Construction and training method, device and screening method of siRNA gene inhibition efficiency screening model

Abstract

The present application provides a construction and training method, device and screening method for a siRNA gene inhibition efficiency screening model. The construction and training method includes obtaining the first sequence information of the target mRNA and the second sequence information of the siRNA; obtaining the first spatial image information of the target mRNA and the second spatial image information of the siRNA; constructing the siRNA gene inhibition efficiency screening model, using the first sequence information and the second sequence information to train the first neural network model to extract sequence features; using the first spatial image information and the second spatial image information to train the second neural network model to extract spatial structure features; fusing the sequence features and the spatial structure features, and using the fused fusion features to train the third neural network model. In this way, the accuracy of screening siRNAs with inhibition efficiency on the targeted gene can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of siRNA design, and specifically to the construction and training method, device and screening method of a siRNA gene inhibition efficiency screening model. Background Art

[0002] siRNA can regulate gene expression through RNA interference (RNAi), and is used to treat diseases such as hypercholesterolemia, hypertension, atherosclerosis, and viral infections, holding great promise in drug development. Currently, siRNA design is primarily done manually by researchers based on their experience and the target gene they wish to silence. However, due to the limitations of these researchers' experience, it's easy to overlook siRNAs with high inhibitory efficiency against the target gene.

[0003] Furthermore, currently, siRNA screening relies heavily on experimental validation methods. For example, Western blotting is used to determine the inhibitory efficiency of siRNAs against their target genes, resulting in low screening efficiency. Therefore, establishing an efficient and accurate screening method from siRNA design to screening for siRNAs with inhibitory efficacy against their target genes has become an urgent challenge. Summary of the Invention

[0004] The present application addresses the above-mentioned technical problems existing in the prior art. The present application aims to provide a method, apparatus, and screening method for constructing and training a siRNA gene inhibition efficiency screening model, which can improve the efficiency of screening siRNAs with targeted gene inhibition efficiency, comprehensively screen siRNAs, and improve both screening efficiency and screening accuracy to ensure that the screened siRNAs have high targeted gene inhibition efficiency.

[0005] According to a first embodiment of the present application, a method for constructing and training a siRNA gene inhibition efficiency screening model is provided, the method comprising: obtaining first sequence information of a target mRNA and second sequence information of a siRNA obtained based on the target mRNA; obtaining first spatial image information of the target mRNA and second spatial image information of the siRNA, the first spatial image information and the second spatial image information being used to characterize the spatial structures of the target mRNA and the siRNA, respectively; constructing a siRNA gene inhibition efficiency screening model, the siRNA gene inhibition efficiency screening model comprising a first neural network model, a second neural network model, and a third neural network model; training the first neural network model using the first sequence information and the second sequence information to extract sequence features; training the second neural network model using the first spatial image information and the second spatial image information to extract spatial structure features; fusing the sequence features and the spatial structure features, and training the third neural network model using the fused features, so that the screening accuracy of the siRNA gene inhibition efficiency screening model is higher than a threshold.

[0006] According to a second embodiment of the present application, a device for constructing and training a siRNA gene inhibition efficiency screening model is provided, comprising an interface configured to: obtain first sequence information of a target mRNA and second sequence information of a siRNA obtained based on the target mRNA; obtain first spatial image information of the target mRNA and second spatial image information of the siRNA, wherein the first spatial image information and the second spatial image information are used to characterize the spatial structures of the target mRNA and the siRNA, respectively; and at least one processor configured to implement the method for constructing and training the siRNA gene inhibition efficiency screening model as described in each embodiment of the present application.

[0007] According to the third scheme of the present application, a screening method for screening siRNA with targeted gene inhibition efficiency is provided, the screening method comprising: based on the principle of base complementarity, traversing the target mRNA with a preset wandering interval length and a preset wandering coverage interval length to obtain a siRNA sequence library combined with the target mRNA sequence; using the siRNA gene inhibition efficiency screening model construction and training device described in the embodiment of the present application to predict the targeted gene inhibition efficiency of the siRNA in the siRNA sequence library to screen out siRNAs whose targeted gene inhibition efficiency is greater than the set inhibition efficiency.

[0008] According to the fourth embodiment of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the construction and training method of the siRNA gene inhibition efficiency screening model as described in the various embodiments of the present application.

[0009] According to the fifth embodiment of the present application, a computer program product is provided, comprising computer instructions for causing a computer to execute the method for constructing and training a siRNA gene inhibition efficiency screening model as described in the various embodiments of the present application.

[0010] Compared with the prior art, the embodiments of the present application have the following advantages:

[0011] The construction and training method of the siRNA gene inhibition efficiency screening model provided in the embodiment of the present application is based on the first sequence information of the target mRNA and the second sequence information of the siRNA, and the first neural network model is trained to extract sequence features. Based on the first spatial image information of the target mRNA and the second spatial image information of the siRNA, the second neural network model is trained to extract spatial structure features. After the sequence features and the spatial structure features are fused, the third neural network model is trained to achieve the prediction of the inhibition efficiency of the targeted gene. In this way, it is possible to automatically and efficiently screen siRNAs with targeted inhibition efficiency using a neural network model, greatly improving the screening efficiency of siRNAs, thereby reducing the time for siRNA drug development. Moreover, in the process of intelligent screening of siRNAs, the sequence information and spatial image information of RNA are fused to comprehensively consider the influence of spatial structure on inhibition efficiency, thereby improving the accuracy of predicting the inhibition efficiency of the targeted gene, thereby ensuring the quality of siRNA drug development.

[0012] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above description and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar components in different views. Similar reference numerals with letter suffixes or different letter suffixes may represent different examples of similar components. The accompanying drawings generally illustrate various embodiments by way of example and not by way of limitation, and together with the description and claims, serve to illustrate the disclosed embodiments. Such embodiments are illustrative and exemplary and are not intended to be exhaustive or exclusive embodiments of the present method or non-transitory computer-readable medium having instructions for implementing the method.

[0014] Figure 1 A schematic diagram of the process of constructing and training the siRNA gene inhibition efficiency screening model according to the embodiments of the present application is shown.

[0015] Figure 2 A schematic diagram showing relevant information of some target mRNAs and siRNAs according to the embodiments of the present application is shown.

[0016] Figure 3 A schematic diagram of a siRNA gene inhibition efficiency screening model according to an embodiment of the present application is shown.

[0017] Figure 4 A schematic diagram of the structure of the siRNA gene inhibition efficiency screening model and training device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the technical solution of the present application, the present application is described in detail below in conjunction with the accompanying drawings and specific embodiments. The embodiments of the present application are further described in detail below in conjunction with the accompanying drawings and specific embodiments, but are not intended to limit the present application.

[0019] The words "first", "second" and similar terms used in this application do not indicate any order, quantity or importance, but are only used to distinguish. The words "include" or "comprises" and similar terms used in this application mean that the elements before the word include the elements listed after the word, and do not exclude the possibility of covering other elements. In this application, the arrows shown in the figures of each step are only examples of the execution order, not limitations. The technical solution of this application is not limited to the execution order described in the embodiments. The steps in the execution order can be combined, decomposed, or swapped, as long as the logical relationship of the execution content is not affected.

[0020] All terms used in this application (including technical or scientific terms) have the same meaning as understood by a person of ordinary skill in the art to which this application belongs, unless otherwise specifically defined. It should also be understood that terms defined in common dictionaries, for example, should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and should not be interpreted in an idealized or highly formal sense, unless explicitly defined as such herein. Techniques and devices known to a person of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, such techniques and devices should be considered part of the specification.

[0021] Figure 1A flowchart of the method for constructing and training the siRNA gene inhibition efficiency screening model according to the embodiment of the present application is shown. Specifically, a processor can be used to execute the method shown in steps S101 to S106. In this embodiment, the arrows shown in the figure for each step are merely examples of the execution order, not limitations. The technical solution of the present application is not limited to the execution order described in the embodiment. The steps in the execution order can be combined, decomposed, or swapped, as long as the logical relationship of the execution content is not affected.

[0022] The targeted gene inhibition efficiency of an siRNA refers to the efficiency with which it inhibits the transcription or translation of the targeted gene during its expression. Specifically, the inhibition efficiency reflects the degree to which the siRNA inhibits the target gene's mRNA or protein levels. If the siRNA effectively degrades the target gene's mRNA, the inhibition efficiency is high. The inhibition efficiency is usually expressed as a percentage, with higher values ​​indicating a stronger targeting effect of the siRNA and a more effective reduction in target gene expression.

[0023] like Figure 1As shown, in step S101, first sequence information of a target mRNA and second sequence information of a siRNA obtained based on the target mRNA are obtained. Wherein, the target mRNA and siRNA can both come from a dataset comprising mRNA and siRNA. For example, the dataset can be sourced from multiple references, for example, Huesken D, Lange J, Mickanin C, et al. Design of a genome-wide siRNA library using an artificial neural network[J];Nature biotechnology, 2005, 23(8): 995-1001;Reynolds A, Leake D,Boese Q, et al. Rational siRNA design for RNA interference[J];Nature biotechnology, 2004, 22(3): 326-330;Vickers TA, Koo S, Bennett CF, et al.Efficient reduction of target RNAs by small interfering RNA and RNase H-dependent antisense agents: a comparative analysis[J];Journal of Biological Chemistry, 2003, 278(9): 7108-7118;Harborth J, Elbashir SM, Vandenburgh K, et al. Sequence, chemical, and structural variation of small interfering RNAsand short hairpin RNAs and the effect on mammalian gene silencing[J]; Antisense and Nucleic Acid Drug Development, 2003, 13(2): 83-105; Ui‐Tei K,Naito Y, Takahashi F, et al.Guidelines for the selection of highly effective siRNA sequences for mammalian and chick RNA interference[J]; Nucleic acids research, 2004, 32(3): 936-948; Ichihara M, Murakumo Y, Masuda A, et al. Thermodynamic instability of siRNA duplex is a prerequisite for dependableprediction of siRNA activities[J]; Nucleic acids research, 2007, 35(18): e123. like. Figure 2 As shown in the figure, the dataset contains detailed information such as siRNA ID, siRNA sequence, mRNA sequence and inhibition rate (efficacy). Figure 2 The data set shown in FIG is only a part of the data set for exemplary description, and does not constitute the entire data set. Figure 2 The siRNA and mRNA shown in the figure are merely exemplary and do not constitute a specific limitation on the target mRNA and siRNA.

[0024] In this embodiment, the first sequence information of the target mRNA refers to the gene mRNA sequence that you want to interfere with or inhibit. In other embodiments, it can be obtained through gene databases such as NCBI, Ensembl, etc. The first sequence information is used to characterize the target mRNA sequence. There is no limitation on the specific representation method of the first sequence information. For example, it can be digital information obtained by word unit conversion of the base, or other representation methods, as long as it can be used to characterize the target mRNA sequence. Among them, the second sequence information is used to characterize the siRNA sequence designed based on the target mRNA. There is no limitation on the specific representation method of the second sequence information. For example, it can be digital information obtained by word unit conversion of the base, or other representation methods, as long as it can be used to characterize the siRNA sequence.

[0025] That is, in some other embodiments of the present application, the sequence information includes digital information obtained by performing word-unit conversion based on bases in the RNA, wherein the word-unit conversion includes mapping different bases to different digital information, or combining different bases according to a preset method to obtain different word-unit combinations, and mapping the different word-unit combinations to different digital information. In this embodiment, the RNA includes target mRNA and siRNA.

[0026] "Tokens" refer to the smallest units of a gene sequence. In this embodiment, token segmentation refers to breaking down siRNA and mRNA sequences into small fragments or units. These fragments are typically single RNA bases (e.g., A, U, G, C) or base pairs. For mRNA sequences, they may be segmented into individual base pairs (e.g., "AU," "GC," etc.) based on the characteristics of the RNA. For siRNA sequences, they may also be segmented into similar base pairs or short sequence fragments. This is not a specific limitation.

[0027] Specifically, as an implementation method, the four bases in RNA, such as A, G, C, and U, can be directly used as four word units, and then the bases A, G, C, and U can be mapped to four numbers respectively. For example, base A can be mapped to number 1, base U can be mapped to number 2, base T (DNA base) can also be mapped to number 2, base C can be mapped to number 3, and base G can be mapped to number 4 to represent special characters for padding. <pad>Mapped to number 0, ie {' <pad>':0,'A':1,'U':2,'T':2,'C':3,'G':4}.

[0028] In addition, byte pair encoding can be used to merge word units and map the merged word units to numbers. Byte Pair Encoding (BPE) is a commonly used algorithm in natural language processing. Its core idea is to compress words or sequences into a more compact form by repeatedly merging the most frequently occurring character pairs.

[0029] For example, siRNA and mRNA sequences are first segmented into tokens based on bases (or base pairs). For example, the mRNA sequence "AUCG" can be segmented into ['A', 'U', 'C', 'G']. The frequency of these tokens is then calculated. For example, if a base pair (such as "AU") appears particularly frequently throughout the sequence, the two most frequently occurring base pairs are merged into a new token. For example, if "AU" and "CG" are the most common, they are merged into a new token, "AUCG." This process of merging additional tokens based on frequency is continued until a predetermined number of merging is reached or a sufficient number of tokens is obtained. Through BPE, RNA sequences are gradually transformed from single-base fragments into higher-level composite tokens that represent common patterns or structures in RNA sequences. After converting RNA sequences into tokens through methods such as BPE, these tokens need to be further mapped into numbers for use in machine learning or deep learning models.

[0030] In this way, the first sequence information and the second sequence information are obtained respectively by processing and converting the target mRNA and siRNA sequences, so that the first sequence information and the second sequence information can be used in the neural network model in a digital representation.

[0031] Returning to the embodiment of the present application, in step S102, first spatial image information of the target mRNA and second spatial image information of the siRNA are obtained, where the first spatial image information and the second spatial image information are used to characterize the spatial structures of the target mRNA and siRNA, respectively.

[0032] Specifically, spatial image information of siRNA and target mRNA can be obtained based on artificial intelligence models or experiments. In a specific embodiment, the artificial intelligence tool HelixFold 3 can be used to predict the spatial structure of siRNA and target mRNA. For specific implementation methods, please refer to the literature Liu L, Zhang S, Xue Y, et al. Technical Report of HelixFold3 for Biomolecular Structure Prediction[J].arXiv preprint arXiv:2408.16975, 2024.

[0033] This is only an exemplary description and does not constitute a limitation on the specific solution.

[0034] For example, the spatial structure of the target mRNA can be determined experimentally, or predicted using other prediction tools such as RNAfold and RNAstructure. Of course, the spatial structure of siRNA can also be obtained from RNA databases (such as RNAstrand and PDB), or its spatial structure can be predicted using tools such as 3D RNA.

[0035] For example, PyMOL Open-Source 2.5.0 can be used to convert the spatial structure of RNA into spatial image information, or the geometric information of the spatial structure of RNA (such as curvature, angle, hydrogen bond, etc.) can be extracted into a numerical feature vector, or its spatial image information can be presented in PDB format, structure vector, etc. There is no specific limitation on this, as long as the first spatial image information and the second spatial image information can respectively characterize the spatial structure of the target mRNA and siRNA, and can be analyzed and processed by the neural network model.

[0036] In this embodiment of the present application, in step S103, a siRNA gene inhibition efficiency screening model is constructed, comprising a first neural network model, a second neural network model, and a third neural network model. In step S104, the first neural network model is trained using the first and second sequence information to extract sequence features. In step S105, the second neural network model is trained using the first and second spatial image information to extract spatial structural features.

[0037] That is, in this embodiment, the siRNA gene inhibition efficiency screening model includes using a first neural network model to extract sequence features based on the first sequence information and the second sequence information, and using a second neural network model to extract spatial structure features based on the first spatial image information and the second spatial image information.

[0038] When the first sequence information and the second sequence information are input into the first neural network model, the input order of the first sequence information and the second sequence information can be limited. For example, the first sequence information can be placed in front and the second sequence information can be placed in the back. After the front and back splicing operations are performed, they are input into the first neural network model, which is conducive to the first neural network model to quickly extract sequence features and improve the screening efficiency of siRNA by the siRNA gene inhibition efficiency screening model.

[0039] In some other embodiments of the present application, the second neural network model is one or two. Specifically, Figure 3 , the first neural network model is Star Transformer 301. RNA sequences usually have a long sequence length, and the dependencies between sequences are relatively complex. Traditional neural networks or RNNs (such as LSTM) may encounter gradient vanishing or gradient exploding problems when processing long sequences, making it difficult to capture long-range dependencies. Star Transformer301, as a variant of the Transformer network, can process long sequence data more effectively and avoid the above problems. Star Transformer 301 can better model the relationship between different row positions in RNA sequences through the self-attention mechanism, including potential secondary structure dependencies, which is crucial for tasks such as RNA function prediction, RNA-protein interactions, and siRNA design. Through training, Star Transformer 301 can extract meaningful sequence features from RNA sequences and effectively capture complex local and global information in RNA sequences.

[0040] In some other embodiments, when there is only one second neural network model, the second neural network model is trained based on the first spatial image information and the second spatial image information. That is, when there is only one second neural network model, the first spatial image information and the second spatial image information are spliced ​​and input into the second neural network model to extract the spatial structural features of RNA.

[0041] In the case where there are two second neural network models, one of the second neural network models is trained based on the first spatial image information, and the other of the second neural network models is trained based on the second spatial image information. Figure 3 As shown, the second neural network model includes Swin Transformer 1 302 and Swin Transformer 2 303, wherein Swin Transformer 1 302 is trained based on the first spatial image information to extract the spatial structural features of mRNA, and Swin Transformer 2 303 is trained based on the second spatial image information to extract the spatial structural features of siRNA. Swin Transformer 1 302 and Swin Transformer 2 303, whose inputs are the target mRNA and siRNA spatial images, respectively, cut the target mRNA and siRNA spatial images into different windows, and move the windows to interact and calculate the attention of the windows, and there is overlap between the windows. By cutting the image into multiple small windows, Swin Transformer only calculates self-attention within the local window without performing global calculations. The size of each window can be adjusted according to the task requirements, and the windows are overlapped to promote information exchange, which means that the features within and between windows can be transmitted and interacted through the attention mechanism. By translating the window at different levels, the Swin Transformer model can capture the connections between different positions, enabling the Swin Transformer model to retain local details while extracting global context information when processing large images.

[0042] Window overlap is crucial for the model's ability to extract image features, especially when processing long sequences of data. Overlapping windows enhances information transfer and ensures cross-window information sharing. When calculating self-attention, each window shares the overlapping portions of adjacent windows. This allows a local window to not only focus on features in its own region but also to gain complementary global information from neighboring windows.

[0043] By using masks within or between windows, the computational efficiency of windows is improved through masking, and relative position encoding is used instead of absolute position encoding to improve model performance.

[0044] In this way, the efficiency of analyzing and processing spatial image information can be greatly improved, and the accuracy of extracting spatial structural features can be improved, which is conducive to improving the accuracy of the siRNA gene inhibition efficiency screening model in screening siRNA with targeted gene inhibition efficiency.

[0045] Swin Transformer adopts a local window mechanism, which enables it to effectively capture local spatial structural features when processing image or structural data, while integrating information globally. Swin Transformer can better capture the folding, interaction and spatial structural characteristics of RNA in space. Moreover, the second neural network model Swin Transformer is used to process spatial image information, which has higher computational efficiency and can capture the multi-scale spatial structural characteristics of RNA, thereby improving the overall prediction accuracy. That is to say, in some other embodiments of the present application, the first neural network model is a Star Transformer model, the second neural network model is a Swin Transformer model, and the third neural network model is a DPCNN (Deep Pyramid Convolutional Neural Network) model.

[0046] like Figure 3 The output of Star Transformer 301 is input to the MLP Head (Multi-Layer Perceptron Head). The outputs of Swin Transformer 1 302 and Swin Transformer 2 303 are combined and also input to the MLP Head. The MLP Head is used to further extract and integrate features from the data processed by the Transformers. The integrated output is then input to DPCNN 304 for feature fusion, so that the siRNA gene inhibition efficiency is predicted based on the fused features. In other words, in step S106, the sequence features and the spatial structure features are fused, and the fused features are used to train the third neural network model so that the screening accuracy of the siRNA gene inhibition efficiency screening model exceeds a threshold.

[0047] The construction and training method of the siRNA gene inhibition efficiency screening model described in the embodiments of the present application fuses sequence features and spatial structure features to predict the siRNA gene inhibition efficiency. This can improve the prediction efficiency while improving the prediction accuracy, thereby ensuring the quality of the screened siRNA. siRNA (small interfering RNA) guides the RISC complex to cleave mRNA by base pairing with the target mRNA, thereby achieving gene silencing. The spatial structure of siRNA and mRNA has an important impact on the efficiency of targeted gene inhibition. For example, the spatial structure of the target mRNA (such as the stem-loop structure) may mask the target sequence, making it difficult for siRNA to bind to it. In the process of training the siRNA gene inhibition efficiency screening model, the present application fully considers the spatial structure information of RNA, fuses the sequence features and spatial structure features to train the siRNA gene inhibition efficiency screening model, and can improve the comprehensiveness of the predictive factors, thereby improving the accuracy of the prediction of the targeted gene inhibition efficiency.

[0048] In this embodiment, the siRNA gene inhibition efficiency screening model is trained and the siRNA gene inhibition efficiency screening model with the best predicted effect is saved, so that the siRNA gene inhibition efficiency screening model with the best predicted effect can be used to screen siRNA. Not only is the screening efficiency high, but the accuracy is also high, thereby avoiding the omission of siRNA with targeted gene inhibition efficiency.

[0049] In other embodiments of the present application, based on the principle of base complementarity, the target mRNA is traversed with a preset walk interval length and a preset walk coverage interval length to obtain a library of siRNA sequences that bind to the target mRNA sequence. Using a trained siRNA gene inhibition efficiency screening model, the targeted gene inhibition efficiency of siRNAs in the siRNA sequence library is predicted, enabling efficient screening of siRNAs with excellent targeted gene inhibition efficiency in the siRNA sequence library with high screening accuracy, without the problem of siRNAs with excellent targeted gene inhibition efficiency being missed.

[0050] Among them, the preset walking interval length can represent the number of bases skipped each time when sliding on the target mRNA. For example, if the walking interval length is set to 10, each time it will start from the first base in the mRNA sequence, slide 10 bases to the right, and then select the next position to start a new slide.

[0051] The preset walking coverage interval length can represent the length of the target mRNA region covered when designing siRNA. For example, each siRNA is set to cover an area of ​​20 bases in length. By setting the walking interval length and the coverage interval length, you can slide on the target mRNA and select multiple target areas. For each coverage interval, an siRNA sequence complementary to the target area will be designed based on the principle of base complementarity. For example, the four bases A, G, C, and U of RNA are directly used as four words. The complementary words of A, G, C, and U are U, C, G, and A, respectively, that is, A and U bases are complementary, C and G bases are complementary, G and C bases are complementary, and U and A bases are complementary.

[0052] In a specific embodiment, the PCSK9 gene mRNA sequence (NCBI database sequence number NM_174936.4) is used as the input sequence, and the base T (DNA base) in the mRNA sequence is replaced with the base U (RNA base) to generate a double-stranded sequence library of siRNA binding to PCSK9.

[0053] By traversing the entire target mRNA sequence, a series of siRNA sequences complementary to different regions can be obtained, forming an siRNA sequence library. Compared with manually designed siRNAs, this not only improves the efficiency from design to screening of siRNAs, but also ensures the comprehensiveness of the siRNA sequences obtained based on the target mRNA, avoiding the omission of siRNAs with targeted inhibition efficiency, thus helping to ensure the quality of the final screened siRNAs.

[0054] In other embodiments of the present application, siRNAs in the sequence library are preliminarily screened based on one or more of the number of consecutive bases, GC content, and similarity to a reference siRNA sequence. The siRNA gene inhibition efficiency screening model is then used to predict the targeted gene inhibition efficiency of the siRNAs in the preliminarily screened sequence library. The number of consecutive bases refers to the consistency of the base sequence within the siRNA sequence. Generally, siRNA design avoids long, continuous repeats of bases (e.g., consecutive A, G, C, or U), as these can lead to nonspecific binding or reduce siRNA stability and efficiency. Sequences containing excessively long, continuous repeats of bases can be excluded during the preliminarily screening. GC content refers to the ratio of G (guanine) to C (cytosine) in the siRNA sequence. For example, the GC content can be controlled within a threshold GC content range. Controlling the GC content can help stabilize the siRNA binding to the target mRNA while avoiding the adverse effects of excessively high or low GC content. Adjusting the GC content range during the preliminarily screening process can improve the siRNA efficiency. Reference siRNA sequences are proven effective siRNA sequences (e.g., experimentally validated siRNAs). By comparing the siRNA to be screened with these reference sequences, its potential inhibitory effect can be assessed based on similarity. If an siRNA sequence has a high degree of similarity to a reference sequence, it is likely to be more functionally effective in binding to the target gene and exerting a gene inhibitory effect. Therefore, during initial screening, siRNAs with a high degree of similarity to reference siRNA sequences can be prioritized.

[0055] In this way, the amount of predicted data can be further reduced through preliminary screening, thereby further improving the efficiency of screening siRNAs with inhibitory efficiency on targeted genes.

[0056] In some other embodiments of the present application, when there are two second neural network models, the spatial image shape of the target mRNA determined based on the first spatial image information is consistent with the spatial image shape of the siRNA determined based on the second spatial image information, so that the spatial image shapes of the siRNA and the target mRNA input to the second neural network model are consistent. Specifically, the spatial image shape can be reflected from three dimensions: length, width, and channel. By controlling the consistency of the spatial image shapes of the siRNA and the target mRNA input to two different second neural network models, it is beneficial to align or map the spatial image shapes of the target mRNA and siRNA, so that they can be effectively compared and predicted in the neural network. This alignment may involve normalizing the three-dimensional structures of the two to ensure that they have the same spatial features (for example, spatial position, folding pattern, etc.). In addition, it can improve training efficiency, and it is less difficult to fuse the spatial structural features extracted by the two different second neural network models.

[0057] In some other embodiments of the present application, when the spatial structural features of the target mRNA and siRNA are extracted using the second neural network model, the size of the image block cut from the siRNA image is larger than the size of the image block cut from the target mRNA image, and the model window size of the siRNA is larger than the model window size of the target mRNA. Since the target mRNA image contains more spatial information than the siRNA image, the size of the image block cut from the siRNA image is larger than the size of the image block cut from the target mRNA image, and the window size of the second neural network model of the siRNA is also larger than the window size of the second neural network model of the target mRNA, so as to capture more details of the local spatial structure of the siRNA. Since the spatial structure of siRNA is relatively simple, the use of larger image blocks and windows can ensure the integrity of its spatial information and improve the predictive performance of the siRNA gene inhibition efficiency screening model. For the target mRNA, due to its complex structure and long length, smaller blocks and windows help reduce over-smoothing and information loss, which is conducive to capturing subtle differences and local features in the structure.

[0058] This example takes into account the structural differences between siRNA and target mRNA. Larger image blocks and window sizes better capture the global features of siRNA, while smaller ones are more suitable for processing the complex structure of target mRNA to extract more detailed local features, thereby improving the prediction accuracy of the siRNA gene inhibition efficiency screening model.

[0059] Figure 4 A schematic diagram of the structure of a device for constructing and training an siRNA gene inhibition efficiency screening model according to an embodiment of the present application is shown. The device 400 for constructing and training an siRNA gene inhibition efficiency screening model includes an interface 401 and a processor 402. The interface 401 is configured to obtain first sequence information of a target mRNA and second sequence information of an siRNA obtained based on the target mRNA; obtain first spatial image information of the target mRNA and second spatial image information of the siRNA, wherein the first spatial image information and the second spatial image information are used to characterize the spatial structures of the target mRNA and the siRNA, respectively; and at least one processor 402 is configured to implement the method for constructing and training an siRNA gene inhibition efficiency screening model as described in various embodiments of the present application.

[0060] This enables the automated and efficient screening of siRNAs with targeted inhibitory efficacy using a neural network model, significantly improving screening efficiency and reducing the time required for siRNA drug development. Furthermore, during the intelligent siRNA screening process, RNA sequence information and spatial image information are integrated to comprehensively consider the impact of spatial structure on inhibitory efficiency, thereby improving the accuracy of predicting inhibitory efficiency against the target gene and ensuring the quality of siRNA drug development.

[0061] In other embodiments of the present application, a screening method for screening siRNAs with targeted gene inhibition efficiency is provided, the screening method comprising: based on the principle of base complementarity, traversing the target mRNA with a preset walk interval length and a preset walk coverage interval length to obtain a library of siRNA sequences that bind to the target mRNA sequence; using the siRNA gene inhibition efficiency screening model construction and training device described in various embodiments of the present application to predict the targeted gene inhibition efficiency of siRNAs in the siRNA sequence library, so as to screen out siRNAs with a targeted gene inhibition efficiency greater than the set inhibition efficiency. The screening method provided by this embodiment not only has high screening efficiency, but also has high screening accuracy, which can ensure the quality of the screened siRNAs.

[0062] The processor may be a processing device including one or more general-purpose processing devices, such as a microprocessor, a central processing unit (CPU), or the like. More specifically, the processor may be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor running other instruction sets, or a processor running a combination of instruction sets. The processor may also be one or more special-purpose processing devices, such as an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), a system on a chip (SoC), or the like.

[0063] This application describes various operations or functions that can be implemented as software code or instructions or defined as software code or instructions. Such content can be source code or differential code ("incremental" or "patch" code) that can be directly executed ("object" or "executable" form). The software code or instructions can be stored in a computer-readable storage medium and, when executed, can cause a machine to perform the described functions or operations, and include any mechanism for storing information in a form accessible to a machine (e.g., a computing device, an electronic system, etc.), such as recordable or non-recordable media (e.g., read-only memory (ROM), random access memory (RAM), magnetic disk storage media, optical storage media, flash memory devices, etc.).

[0064] The exemplary methods described herein can be implemented, at least in part, by a machine or computer. In some embodiments, a computer-readable storage medium stores computer program instructions that, when executed by a processor, cause the processor to execute the methods for constructing and training a siRNA gene inhibition efficiency screening model as described in various embodiments of the present invention.

[0065] Implementations of such methods may include software code, such as microcode, assembly language code, high-level language code, and the like. Various software programming techniques may be used to create various programs or program modules. For example, program portions or program modules may be designed using or with the aid of Java, Python, C, C++, assembly language, or any other known programming language. One or more of such software portions or modules may be integrated into a computer system and / or computer-readable media. Such software code may include computer-readable instructions for performing the various methods. The software code may form part of a computer program product or computer program module. Furthermore, in an example, the software code may be tangibly stored on one or more volatile, non-transitory, or non-volatile tangible computer-readable media, such as during execution or at other times. Examples of such tangible computer-readable media may include, but are not limited to, hard disks, removable disks, removable optical disks (e.g., optical disks and digital video disks), magnetic cassettes, memory cards or sticks, random access memory (RAM), read-only memory (ROM), and the like.

[0066] According to an embodiment of the present application, a computer program product is also provided, which includes computer instructions, and the computer instructions are used to enable a computer to execute the steps in the construction and training method of the siRNA gene inhibition efficiency screening model as described in each embodiment of the present application.

[0067] Furthermore, although exemplary embodiments have been described herein, the scope includes any and all embodiments based on the present application with equivalent elements, modifications, omissions, combinations (e.g., solutions that intersect various embodiments), adaptations, or changes. The elements of the claims are to be interpreted broadly based on the language employed in the claims and are not limited to the examples described in this specification or during the prosecution of the application, which examples are to be construed as non-exclusive. Therefore, it is intended that the specification and examples be considered merely as examples, with the true scope and spirit being indicated by the claims and their full scope of equivalents.

[0068] The above description is intended to be illustrative and not restrictive. For example, the above examples (or one or more of their embodiments) may be used in combination with each other. For example, a person of ordinary skill in the art may use other embodiments when reading the above description. In addition, in the above detailed description, various features may be grouped together to simplify the application. This should not be interpreted as an intention that a disclosed feature that is not claimed for protection is essential to any claim. On the contrary, the subject matter of the present application may have less than all the features of a particular disclosed embodiment. Thus, the claims are incorporated herein into the detailed description as examples or embodiments, with each claim independently serving as a separate embodiment, and it is contemplated that these embodiments may be combined with each other in various combinations or arrangements. The scope of the present application should be determined with reference to the appended claims and the full scope of equivalents to which such claims are entitled.

[0069] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.< / pad> < / pad>

Claims

1. A method for constructing and training a siRNA gene inhibition efficiency screening model, characterized in that: The construction and training method includes: Obtaining first sequence information of a target mRNA and second sequence information of a siRNA obtained based on the target mRNA; Acquire first spatial image information of the target mRNA and second spatial image information of the siRNA, wherein the first spatial image information and the second spatial image information are used to characterize the spatial structures of the target mRNA and the siRNA, respectively; Constructing a siRNA gene inhibition efficiency screening model, wherein the siRNA gene inhibition efficiency screening model comprises a first neural network model, a second neural network model and a third neural network model, wherein the first neural network model is a StarTransformer model, the second neural network model is a Swin Transformer model, and the third neural network model is a DPCNN model; Using the first sequence information and the second sequence information to train the first neural network model to extract sequence features; Using the first spatial image information and the second spatial image information to train the second neural network model to extract spatial structure features; The sequence features and the spatial structure features are fused, and the fused features are used to train the third neural network model, so that the screening accuracy of the siRNA gene inhibition efficiency screening model is higher than a threshold.

2. The construction and training method according to claim 1, characterized in that: The construction and training method also includes: based on the principle of base complementarity, traversing the target mRNA with a preset walking interval length and a preset walking coverage interval length to obtain a siRNA sequence library combined with the target mRNA sequence.

3. The construction and training method according to claim 2, characterized in that: The siRNAs in the sequence library are preliminarily screened based on one or more of the number of consecutive bases, GC content and similarity to the reference siRNA sequence, so as to predict the targeted gene inhibition efficiency of the siRNAs in the sequence library after the preliminarily screening using the siRNA gene inhibition efficiency screening model.

4. The construction and training method according to claim 1, characterized in that: Sequence information includes digital information obtained by converting words based on the bases in RNA; The word-unit conversion includes: mapping different bases to different digital information respectively, or combining different bases in a preset manner to obtain different word-unit combinations, and mapping different word-unit combinations to different digital information.

5. The construction and training method according to claim 1, characterized in that: The second neural network model is one or two, In the case where there is only one second neural network model, the second neural network model is trained based on the first spatial image information and the second spatial image information; In the case that there are two second neural network models, one of the second neural network models is trained based on the first spatial image information, and the other of the second neural network models is trained based on the second spatial image information.

6. The construction and training method according to claim 5, characterized in that: When there are two second neural network models, the spatial image shape of the target mRNA determined based on the first spatial image information is consistent with the spatial image shape of the siRNA determined based on the second spatial image information.

7. The construction and training method according to claim 1, characterized in that: When the second neural network model is used to extract spatial structural features of the target mRNA and siRNA, the size of the image block of the cut siRNA image is larger than the size of the image block of the cut target mRNA image, and the model window size of the siRNA is larger than the model window size of the target mRNA.

8. A construction and training device for a siRNA gene inhibition efficiency screening model, characterized in that: include: Interface, which is configured as: Obtaining first sequence information of a target mRNA and second sequence information of a siRNA obtained based on the target mRNA; Acquire first spatial image information of the target mRNA and second spatial image information of the siRNA, wherein the first spatial image information and the second spatial image information are used to characterize the spatial structures of the target mRNA and the siRNA, respectively; At least one processor is configured to implement the construction and training method of the siRNA gene inhibition efficiency screening model as described in any one of claims 1-7.

9. A method for screening siRNA with targeted gene inhibition efficiency, characterized in that: The screening method comprises: Based on the principle of base complementarity, the target mRNA is traversed with a preset walking interval length and a preset walking coverage interval length to obtain a siRNA sequence library combined with the target mRNA sequence; The construction and training device of the siRNA gene inhibition efficiency screening model described in claim 8 is used to predict the targeted gene inhibition efficiency of siRNA in the siRNA sequence library to screen out siRNAs whose targeted gene inhibition efficiency is greater than the set inhibition efficiency.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the construction and training method of the siRNA gene inhibition efficiency screening model as described in any one of claims 1-7.

11. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions are used to enable a computer to execute the construction and training method of the siRNA gene inhibition efficiency screening model as described in any one of claims 1 to 7.

Images