Information processing method and device, electronic equipment and computer readable storage medium

Abstract

Embodiments of the present application provide an information processing method and device, electronic equipment and computer readable storage medium, and relate to the field of information processing. Embodiments of the present application obtain the first subsequence feature and the second subsequence feature of the target ribonucleic acid sequence; and through a first pre-training model, the first subsequence feature and the second subsequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first subsequence feature is determined based on a segment sequence containing one base, and the second subsequence feature is determined based on a segment sequence containing at least two bases, the first subsequence feature and the second subsequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, the sequence representation vector of the target ribonucleic acid sequence is determined through the first subsequence feature and the second subsequence feature, and the feature representation capability of the determined sequence representation vector is improved.

Description

Technical Field

[0001] This application relates to the field of information processing technology, and more specifically, to an information processing method, apparatus, electronic device, and computer-readable storage medium. Background Technology

[0002] Ribonucleic acid (RNA) is the carrier of genetic information found in biological cells and some viruses and viroids. RNA is a long, chain-like molecule formed by the condensation of ribonucleotides via phosphodiester bonds. A ribonucleotide molecule consists of a phosphate group, a sugar sugar, and a base. RNA primarily uses four bases: A (adenine), G (guanine), C (cytosine), and U (uracil). The main function of RNA in the body is to guide protein synthesis.

[0003] Currently, the problem of vector representation of proteins has been solved in related technologies. However, the vector representation capability of related solutions for RNA sequences needs to be improved. Summary of the Invention

[0004] The purpose of this application is to address at least one of the aforementioned technical deficiencies, particularly the technical deficiency of low vector representation capability.

[0005] According to one aspect of this application, an information processing method is provided, the method comprising:

[0006] Obtain the first and second subsequence features of the target ribonucleic acid sequence;

[0007] The first sub-sequence feature is determined based on the first sequence segment, and the second sub-sequence feature is determined based on the second sequence segment; the first sequence segment includes one base, the second sequence segment includes at least two bases, and the target ribonucleic acid sequence includes multiple bases;

[0008] The first pre-trained model processes the first sub-sequence features and the second sub-sequence features to obtain the sequence representation vector of the target ribonucleic acid sequence.

[0009] Optionally, the step of processing the first sub-sequence features and the second sub-sequence features using a first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence includes:

[0010] The first sub-sequence feature and the second sub-sequence feature are fused to obtain the fusion feature corresponding to the target ribonucleic acid sequence;

[0011] The fusion feature is input into the first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence; wherein, the first pre-trained model performs masking processing on the feature part corresponding to the target base in the fusion feature to obtain mask features, and predicts the target ribonucleic acid sequence based on the mask features.

[0012] Optionally, the method further includes:

[0013] The first pre-trained model is obtained by performing masked training on the preset natural language processing model.

[0014] Optionally, the step of performing masked training on a preset natural language processing model to obtain the first pre-trained model includes:

[0015] Obtain a training sample set, wherein each training sample in the training sample set includes a sample ribonucleic acid sequence;

[0016] For each of the sample ribonucleic acid sequences, obtain the sample fusion features corresponding to the sample ribonucleic acid sequence;

[0017] The feature portion corresponding to the target base in the sample fusion feature is masked to obtain the sample mask feature;

[0018] The sample mask features are input into the preset natural language processing model to obtain the predicted ribonucleic acid sequence;

[0019] The total training loss value is determined based on the sample ribonucleic acid sequence and the predicted ribonucleic acid sequence.

[0020] Based on the sample ribonucleic acid sequence and the corresponding total training loss value, the preset natural language processing model is repeatedly trained until the first pre-trained model that meets the training termination condition is obtained.

[0021] Optionally, inputting the sample mask features into the preset natural language processing model to obtain the predicted ribonucleic acid sequence includes:

[0022] The sample mask features are input into the preset natural language processing model to obtain the prediction result corresponding to each sample ribonucleic acid sequence. The prediction result includes the prediction probability of the prediction type corresponding to each sample target base in the sample ribonucleic acid sequence.

[0023] Based on the prediction results, the predicted ribonucleic acid sequence is determined.

[0024] Optionally, determining the total training loss value includes:

[0025] The first loss value is determined based on the input type of the target base in the sample and the prediction type in the prediction result;

[0026] A second loss value is determined based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence;

[0027] The total training loss value is determined based on the first loss value and the second loss value.

[0028] Optionally, determining the first loss value based on the input type of the target base in the sample and the prediction type in the prediction result includes:

[0029] The first loss value is obtained based on the input type of the target base of the sample, the prediction type in the prediction result, and the first data relationship;

[0030] The first data relationship is as follows:

[0031]

[0032] Where, loss seq This represents the first loss value;

[0033] p represents the matrix corresponding to the prediction type of the target base of the sample;

[0034] p[i] represents the matrix corresponding to the prediction type of the i-th target base in the sample;

[0035] N represents the product of the number of training samples and the length of the sample ribonucleic acid sequence;

[0036] y represents the matrix corresponding to the true type of the target base in the sample;

[0037] y[i] represents the index value of the i-th position in the y matrix.

[0038] Optionally, determining the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence includes:

[0039] The second loss value is determined based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence;

[0040] The first structural information is obtained by predicting the predicted ribonucleic acid sequence.

[0041] Optionally, determining the second loss value based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence includes:

[0042] The second loss value is obtained based on the original structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the second data relationship;

[0043] The second data relationship is as follows:

[0044]

[0045] Where, loss ss This represents the second loss value;

[0046] sgn(ρ, σ) is the sign function;

[0047] N represents the length of the sample's ribonucleic acid sequence;

[0048] c represents the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence;

[0049] This represents the original structural information of the i-th position in the sample's ribonucleic acid sequence;

[0050] This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

[0051] Optionally, determining the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence includes:

[0052] The second loss value is determined based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence;

[0053] The first structural information is obtained by predicting the predicted ribonucleic acid sequence.

[0054] The second structural information is obtained by predicting the ribonucleic acid sequence of the sample.

[0055] Optionally, determining the second loss value based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence includes:

[0056] The second loss value is obtained based on the second structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the third data relationship;

[0057] The third data relationship is as follows:

[0058]

[0059] Where, loss ss This represents the second loss value;

[0060] N represents the length of the sample's ribonucleic acid sequence;

[0061] This indicates the prediction reliability of the second structural information at the i-th position of the sample ribonucleic acid sequence;

[0062] This indicates the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence;

[0063] This represents the second structural information at the i-th position of the sample's ribonucleic acid sequence;

[0064] This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

[0065] Optionally, the method further includes:

[0066] Obtain the protein representation vector of the target protein;

[0067] The sequence representation vector and the protein representation vector are concatenated to obtain a concatenated vector.

[0068] The spliced ​​vector is input into the second pre-trained model to obtain the recognition and representation result, which indicates whether the sequence representation vector and the protein representation vector interact.

[0069] Optionally, the original structural information includes the original secondary structure information of the sample ribonucleic acid sequence;

[0070] The first structural information includes the predicted secondary structure information of the predicted ribonucleic acid sequence;

[0071] The second structural information includes the predicted secondary structure information of the sample ribonucleic acid sequence.

[0072] According to another aspect of this application, an information processing apparatus is provided, the apparatus comprising:

[0073] The acquisition module is used to acquire the first subsequence features and the second subsequence features of the target ribonucleic acid sequence;

[0074] The first sub-sequence feature is determined based on the first sequence segment, and the second sub-sequence feature is determined based on the second sequence segment; the first sequence segment includes one base, the second sequence segment includes at least two bases, and the target ribonucleic acid sequence includes multiple bases;

[0075] The processing module is used to process the first sub-sequence features and the second sub-sequence features through a first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence.

[0076] According to another aspect of this application, an electronic device is provided, the electronic device comprising:

[0077] One or more processors;

[0078] Memory;

[0079] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: perform the information processing method according to any one of the first aspects of this application.

[0080] For example, in a third aspect of this application, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus;

[0081] The memory is used to store at least one executable instruction that causes the processor to perform an operation corresponding to the information processing method shown in the first aspect of this application.

[0082] According to another aspect of this application, a computer-readable storage medium is provided, wherein the computer program, when executed by a processor, implements the information processing method described in any of the first aspects of this application.

[0083] For example, in a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the information processing method shown in the first aspect of the present application.

[0084] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in various alternative implementations of the first aspect described above.

[0085] The beneficial effects of the technical solution provided in this application are:

[0086] This application embodiment obtains a first sub-sequence feature and a second sub-sequence feature of the target ribonucleic acid sequence; through a first pre-trained model, the first sub-sequence feature and the second sub-sequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first sub-sequence feature is determined based on a segment sequence containing one base, and the second sub-sequence feature is determined based on a segment sequence containing at least two bases, the first sub-sequence feature and the second sub-sequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, by determining the sequence representation vector of the target ribonucleic acid sequence through the first sub-sequence feature and the second sub-sequence feature, the feature representation capability of the determined sequence representation vector is improved. Attached Figure Description

[0087] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0088] Figure 1 A system architecture diagram of an information processing method provided in an embodiment of this application;

[0089] Figure 2 This is a schematic diagram of the system structure of an information processing method provided in an embodiment of this application;

[0090] Figure 3 A flowchart illustrating an information processing method provided in an embodiment of this application;

[0091] Figure 4 This is a schematic diagram of the fusion processing of an information processing method provided in an embodiment of this application;

[0092] Figure 5 This is a schematic diagram of downstream task processing in an information processing method provided in an embodiment of this application;

[0093] Figure 6 A schematic diagram of the model structure of an information processing method provided in an embodiment of this application;

[0094] Figure 7 A training diagram illustrating an information processing method provided in an embodiment of this application;

[0095] Figure 8 This is a schematic diagram of the structure of an information processing device provided in an embodiment of this application;

[0096] Figure 9 This is a schematic diagram of the structure of an electronic device for information processing provided in an embodiment of this application. Detailed Implementation

[0097] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0098] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0099] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0100] At least some of the information processing methods provided in this application involve fields such as machine learning in the field of artificial intelligence, as well as various fields of cloud technology, such as cloud computing, cloud services, and related data computing and processing in the field of big data.

[0101] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0102] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0103] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close relationship with linguistic research. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0104] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0105] With the research and advancement of artificial intelligence (AI) technology, AI is being studied and applied in various fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, robots, smart healthcare, and smart customer service. It is believed that with the development of technology, AI will be applied in more fields and play an increasingly important role.

[0106] The information processing method in this application can be applied to fields such as biomedicine.

[0107] Specifically, the following is an explanation of the technical terms used in this application:

[0108] 1. RNA secondary structure: RNA molecules have multiple planar structures formed by various components such as single-stranded regions, stem-loop structures, and double-stranded structures composed of complementary base pairs. The structure formed by self-folding through these structures is the secondary structure of RNA.

[0109] Dot-bracket annotation is a symbolic string used in RNA to represent secondary structures. The symbols contained therein and their meanings are as follows:

[0110]

[0111]

[0112] 2. RNAfold: RNAfold is a method that finds the minimum free energy of RNA folding and compares sequence analysis. It is also a tool for predicting the secondary structure of RNA sequences. RNAfold can output the reliability score c of the predicted secondary structure at each position of the RNA sequence.

[0113] 3. Self-Supervised Pre-training: A commonly used pre-training method is masked language models. The basic idea is to learn the structural information hidden in a sequence by learning sequence completion. This involves manually masking some residues in the sequence and requiring the model to predict the types of these masked residues based on the remaining residues. The training data is unlabeled, requiring the learning algorithm to automatically learn valuable information from a large amount of unlabeled data. Common learning algorithms include Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs).

[0114] 4. k-mer subsequences: Since ribonucleic acid (RNA) sequences contain four bases, if a single base is used to represent the smallest unit of nucleic acid, there are four possible combinations, limiting the amount of information that can be represented. In bioinformatics, the k-mer analysis method is commonly used, which divides the RNA sequence into k-mer fragments. 'mer' represents a single base, and 'k-mer' represents a sequence segment composed of k adjacent bases, called a "word". k-mer is also known as k-word or k-tuple. K can take any positive integer from 1 to 3. For any value of K, there are a total of 4... k There are 16 types of k-mers, for example, when k=2.

[0115] To further illustrate the technical solutions provided in the embodiments of this application, a detailed description is provided below in conjunction with the accompanying drawings and specific implementation methods. Although the embodiments of this application provide method operation steps as shown in the following embodiments or drawings, the method may include more or fewer operation steps based on conventional or non-inventive methods. In steps where there is no logically necessary causal relationship, the execution order of these steps is not limited to the execution order provided in the embodiments of this application.

[0116] First, combine Figure 1 This is a system architecture diagram of the information processing method provided in the embodiments of this application. The system may include a server 101 and a terminal cluster, wherein the server 101 can be considered as a backend server for information processing.

[0117] The terminal cluster may include: terminal 102, terminal 103, terminal 104, ..., wherein each terminal may have a client installed that supports information processing. Communication connections may exist between the terminals; for example, there may be a communication connection between terminal 102 and terminal 103, and a communication connection between terminal 103 and terminal 104.

[0118] Meanwhile, server 101 can provide services to the terminal cluster through communication connection function. Any terminal in the terminal cluster can have a communication connection with server 101. For example, terminal 102 has a communication connection with server 101, and terminal 103 has a communication connection with server 101. The above-mentioned communication connection is not limited to the connection method. It can be directly or indirectly connected through wired communication, or directly or indirectly connected through wireless communication, or through other methods.

[0119] The network for the aforementioned communication connection can be a wide area network (WAN), a local area network (LAN), or a combination of both. This application does not impose any restrictions on this.

[0120] The information processing method of this application embodiment can be executed on the server side or the terminal side, and the execution subject is not limited in this application embodiment. In the information processing process, the first sub-sequence features and the second sub-sequence features of the target ribonucleic acid sequence can be obtained; through a first pre-trained model, the first sub-sequence features and the second sub-sequence features are processed to obtain the sequence representation vector of the target ribonucleic acid sequence.

[0121] Therefore, the method provided in this application embodiment can be executed by a computer device, which includes, but is not limited to, a terminal (including the aforementioned user terminal) or a server (including the aforementioned server 101). The aforementioned server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The aforementioned terminal can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0122] Of course, the methods provided in the embodiments of this application are not limited to... Figure 1 The application scenarios shown can also be used in other possible scenarios, and this application does not impose any limitations. Figure 1 The functions that each device in the application scenario shown can achieve will be described in subsequent method embodiments, and will not be elaborated on here.

[0123] As an optional implementation method, Figure 2 The diagram shows a framework structure of an information processing system provided in this application, such as... Figure 2 As shown, the system may include a feature encoding module, a training module (e.g., a BERT training module), and a downstream task module (e.g., an RNA-protein interaction module). Specifically, the feature encoding module is used to encode the RNA sequence. For example, it encodes the first and second sequences of the present application embodiment to obtain first subsequence features and second subsequence features. Then, the first subsequence features and second subsequence features are processed by a first pre-trained model trained by the BERT training module to obtain the embedding vector of the RNA sequence. Subsequently, the embedding vector of the RNA sequence can be applied to the downstream task module, i.e., to predict whether RNA and protein interact.

[0124] This application provides a possible implementation, which can be executed by any electronic device. Optionally, any electronic device can be a server device with information processing capabilities, or a device or chip integrated into these devices. Figure 3 As shown, this is one of the flowcharts of an information processing method provided in an embodiment of this application. The method includes the following steps:

[0125] Step S301: Obtain the first subsequence features and the second subsequence features of the target ribonucleic acid sequence.

[0126] The first sub-sequence feature is determined based on the first sequence segment, and the second sub-sequence feature is determined based on the second sequence segment; the first sequence segment includes one base, the second sequence segment includes at least two bases, and the target ribonucleic acid sequence includes multiple bases.

[0127] Specifically, the information processing method of this application embodiment can be applied to the field of bioinformatics to process target ribonucleic acid sequences. The target ribonucleic acid sequence, i.e., the target RNA sequence, can include any RNA sequence to be processed. For example, the target RNA sequence can include a target messenger RNA (mRNA) sequence, a target transfer RNA (tRNA) sequence, a target ribosomal RNA (rRNA) sequence, etc.

[0128] The first subsequence characteristic of the target ribonucleic acid (RNA) sequence is a characteristic determined based on the first sequence segment. The first sequence segment may include a single base from the target RNA sequence. Bases are nitrogen-containing compounds that form nucleosides, also known as nucleobases or nitrogenous bases. Bases are an important component of RNA sequences, and there are four main types of bases in RNA sequences: A (adenine), G (guanine), C (cytosine), and U (uracil).

[0129] Specifically, when obtaining the features of the first subsequence, you can first perform step one: divide the target ribonucleic acid sequence into multiple first segments, and then perform step two: extract the features of each first segment.

[0130] In step one, optionally, k-mer (k-word) analysis can be used to divide the target ribonucleic acid sequence into 1-mer segments, resulting in multiple 1-mer sequence fragments, each containing one base. As an example, let's take the target ribonucleic acid sequence L as an example, where sequence L is AUAUCCGGA… (A, G, C, U represent the base types: A for adenine, G for guanine, C for cytosine, and U for uracil), and sequence L is 1 unit in length. After dividing sequence L, multiple 1-mer sequence fragments are obtained, namely A, U, A, U, C, C… etc.

[0131] When performing step two, one-hot encoding or randomization can be used to vectorize each 1-mer segment sequence. For example, in conjunction with Example 1, the vector size of each 1-mer segment sequence of sequence L can be d1. Since the length of sequence L is l (sequence L contains l bases, which means it contains l 1-mer segments), the vector size of the first subsequence feature of sequence L, i.e., the 1-mer feature, is d1*l.

[0132] Furthermore, the second subsequence feature of the target ribonucleic acid sequence is a feature determined based on the second sequence segment. The second sequence segment may include at least two bases from the target ribonucleic acid sequence, wherein the bases in the second sequence segment may be consecutive bases in the target ribonucleic acid sequence; additionally, in some embodiments, the bases in the second sequence segment may be non-consecutive bases in the target ribonucleic acid sequence.

[0133] Specifically, when obtaining the features of the second subsequence, you can first perform step one: divide the target ribonucleic acid sequence into multiple second segments, and then perform step two: extract the features of each second segment.

[0134] In performing step one, optionally, the k-mer (k-word) analysis method can still be used to divide the target ribonucleic acid sequence into multiple k-mer sequence fragments (k is an integer greater than 1), that is, each k-mer sequence fragment contains k bases. Optionally, since the 3-mer sequence has good statistical characteristics, a certain degree of conservation and distribution consistency, and contains nucleic acid codon information, this embodiment of the application can divide the target ribonucleic acid sequence into multiple 3-mer sequence fragments, using the 3-mer sequence fragments as the second sequence fragment.

[0135] Taking Example 1 as an example, with the target ribonucleic acid sequence L, after dividing the sequence L: AUAUCCGGA...... into 3-mer segments, multiple 3-mer sequence fragments are obtained, namely AUA, UCC, GGA...... etc.

[0136] In step two, one-hot encoding or a randomization method can be optionally used to vectorize each 3-mer segment sequence. For example, referring to Example 1, the vector size of each 3-mer segment sequence of sequence L can be d2. Since the sequence length is l (sequence L contains l bases, which is l / 3 of the 3-mer segments), the vector size of the second subsequence feature of sequence L, i.e., the 3-mer feature, is d2*l / 3. Where d1 = d2.

[0137] It should be noted that if the sequence length l is not an integer multiple of 3, the missing positions in the vector corresponding to the last incomplete 3-mer segment can be filled with the value 0.

[0138] Step S302: Using the first pre-trained model, process the first sub-sequence features and the second sub-sequence features to obtain the sequence representation vector of the target ribonucleic acid sequence.

[0139] Specifically, in this embodiment of the application, the first pre-trained model can be trained by a self-supervised pre-training method. Optionally, the first pre-trained model can be trained by masking a preset natural language processing model. The preset natural language processing model can be a Bidirectional Encoder Representations from Transformers (BERT) model.

[0140] When processing information, it can be based on two sub-sequence features: the first sub-sequence feature and the second sub-sequence feature. Alternatively, the first sub-sequence feature and the second sub-sequence feature can be fused together, and the fused feature can be processed to obtain the sequence representation vector of the target ribonucleic acid sequence.

[0141] The sequence representation vector includes a vector representing the target ribonucleic acid sequence, i.e., the embedding vector of the target ribonucleic acid sequence.

[0142] This application embodiment obtains a first sub-sequence feature and a second sub-sequence feature of the target ribonucleic acid sequence; through a first pre-trained model, the first sub-sequence feature and the second sub-sequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first sub-sequence feature is determined based on a segment sequence containing one base, and the second sub-sequence feature is determined based on a segment sequence containing at least two bases, the first sub-sequence feature and the second sub-sequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, by determining the sequence representation vector of the target ribonucleic acid sequence through the first sub-sequence feature and the second sub-sequence feature, the feature representation capability of the determined sequence representation vector is improved.

[0143] In another embodiment of this application, the step of processing the first sub-sequence features and the second sub-sequence features using a first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence includes:

[0144] The first sub-sequence feature and the second sub-sequence feature are fused to obtain the fusion feature corresponding to the target ribonucleic acid sequence;

[0145] The fusion feature is input into the first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence; wherein, the first pre-trained model performs masking processing on the feature part corresponding to the target base in the fusion feature to obtain mask features, and predicts the target ribonucleic acid sequence based on the mask features, thereby obtaining the sequence representation vector of the target ribonucleic acid sequence.

[0146] Specifically, since the 3-mer segment sequence has good statistical characteristics, exhibits a certain degree of conservation and distribution consistency, and contains nucleic acid codon information, in this embodiment, the characteristics of the 3-mer segment sequence can be used as the second subsequence characteristics. Combined with... Figure 4 As shown, in order to enhance the feature representation capability of the sequence representation vector, the features of the first subsequence, such as 1-mer features, and the features of the second subsequence, such as 3-mer features, can be fused to obtain a fused feature that combines the features of the first and second sequences. The fused feature is then processed by the first pre-trained model to obtain a sequence representation vector with stronger feature representation capability.

[0147] Optionally, the fusion process may include adding the first sub-sequence features and the second sub-sequence features together.

[0148] In the process of predicting the sequence representation vector of the target ribonucleic acid sequence, this application embodiment improves the feature representation capability of the sequence representation vector by fusing the first sub-sequence features, such as 1-mer features, and the second sub-sequence features, such as 3-mer features.

[0149] In another embodiment of this application, the first pre-trained model can be obtained by performing masked training on the preset natural language processing model, which may specifically include:

[0150] Obtain a training sample set, wherein each training sample in the training sample set includes a sample ribonucleic acid sequence;

[0151] For each of the sample ribonucleic acid sequences, obtain the sample fusion features corresponding to the sample ribonucleic acid sequence;

[0152] The feature portion corresponding to the target base in the sample fusion feature is masked to obtain the sample mask feature;

[0153] The sample mask features are input into the preset natural language processing model to obtain the predicted ribonucleic acid sequence;

[0154] The total training loss value is determined based on the sample ribonucleic acid sequence and the predicted ribonucleic acid sequence.

[0155] Based on the sample ribonucleic acid sequence and the corresponding total training loss value, the preset natural language processing model is repeatedly trained until the first pre-trained model that meets the training termination condition is obtained.

[0156] Specifically, the first pre-trained model in this application embodiment can be obtained by masking a preset natural language processing model, such as the BERT model.

[0157] The RNA sequences in the training sample set can include any RNA sequence, and the training sample set can be obtained from a pre-defined RNA sequence database.

[0158] During training, the feature portion corresponding to the target base in the sample ribonucleic acid sequence (i.e., the sample fusion feature) can be masked to obtain sample mask features. Furthermore, by inputting the sample mask features into the preset natural language processing model, a predicted ribonucleic acid sequence is obtained.

[0159] In one embodiment of this application, the step of inputting the sample mask features into the preset natural language processing model to obtain the predicted ribonucleic acid sequence includes:

[0160] The sample mask features are input into the preset natural language processing model to obtain the prediction result corresponding to each sample ribonucleic acid sequence. The prediction result includes the prediction probability of the prediction type corresponding to each sample target base in the sample ribonucleic acid sequence.

[0161] Based on the prediction results, the predicted ribonucleic acid sequence is determined.

[0162] Specifically, in this embodiment, the preset natural language processing model predicts the type of the target base corresponding to the masked feature portion of the sample using sample mask features, obtaining the prediction probability of the prediction type corresponding to each target base. Based on the prediction probability of the prediction type corresponding to each target base, the prediction type of each target base is determined. Optionally, the prediction type with the highest prediction probability can be taken as the prediction type corresponding to the target base.

[0163] Furthermore, based on the prediction type corresponding to the determined target base of the sample, the predicted ribonucleic acid sequence is determined, i.e., the predicted ribonucleic acid sequence.

[0164] After determining the predicted ribonucleic acid sequence, the total training loss value is determined based on the sample ribonucleic acid sequence and the predicted ribonucleic acid sequence.

[0165] Furthermore, based on the sample ribonucleic acid sequence and the corresponding total training loss value, the preset natural language processing model is repeatedly trained. If the preset training termination condition is met, the training ends, and the preset natural language processing model at the end of training is used as the first pre-trained model. If the training termination condition is not met, the model parameters of the preset natural language processing model are adjusted, and the adjusted model is trained again based on each training sample until the preset training termination condition is met.

[0166] In another embodiment of this application, determining the total training loss value includes:

[0167] The first loss value is determined based on the input type of the target base in the sample and the prediction type in the prediction result;

[0168] A second loss value is determined based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence;

[0169] The total training loss value is determined based on the first loss value and the second loss value.

[0170] In order to enable the first pre-trained model to learn the sequence features of the RNA sequence (the features of the bases contained in the RNA sequence) and the structural information features of the RNA sequence, the total training loss value of the preset natural language processing model in this embodiment of the application may include two parts, namely the first loss value and the second loss value.

[0171] The first loss value can be determined based on the input type of the target base of the sample and the prediction type in the prediction result; wherein the input type of the target base of the sample can be the actual type of the target base of the sample; the input type of the target base of the sample and the prediction type in the prediction result can each include one of A, G, C, and U.

[0172] The second loss value can be determined based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence; wherein, the structural information may include the secondary structure information of the RNA sequence.

[0173] Optionally, the data relationship between the total loss value and the first loss value and the second loss value can be specifically expressed as follows:

[0174] Loss = w1loss seq +w2loss ss

[0175] Where Loss represents the total loss value; loss seq This represents the first loss value; loss ss w1 and w2 represent the second loss value; w1 and w2 represent the weight coefficients, which can be determined by continuous adjustment during training.

[0176] Optionally, in another embodiment of this application, determining the first loss value based on the input type of the target base of the sample and the prediction type in the prediction result includes:

[0177] The first loss value is obtained based on the input type of the target base of the sample, the prediction type in the prediction result, and the first data relationship;

[0178] The first data relationship is as follows:

[0179]

[0180] Where, loss seq This represents the first loss value;

[0181] p represents the matrix corresponding to the prediction type of the target base of the sample;

[0182] p[i] represents the matrix corresponding to the prediction type of the i-th target base in the sample;

[0183] N represents the product of the number of training samples and the length of the sample ribonucleic acid sequence;

[0184] y represents the matrix corresponding to the true type of the target base in the sample;

[0185] y[i] represents the index value of the i-th position in the y matrix.

[0186] Specifically, the first loss value seq The cross-entropy loss of the multi-class classification of the preset natural language processing model can be determined. In the first data relationship, p represents the matrix corresponding to the predicted type of the target base of the sample. The size of matrix p is N*C, where N represents the product of the number of training samples and the length of the sample ribonucleic acid sequence, optionally, the number of training samples can be 1; C represents the number of types of sample fusion features. y represents the matrix corresponding to the true type of the target base of the sample; the size of matrix y is N. y[i] represents the index value of the i-th position in matrix y; the value range is 0≤y[i]≤C-1.

[0187] It is understandable that when the actual type of the target base in the sample is consistent with the predicted type of the target base in the sample, the first loss value is smaller; conversely, when the actual type of the target base in the sample is inconsistent with the predicted type of the target base in the sample, the first loss value is larger.

[0188] In another embodiment of this application, determining the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence includes:

[0189] The second loss value is determined based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence;

[0190] The first structural information is obtained by predicting the predicted ribonucleic acid sequence.

[0191] Specifically, the original structural information may include the actual structural information of the sample ribonucleic acid sequence, and the first structural information may be the structural information obtained by predicting the predicted ribonucleic acid sequence.

[0192] In this embodiment of the application, the original structural information may include the original secondary structure information of the sample ribonucleic acid sequence, that is, the actual secondary structure information of the sample ribonucleic acid sequence; the first structural information includes the predicted secondary structure information of the predicted ribonucleic acid sequence.

[0193] Optionally, the original structural information can be determined based on the RNA secondary structure information obtained from experimental analysis in the URSDB database, where the RNA secondary structure information can be represented by σ, where σ∈{“.”, “(”, “)”, “[”, “]”, “{”, “}”, “<”, “>”, “-”, “*”}.

[0194] The first structural information can be determined by predicting the secondary structure of the predicted ribonucleic acid sequence using RNAfold. The first structural information can be represented by (ρ, c), where ρ ∈ {".", "(", ")", "[", "]", "{", "}", "<", ">", "-", "*"}, and c represents the prediction reliability of the secondary structure at the corresponding site in the RNA sequence, c ∈ (0, 1); the larger the value of c, the higher the prediction reliability.

[0195] Understandably, when determining the second loss value, for locations with high secondary structure prediction reliability, the higher the similarity between the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence, the smaller the second loss value; conversely, the lower the similarity between the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence, the larger the second loss value. Furthermore, for locations with low secondary structure prediction reliability, the similarity between the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence has a relatively small impact on the second loss value.

[0196] Optionally, in another embodiment of this application, determining the second loss value based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence includes:

[0197] The second loss value is obtained based on the original structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the second data relationship;

[0198] The second data relationship is as follows:

[0199]

[0200] Where, loss ss This represents the second loss value;

[0201] sgn(ρ, σ) is the sign function;

[0202] N represents the length of the sample's ribonucleic acid sequence;

[0203] c represents the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence;

[0204] This represents the original structural information of the i-th position in the sample's ribonucleic acid sequence;

[0205] This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

[0206] In another embodiment of this application, determining the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence includes:

[0207] The second loss value is determined based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence;

[0208] The first structural information is obtained by predicting the predicted ribonucleic acid sequence.

[0209] The second structural information is obtained by predicting the ribonucleic acid sequence of the sample.

[0210] In another embodiment of this application, the second loss value can also be determined based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence.

[0211] The first structural information may include the predicted secondary structure information of the predicted ribonucleic acid sequence; the second structural information may include the predicted secondary structure information of the sample ribonucleic acid sequence.

[0212] Optionally, the first structural information and the second structural information can be determined by RNAfold to predict the secondary structure of the predicted ribonucleic acid sequence and the secondary structure of the sample ribonucleic acid sequence, respectively.

[0213] Optionally, in another embodiment of this application, determining the second loss value based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence includes:

[0214] The second loss value is obtained based on the second structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the third data relationship;

[0215] The third data relationship is as follows:

[0216]

[0217] Where, loss ss This represents the second loss value;

[0218] N represents the length of the sample's ribonucleic acid sequence;

[0219] This indicates the prediction reliability of the second structural information at the i-th position of the sample ribonucleic acid sequence;

[0220] This indicates the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence;

[0221] This represents the second structural information at the i-th position of the sample's ribonucleic acid sequence;

[0222] This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

[0223] This application embodiment determines the first loss value, namely sequence loss, and the second loss value, namely secondary structure loss, and then determines the total loss value, thereby enabling the preset natural language processing model to learn the sequence features and secondary structure features of the RNA sequence.

[0224] In another embodiment of this application, the method further includes:

[0225] Obtain the protein representation vector of the target protein;

[0226] The sequence representation vector and the protein representation vector are concatenated to obtain a concatenated vector.

[0227] The spliced ​​vector is input into the second pre-trained model to obtain the recognition and representation result, which indicates whether the sequence representation vector and the protein representation vector interact.

[0228] Specifically, the protein representation vector may include a vector representing the target protein, i.e., the embedding vector of the target protein.

[0229] Combination Figure 5 As shown, to verify the representational power of the sequence representation vector, this embodiment may further include a downstream task to verify the representational power of the sequence representation vector, namely, inputting the concatenated vector obtained by concatenating the sequence representation vector (RNA embedding vector in the figure) and the protein representation vector (protein embedding vector in the figure) into a second pre-trained model, such as the feedforward neural network in the figure, to predict whether the target ribonucleic acid sequence corresponding to the sequence representation vector interacts with the target protein, and comparing the prediction result with the known true value, calculating indicators such as prediction accuracy, thereby judging the representational power of the sequence representation vector output by the first pre-trained model.

[0230] The following is combined with Figure 6 and Figure 7 The embodiments of this application will be described in general as follows:

[0231] Taking RNA sequence M: AUCGAUCG...... as an example, when determining the embedding vector of sequence M, i.e., the sequence representation vector in this embodiment, sequence M can be divided into 1-mer and 3-mer segments, and the 1-mer and 3-mer features of sequence M can be obtained respectively through one-hot encoding methods, i.e., the first sub-sequence features and the second sub-sequence features in this embodiment. For example Figure 6 As shown, the 1-mer features of sequence M can be obtained through the features h0, h1, ... h1 corresponding to each 1-mer segment of the sequence. n Determined; the 3-mer features of sequence M can be obtained through the features h′0, h′1, ..., h′ corresponding to each 3-mer segment sequence. m Sure.

[0232] Then, by fusing 1-mer features and 3-mer features, that is, by fusing the features h0, h1, ... h0 corresponding to each 1-mer segment sequence. n Features h′0, h′1, ..., h′ corresponding to the 3-mer segment sequence m Adding them together yields the fusion characteristics.

[0233] The fused features are then input into the first training model obtained through masking training of the BERT model to determine the embedding vector of sequence M. When training the BERT model using a mask, the total loss of the model can be composed of two parts: sequence loss and sequence loss. seq and secondary structure loss ss .

[0234] The training process can be combined with Figure 7 As shown, the fused features are input into the feedforward network of the BERT model for masked training to obtain the predicted RNA sequence. Then, RNAfold is used to predict the secondary structure of both the predicted RNA sequence and the original sequence M, obtaining the secondary structure information of both. The sequence loss can then be calculated. seq and secondary structure loss ss And it is trained and adjusted through the BERT model feedback network.

[0235] After obtaining the embedding vector of sequence M, it can be applied to a downstream task: RNA-protein interaction prediction. This is achieved by concatenating the embedding vectors of sequence M and the protein, and then inputting the concatenated vector into a second pre-trained model to predict whether the two interact.

[0236] This application embodiment obtains a first sub-sequence feature and a second sub-sequence feature of the target ribonucleic acid sequence; through a first pre-trained model, the first sub-sequence feature and the second sub-sequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first sub-sequence feature is determined based on a segment sequence containing one base, and the second sub-sequence feature is determined based on a segment sequence containing at least two bases, the first sub-sequence feature and the second sub-sequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, by determining the sequence representation vector of the target ribonucleic acid sequence through the first sub-sequence feature and the second sub-sequence feature, the feature representation capability of the determined sequence representation vector is improved.

[0237] This application provides an information processing device, such as... Figure 8 As shown, the information processing device 80 may include: an acquisition module 801 and a processing module 802, wherein,

[0238] The acquisition module 801 is used to acquire the first subsequence features and the second subsequence features of the target ribonucleic acid sequence;

[0239] The first sub-sequence feature is determined based on the first sequence segment, and the second sub-sequence feature is determined based on the second sequence segment; the first sequence segment includes one base, the second sequence segment includes at least two bases, and the target ribonucleic acid sequence includes multiple bases;

[0240] The processing module 802 is used to process the first sub-sequence features and the second sub-sequence features through a first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence.

[0241] In another embodiment of this application, the processing module is specifically used to perform fusion processing on the first sub-sequence feature and the second sub-sequence feature to obtain the fusion feature corresponding to the target ribonucleic acid sequence;

[0242] The fusion feature is input into the first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence; wherein, the first pre-trained model performs masking processing on the feature part corresponding to the target base in the fusion feature to obtain mask features, and predicts the target ribonucleic acid sequence based on the mask features.

[0243] In another embodiment of this application, the device further includes:

[0244] The training module is used to perform mask training on a preset natural language processing model to obtain the first pre-trained model.

[0245] In another embodiment of this application, the training module is specifically used to obtain a training sample set, wherein each training sample in the training sample set includes a sample ribonucleic acid sequence;

[0246] For each of the sample ribonucleic acid sequences, obtain the sample fusion features corresponding to the sample ribonucleic acid sequence;

[0247] The feature portion corresponding to the target base in the sample fusion feature is masked to obtain the sample mask feature;

[0248] The sample mask features are input into the preset natural language processing model to obtain the predicted ribonucleic acid sequence;

[0249] The total training loss value is determined based on the sample ribonucleic acid sequence and the predicted ribonucleic acid sequence.

[0250] Based on the sample ribonucleic acid sequence and the corresponding total training loss value, the preset natural language processing model is repeatedly trained until the first pre-trained model that meets the training termination condition is obtained.

[0251] In another embodiment of this application, the training module is specifically used to input the sample mask features into the preset natural language processing model to obtain the prediction result corresponding to each sample ribonucleic acid sequence. The prediction result includes: the prediction probability of the prediction type corresponding to each sample target base in the sample ribonucleic acid sequence.

[0252] Based on the prediction results, the predicted ribonucleic acid sequence is determined.

[0253] In another embodiment of this application, the training module is specifically used to determine a first loss value based on the input type of the target base of the sample and the prediction type in the prediction result;

[0254] A second loss value is determined based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence;

[0255] The total training loss value is determined based on the first loss value and the second loss value.

[0256] In another embodiment of this application, the training module is specifically used to obtain the first loss value based on the input type of the target base of the sample, the prediction type in the prediction result, and the first data relationship;

[0257] The first data relationship is as follows:

[0258]

[0259] Where, loss seq This represents the first loss value;

[0260] p represents the matrix corresponding to the prediction type of the target base of the sample;

[0261] p[i] represents the matrix corresponding to the prediction type of the i-th target base in the sample;

[0262] N represents the product of the number of training samples and the length of the sample ribonucleic acid sequence;

[0263] y represents the matrix corresponding to the true type of the target base in the sample;

[0264] y[i] represents the index value of the i-th position in the y matrix.

[0265] In another embodiment of this application, the training module is specifically used to determine a second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence, including:

[0266] The second loss value is determined based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence;

[0267] The first structural information is obtained by predicting the predicted ribonucleic acid sequence.

[0268] In another embodiment of this application, the training module is specifically used to determine the second loss value based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence, including:

[0269] The second loss value is obtained based on the original structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the second data relationship;

[0270] The second data relationship is as follows:

[0271]

[0272] Where, loss ss This represents the second loss value;

[0273] sgn(ρ, σ) is the sign function;

[0274] N represents the length of the sample's ribonucleic acid sequence;

[0275] c represents the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence;

[0276] This represents the original structural information of the i-th position in the sample's ribonucleic acid sequence;

[0277] This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

[0278] In another embodiment of this application, the training module is specifically used to determine the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence, including:

[0279] The second loss value is determined based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence;

[0280] The first structural information is obtained by predicting the predicted ribonucleic acid sequence.

[0281] The second structural information is obtained by predicting the ribonucleic acid sequence of the sample.

[0282] In another embodiment of this application, the training module is specifically used to determine the second loss value based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence, including:

[0283] The second loss value is obtained based on the second structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the third data relationship;

[0284] The third data relationship is as follows:

[0285]

[0286] Where, loss ss This represents the second loss value;

[0287] N represents the length of the sample's ribonucleic acid sequence;

[0288] This indicates the prediction reliability of the second structural information at the i-th position of the sample ribonucleic acid sequence;

[0289] This indicates the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence;

[0290] This represents the second structural information at the i-th position of the sample's ribonucleic acid sequence;

[0291] This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

[0292] In another embodiment of this application, the device further includes:

[0293] The verification task module is used to obtain the protein representation vector of the target protein;

[0294] The sequence representation vector and the protein representation vector are concatenated to obtain a concatenated vector.

[0295] The spliced ​​vector is input into the second pre-trained model to obtain the recognition and representation result, which indicates whether the sequence representation vector and the protein representation vector interact.

[0296] In another embodiment of this application, the original structural information includes the original secondary structure information of the sample ribonucleic acid sequence;

[0297] The first structural information includes the predicted secondary structure information of the predicted ribonucleic acid sequence;

[0298] The second structural information includes the predicted secondary structure information of the sample ribonucleic acid sequence.

[0299] The information processing apparatus of this embodiment can execute the information processing apparatus method shown in the above embodiments of this application, and its implementation principle is similar, so it will not be described again here.

[0300] This application embodiment obtains a first sub-sequence feature and a second sub-sequence feature of the target ribonucleic acid sequence; through a first pre-trained model, the first sub-sequence feature and the second sub-sequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first sub-sequence feature is determined based on a segment sequence containing one base, and the second sub-sequence feature is determined based on a segment sequence containing at least two bases, the first sub-sequence feature and the second sub-sequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, by determining the sequence representation vector of the target ribonucleic acid sequence through the first sub-sequence feature and the second sub-sequence feature, the feature representation capability of the determined sequence representation vector is improved.

[0301] This application provides an electronic device, which includes a memory and a processor; at least one program stored in the memory, which, when executed by the processor, can achieve the following compared to the prior art: This application obtains a first sub-sequence feature and a second sub-sequence feature of a target ribonucleic acid sequence; through a first pre-trained model, the first sub-sequence feature and the second sub-sequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first sub-sequence feature is determined based on a segment sequence containing one base, and the second sub-sequence feature is determined based on a segment sequence containing at least two bases, the first sub-sequence feature and the second sub-sequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, by determining the sequence representation vector of the target ribonucleic acid sequence through the first sub-sequence feature and the second sub-sequence feature, the feature representation capability of the determined sequence representation vector is improved.

[0302] In one alternative embodiment, an electronic device is provided, such as Figure 9 As shown, Figure 9 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this application.

[0303] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0304] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0305] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0306] The memory 4003 stores application code (computer program) that executes the solution of this application, and its execution is controlled by the processor 4001. The processor 4001 executes the application code stored in the memory 4003 to implement the content shown in the foregoing method embodiments.

[0307] Electronic devices include, but are not limited to: mobile phones, laptops, multimedia players, desktop computers, etc.

[0308] This application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0309] This application embodiment obtains a first sub-sequence feature and a second sub-sequence feature of the target ribonucleic acid sequence; through a first pre-trained model, the first sub-sequence feature and the second sub-sequence feature are processed to obtain a sequence representation vector of the target ribonucleic acid sequence. Since the first sub-sequence feature is determined based on a segment sequence containing one base, and the second sub-sequence feature is determined based on a segment sequence containing at least two bases, the first sub-sequence feature and the second sub-sequence feature respectively reflect different characteristics of the target ribonucleic acid sequence. Thus, by determining the sequence representation vector of the target ribonucleic acid sequence through the first sub-sequence feature and the second sub-sequence feature, the feature representation capability of the determined sequence representation vector is improved.

[0310] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0311] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An information processing method, characterized in that, include: Obtain the first and second subsequence features of the target ribonucleic acid sequence; The first sub-sequence feature is determined based on the first sequence segment, and the second sub-sequence feature is determined based on the second sequence segment; the first sequence segment includes one base, the second sequence segment includes at least two bases, and the target ribonucleic acid sequence includes multiple bases; The first pre-trained model is used to process the first sub-sequence features and the second sub-sequence features to obtain the sequence representation vector of the target ribonucleic acid sequence. The method further includes: The first pre-trained model is obtained by performing masked training on the preset natural language processing model. The step of performing masked training on a preset natural language processing model to obtain the first pre-trained model includes: Obtain a training sample set, wherein each training sample in the training sample set includes a sample ribonucleic acid sequence; For each of the sample ribonucleic acid sequences, obtain the sample fusion features corresponding to the sample ribonucleic acid sequence; The feature portion corresponding to the target base in the sample fusion feature is masked to obtain the sample mask feature; The sample mask features are input into the preset natural language processing model to obtain the predicted ribonucleic acid sequence; The total training loss value is determined based on the sample ribonucleic acid sequence and the predicted ribonucleic acid sequence. Based on the sample ribonucleic acid sequence and the total training loss value corresponding to the sample ribonucleic acid sequence, the preset natural language processing model is repeatedly trained until the first pre-trained model that meets the training termination condition is obtained. Determining the total training loss value includes: The first loss value is determined based on the input type of the target base in the sample and the prediction type in the prediction result; A second loss value is determined based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence; The total training loss value is determined based on the first loss value and the second loss value; The step of determining the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence includes: The second loss value is determined based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence; The first structural information is obtained by predicting the predicted ribonucleic acid sequence; Determining the second loss value based on the original structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence includes: The second loss value is obtained based on the original structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the second data relationship; The second data relationship is as follows: ； in, This represents the second loss value; It is a symbolic function; ; N represents the length of the sample ribonucleic acid sequence; C represents the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence. This represents the original structural information of the i-th position in the sample's ribonucleic acid sequence; This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

2. The information processing method according to claim 1, characterized in that, The step of processing the first sub-sequence features and the second sub-sequence features using a first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence includes: The first sub-sequence feature and the second sub-sequence feature are fused to obtain the fusion feature corresponding to the target ribonucleic acid sequence; The fusion feature is input into the first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence; wherein, the first pre-trained model performs masking processing on the feature part corresponding to the target base in the fusion feature to obtain mask features, and predicts the target ribonucleic acid sequence based on the mask features.

3. The information processing method according to claim 1, characterized in that, The step of inputting the sample mask features into the preset natural language processing model to obtain the predicted ribonucleic acid sequence includes: The sample mask features are input into the preset natural language processing model to obtain the prediction result corresponding to each sample ribonucleic acid sequence. The prediction result includes the prediction probability of the prediction type corresponding to each sample target base in the sample ribonucleic acid sequence. Based on the prediction results, the predicted ribonucleic acid sequence is determined.

4. The information processing method according to claim 1, characterized in that, The step of determining the first loss value based on the input type of the target base in the sample and the prediction type in the prediction result includes: The first loss value is obtained based on the input type of the target base of the sample, the prediction type in the prediction result, and the first data relationship; The first data relationship is as follows: ; in, This represents the first loss value; p represents the matrix corresponding to the prediction type of the target base of the sample; The matrix represents the prediction type corresponding to the i-th target base of the sample; N represents the product of the number of training samples and the length of the sample ribonucleic acid sequence; y represents the matrix corresponding to the true type of the target base in the sample; This represents the index value of the i-th position in the y matrix.

5. The information processing method according to claim 1, characterized in that, The step of determining the second loss value based on the structural information of the sample ribonucleic acid sequence and the structural information of the predicted ribonucleic acid sequence includes: The second loss value is determined based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence; The first structural information is obtained by predicting the predicted ribonucleic acid sequence. The second structural information is obtained by predicting the ribonucleic acid sequence of the sample.

6. The information processing method according to claim 5, characterized in that, Determining the second loss value based on the second structural information of the sample ribonucleic acid sequence and the first structural information of the predicted ribonucleic acid sequence includes: The second loss value is obtained based on the second structural information of the sample ribonucleic acid sequence, the first structural information of the predicted ribonucleic acid sequence, and the third data relationship; The third data relationship is as follows: ； in, This represents the second loss value; N represents the length of the sample's ribonucleic acid sequence; This indicates the prediction reliability of the second structural information at the i-th position of the sample ribonucleic acid sequence; This indicates the prediction reliability of the first structural information at the i-th position of the predicted ribonucleic acid sequence; This represents the second structural information at the i-th position of the sample's ribonucleic acid sequence; This represents the first structural information at the i-th position of the predicted ribonucleic acid sequence.

7. The information processing method according to any one of claims 1 to 6, characterized in that, The method further includes: Obtain the protein representation vector of the target protein; The sequence representation vector and the protein representation vector are concatenated to obtain a concatenated vector. The spliced ​​vector is input into the second pre-trained model to obtain the recognition and representation result, which indicates whether the sequence representation vector and the protein representation vector interact.

8. The information processing method according to any one of claims 5 to 6, characterized in that, The original structural information includes the original secondary structure information of the sample ribonucleic acid sequence; The first structural information includes the predicted secondary structure information of the predicted ribonucleic acid sequence; The second structural information includes the predicted secondary structure information of the sample ribonucleic acid sequence.

9. An information processing apparatus, applied to the method according to any one of claims 1-8, characterized in that, include: The acquisition module is used to acquire the first subsequence features and the second subsequence features of the target ribonucleic acid sequence; The first sub-sequence feature is determined based on the first sequence segment, and the second sub-sequence feature is determined based on the second sequence segment; the first sequence segment includes one base, the second sequence segment includes at least two bases, and the target ribonucleic acid sequence includes multiple bases; The processing module is used to process the first sub-sequence features and the second sub-sequence features through a first pre-trained model to obtain the sequence representation vector of the target ribonucleic acid sequence.

10. An electronic device, characterized in that, The electronic device includes: One or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the one or more processors, the one or more applications being configured to: perform the information processing method according to any one of claims 1 to 8.

11. 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 information processing method according to any one of claims 1 to 8.

Images