mer, RNA 4-mer, protein 3-mer, and protein 4-mer subsequences. These subsequences can be paired with protein 3-mer and 4-mer sequences to obtain various 3-mer and 4-mer subsequence pairs. Then, based on the original sequence feature set, feature calculations can be performed on the RNA and protein sequences in the RNA-protein pair to be predicted, yielding the sequence features of that RNA-protein pair.

[0191] Specifically, for the original sequence feature set [CUG_122, AAU_122, ..., CUUU_1312, UCUG_1312, ...], which includes 370 feature dimensions, each feature dimension corresponds to a k-mer subsequence pair. For example, the subsequence pair CUG_122 is a feature of the first feature dimension. All subsequence pairs of the RNA-protein pair to be predicted can be searched in the original sequence feature set, and the presence of a feature in each feature dimension of the original sequence feature set can be determined based on the search results. If it exists, the feature value in that feature dimension is 1; otherwise, it is 0. For example, if the RNA sequence of the RNA-protein pair to be predicted is AUCUGAAAU and the protein sequence is 512261312, it can be seen that CUG_122, AAU_122, and UCUG_1312 in the subsequence pairs of this RNA-protein pair exist in the original sequence feature set. Therefore, the feature value in the corresponding feature dimension of the original sequence feature set can be recorded as 1. Finally, a 370-dimensional feature vector [1, 1, ..., 0, 1, ...] can be calculated, which represents the sequence features of the RNA-protein pair to be predicted. Similarly, each feature value in this feature vector corresponds one-to-one with the feature value of each feature dimension in the original sequence feature set.

[0192] In another example implementation, after obtaining the original sequence feature set, the RNA and protein sequences in the RNA-protein pair to be predicted can be converted into k-mer subsequences, respectively. Each k-mer subsequence is then searched in the original sequence feature set to obtain the first sequence feature. Next, the RNA and protein k-mer subsequences can be combined to obtain multiple RNA-protein k-mer subsequence pairs. Each RNA-protein k-mer subsequence pair is then searched in the original sequence feature set to obtain the second sequence feature. Finally, the first and second sequence features can be combined to form the sequence feature of the RNA-protein pair to be predicted.

[0193] For example, the original sequence feature set may include two feature subsets, each containing 560 k-mer subsequences [CCC, ..., CCCC, ..., 777, ..., 7774, ...] and 370 frequent itemset features [CUG_122, AAU_122, ..., CUUU_1312, UCUG_1312, ...]. The RNA-protein pair to be predicted can be converted into RNA 3-mer subsequences, RNA 4-mer subsequences, protein 3-mer subsequences, and protein 4-mer subsequences. Simultaneously, RNA 3-mer subsequences can be paired with protein 3-mer subsequences, and RNA 4-mer subsequences can be paired with protein 4-mer subsequences to obtain various 3-mer and 4-mer subsequence pairs. Then, feature calculations can be performed on the RNA and protein sequences in the RNA-protein pair to be predicted based on the original sequence feature set to obtain the sequence features of the RNA-protein pair to be predicted.

[0194] Specifically, all subsequences and subsequence pairs of the RNA-protein pair to be predicted can be searched in the original sequence feature set, and the existence of features in each feature dimension of the original sequence feature set can be determined based on the search results. For example, by searching all subsequences of the RNA-protein pair to be predicted, a 560-dimensional feature vector [1, ..., 1, ..., 1, ..., 0, ...] can be calculated, which is the first sequence feature. By searching all subsequence pairs of the RNA-protein pair, a 370-dimensional feature vector [1, 1, ..., 0, 1, ...] can be calculated, which is the second sequence feature. For example, by concatenating the two feature vectors, a 930-dimensional feature vector can be obtained, which is the sequence feature of the RNA-protein pair to be predicted. Alternatively, both feature vectors can be directly input into the interaction prediction model simultaneously, which is not specifically limited in this disclosure.

[0195] In other examples, a 930-dimensional original sequence feature set can be composed of 560 k-mer subsequences and 370 frequent itemset features. For example, the original sequence feature set could be [CCC, ..., CCCC, ..., 777, ..., 7774, ..., CCA_121, ..., UCUG_1312, ..., AAU_122, ..., CUUU_1312, ...]. All subsequences and subsequence pairs of the RNA-protein pair can be searched within the original sequence feature set. Based on the search results, it can be determined whether a feature exists in each feature dimension of the original sequence feature set, resulting in a 930-dimensional feature vector [1, ..., 1, ..., 1, ..., 0, ..., 1, ..., 1, ..., 1, ..., 0, ...]. This feature vector represents the sequence features of the RNA-protein pair to be predicted.

[0196] To further explore the relationships between RNA and protein sequences, frequent itemset features can be extracted from each RNA-protein pair in the original dataset. These frequent itemset features form the original sequence feature set. Frequent itemset features combine the k-mer features of the RNA and protein sequences. Therefore, using frequent itemset features can better distinguish between interacting and non-interacting RNA-protein pairs. Alternatively, k-mer features and frequent itemset features can be extracted simultaneously and combined to form the original sequence feature set. By combining the properties of k-mer features and frequent itemset features, the interactions between RNA and protein in unknown RNA-protein pairs can be predicted more accurately.

[0197] In step S230, the RNA-protein pair to be predicted is vectorized to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted.

[0198] After feature extraction of the RNA-protein pair to be predicted, the obtained sequence features can be used as input to the first interaction prediction model. In order to obtain the sequence information of the RNA and protein sequences, as well as the information between adjacent bases and amino acids, the RNA-protein pair to be predicted can also be vectorized, and the resulting vector can be used as input to at least one second interaction prediction model.

[0199] In one example implementation, the RNA and protein sequences in the RNA-protein pair to be predicted can be converted into k-mer subsequences, respectively. For example, the non-overlapping RNA sequences can be divided into M RNA k-mer subsequences, and the non-overlapping protein sequences can be divided into N protein k-mer subsequences. For instance, if the RNA sequence is AUCUGAAAU, it can be divided into three RNA k-mer subsequences: AUC, UGA, and AAU. It is understood that dividing the non-overlapping RNA and protein sequences into multiple k-mer subsequences is for vectorizing the RNA and protein sequences, that is, vectorizing the bases in the RNA sequence and the amino acids in the protein sequence in a k-connected manner. Similarly, in other examples, each base in the RNA sequence of the RNA-protein pair can be vectorized to obtain multiple nucleotide vectors, and these multiple base vectors can be concatenated to obtain the representation vector of the RNA sequence. Simultaneously, each amino acid in the protein sequence of the RNA-protein pair can be vectorized to obtain multiple amino acid vectors, and these multiple amino acid vectors can be concatenated to obtain the representation vector of the protein sequence. Furthermore, overlapping RNA sequences can be divided into P RNA k-mer subsequences, and overlapping protein sequences can be divided into Q protein k-mer subsequences; this disclosure does not impose specific limitations on these divisions. Then, the RNA sequence representation vector and the protein sequence representation vector can be input into the second interaction prediction model, respectively.

[0200] For example, each k-mer subsequence of the RNA and protein sequences can be encoded first. For instance, when k=3, there can be 64 RNA 3-mer subsequences and 343 protein 3-mer subsequences. Each RNA and protein 3-mer subsequence can be sequentially embedded (vector mapping) encoded, representing each 3-mer subsequence with a low-dimensional vector, resulting in multiple corresponding 3-mer subsequence vectors. In one example, each 3-mer subsequence can be one-hot encoded. One-hot encoding, also known as one-bit valid encoding, uses an N-bit state register to encode N states, each state having an independent register bit, and at any given time, only one bit in the register is valid. For example, for the i-th RNA 3-mer subsequence (i.e., the RNA 3-mer subsequence with index i), a 64-dimensional One-Hot vector can be encoded, with the i-th element set to 1 and all other elements set to 0, in the form [0, 1, 0, 0, ..., 0]. For the j-th protein 3-mer subsequence (i.e., the protein 3-mer subsequence with index j), a 343-dimensional One-Hot vector can be encoded, with the j-th element set to 1 and all other elements set to 0. Similarly, each RNA and protein 3-mer subsequence can correspond to a 3-mer One-Hot vector. In other examples, dense vectors can also be used to represent each 3-mer subsequence. For example, the Word2vec algorithm can be used to map each 3-mer subsequence to a vector space, and each 3-mer subsequence can be represented by a subsequence vector in that vector space. Each 3-mer subsequence can also be encoded using a BERT (Bidirectional Encoded Representation from Transformer) pre-trained model to obtain multiple corresponding 3-mer subsequence vectors. Specifically, large-scale RNA sequence data can be obtained, and a BERT pre-trained model can be used for training. After training, a high-dimensional vector of the RNA sequence can be obtained by inputting a certain RNA sequence into the trained model. This disclosure does not impose any specific limitations on this.

[0201] After obtaining all 3-mer One-Hot vectors for the RNA and protein sequences, the RNA and protein sequences in the RNA-protein pair to be predicted can be converted into 3-mer subsequences, resulting in M ​​RNA 3-mer subsequences and N protein 3-mer subsequences. Then, by querying, the M 3-mer One-Hot vectors corresponding to the M RNA 3-mer subsequences can be determined. These M 3-mer One-Hot vectors are then concatenated sequentially, for example, along rows, to obtain an M*64 two-dimensional matrix, as shown below:

[0202] [[0,1,0,0,....,0A;[0,0,0,1,....,0];…;[1,0,0,0,....,0]]

[0203] This two-dimensional matrix is ​​the 3-mer One-Hot representation vector of the RNA sequence. By querying, N 3-mer subsequences of proteins can be identified as N 3-mer One-Hot vectors. Concatenating these N 3-mer One-Hot vectors sequentially in the row direction yields an N*343 two-dimensional matrix, which is the 3-mer One-Hot representation vector of the protein sequence. It is understood that M or N 3-mer One-Hot vectors can also be concatenated column-wise, or directly (i.e., concatenated at the tail) to obtain the 3-mer One-Hot vector of the sequence; this disclosure does not impose specific limitations on these methods.

[0204] In the exemplary embodiments of this disclosure, by vectorizing the RNA sequence and protein sequence in the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector can be used as input to a deep learning model to further discover feature combinations that occur infrequently or are novel in the data, thereby revealing the interaction between implicit features.

[0205] In step S240, based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, multiple interaction prediction models are used to obtain multiple interaction prediction values ​​of the RNA-protein pair to be predicted.

[0206] In one example implementation, multiple interaction prediction models are obtained through joint training. These multiple interaction prediction models can all be traditional machine learning models, all be deep learning models, or simultaneously include at least one traditional machine learning model and at least one deep learning model. Traditional machine learning models refer to processing natural data in its raw form. For example, constructing a pattern recognition or machine learning system requires using expertise to extract features from raw data (such as pixel values ​​of an image) and transforming them into an appropriate feature representation. Exemplary examples of traditional machine learning models include linear regression models, logistic regression models, support vector machine models, decision tree models, K-Nearest Neighbor (KNN) models, random forest models, and Naive Bayes models, etc. Deep learning models, on the other hand, have the ability to automatically extract features. They can consist of complex computational models composed of multiple processing layers, thereby automatically acquiring data representations and multiple levels of abstraction; they represent a learning process focused on feature representations. Exemplary examples of deep learning models include convolutional neural network models and recurrent neural network models, etc. When multiple interaction prediction models are all traditional machine learning models, it is not necessary to vectorize the RNA-protein pairs to be predicted; instead, the sequence features obtained from feature extraction of the RNA-protein pairs to be predicted can be used as inputs to each traditional machine learning model. When multiple interaction prediction models are all deep learning models, it is not necessary to extract features from the RNA-protein pairs to be predicted; instead, the RNA sequence representation vector and protein sequence representation vector obtained from vectorizing the RNA-protein pairs to be predicted can be used as inputs to each deep learning model.

[0207] After obtaining the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector, and the protein sequence representation vector in the RNA-protein pair, the sequence features of the RNA-protein pair can be input into at least one first interaction prediction model to obtain at least one first interaction prediction value. Simultaneously, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted can be input into at least one second interaction prediction model to obtain at least one second interaction prediction value. It should be noted that the first interaction prediction model and at least one second interaction prediction model are obtained based on joint learning. Joint learning refers to combining multiple sub-models into one model to complete the final target task. Similarly, in the exemplary embodiments of this disclosure, at least one first interaction prediction model and at least one second interaction prediction model can be combined, and the final prediction result can be obtained by fusing the outputs of each model. Moreover, during model training, the results of each model and the weighted sum can be considered simultaneously, and the model parameters of each model can be optimized simultaneously to obtain the optimal overall model, thereby improving the overall model's predictive ability.

[0208] In one example implementation, the first interaction model can be a conventional machine learning model, and the second interaction model can be a deep learning model. Correspondingly, the sequence features of the RNA-protein pair to be predicted can be input into at least one conventional machine learning model to obtain at least one first interaction prediction value. The RNA sequence representation vector and protein sequence representation vector of the RNA-protein pair to be predicted can be input into at least one deep learning model to obtain at least one second interaction prediction value. It should be noted that each deep learning model can include at least two sub-deep learning models, respectively used to process the RNA sequence representation vector and protein sequence representation vector of the RNA-protein pair to be predicted. For example, the RNA sequence representation vector can be input into the first sub-deep learning model to obtain the first sequence feature. Simultaneously, the protein sequence representation vector can be input into the second sub-deep learning model to obtain the second sequence feature. Finally, the first sequence feature and the second sequence feature can be fused through the fully connected layer of each deep learning model to obtain the second interaction prediction value based on the fused features.

[0209] In the exemplary embodiments of this disclosure, the conventional machine learning model may include at least one of LR (Logistic Regression) model, SVM model and decision tree model, and the deep learning model may include at least one of CNN model and recurrent neural network model, wherein the recurrent neural network model may be LSTM model or BiLSTM (Bi-directional LSTM) model.

[0210] This can be illustrated by using a joint learning approach combining an LR model, a CNN model, and a BiLSTM model to predict RNA-protein pair interactions. For example, a 930-dimensional original sequence feature set can be constructed from 560 k-mer subsequences and 370 frequent itemset features. Feature calculations are then performed on this original sequence feature set based on the k-mer subsequences of the RNA-protein pair to be predicted, resulting in a 930-dimensional feature vector. This vector is then used as input to the LR model to obtain the predicted interaction value y1. Simultaneously, the RNA-protein pair to be predicted can be vectorized to obtain k-merOne-Hot vectors for the RNA and protein sequences, respectively. These vectors are then used as input to the CNN and BiLSTM models to obtain the predicted interaction values ​​y2 and y3, respectively.

[0211] In the exemplary embodiments of this disclosure, firstly, the LR model has good memory capabilities, enabling it to learn the correlations between sequences or features from data. CNN and BiLSTM models have good generalization abilities, capable of discovering infrequent or novel feature combinations in the data, thereby revealing the interactions between implicit features. Secondly, while CNN models can better capture features but neglect the location information of features, BiLSTM models have better memory capabilities, utilizing the sequence and location information of the data to compensate for the memory limitations of CNN models. Finally, the LR model is simple and has good interpretability; jointly learning CNN, BiLSTM, and LR models can enhance the interpretability of the overall model's RPI predictions. In summary, this disclosure effectively combines the characteristics of various interaction prediction models, thereby improving the overall model's predictive power.

[0212] In step S250, the interaction between the RNA and the protein is determined based on the plurality of interaction prediction values.

[0213] After obtaining multiple interaction prediction values, a weighted sum can be calculated, and the interaction between RNA and protein can be determined based on the result. If the calculated result is greater than a preset interaction prediction threshold, an interaction between RNA and protein can be determined. If the calculated result is less than or equal to the preset interaction prediction threshold, no interaction between RNA and protein can be determined. By fusing the output values ​​of each interaction prediction model to obtain the final prediction result, the interaction between RNA and protein in the RNA-protein pair to be predicted can be determined more accurately.

[0214] For example, when predicting RNA-protein pair interactions, one can base it on:

[0215] y out =α*y1+β*y2+γ*y3

[0216] The predicted value y of the RNA-protein pair interaction to be predicted was calculated. out Where y1 is the output value of the logistic regression model, y2 is the output value of the convolutional neural network model, y3 is the output value of the recurrent neural network model, and α, β, and γ are the weight parameters of the logistic regression model, the convolutional neural network model, and the recurrent neural network model, respectively. out It can be any value between 0 and 1. For example, the interaction prediction threshold can be preset to 0.5, when y... out When the value is greater than 0.5, the prediction result can be marked as 1, indicating that the RNA-protein pair interacts. When y out When the value is ≤0.5, the prediction result can be marked as 0, which means that the RNA-protein pair does not interact.

[0217] For example, the weight parameters α, β, and γ mentioned above can be obtained based on joint learning training.

[0218] In the exemplary embodiments of this disclosure, reference is made to Figure 6 As shown, multiple interaction prediction models and related parameters can be pre-trained jointly according to steps S610 to S650 to optimize all model parameters in each prediction model. The final model obtained from the training can be used to predict RNA-protein pairs with unknown interactions.

[0219] Step S610. Obtain a training dataset, which includes positive RNA-protein pairs and negative RNA-protein pairs.

[0220] You can use all RNA-protein pairs from the original dataset as the training dataset, or you can use only a subset of RNA-protein pairs from the original dataset. Taking the RPI1807 dataset as an example, this dataset contains 3243 RNA-protein pairs, specifically 1807 positive pairs and 1436 negative pairs. For instance, you can select 1200 positive pairs and 1000 negative pairs as the training dataset. It's important to understand that the number of RNA-protein pairs in the training dataset is merely illustrative; you can obtain any number of RNA-protein pairs to train each interaction prediction model multiple times to improve the performance of each model. It should be noted that positive RNA-protein pairs can be labeled with a value of "1", indicating that the RNA-protein pairs interact. Negative RNA-protein pairs can be labeled with a value of "0", indicating that the RNA-protein pairs do not interact.

[0221] Step S620. Use the multiple interaction prediction models to obtain multiple interaction prediction values ​​for each RNA-protein pair in the training dataset.

[0222] For example, multiple interaction prediction models can be LR (Regression-Linguistic Programming) models, CNN (Convolutional Network) models, and BiLSTM (Bilateral LSTM) models, respectively. Similarly, feature extraction is performed on each RNA-protein pair in the training dataset, and the extracted sequence features can be sequentially input into the LR model. Vectorization is performed on each RNA-protein pair, and the resulting RNA sequence representation vector and protein sequence representation vector can be input into the CNN and BiLSTM models. For example, let's take training each prediction model using the i-th RNA-protein pair. The sequence features of this RNA-protein pair can be input into the LR model, and the predicted value y1 is output. The RNA sequence representation vector and protein sequence representation vector of this RNA-protein pair can be input into two CNN sub-models respectively, and the outputs of the two CNN sub-models are concatenated to finally output the predicted value y2. Similarly, the RNA sequence representation vector and protein sequence representation vector of this RNA-protein pair can be input into the BiLSTM model, and the predicted value y3 is output. It can be understood that for each RNA-protein pair in the training dataset, three interaction prediction values ​​can be obtained through these three interaction prediction models.

[0223] Step S630. Obtain the joint prediction value for each RNA-protein pair in the training dataset based on the multiple interaction prediction values.

[0224] After obtaining multiple interaction prediction values ​​for each RNA-protein pair in the training dataset, these prediction values ​​can be weighted and summed to obtain the joint prediction value for each RNA-protein pair. Taking the i-th RNA-protein pair in the training dataset as an example, we can proceed as follows:

[0225] y out =α*y1+β*y2+γ*y3

[0226] The joint prediction value y of this RNA-protein pair was calculated. out , where y1 is the output value of the LR model, y2 is the output value of the CNN model, y3 is the output value of the BiLSTM model, and α, β, and γ are the weight parameters of the LR model, CNN model, and BiLSTM model, respectively.

[0227] Step S640. Calculate the joint predicted value and label value of each RNA-protein pair in the training dataset using the loss function to obtain the corresponding loss value.

[0228] Each RNA-protein pair in the training dataset has a label value, such as 1 for each positive pair and 0 for each negative pair. For example, the i-th RNA-protein pair is a positive example, and its label value is 1. The joint prediction value y for this RNA-protein pair can be used as a basis for further analysis. out The loss function is calculated using the label value 1 to obtain the corresponding loss value. During the training of the prediction model, it is necessary to make the joint predicted value infinitely close to the label value, i.e., to minimize the objective function. In one example, when minimizing the objective function, the cross-entropy loss function can be chosen as the objective function. When calculating the cross-entropy loss function, if the label value is 1, y... out The closer y is to 1, the smaller the calculated loss value. out The closer the value is to 0, the greater the calculated loss value. If the label value is 0, y out The closer y is to 0, the smaller the calculated loss value. out The closer the value is to 1, the larger the calculated loss value. Understandably, the cross-entropy loss function is a performance function in the predictive model, used to estimate the degree of discrepancy between the model's predicted values ​​and the labeled values. The smaller the calculated cross-entropy loss function value, the better the model's predictive performance.

[0229] Step S650. Adjust the model parameters of the plurality of interaction prediction models according to the loss value.

[0230] The model parameters of multiple interaction prediction models can be iteratively updated based on the calculated loss values. When the iteration termination condition is met, the training of the model parameters and weight parameters of multiple interaction prediction models is complete. For example, the stochastic gradient descent algorithm can be used to update the parameters. Based on the backpropagation principle, the objective function, such as the cross-entropy loss function, is continuously calculated, and the model parameters and weight parameters of each interaction prediction model are updated simultaneously based on the calculated loss values. When the objective function converges to its minimum value, the training of all model parameters is complete. Alternatively, the model parameters can be updated iteratively in reverse order. When a preset number of iterations is met, the training of all model parameters is complete. For example, the preset number of iterations can be 20. During the 20 reverse iterations, each interaction prediction model continuously updates its model parameters. After iteration, the optimized model parameters can be obtained. In other examples, the objective function can be minimized alternately using least squares, Adam optimization algorithms, etc., and the model parameters and weight parameters are updated sequentially from back to front to optimize the parameters. It should be noted that the weight parameters α, β, and γ are also part of the model parameters. During the joint training of multiple interaction prediction models, the weight parameters will also be trained and continuously optimized.

[0231] In the exemplary embodiments of this disclosure, when using multiple interaction prediction models to predict the interaction of unknown RNA-protein pairs, by combining a logistic regression model with a deep learning model, the logistic regression model can learn the correlation between each k-mer feature and / or each frequent itemset feature from the original dataset, while the deep learning model can reveal the interaction between implicit features. Furthermore, by combining the characteristics of CNN models and BiLSTM models, the model parameters of each model are optimized simultaneously during the joint training of multiple models, thereby improving the overall accuracy of the model in predicting the interaction of unknown RNA-protein pairs.

[0232] In another example implementation, when jointly training multiple interaction prediction models in advance, the original dataset can be divided into training, validation, and test datasets proportionally. For instance, the original dataset can be the RPI1807 dataset. This dataset contains 3243 RNA-protein pairs, specifically 1807 positive pairs and 1436 negative pairs. For example, this dataset can be divided into training, validation, and test datasets in a 7:2:1 ratio. The ratio of positive to negative examples in each dataset can be consistent with the overall dataset distribution, i.e., 1807:1436, approximately 1.25:1. For example, 1250 positive and 1000 negative examples can be selected as the training dataset, 360 positive and 280 negative examples as the validation dataset, and 180 positive and 140 negative examples as the test dataset.

[0233] The training dataset can be input into multiple interaction prediction models, and the corresponding model parameters can be determined using the backpropagation algorithm to obtain the first joint model. For example, the training dataset can be input into each interaction prediction model, and the model parameters can be adjusted. These model parameters can include weight parameters, bias parameters, and intercept parameters. For example, the stochastic gradient descent algorithm can be used to update the parameters of each model. Based on the backpropagation principle, the objective function is continuously calculated, and the model parameters are updated according to the objective function. When the objective function converges to its minimum value, the training of the model parameters is complete, thus obtaining the first joint model.

[0234] A validation dataset can be input into the first joint model to validate its performance, and a second joint model can be obtained based on the validation results. For example, when optimizing model parameters, a set of hyperparameters can be initialized first, and multiple interaction prediction models can be continuously trained using the training dataset to obtain the first joint model. These hyperparameters can include the learning rate, the number of CNN layers, and the kernel size. Then, the validation dataset can be input into the trained first joint model to validate its prediction accuracy. When the prediction accuracy reaches a preset accuracy threshold, the current first joint model can be used as the second joint model, resulting in the final trained model. Finally, the final performance of the model can be tested using a test dataset. Understandably, if the test dataset shows that the second joint model's prediction performance is poor, a new set of hyperparameters can be set, and the interaction prediction model can be trained and validated again using the training and validation datasets. When the prediction accuracy of the trained interaction prediction model on the validation dataset reaches the preset accuracy threshold, the final performance of the prediction model can be tested using a new test dataset.

[0235] For example, the accuracy of the second joint model can be determined based on the test dataset. When the accuracy exceeds a preset threshold, a third joint model is obtained, which includes multiple pre-trained interaction prediction models. For instance, after obtaining the second joint model, each RNA-protein pair from the test dataset can be input into the second joint model to assess its accuracy. If the model's accuracy exceeds a preset accuracy threshold, the third joint model is obtained. It is understood that the third joint model includes multiple pre-trained interaction prediction models, which can then be used to predict interactions between unknown RNA-protein pairs. The Matthews correlation coefficient of the second joint model can also be assessed using the test dataset. The Matthews correlation coefficient is the correlation coefficient between the actual classification and the predicted classification, and its value ranges from [0, 1]. A larger value indicates a stronger correlation between the predicted and actual values, while a value of 1 indicates a completely correct prediction. If the Matthews correlation coefficient of this model exceeds a preset threshold, the third joint model is obtained. The specificity and recall of the second joint model can also be assessed using the test dataset, but this disclosure does not specifically limit this assessment. For example, if the accuracy of the second joint model is not greater than a preset accuracy threshold, a new training dataset can be obtained to retrain the model parameters of each interaction prediction model to continuously improve model performance. In the example implementation of this disclosure, the optimal prediction model can be trained by adjusting the parameters using a training dataset and a validation dataset, and then the generalization performance of the prediction model can be tested using a test dataset.

[0236] In the training process described above, the model parameters of the LR model, CNN model, and BiLSTM model can be trained simultaneously. For example, based on the objective function, the model parameters of the three interaction prediction models can be adjusted simultaneously using the loss value calculated from the joint predicted value and label value of each RNA-protein pair in the training dataset. Through multiple backpropagation iterations, the model parameters of the three interaction prediction models can eventually converge, or training can be terminated after a certain number of iterations. This training method allows for the simultaneous training of the three interaction prediction models (LR, CNN, and BiLSTM), achieving joint learning of the three models. Furthermore, it not only ensures higher accuracy and precision in predicting RNA-protein pair interactions but also improves the training efficiency of each interaction prediction model.

[0237] After each model is trained, the final interaction prediction model can be used to predict the interaction between unknown RNA-protein pairs, and the prediction results of the RNA-protein pair interaction can be output to the terminal device for users to view.

[0238] In one example implementation, the same training, validation, and test datasets can be used to train and test individual LR models, CNN models, BiLSTM models, and a joint learning model of the three. For example, the performance of each prediction model can be evaluated using two performance metrics: accuracy and Matthews correlation coefficient. When training and testing the model using the RPI1807 dataset, the joint learning model outperformed the LR, CNN, and BiLSTM models in both accuracy and Matthews correlation coefficient, indicating that the joint learning model performs better than the individual models, and therefore has superior predictive ability.

[0239] In the exemplary embodiments of this disclosure, at least one RNA sequence can also be obtained, and multiple interaction prediction models can be used to search a database for protein sequences that interact with each input RNA sequence. Specifically, after obtaining the original dataset, one can refer to... Figure 6 Multiple interaction prediction models are pre-trained jointly using the original dataset. After training, all protein sequences participating in the joint training can be stored in a database. It should be noted that the database may also include other protein sequences that did not participate in the joint training; that is, the number of protein sequences in the database can be arbitrary. The database may also include any number of RNA sequences, such as including but not limited to all RNA sequences participating in the joint training. This disclosure does not impose specific limitations on this.

[0240] After a user inputs at least one RNA sequence, each input RNA sequence can be combined with all protein sequences in the database to form several RNA-protein pairs. Further, according to steps S220 to S250, multiple interaction models can be jointly learned to predict the interactions of each RNA-protein pair. Specifically, feature extraction and vectorization can be performed on each RNA-protein pair. The obtained sequence features, RNA sequence representation vectors, and protein sequence representation vectors in the RNA-protein pair are input into multiple interaction prediction models to obtain the interaction prediction value for each RNA-protein pair. An interaction prediction value of 1 indicates that the RNA-protein pair interacts, and an interaction prediction value of 0 indicates that the RNA-protein pair does not interact. Then, all RNA-protein pairs with an interaction prediction value of 1 can be filtered out, and the protein sequences in each RNA-protein pair can be output to the terminal device for the user to view the protein sequences that interact with the input RNA sequence.

[0241] Similarly, in the exemplary embodiments of this disclosure, at least one protein sequence can be obtained, and multiple interaction prediction models can be used to search a database for RNA sequences that interact with each input protein sequence. For example, after a user inputs at least one protein sequence, each input protein sequence can be combined with all RNA sequences in the database to form several RNA-protein pairs. Further, multiple interaction models can be jointly learned according to steps S220 to S250 to predict the interaction of each RNA-protein pair. Specifically, feature extraction and vectorization can be performed on each RNA-protein pair, and the obtained sequence features, RNA sequence representation vector, and protein sequence representation vector in the RNA-protein pair can be input into multiple interaction prediction models to obtain the interaction prediction value for each RNA-protein pair. An interaction prediction value of 1 indicates that the RNA-protein pair interacts, and an interaction prediction value of 0 indicates that the RNA-protein pair does not interact. Then, all RNA-protein pairs with an interaction prediction value of 1 can be filtered out, and the RNA sequences in each RNA-protein pair can be output to a terminal device for the user to view the RNA sequences that interact with the input protein sequence.

[0242] In the RNA-protein interaction prediction method provided in the exemplary embodiments of this disclosure, an RNA-protein pair to be predicted is obtained; features are extracted from the RNA-protein pair to obtain its sequence features; the RNA-protein pair is vectorized to obtain its RNA sequence representation vector and protein sequence representation vector; based on the sequence features, RNA sequence representation vector, and protein sequence representation vector of the RNA-protein pair, multiple interaction prediction models are used to obtain multiple interaction prediction values ​​for the RNA-protein pair; and the interaction between the RNA and protein is determined based on the multiple interaction prediction values. On the one hand, by performing feature extraction and vectorization on the RNA-protein pair, the connection between RNA and protein sequences can be fully explored, so as to accurately predict the interaction between RNA and protein; on the other hand, effectively combining the characteristics of multiple interaction prediction models can further improve the accuracy of predicting RNA-protein interactions.

[0243] It should be noted that although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0244] Furthermore, this example embodiment also provides an RNA-protein interaction prediction device. This device can be applied to a server or terminal device. (Reference) Figure 7 As shown, the RNA-protein interaction prediction device 700 may include a data acquisition module 710, a feature extraction module 720, a data vectorization module 730, an interaction prediction module 740, and an interaction determination module 750, wherein:

[0245] The data acquisition module 710 is used to acquire the RNA-protein pairs to be predicted;

[0246] The feature extraction module 720 is used to extract features from the RNA-protein pair to be predicted, and obtain the sequence features of the RNA-protein pair to be predicted.

[0247] The data vectorization module 730 is used to vectorize the RNA-protein pair to be predicted to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted.

[0248] The interaction prediction module 740 is used to obtain multiple interaction prediction values ​​of the RNA-protein pair to be predicted based on the sequence features of the RNA-protein pair to be predicted, the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, using multiple interaction prediction models respectively.

[0249] Interaction determination module 750 is used to determine the interaction between the RNA and the protein based on the plurality of interaction prediction values.

[0250] In one alternative implementation, the feature extraction module 720 includes:

[0251] The feature set acquisition module is used to acquire the original sequence feature set;

[0252] The feature determination module is used to determine the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set.

[0253] In one optional implementation, the feature determination module includes:

[0254] A sequence conversion unit is used to convert the RNA sequence and protein sequence in the RNA-protein pair to be predicted into k-mer subsequences, respectively.

[0255] The first sequence lookup unit is used to look up each k-mer subsequence in the original sequence feature set and obtain the sequence features of the RNA-protein pair to be predicted based on the lookup results.

[0256] In one optional implementation, the feature determination module includes:

[0257] A sequence conversion unit is used to convert the RNA sequence and protein sequence in the RNA-protein pair to be predicted into k-mer sub-sequences, respectively, wherein the k-mer sub-sequences include RNA k-mer sub-sequences and protein k-mer sub-sequences;

[0258] A sequence combination unit is used to combine the RNA k-mer subsequence and the protein k-mer subsequence to obtain multiple RNA-protein k-mer subsequence pairs;

[0259] The second sequence search unit is used to search for each RNA-protein k-mer subsequence pair in the original sequence feature set, and to obtain the sequence features of the RNA-protein pair to be predicted based on the search results.

[0260] In one optional implementation, the feature determination module includes:

[0261] A sequence conversion unit is used to convert the RNA sequence and protein sequence in the RNA-protein pair to be predicted into k-mer sub-sequences, respectively, wherein the k-mer sub-sequences include RNA k-mer sub-sequences and protein k-mer sub-sequences;

[0262] The first sequence search unit is used to search for each k-mer subsequence in the original sequence feature set to obtain the first sequence feature;

[0263] A sequence combination unit is used to combine the RNA k-mer subsequence and the protein k-mer subsequence to obtain multiple RNA-protein k-mer subsequence pairs;

[0264] The second sequence lookup unit is used to search for each RNA-protein k-mer subsequence pair in the original sequence feature set to obtain the second sequence feature;

[0265] A feature splicing unit is used to compose the sequence features of the RNA-protein pair to be predicted by the first sequence feature and the second sequence feature.

[0266] In one optional implementation, the feature set acquisition module includes:

[0267] The dataset acquisition module is used to acquire the original dataset;

[0268] The feature extraction module is used to extract features from each RNA-protein pair in the original dataset to obtain the original sequence feature set.

[0269] In one optional implementation, the feature extraction module includes:

[0270] Sequence generation unit, used to arrange and combine basic units of RNA and protein to obtain k-mer subsequences;

[0271] A variance calculation unit is used to count the frequency of occurrence of each k-mer subsequence in the original dataset and calculate the variance of each k-mer subsequence based on the frequency of occurrence.

[0272] The dataset determination unit is used to determine the original sequence feature set based on the variance of each k-mer sequence.

[0273] In one optional implementation, the variance calculation unit includes:

[0274] The frequency statistics subunit is used to count the frequency of occurrence of each k-mer subsequence in the original dataset.

[0275] A frequency calculation subunit is used to calculate the frequency of occurrence of each k-mer subsequence in the original dataset based on the occurrence frequency.

[0276] A sequence marker subunit is used to traverse the original dataset and mark whether each k-mer subsequence appears in each RNA-protein pair;

[0277] The variance calculation subunit is used to calculate the variance of each k-mer subsequence based on the frequency of occurrence of each k-mer subsequence in the original dataset and the label value in each RNA-protein pair.

[0278] In one alternative implementation, the variance calculation subunit is configured to calculate based on:

[0279]

[0280] Calculate the variance Var of the i-th k-mer subsequence. i .in, Freq represents the label value of the i-th k-mer subsequence in the n-th RNA-protein pair. i Let represent the frequency of the i-th k-mer subsequence in the original dataset, and N be the total number of RNA-protein pairs in the original dataset.

[0281] In one alternative implementation, the dataset determination unit is configured to determine k-mer subsequences that satisfy preset conditions based on the variance of each k-mer subsequence, and to form the original sequence feature set from the k-mer subsequences that satisfy the preset conditions.

[0282] In an optional implementation, the feature extraction module further includes:

[0283] The first item set generation unit is used to convert the RNA sequence and protein sequence in each RNA-protein pair into k-mer sub-sequences respectively, and the k-mer sub-sequences form a first candidate item set, wherein the k-mer sub-sequences include RNA k-mer sub-sequences and protein k-mer sub-sequences;

[0284] The frequent itemset generation unit is used to count the frequency of each k-mer subsequence in the first candidate itemset in the original dataset, and to form a frequent itemset by k-mer subsequences that meet a preset frequency threshold.

[0285] The second itemset generation unit is used to cross-combine the RNA k-mer subsequences and protein k-mer subsequences in the frequent itemset, and the resulting k-mer subsequence pairs form the second candidate itemset.

[0286] The support determination unit is used to count the frequency of occurrence of each k-mer subsequence pair in the second candidate set in the original dataset, and to obtain the support of each k-mer subsequence pair.

[0287] The feature set acquisition unit is used to form the original sequence feature set by k-mer sub-sequence pairs whose support satisfies the preset conditions.

[0288] In one alternative implementation, the data vectorization module 730 includes:

[0289] A sequence conversion unit is used to convert the RNA sequence and protein sequence in the RNA-protein pair to be predicted into k-mer sub-sequences, respectively, wherein the k-mer sub-sequences include M RNA k-mer sub-sequences and N protein k-mer sub-sequences;

[0290] The first vectorization unit is used to vectorize each RNA k-mer subsequence to obtain M RNA k-mer vectors;

[0291] The first splicing unit is used to splice the M RNA k-mer vectors to obtain the RNA sequence representation vector;

[0292] The second vectorization unit is used to vectorize each protein k-mer sequence to obtain N protein k-mer vectors;

[0293] The second splicing unit is used to splice the N protein k-mer vectors to obtain the protein sequence representation vector.

[0294] In one alternative implementation, the interaction prediction module 740 includes:

[0295] The first prediction unit is used to input the sequence features of the RNA-protein pair to be predicted into at least one first interaction prediction model to obtain at least one first interaction prediction value.

[0296] The second prediction unit is used to input the RNA sequence representation vector and the protein sequence representation vector of the RNA-protein pair to be predicted into at least one second interaction prediction model to obtain at least one second interaction prediction value.

[0297] In one alternative implementation, the interaction prediction module 740 includes:

[0298] The third prediction unit is used to input the sequence features of the RNA-protein pair to be predicted into at least one conventional machine learning model to obtain at least one first interaction prediction value.

[0299] The fourth prediction unit is used to input the RNA sequence representation vector and the protein sequence representation vector of the RNA-protein pair to be predicted into at least one deep learning model to obtain at least one second interaction prediction value.

[0300] In one optional implementation, each deep learning model includes at least two sub-deep learning models; the fourth prediction unit includes:

[0301] The first feature generation unit is used to input the RNA sequence representation vector in the RNA-protein pair to be predicted into the first sub-deep learning model to obtain the first sequence feature;

[0302] The second feature generation unit is used to input the protein sequence representation vector in the RNA-protein pair to be predicted into the second sub-deep learning model to obtain the second sequence feature;

[0303] The feature fusion unit is used to fuse the first sequence feature and the second sequence feature, and obtain the second interaction prediction value based on the fused features.

[0304] In one alternative implementation, the conventional machine learning model includes at least one of a logistic regression model, a support vector machine model, and a decision tree model, and the deep learning model includes at least one of a convolutional neural network model and a recurrent neural network model.

[0305] In one alternative implementation, the interaction determination module 750 includes:

[0306] A weighted calculation unit is used to perform a weighted summation calculation on the multiple interaction prediction values;

[0307] An interaction determination unit is used to determine the interaction between the RNA and the protein based on calculation results.

[0308] In one optional implementation, the interaction determination unit includes:

[0309] The first interaction determination subunit is used to determine that there is an interaction between the RNA and the protein if the calculation result is greater than a preset interaction prediction value threshold.

[0310] The second interaction determination subunit is used to determine that there is no interaction between the RNA and the protein if the calculation result is less than or equal to a preset interaction prediction value threshold.

[0311] In an optional embodiment, the RNA-protein interaction prediction device 700 further includes:

[0312] A joint training module is used to jointly train the multiple interaction prediction models.

[0313] In one alternative implementation, the joint training module includes:

[0314] The training data acquisition unit is used to acquire a training dataset, which includes positive RNA-protein pairs and negative RNA-protein pairs.

[0315] The first prediction value output unit is used to obtain multiple interaction prediction values ​​for each RNA-protein pair in the training dataset using the multiple interaction prediction models.

[0316] The second prediction value output unit is used to obtain the joint prediction value of each RNA-protein pair in the training dataset based on the multiple interaction prediction values;

[0317] The loss value calculation unit is used to calculate the joint prediction value and label value of each RNA-protein pair in the training dataset using a loss function to obtain the corresponding loss value;

[0318] The model parameter adjustment unit is used to adjust the model parameters of the plurality of interaction prediction models according to the loss value.

[0319] In one optional implementation, the second predicted value output unit includes:

[0320] The second prediction value output subunit is used to perform a weighted summation of multiple interaction prediction values ​​for each RNA-protein pair in the training dataset to obtain a joint prediction value for each RNA-protein pair.

[0321] In one alternative implementation, the second predicted value output subunit is configured to:

[0322] according to:

[0323] y out =α*y1+β*y2+γ*y3

[0324] The joint prediction value y for each RNA-protein pair in the training dataset was calculated. out Where y1 is the output value of the traditional machine learning model, y2 is the output value of the convolutional neural network model, y3 is the output value of the recurrent neural network model, and α, β, and γ are the weight parameters of the traditional machine learning model, the convolutional neural network model, and the recurrent neural network model, respectively.

[0325] In one optional implementation, the model parameter adjustment unit is configured to iteratively update the model parameters of the plurality of interaction prediction models based on the loss value, and when the iteration termination condition is met, complete the training of the model parameters of the plurality of interaction prediction models, so as to use the trained plurality of interaction prediction models to predict the interaction of the RNA-protein pair to be predicted.

[0326] In an optional embodiment, the RNA-protein interaction prediction device 700 further includes:

[0327] The data output module is used to output the predicted results of the interaction between the RNA and the protein.

[0328] The specific details of each module in the above-mentioned RNA-protein interaction prediction device have been described in detail in the corresponding RNA-protein interaction prediction methods, so they will not be repeated here.

[0329] The modules in the above-described device can be general-purpose processors, including central processing units (CPUs), network processors, etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. Each module can also be implemented using software, firmware, etc. The processors in the above-described device can be independent processors or integrated together.

[0330] Exemplary embodiments of this disclosure also provide a computer-readable storage medium having a program product stored thereon capable of implementing the methods described above in this specification. In some possible embodiments, various aspects of this disclosure may also be implemented as a program product including program code that, when run on an electronic device, causes the electronic device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. This program product may be a portable compact disc read-only memory (CD-ROM) including program code and may run on an electronic device, such as a personal computer. However, the program product of this disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0331] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0332] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0333] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0334] Program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing devices can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0335] Exemplary embodiments of this disclosure also provide an electronic device capable of implementing the above-described method. Referring below... Figure 8 To describe an electronic device 800 according to such an exemplary embodiment of the present disclosure. Figure 8 The electronic device 800 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0336] like Figure 8 As shown, the electronic device 800 can be represented as a general-purpose computing device. The components of the electronic device 800 may include, but are not limited to: at least one processing unit 810, at least one storage unit 820, a bus 830 connecting different system components (including storage unit 820 and processing unit 810), and a display unit 840.

[0337] Storage unit 820 stores program code that can be executed by processing unit 810, causing processing unit 810 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, processing unit 810 can perform... Figures 2 to 6 Any one or more of the method steps.

[0338] Storage unit 820 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 821 and / or cache memory 822, and may further include a read-only memory (ROM) 823.

[0339] The storage unit 820 may also include a program / utility 824 having a set (at least one) of program modules 825, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0340] Bus 830 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0341] Electronic device 800 can also communicate with one or more external devices 900 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 800, and / or with any device that enables electronic device 800 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 850. Furthermore, electronic device 800 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 860. As shown, network adapter 860 communicates with other modules of electronic device 800 via bus 830. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 800, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0342] In some embodiments, the RNA-protein interaction prediction method described herein can be executed by the processing unit 810 of an electronic device. In some embodiments, the RNA-protein pair to be predicted / RNA sequence to be predicted / protein sequence to be predicted, the original dataset, and the training dataset for training each interaction prediction model can be input through the input interface 850. For example, the RNA-protein pair to be predicted, the original dataset, and the training dataset for training each interaction prediction model can be input through the user interface of the electronic device. In some embodiments, the prediction result of the interaction of the RNA-protein pair to be predicted can be output to an external device 900 for user viewing through the output interface 850.

[0343] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the method according to the exemplary embodiments of this disclosure.

[0344] Furthermore, the above figures are merely illustrative representations of the processes included in the methods according to exemplary embodiments of this disclosure, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0345] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0346] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for predicting RNA-protein interactions, characterized in that, include: Obtain the RNA-protein pairs to be predicted; Feature extraction is performed on the RNA-protein pair to be predicted to obtain the sequence features of the RNA-protein pair to be predicted; Vectorize the RNA-protein pair to be predicted to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted; The sequence features of the RNA-protein pair to be predicted are input into at least one conventional machine learning model to obtain at least one first interaction prediction value; The RNA sequence representation vector in the RNA-protein pair to be predicted is input into the first sub-deep learning model to obtain the first sequence feature; The protein sequence representation vector in the RNA-protein pair to be predicted is input into the second sub-deep learning model to obtain the second sequence feature; The first sequence features and the second sequence features are fused, and the second interaction prediction value is obtained based on the fused features; wherein, the first sub-deep learning model and the second sub-deep learning model form the deep learning model; the traditional machine learning model and the deep learning model are obtained based on a joint learning approach; The multiple interaction prediction values ​​are weighted and summed. The interaction between the RNA and the protein was determined based on the calculation results.

2. The method for predicting RNA-protein interactions according to claim 1, characterized in that, The feature extraction of the RNA-protein pair to be predicted, to obtain the sequence features of the RNA-protein pair to be predicted, includes: Obtain the original sequence feature set; The sequence features of the RNA-protein pair to be predicted are determined based on the original sequence feature set.

3. The method for predicting RNA-protein interactions according to claim 2, characterized in that, The step of determining the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set includes: The RNA and protein sequences in the RNA-protein pair to be predicted are converted into k-mer subsequences, respectively. Each k-mer subsequence is searched in the original sequence feature set, and the sequence features of the RNA-protein pair to be predicted are obtained based on the search results.

4. The method for predicting RNA-protein interactions according to claim 2, characterized in that, The step of determining the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set includes: The RNA sequence and protein sequence in the RNA-protein pair to be predicted are converted into k-mer sub-sequences, which include RNA k-mer sub-sequences and protein k-mer sub-sequences. By combining the RNA k-mer subsequences and protein k-mer subsequences, various RNA-protein k-mer subsequence pairs can be obtained; Search for each RNA-protein k-mer subsequence pair in the original sequence feature set, and obtain the sequence features of the RNA-protein pair to be predicted based on the search results.

5. The method for predicting RNA-protein interactions according to claim 2, characterized in that, The step of determining the sequence features of the RNA-protein pair to be predicted based on the original sequence feature set includes: The RNA sequence and protein sequence in the RNA-protein pair to be predicted are converted into k-mer sub-sequences, which include RNA k-mer sub-sequences and protein k-mer sub-sequences. The first sequence feature is obtained by searching for each k-mer subsequence in the original sequence feature set; By combining the RNA k-mer subsequences and protein k-mer subsequences, various RNA-protein k-mer subsequence pairs can be obtained; The second sequence feature is obtained by searching for each RNA-protein k-mer subsequence pair in the original sequence feature set; The sequence features of the RNA-protein pair to be predicted consist of the first sequence features and the second sequence features.

6. The method for predicting RNA-protein interactions according to claim 2, characterized in that, The process of obtaining the original sequence feature set includes: Obtain the original dataset; Feature extraction is performed on each RNA-protein pair in the original dataset to obtain the original sequence feature set.

7. The method for predicting RNA-protein interactions according to claim 6, characterized in that, The step of extracting features from each RNA-protein pair in the original dataset to obtain the original sequence feature set includes: The basic units of RNA and protein are arranged and combined to obtain k-mer subsequences; The frequency of occurrence of each k-mer subsequence in the original dataset is counted, and the variance of each k-mer subsequence is calculated based on the frequency of occurrence. The original sequence feature set is determined based on the variance of each k-mer subsequence.

8. The method for predicting RNA-protein interactions according to claim 7, characterized in that, The process of statistically analyzing the frequency of each k-mer subsequence in the original dataset and calculating the variance of each k-mer subsequence based on the frequency of occurrence includes: Calculate the frequency of occurrence of each k-mer subsequence in the original dataset; The frequency of occurrence of each k-mer subsequence in the original dataset is calculated based on the frequency of occurrence. The original dataset is traversed, and each k-mer subsequence is marked as appearing in each RNA-protein pair. The variance of each k-mer subsequence was calculated based on its frequency of occurrence in the original dataset and its label value in each RNA-protein pair.

9. The method for predicting RNA-protein interactions according to claim 8, characterized in that, The calculation of the variance of each k-mer subsequence based on its frequency of occurrence in the original dataset and its label value in each RNA-protein pair includes: according to: Calculate the first i Variance of k-mer subsequences ;in, For the first i k-mer subsequences in the th n The label value in each RNA-protein pair For the first i The frequency of occurrence of k-mer subsequences in the original dataset. N This represents the total number of RNA-protein pairs in the original dataset.

10. The method for predicting RNA-protein interactions according to claim 7, characterized in that, Determining the original sequence feature set based on the variance of each k-mer sequence includes: Based on the variance of each k-mer subsequence, k-mer subsequences that meet preset conditions are determined, and the original sequence feature set is composed of the k-mer subsequences that meet the preset conditions.

11. The method for predicting RNA-protein interactions according to claim 6, characterized in that, The step of extracting features from each RNA-protein pair in the original dataset to obtain the original sequence feature set further includes: The RNA sequence and protein sequence in each RNA-protein pair are converted into k-mer sub-sequences, and the k-mer sub-sequences form a first candidate set, which includes RNA k-mer sub-sequences and protein k-mer sub-sequences. The frequency of each k-mer subsequence in the first candidate itemset is counted in the original dataset, and frequent itemsets are formed from the k-mer subsequences that meet the preset frequency threshold. The RNA k-mer subsequences and protein k-mer subsequences in the frequent itemset are cross-combined to form a second candidate itemset. The frequency of occurrence of each k-mer subsequence pair in the second candidate set in the original dataset is counted to obtain the support of each k-mer subsequence pair. The original sequence feature set is composed of k-mer subsequence pairs whose support meets the preset conditions.

12. The method for predicting RNA-protein interactions according to claim 1, characterized in that, The vectorization of the RNA-protein pair to be predicted, to obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted, includes: The RNA sequence and protein sequence in the RNA-protein pair to be predicted are converted into k-mer sub-sequences, which include M RNA k-mer sub-sequences and N protein k-mer sub-sequences. Each RNA k-mer subsequence is vectorized to obtain M RNA k-mer vectors; The RNA sequence representation vector is obtained by concatenating the M RNA k-mer vectors. Each protein k-mer sequence is vectorized to obtain N protein k-mer vectors; The protein sequence representation vector is obtained by concatenating the N protein k-mer vectors.

13. The method for predicting RNA-protein interactions according to claim 1, characterized in that, The traditional machine learning models include at least one of logistic regression models, support vector machine models, and decision tree models, while the deep learning models include at least one of convolutional neural network models and recurrent neural network models.

14. The method for predicting RNA-protein interactions according to claim 1, characterized in that, The determination of the interaction between the RNA and the protein based on the calculation results includes: If the calculation result is greater than the preset interaction prediction threshold, it is determined that there is an interaction between the RNA and the protein; If the calculation result is less than or equal to a preset interaction prediction threshold, it is determined that there is no interaction between the RNA and the protein.

15. The method for predicting RNA-protein interactions according to any one of claims 1-14, characterized in that, The method further includes: The multiple interaction prediction models are jointly trained.

16. The method for predicting RNA-protein interactions according to claim 15, characterized in that, The joint training of the multiple interaction prediction models includes: Obtain a training dataset, which includes positive RNA-protein pairs and negative RNA-protein pairs; The multiple interaction prediction models are used to obtain multiple interaction prediction values ​​for each RNA-protein pair in the training dataset; The joint prediction value for each RNA-protein pair in the training dataset is obtained based on the multiple interaction prediction values; The loss function is used to calculate the joint predicted value and label value of each RNA-protein pair in the training dataset to obtain the corresponding loss value; The model parameters of the multiple interaction prediction models are adjusted based on the loss value.

17. The method for predicting RNA-protein interactions according to claim 16, characterized in that, The step of obtaining the joint prediction value for each RNA-protein pair in the training dataset based on the multiple interaction prediction values ​​includes: The weighted summation of multiple interaction predictions for each RNA-protein pair in the training dataset yields the joint prediction value for each RNA-protein pair.

18. The method for predicting RNA-protein interactions according to claim 17, characterized in that, The step of weighted summing of multiple interaction predictions for each RNA-protein pair in the training dataset to obtain a joint prediction value for each RNA-protein pair includes: according to: The joint prediction value for each RNA-protein pair in the training dataset was calculated. ;in, The output value of a traditional machine learning model. This represents the output value of the convolutional neural network model. This represents the output value of the recurrent neural network model. These are the weight parameters for traditional machine learning models, convolutional neural network models, and recurrent neural network models, respectively.

19. The method for predicting RNA-protein interactions according to claim 16, characterized in that, The step of adjusting the model parameters of the multiple interaction prediction models based on the loss value includes: Based on the loss value, the model parameters of the multiple interaction prediction models are iteratively updated. When the iteration termination condition is met, the training of the model parameters of the multiple interaction prediction models is completed, so as to use the trained multiple interaction prediction models to predict the interaction of the RNA-protein pair to be predicted.

20. The method for predicting RNA-protein interactions according to claim 1, characterized in that, The method further includes: Output the predicted results of the interaction between the RNA and the protein.

21. An RNA-protein interaction prediction device, characterized in that, include: The data acquisition module is used to acquire the RNA-protein pairs to be predicted; The feature extraction module is used to extract features from the RNA-protein pair to be predicted, and obtain the sequence features of the RNA-protein pair to be predicted. The data vectorization module is used to vectorize the RNA-protein pair to be predicted, and obtain the RNA sequence representation vector and the protein sequence representation vector in the RNA-protein pair to be predicted. An interaction prediction module is used to input the sequence features of the RNA-protein pair to be predicted into at least one conventional machine learning model to obtain at least one first interaction prediction value; and to input the RNA sequence representation vector in the RNA-protein pair to be predicted into a first sub-deep learning model to obtain a first sequence feature. The protein sequence representation vector in the RNA-protein pair to be predicted is input into the second sub-deep learning model to obtain the second sequence feature; The first sequence features and the second sequence features are fused, and the second interaction prediction value is obtained based on the fused features; wherein, the first sub-deep learning model and the second sub-deep learning model form the deep learning model; the traditional machine learning model and the deep learning model are obtained based on a joint learning approach; An interaction determination module is used to perform a weighted summation calculation on the multiple interaction prediction values ​​and determine the interaction between the RNA and the protein based on the calculation results.

22. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1-20.

23. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the method of any one of claims 1-20 by executing the executable instructions.

Images