An activity-guided deep implicit evolution-based aptamer design method and its application

Abstract

This invention discloses a novel aptamer design method based on activity-guided deep implicit evolution and its application. Specifically, this invention discloses an aptamer optimization method based on activity-guided deep implicit evolution, comprising the following steps: (s1) providing a high-throughput screening dataset of aptamer sequences; (s2) data processing; (s3) encoder-decoder training; (s4) data dimensionality reduction and clustering; (s5) selection of active sequence regions and deep implicit evolution; (s6) obtaining the designed and optimized aptamer sequences; and (s7) decoder output. The method of this invention can efficiently screen optimized novel aptamer sequences with high activity and high affinity.

Description

Technical Field

[0001] This invention belongs to the field of molecular biology, specifically, it relates to a novel aptamer design method based on activity-guided deep implicit evolution and its application. Background Technology

[0002] Aptamers are typically obtained using the phylogenetic technique of exponential enrichment (SELEX), but traditional aptamer screening and design methods have several limitations. In SELEX, aptamer selection based on read counts is susceptible to interference from polymerase chain reaction (PCR) amplification bias, meaning that high-abundance sequences do not necessarily correspond to high-affinity sequences. Furthermore, early clustering techniques, such as AptaCluster and FASTAptamer, while attempting to reveal commonalities in aptamer sequences, neglected the secondary structure of the sequences. Similarly, motif-finding methods such as MEMRIS and Aptamotif, although focusing on high-affinity structural patterns, also fail to adequately consider the influence of overall secondary structure. More importantly, these methods are inefficient when processing large HT-SELEX libraries, making it difficult to quickly and accurately screen for high-quality aptamers from massive datasets.

[0003] With the development of deep sequencing technology, the amount of data generated for aptamer screening has increased rapidly. However, traditional methods struggle to effectively utilize this data to advance aptamer research. Studies have shown that biases generated during PCR amplification mean that high-abundance sequences do not always reflect their affinity for target molecules, further reducing screening accuracy. Furthermore, traditional clustering methods are computationally intensive and inefficient in capturing high-affinity aptamer sequences when dealing with large-scale data. While screening strategies based on high-abundance sequences can identify some sequences in the short term, they fail to effectively improve the activity and selectivity of the final aptamers. Therefore, new technological solutions are urgently needed to overcome the bottlenecks in data processing and screening efficiency of existing methods.

[0004] In actual screening, although next-generation sequencing technology is used to sort and verify nucleic acid sequences by copy number, the discovery efficiency of aptamers in the early stages remains low because there is no direct positive correlation between copy number, fluorescence activation fold, and affinity. To complicate matters further, a large number of disordered sequences often exist in the early rounds of screening; while in the later stages, aptamer sequences have clustered into families, making it more difficult to identify the active sequences within them.

[0005] These problems further highlight the shortcomings of traditional methods when dealing with large-scale data. Therefore, there is an urgent need in the field to develop more advanced technical solutions to improve screening efficiency and effectively enhance the activity and applicability of aptamers. Summary of the Invention

[0006] This invention provides a method to improve aptamer screening efficiency and effectively enhance the discovery of highly active aptamers.

[0007] In a first aspect of the present invention, a method for aptamer design and optimization based on activity-guided deep implicit evolution is provided, comprising the following steps:

[0008] (s1) Provide a high-throughput screening aptamer sequence dataset, the dataset including nucleic acid sequence information and copy number information of aptamers; and randomly divide the aptamer sequence dataset into a training set and a test set according to a certain proportion;

[0009] (s2) Data processing: Each nucleotide character in the nucleic acid sequence in the dataset is converted into a corresponding numerical index to obtain the numerical form of each nucleic acid sequence; and the numerical index is expanded into a vector to obtain a nucleic acid sequence representation composed of vectors;

[0010] In this context, the vector representation of nucleotides in the nucleic acid sequences of the training set is called the first vector, and the vector representation of nucleotides in the nucleic acid sequences of the test set is called the second vector.

[0011] (s3) Encoder-decoder training: The nucleic acid sequence representation formed by the first vector is input into the Gaussian mixture variational encoder (GM-VAE) for training, so that the trained encoder can capture the latent structural features and aptamer family distribution in the aptamer nucleic acid sequence and map these features to a low-dimensional latent space; and enable the decoder to generate the original input sequence; and then use the nucleic acid sequence representation formed by the second vector for testing;

[0012] (s4) Data dimensionality reduction and clustering: The encoder trained in (s3) is used to perform embedded dimensionality reduction on the nucleic acid sequence of the target round number, thereby mapping the nucleic acid sequence into a low-dimensional latent space to capture the potential similarity and structural features between nucleic acid sequences; and based on the potential similarity and structural features, different clusters (or families) of nucleic acid sequences are obtained.

[0013] (s5) Selection of active sequence regions and deep implicit evolution: select clusters corresponding to high copy number nucleic acid sequences and define them as active sequence regions; and perform implicit spatial genetic evolution strategy on the active sequence regions;

[0014] (s6) Obtain the optimized aptamer sequence: After the evolution in (s5), the designed and optimized aptamer sequence is obtained; and

[0015] (s7) Decoder output: The designed and optimized aptamer sequence obtained using the decoder output (s6).

[0016] In another preferred embodiment, the different clusters of the nucleic acid sequences refer to the family distribution of the corresponding nucleic acid sequences.

[0017] In another preferred embodiment, the different clusters of the nucleic acid sequences are aptamer family distributions.

[0018] In another preferred embodiment, the method further includes the following steps:

[0019] (s0) Provides single-round and / or multi-round nucleic acid aptamer sequence information and related experimental data obtained by SELEX technology (including HT-SELEX), wherein the experimental data includes copy number information; and preprocesses the data to obtain a high-throughput screened aptamer sequence dataset;

[0020] The preprocessing includes removing redundancy and measurement errors from the data.

[0021] In another preferred embodiment, the SELEX technology includes HT-SELEX technology.

[0022] In another preferred embodiment, step (s1) includes using the "train_test_split" function in the "sklearn.model_selection" module to randomly divide the aptamer sequence dataset into a training set and a test set in a 9:1 ratio.

[0023] In another preferred embodiment, step (s2) includes the following sub-steps:

[0024] (s2a) Convert each character in the nucleic acid sequence in the dataset into a corresponding numerical index, that is, map nucleotides A, T / U, G, and C to 0, 1, 2, and 3 respectively;

[0025] (s2b) Using the “nn.Embedding” module in the PyTorch framework (version 1.5.0), each numerical index is expanded into a 32-dimensional vector to obtain a vector representation of each nucleotide; thus, a nucleic acid sequence representation composed of vectors is obtained; wherein, the vector representation of nucleotides in the nucleic acid sequences in the training set is the first vector, and the vector representation of nucleotides in the nucleic acid sequences in the test set is the second vector.

[0026] In another preferred embodiment, in step (s3), the data information input to the Gaussian hybrid variable encoder includes: the number of nucleic acid sequences, the dimension of the nucleic acid sequences, and the length of the nucleic acid sequences.

[0027] In another preferred embodiment, the Gaussian hybrid variable self-dividing encoder (GM-VAE) includes an encoder and a decoder.

[0028] In another preferred embodiment, the Gaussian mixture variable self-divided encoder (GM-VAE) further includes a Gaussian mixture model (GMM).

[0029] In another preferred embodiment, the encoder is a convolutional application network (CNN) architecture containing multiple skip connection layers.

[0030] In another preferred embodiment, each skip connection layer consists of a convolutional layer, a batch normalization layer, and a Leaky ReLU nonlinear activation function.

[0031] In another preferred embodiment, the input latent space sampling vector z is first passed through the first fully connected layer FC. D，32 (D represents the dimension of z), and the feature vector X1 is obtained by sequentially passing it through batch normalization and the Leaky ReLU nonlinear activation function; then X1 is used as input to pass through the second fully connected layer FC. 22，64 Similarly, after batch normalization and activation functions, the feature vector X2 is obtained; subsequently, the feature vector X2 is passed through a third fully connected layer (FC). 64，32 The feature vector X1 and X3 are then processed similarly to obtain feature vector X3. It is worth noting that, to further fuse and extract the feature information of the sequence, this invention adds feature vectors X1 and X3 together and passes them through a fourth fully connected layer to obtain feature vector X4. Next, feature vector X4... 依 The output is obtained by passing the data through three transposed convolutional layers, each of which includes batch normalization and Leaky ReLU activation function.

[0032] In another preferred embodiment, the loss function of the variational autoencoder of the present invention, which incorporates a Gaussian mixture model, is:

[0033]

[0034] In another preferred embodiment, in step (s5), the genetic evolution strategy protects the design and optimization of active sequences.

[0035] In another preferred embodiment, in step (s5), the genetic evolution strategy includes elitism, crossover, and mutation.

[0036] In another preferred embodiment, in step (s5), the elitism is used for active sequence design, and the crossover and mutation are used for active sequence optimization.

[0037] In another preferred embodiment, the elitism includes two algorithms: Gaussian Mixture Model (GMM) and K-means clustering algorithm.

[0038] In another preferred embodiment, the elitism is implemented using the sklearn.mixture.GaussianMixture and sklearn.cluster.KMeans packages in the Scikit-learn library (version 1.0.2).

[0039] In another preferred embodiment, the elitism also introduces a silhouette coefficient to evaluate the number of reasonable aptamer families calculated by both algorithms.

[0040] In another preferred embodiment, the crossover includes the following steps: generating a random number between 0 and 1, and when the value is lower than a preset crossover rate threshold, performing a uniform order crossover (UOX) operation.

[0041] In another preferred embodiment, the crossing further includes the following steps: creating a random Boolean mask of the same length as the parent sequence, and exchanging elements at corresponding positions between the parent sequences based on the mask information, thereby generating two completely new child sequences.

[0042] In another preferred embodiment, the mutation includes: performing a mutation on the selected offspring; and replacing the random index of the latent variable z of the offspring with a random normally distributed value.

[0043] In another preferred embodiment, step (s3) further includes optimizing the trained encoder, including optimizing the encoder using a standard optimization algorithm (Adam optimizer).

[0044] In a second aspect of the invention, an apparatus for aptamer design and optimization is provided, comprising:

[0045] (a) A data input module, configured to input a high-throughput filtered aptamer sequence dataset, the dataset including nucleic acid sequence information and copy number information of the aptamers; and to randomly divide the aptamer sequence dataset into a training set and a test set according to a certain proportion;

[0046] (b) A data processing module configured to convert each nucleotide character in the nucleic acid sequence in the dataset into a corresponding numerical index, thereby obtaining a numerical form of each nucleic acid sequence; and to expand the numerical index into a first vector, thereby obtaining a nucleic acid sequence representation composed of the first vector;

[0047] In this context, the vector representation of nucleotides in the nucleic acid sequences of the training set is called the first vector, and the vector representation of nucleotides in the nucleic acid sequences of the test set is called the second vector.

[0048] (c) An analysis module, which is configured to perform the following operations:

[0049] (i) Encoder-decoder training: The nucleic acid sequence representation composed of the first vector is input into the Gaussian mixture variational encoder (GM-VAE) for training, so that the trained encoder can capture the latent structural features and aptamer family distribution in the aptamer nucleic acid sequence and map these features to a low-dimensional latent space; and enable the decoder to generate the original input sequence; and then use the nucleic acid sequence representation composed of the second vector for testing;

[0050] (ii) Data dimensionality reduction and clustering: The trained encoder is used to perform embedded dimensionality reduction on the nucleic acid sequence of the target round number, thereby mapping the nucleic acid sequence into a low-dimensional latent space to capture the potential similarity and structural features between nucleic acid sequences; and different clusters (or families) of nucleic acid sequences are obtained based on the potential similarity and structural features.

[0051] (iii) Selection of active sequence regions and deep implicit evolution: select clusters corresponding to high copy number nucleic acid sequences and define them as active sequence regions; and perform implicit spatial genetic evolution strategy on the active sequence regions;

[0052] (iv) Obtain the optimized aptamer sequence: After the evolution in (iii), the designed and optimized aptamer sequence is obtained;

[0053] (d) Output module: The output module is configured to output the designed and optimized aptamer sequence.

[0054] In another preferred embodiment, the different clusters of the nucleic acid sequences refer to the family distribution of the corresponding nucleic acid sequences.

[0055] In another preferred embodiment, the different clusters of the nucleic acid sequences are aptamer family distributions.

[0056] In another preferred embodiment, the apparatus further includes (a0) a data preprocessing module, which is configured to: preprocess the nucleic acid aptamer sequence information and related experimental data obtained by SELEX technology screening in one round and / or multiple rounds for a specific target, thereby obtaining a high-throughput screened aptamer sequence dataset;

[0057] The experimental data includes copy number information; the preprocessing includes removing redundancy and sequences with measurement errors from the data.

[0058] In another preferred embodiment, the following procedure is performed in the data processing module:

[0059] (b1) Convert each character in the nucleic acid sequence in the dataset into a corresponding numerical index, that is, map nucleotides A, T / U, G, and C to 0, 1, 2, and 3 respectively;

[0060] (b2) Using the “nn.Embedding” module in the PyTorch framework (version 1.5.0), each numerical index is expanded into a 32-dimensional vector to obtain a vector representation of each nucleotide; thus obtaining a nucleic acid sequence representation composed of vectors; wherein, the vector representation of nucleotides in the nucleic acid sequences in the training set is the first vector, and the vector representation of nucleotides in the nucleic acid sequences in the test set is the second vector.

[0061] In another preferred embodiment, the Gaussian hybrid variable self-dividing encoder (GM-VAE) includes an encoder and a decoder.

[0062] In another preferred embodiment, the genetic evolution strategy includes elitism, crossover, and mutation.

[0063] In another preferred embodiment, the elitism includes two algorithms: Gaussian Mixture Model (GMM) and K-means clustering algorithm.

[0064] In another preferred embodiment, the elitism is implemented using the sklearn.mixture.GaussianMixture and sklearn.cluster.KMeans packages in the Scikit-learn library (version 1.0.2).

[0065] In another preferred embodiment, the elitism also introduces a silhouette coefficient to evaluate the number of reasonable aptamer families calculated by both algorithms.

[0066] In another preferred embodiment, the crossover includes the following steps: generating a random number between 0 and 1, and when the value is lower than a preset crossover rate threshold, performing a uniform order crossover (UOX) operation.

[0067] In another preferred embodiment, the crossing further includes the following steps: creating a random Boolean mask of the same length as the parent sequence, and exchanging elements at corresponding positions between the parent sequences based on the mask information, thereby generating two completely new child sequences.

[0068] In another preferred embodiment, the mutation includes: performing a mutation on the selected offspring; and replacing the random index of the latent variable z of the offspring with a random normally distributed value.

[0069] In another preferred embodiment, the encoder-decoder training process also includes optimizing the trained encoder, including optimizing the encoder using a standard optimization algorithm (Adam optimizer).

[0070] In a third aspect of the invention, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the method described in the first aspect of the invention.

[0071] It should be understood that, within the scope of this invention, the above-described technical features of this invention and the technical features specifically described below (such as in the embodiments) can be combined with each other to form new or preferred technical solutions. Due to space limitations, they will not be described in detail here. Attached Figure Description

[0072] Figure 1 The overall flowchart of a novel aptamer design method based on activity-guided deep implicit evolution is shown.

[0073] Figure 2 After training the model on the public database dataset DRA009383 (dataset 1), the sequence space was visualized using the encoder, and the distribution of sequence families was visualized based on two elitist algorithms.

[0074] Figure 3 After displaying the training model using an unpublished dataset (dataset 2) provided by the collaborators, the sequence space was visualized using an encoder, and the distribution of sequence families was visualized based on two elitist algorithms.

[0075] Figure 4 The results show the novel aptamer sequences and secondary structure information with potential activity designed by our inventive method after training the model on dataset 1.

[0076] Figure 5 The experimental results show the results of the novel aptamer sequences with potential activity designed by our inventive method after training the model on dataset 2. Detailed Implementation

[0077] Through extensive and in-depth research, the inventors have developed for the first time a method for designing and optimizing aptamers based on activity-guided deep implicit evolution. This invention employs unsupervised machine learning and generative models, mapping nucleic acid sequences to a low-dimensional latent space to deeply learn the characteristics of existing aptamer sequences and activity sequences, thereby generating novel aptamer sequences with high activity and affinity.

[0078] Specifically, this invention utilizes single-round or multi-round nucleic acid aptamer sequence information and experimental data (including copy number information) from high-throughput sequencing using SELEX technology (including HT-SELEX) to train a Gaussian mixture variational encoder (GM-VAE). This allows the encoder to map the nucleic acid aptamer sequences into a low-dimensional latent space and learn the potential similarities and structural features between nucleic acid sequences. Furthermore, it selects active sequence regions for guidance, employing a genetic evolution strategy on these active sequence regions to ultimately obtain optimized novel aptamer sequences. Based on this, the invention was completed.

[0079] the term

[0080] To facilitate a clearer understanding of this disclosure, certain terms are first defined. As used herein, unless otherwise expressly specified herein, each of the following terms shall have the meaning given below. Other definitions are set forth throughout the application.

[0081] As used herein, the terms “containing” or “including (comprise)” can be open-ended, semi-closed, or closed. In other words, the terms also include “consistently made of” or “made of”.

[0082] As used herein, unless otherwise stated, any concentration range, percentage range, proportion range, or integer range shall be understood to include any integer value within the range and, where appropriate, its fractional value (e.g., one-tenth and one-hundredth of an integer).

[0083] As used herein, the term “and / or” refers to and covers any and all possible combinations of one or more of the related listed items.

[0084] As used in this paper, the term "Gaussian mixture variational autoencoder (GM-VAE)" is an extended model of variational autoencoder (VAE) that can better model multimodal data distributions by introducing a Gaussian mixture distribution into the implicit space.

[0085] As used in this article, SELEX (Systematic Evolution of Ligands by Exponential Enrichment) is a method for obtaining high-affinity nucleic acid aptamers through in vitro screening. HT-SELEX (High-Throughput SELEX) is a high-throughput version of SELEX technology.

[0086] The main advantages of this invention include:

[0087] (a) The method of the present invention overcomes the shortcomings of the prior art in handling complex sequence distributions and aptamer optimization.

[0088] (b) The method of this invention improves the efficiency of aptamer discovery: By combining a deep variational autoencoder (VAE) with a Gaussian mixture model (GMM), this invention can accurately learn the low-dimensional latent space distribution characteristics of aptamer sequences. This method not only solves the inefficiency problem caused by data complexity and dimensionality issues in traditional methods, but also efficiently mines potential highly active aptamer sequences.

[0089] (c) The method of this invention avoids interference from early invalid sequences: This invention avoids interference from disordered and low-affinity sequences in early screening by using active region guidance and latent space genetic evolution strategies, ensuring a more accurate and efficient screening process. This technology can accurately identify active sequences in the early screening stage, thereby improving the accuracy and efficiency of screening.

[0090] (d) The method of this invention ensures dataset compatibility and universality: This invention can adapt to different sequencing datasets (including single-round or multi-round sequencing data) and has broad application compatibility. This enables the technology to play a role in a variety of biological research and applications, whether in cancer screening, antibody development, or other fields requiring the screening of high-affinity molecules.

[0091] (e) The method of the present invention promotes the application of aptamers in multiple fields: The present invention provides a novel and efficient aptamer design and screening technology, which provides strong technical support for aptamer research and can accelerate the application and development of aptamers in multiple fields such as biomedicine, environmental monitoring, and drug development.

[0092] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Experimental methods in the following embodiments, unless otherwise specified, are generally performed under conventional conditions, such as those described in Sambrook et al., Molecular Cloning: A Laboratory Manual (New York: Cold Spring Harbor Laboratory Press, 1989), or as recommended by the manufacturer. Unless otherwise stated, percentages and parts are weight percentages and parts by weight.

[0093] Example 1

[0094] 1.1 Provide experimental datasets and data processing

[0095] The first step of this invention is to provide a high-throughput screening aptamer sequence dataset. This dataset typically consists of single-round or multi-round nucleic acid aptamer sequence information and experimental data (e.g., copy number information) from high-throughput sequencing using HT-SELEX technology.

[0096] This dataset contains nucleic acid aptamer sequence information from one or more sequencing epochs obtained through HT-SELEX technology, along with related experimental data (e.g., copy number information). Specifically, the experimental dataset includes not only sequence information but also enrichment information (copy number) within the current sequencing epoch. This dataset provides ample input information for the model, enabling more accurate subsequent sequence analysis and optimization.

[0097] Data processing includes the following steps:

[0098] (1) Data Preprocessing: To ensure the accuracy and usability of the data, the raw data must first be preprocessed. Preprocessing steps include removing redundant and erroneous sequences from the raw sequence data. By removing inaccurate sequencing data (e.g., characters other than ACGU) and duplicate data, the reliability of the screening results is ensured.

[0099] (2) Dataset Partitioning: The preprocessed sequence data was randomly divided into a training set and a test set in a 9:1 ratio using the "train_test_split" function from the "sklearn.model_selection" module. The model with the smallest test loss was selected iteratively. The training set, with its large amount of data, helps the model learn the distribution and clustering characteristics of the sequences, while the test set, with its smaller amount of data, verifies the model's training results and helps adjust the learning method and training parameters. This data partitioning method ensures that the model obtains a sufficient sample size during training while reserving some data for verifying the model's performance, thereby improving the model's generalization ability and prediction accuracy.

[0100] (3) Sequence characterization: The nucleic acid sequence is first converted into a one-hot encoded form suitable for machine learning algorithms. First, each character in the nucleotide sequence is converted into a corresponding numerical index. Specifically, nucleotides A, T / U, G, and C are mapped to 0, 1, 2, and 3, respectively.

[0101] This conversion process transforms nucleotide characters into numerical forms, which are then expanded into 32-dimensional vectors using the "nn.Embedding" module in the PyTorch framework (version 1.5.0) and input into the model for training.

[0102] 1.2 Constructing a pre-trained encoder-decoder network architecture

[0103] The core model framework of this invention consists of an encoder and decoder from a Convolutional Neural Network (CNN), and an implicit space that captures the distribution characteristics of aptamer sequence families in single or multiple rounds of sequencing data. These three elements together form a deep variational autoencoder (VAE). The implicit space establishes a probabilistic model using a parameterized Gaussian distribution. Its dimensional features are learned and generated by the inference network in the encoder, effectively capturing the evolutionary patterns of aptamer sequence families in multiple rounds of SELEX screening. The decoder network parameterizes the latent variable z into a conditional probability distribution of nucleotides through a conditional generation mechanism, reconstructing the nucleic acid sequence using deconvolution operations. Its random sampling process effectively generates new sequence variants that possess both family conservatism and reasonable variability. In the vector generation and distribution training stages within the latent space, to capture the multimodal distribution of the sequences, this invention incorporates a Gaussian Mixture Model (GMM).

[0104] Specifically, the Gaussian mixture model enhances the ability to express latent structures in sequence data of different truncation numbers by representing the implicit space as a mixture of multiple Gaussian distributions, making the distribution of aptamer sequencing data in low-dimensional space more intuitive and clear. The construction and implementation of this model can effectively capture complex structures and features in sequences.

[0105] The specific construction process is as follows:

[0106] (1) Encoder Network Structure: The encoder network is a convolutional neural network (CNN) architecture containing multiple skip connection layers. First, the unique encoding form of each nucleic acid sequence is embedded as a 32-dimensional vector, which serves as the input to the model. Subsequently, the input data passes through up to 6 skip connection layers, from which the structure and interaction information of the sequence are learned.

[0107] Specifically, in the design that skips the connection layer, each layer consists of a convolutional layer, a batch normalization layer, and a Leaky ReLU nonlinear activation function.

[0108] First, the input latent space sampling vector z passes through the first fully connected layer (FC). D，32 (D represents the dimension of z), and the feature vector X1 is obtained by sequentially passing it through batch normalization and the Leaky ReLU nonlinear activation function; then X1 is used as input to pass through the second fully connected layer FC. 32，64Similarly, after batch normalization and activation functions, the feature vector X2 is obtained; subsequently, the feature vector X2 is passed through a third fully connected layer PC. 64，32 The feature vector X1 and X3 are then processed similarly to obtain feature vector X3. Notably, to further fuse and extract the sequence's feature information, this invention adds feature vectors X1 and X3 together and passes them through a fourth fully connected layer to obtain feature vector X4. Next, feature vector X4 is passed through three transposed convolutional layers, each including batch normalization and Leaky ReLU activation function processing, ultimately yielding the output.

[0109] (2) Decoder network structure: The decoder network is a multi-layer structure composed of fully connected layers and transposed convolutional layers. The final output is the nucleic acid probability of each position in the sequence in this round of sequencing, that is, each position of the sequence is assigned a classification distribution.

[0110] (3) Introduction of Gaussian Mixture Model: A binary vector c∈{0,1} is introduced. k×1 This is used to represent which Gaussian component the latent variable z belongs to. Ultimately, this invention introduces a new distribution. To approximate the posterior distribution p θ (z, c|x). Here, c is the categorical variable of the aptamer family to which each nucleic acid sequence belongs, and its probability follows a discrete distribution p(c|φ). And z is a latent variable.

[0111] Therefore, the overall depth variational autoencoder is specifically manifested as an encoder with skip connections that transforms a sequence X into a latent distribution in the implicit space. The visualization of z is used to simulate the distribution of the high-dimensional original dataset p(x). The decoder network architecture learns p... e (x, c|z) thus reconstructs the latent space z to the original input data. p(z|c) is a function that, given c categories, is derived from μ. c and σ z Parameterized Gaussian mixture distribution.

[0112] Assuming x and c are conditional on z, the joint probability p(x, z, c) can be decomposed as follows:

[0113] p(x,z,c)=p(x|z)p(z|c)p(c) (Formula 1)

[0114] The various probability models are defined as follows:

[0115]

[0116] p(x|z)=Ber(x|μ x (Equation 4)

[0117] Furthermore, the present invention assumes It can be broken down into:

[0118]

[0119] The present invention then defines the following formula to describe the specific probability distribution:

[0120]

[0121] Finally, the loss function of the variational autoencoder combining the Gaussian mixture model in this invention is:

[0122]

[0123] This model can not only effectively learn the distribution of aptamer sequencing data from different epochs, but also maximize the log-likelihood of the current epoch sequencing data, ensuring that the reconstructed estimated data is similar to the input nucleic acid sequence data. This model can adaptively adjust weights to handle sequencing data with different aggregation distribution characteristics, making the distribution of aptamer sequencing data from different epochs clearer, thereby improving the model's generalization ability and predictive performance. Simultaneously, regularization using Kullback-Leibler divergence (KL divergence) helps prevent model overfitting and ensures a reasonable distribution of latent variables.

[0124] (4) Model Training and Fine-tuning Strategy: The training and fine-tuning strategy of the method in this invention adopts a round-by-round training and target round-by-round data fine-tuning strategy. This strategy aims to improve the model's ability to encode and decode nucleic acid sequences, mimicking the SELEX screening process to gradually learn the data distribution of each round and optimize the model's performance under specific data distributions during the fine-tuning phase. Specifically, this strategy includes the following two main steps:

[0125] 1) Pre-training with incrementally increasing datasets: First, train the model using datasets with increasing round numbers (e.g., Round 1, Round 2, Round 3). The purpose of this process is to allow the model to gradually learn and adapt to the distribution characteristics of different datasets and improve the model's ability to encode and decode nucleic acid sequences. This helps the model further improve its generalization ability and its deeper understanding of nucleic acid sequences.

[0126] 2) Fine-tuning with high-copy datasets of small samples from the target round: The fine-tuning phase uses a high-copy sequence dataset (typically the first 5000–10000 sequences) of the target round (e.g., Round 3) to further fine-tune the model. This allows the model to better adapt to and generalize the distribution of the dataset for that specific round. The fine-tuning process further optimizes the model's response to the target data by adjusting its parameters, while ensuring that the model performs optimally when studying the data from the target round.

[0127] 1.3 Activity-guided aptamer discovery strategy

[0128] Because amplification bias and non-specific adsorption can lead to the inclusion of sequences unrelated to the target molecule in the later-enriched sequence data, these unrelated sequences may cluster in different groups within the sequence distribution, similar to the active aptamer family. Therefore, the core assumption of this invention's strategy is that the sequence clusters mapped by high-copy-count data represent active sequence regions. Within these active sequence regions, there exist sequences with high activity and high fluorescence activation folds.

[0129] In this invention, aptamer sequence data is first trained using a Gaussian mixture variational encoder (GM-VAE) to learn a low-dimensional latent space representation of the sequence. The trained encoder can effectively capture the latent structural features in the aptamer sequence and map these features to a low-dimensional latent space.

[0130] Subsequently, the active sequence region was defined by mapping the "aptamer family" to which the high-copy sequence resides, thereby minimizing interference from false positive sequences in the selection process. Following this, a genetic evolution study was conducted focusing on the aptamer family of the active sequence region to design aptamers with high activity and high affinity.

[0131] The core advantage of this strategy lies in its ability to intelligently screen and optimize aptamers during the initial discovery process, guided by activity data. Compared to traditional screening methods that rely on single copy number and sequence matching, the activity-guided strategy significantly improves the experimental success rate and practical application value of aptamers, ensuring that the aptamers obtained in real-world applications have better performance and reliability.

[0132] Specifically, the following steps are included:

[0133] (1) Sequence space mapping: First, the first 5,000 to 10,000 sequences in SELEX sequencing are mapped into the latent space by the trained encoder.

[0134] (2) Adaptor family classification: Based on Gaussian mixture model (GMM) and K-means clustering algorithm, the sequences in the implicit space are divided into different aptor families.

[0135] (3) Define the active sequence region: Define the active sequence region based on the distribution of high copy number sequences and their corresponding aptamer families.

[0136] (4) Deep implicit evolution of active sequence region: Genetic evolution research on aptamer families in active sequence region, and optimization of sequence design through crossover, mutation and other operations to further enhance the activity and affinity of aptamers.

[0137] 1.4 Deep Implicit Evolution of Active Sequence Regions

[0138] Based on the active region-guided aptamer discovery strategy, after determining the active sequence regions from data of different rounds, this invention performs a implicit spatial genetic evolution strategy on these active regions.

[0139] This strategy mainly involves three Darwinian evolutionary operations: elitism, crossover, and mutation, aiming to efficiently screen for high-quality aptamer sequences with high activity and high fluorescence activation folds through deep evolutionary strategies.

[0140] Specifically, the steps include the following:

[0141] (1) The first evolutionary operation is elitism. This is implemented using the `sklearn.mixture.GaussianMixture` and `sklearn.cluster.KMeans` packages in the Scikit-learn library (version 1.0.2). The relevant parameters are shown in Table 1 below:

[0142] Table 1

[0143]

[0144] This paper introduces a parameter for rationally evaluating clustering performance—the silhouette coefficient—to assess the number of reasonable aptamer families calculated by two clustering algorithms. The silhouette coefficient works by calculating the average intra-cluster distance (a) and average nearest cluster distance (b) for each nucleic acid sequence, measuring the closeness of the sequence to its own cluster and its distance from other clusters, thus quantifying the quality of the clustering performance. Here, a cluster represents the aptamer family distribution predicted by this invention. The average intra-cluster distance (a) assesses the average distance between the nucleic acid sequence and other sequences within the same cluster, measuring the closeness of the nucleic acid sequence to the sequences in its own cluster. The average nearest cluster distance (b) assesses the average distance between the nucleic acid sequence data and all nucleic acid sequences in its nearest different clusters. The final comprehensive evaluation ranges from [-1, 1], with a larger value indicating a better current clustering effect and a more reasonable aptamer family distribution.

[0145]

[0146] Subsequently, based on two clustering algorithms, Gaussian Mixture Model (GMM) and K-means clustering, the cyclic clustering parameters "n_components" and "n_clusters" were calculated with values ​​ranging from 5 to 15, meaning each method was run 11 times. Furthermore, to avoid potential errors in model fitting, each evaluation was repeated 200 times, retaining the optimal clustering model, and outputting the average silhouette_score calculated by the optimal model with the current number of clusters. This provides a rational evaluation of the classification results of the "aptamer families" under the current clustering parameters.

[0147] Finally, based on the silhouette_scores obtained from 11 calculations, the optimal model is determined by the highest silhouette_score. This optimal model is then selected to predict the distribution of aptamer families and the elite sequences within each aptamer family. These elite sequences will play a crucial role in subsequent genetic evolution, exploring potential highly active aptamers in a larger sequence space through genetic operations such as crossover and mutation. Specifically, this is implemented using `sklearn.metrics.silhouette_score` in the Scikit-learn library (version 1.0.2).

[0148] (2) Uniform Order Crossover (UOX) operation determined by crossover rate: Generate a random number between 0 and 1. When the value is lower than the preset crossover rate threshold, perform a uniform order crossover (UOX) operation. The UOX operation ensures that the gene segments of the parent sequence are exchanged in a uniform and random manner, which preserves the characteristics of the parent sequence while introducing new variations.

[0149] Random Boolean mask-assisted sequence recombination: First, a random Boolean mask of the same length as the parent sequence is created. Then, based on the mask information, elements at corresponding positions are swapped between the parent sequences, resulting in two entirely new child sequences. This process increases the diversity of aptamer sequences. This cross-fusion strategy allows the characteristics of superior parents to be passed on to offspring and increases the diversity of aptamer sequences, thereby accelerating the evolutionary process of the entire aptamer family.

[0150] (3) Set the mutation probability and perform mutation on the selected offspring; randomly index the latent variable z of the offspring and replace it with a random normal distribution value. The mutation rate should be very small, considering that a large mutation rate will cause the algorithm to perform random search, thereby destroying the overall genetic evolution, so the mutation rate is set very small.

[0151] 1.5 Model Training and Optimization

[0152] This invention employs standard optimization algorithms (such as the Adam optimizer) to train the model, and selects a loss function suitable for the objectives of this invention (as shown in Equation 8). The training process uses common learning rate adjustment strategies and introduces an early stopping mechanism to prevent overfitting.

[0153] During the pre-training phase, the maximum number of iterations is 2000, and an appropriate learning rate adjustment method is used. In the fine-tuning phase, the learning rate is further reduced, and a stricter early stopping criterion is adopted; training terminates when the model's loss shows little change after 50 consecutive epochs. This method effectively avoids overfitting during training and improves the model's generalization ability.

[0154] Specifically, the following steps are included:

[0155] (1) The lower bound of evidence (ELBO) loss is used as the optimization objective, which consists of a reconstruction term and a regularization term. The reconstruction term uses the cross-entropy loss function to measure the model's ability to reconstruct the original input data, and the regularization term uses Kullback-Leibler divergence (KL divergence) to regularize the latent variables into a GMM manifold.

[0156] In the pre-training phase, the maximum number of iterations was set to 2000, the Adam optimizer was used, the learning rate was 1e-3, the training batch size was 512, and an early termination mechanism was introduced (the training was stopped if there was no significant improvement in the performance on the validation set after 10 consecutive iterations). In the fine-tuning phase, the training batch size remained at 512, the learning rate was reduced to 1e-4, and the early termination criterion was adjusted to no significant improvement in the performance after 15 consecutive iterations. Training and fine-tuning were implemented using the GPU version of the PyTorch framework and the Scikit-learn library.

[0157] Example 2

[0158] This invention proposes an activity-guided deep implicit evolution-based aptamer design method, aiming to improve the efficiency and accuracy of aptamer selection through a unique combination of deep learning and genetic algorithms. This method utilizes a pre-training strategy to learn the distribution of sequences in a low-dimensional latent space and generates efficient aptamer sequences through genetic evolution operations. The implementation process is as follows: Figure 1 As shown. To verify the effectiveness and feasibility of the method of the present invention, the present invention used the dataset DRA009383 in the public database DDBJ for experiments, demonstrating the application effect of the method on real data.

[0159] The specific implementation steps are as follows:

[0160] (1) Data preparation: This case study used two datasets. Dataset 1 was the DRA009383 dataset from the publicly available database DDBJ; Dataset 2 was the internal, unpublished SELEX high-throughput sequencing dataset provided by the collaborative group, which was used as the data for validating the method of this invention. The dataset contained high-throughput screening and sequencing data from a certain round of aptamer SELEX screening, and only contained sequence information.

[0161] Dataset 1 contains sequencing data targeting human glutamine transaminase 2, totaling 80,846 sequences, stored in FASTQ format. Dataset 2 contains sequencing data targeting an undisclosed fluorophore, totaling 1,048,576 sequences, stored in FASTA format.

[0162] (2) Data Preprocessing and Feature Extraction: Based on the overall processing flow of the method of this invention, the sequences in the dataset are first automatically preprocessed to remove duplicates, eliminate inaccurate sequencing information, and extract relevant features for each sequence. Dataset 1 contains 38,513 processed sequences, and Dataset 2 contains 485,735 processed sequences. Subsequently, the data are input into a Gaussian mixture deep variational autoencoder (GM-VAE) model for training, and the trained encoder is used to perform embedded dimensionality reduction on these sequences, mapping the sequences to a low-dimensional latent space, thereby capturing the potential similarities and structural features between sequences.

[0163] (3) Genetic algorithm evolution operation: Using the latent space genetic evolution strategy based on Gaussian mixture model, the dataset 1 ( Figure 2 ) and dataset 2 ( Figure 3 The sequence space is subjected to a three-step genetic evolution: elitism, crossover fusion, and mutation. Based on experimental results, active regions can be further determined, thereby executing the genetic evolution of active sequence regions. Two clustering algorithms are used to implement the division of sequence families and the discovery of center sequences in the elitism phase.

[0164] (4) Generate efficient aptamer sequences: After multiple rounds of evolution, the final generated novel aptamer sequences with potential high activity and rich structural diversity are generated.

[0165] Experimental Results and Analysis:

[0166] The aptamer sequences designed using the method of this invention were compared with those designed using traditional methods. The specific experimental results are as follows:

[0167] Dataset 1: When validating using the public dataset DRA009383, the method of this invention successfully designed 12 potentially highly active aptamer sequences and visualized their secondary structures. Figure 4These sequences represent active sequences from various aptamer families. Furthermore, sequence similarity calculations on the generated sequences show that all generated sequences are novel and did not appear in the training set. This indicates that the method of this invention can design potentially highly active novel aptamer sequences.

[0168] Dataset 2: When validating using an unpublished dataset provided by a collaborative research group, the method of this invention successfully designed 30 potential highly active aptamer sequences. Experimental verification showed that the fluorescence activation fold of 27 sequences was approximately 400 times that of the control sequences, with some sequences reaching approximately 600 times the fluorescence activation fold. This indicates that 86.7% of the sequences designed in this invention have been experimentally verified to have high fluorescence activation folds. Figure 5 Furthermore, sequence similarity calculations on the generated sequences show that all generated sequences are novel and did not appear in the training set. This indicates that the method of the present invention can design potentially highly active novel aptamer sequences.

[0169] discuss

[0170] This invention proposes a novel aptamer design method—activity-guided deep implicit evolution. This invention provides a solution to the aforementioned problems by learning the low-dimensional latent space distribution of sequences through a pre-training strategy and combining it with genetic algorithm evolutionary operations, offering a solution that can efficiently capture the complex distribution characteristics of aptamer sequencing data. On one hand, it accurately captures the complex distribution characteristics of aptamer sequencing data through a unique deep variational autoencoder and Gaussian mixture model; on the other hand, it focuses on discovering highly active aptamers using activity-guided and latent space genetic evolution strategies. Ultimately, it achieves efficient design of highly active nucleic acid sequences with high fluorescence activation folds, improving aptamer discovery efficiency, and demonstrating excellent performance in dataset compatibility and design universality, effectively filling the gaps in traditional methods.

[0171] The aptamer design method of this invention overcomes the limitations of traditional aptamer design methods by combining a Gaussian mixture variable autoencoder (GM-VAE) with a genetic algorithm. This method improves the efficiency and accuracy of aptamer selection by learning the low-dimensional latent space representation of the aptamer sequence. The core innovation of this technical solution lies in learning the latent features of aptamers in the latent space through an autoencoder and optimizing them using a genetic algorithm, thereby generating high-affinity aptamer sequences.

[0172] In this invention, a latent variable in the latent space of high-copy-number sequences is obtained through a trained encoder, constructing an initial population suitable for genetic evolution research. The innovation lies in ensuring that the genetic algorithm can generate a higher-quality initial population based on actual screening data through encoding and learning the latent space, thereby improving the efficiency of subsequent evolutionary processes. This latent space-based initial population construction method effectively avoids the shortcomings of traditional methods that rely on high-copy-number screening.

[0173] This invention proposes an activity-guided aptamer discovery strategy that optimizes the aptamer screening process by combining experimental activity data (such as affinity and fluorescence signals) of aptamers. This method not only relies on sequence information but also integrates experimental data of aptamers, resulting in aptamers with higher affinity and a higher success rate in experimental validation. The innovation lies in introducing activity data into the early stages of aptamer design, significantly improving screening efficiency and accuracy.

[0174] This invention innovatively applies genetic algorithms to the optimization process of aptamers, employing operations such as selection, crossover, and mutation to evolve the initial population. These evolutionary operations enable global search in a low-dimensional latent space, thereby discovering potential, high-affinity aptamer sequences. Genetic algorithms have significant advantages in optimizing the specificity and affinity of aptamers, especially when faced with complex multi-round screening data, effectively improving the efficiency of aptamer identification and optimization.

[0175] The technical solution proposed in this invention combines high-throughput screening data with deep learning technology, using machine learning models to optimize aptamer sequences. Particularly in high-throughput screening, by using deep learning-based models to predict and optimize aptamer sequences, this method improves the efficiency of aptamer discovery and reduces the time and cost required for experimental screening. The innovation lies in the combination of deep learning and high-throughput data, enhancing the ability to process complex data and the accuracy of aptamer prediction.

[0176] This invention learns the latent feature space of aptamer sequences through a pre-trained Gaussian mixture variational encoder (GM-VAE) and uses this model to generate candidate aptamer sequences. This method innovatively utilizes deep learning technology to model and predict the latent space of aptamers, optimizing the aptamer generation process based on manual design or random selection in traditional methods. The key to this technology lies in effectively capturing the structural information of sequences in a lower-dimensional space through a pre-trained model, thereby providing more accurate and efficient input data for genetic algorithms.

[0177] All documents mentioned in this invention are incorporated herein by reference as if each document were individually incorporated by reference. Furthermore, it should be understood that after reading the foregoing teachings of this invention, those skilled in the art can make various alterations or modifications to this invention, and these equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A method for aptamer design and optimization based on activity-guided deep implicit evolution, characterized in that, Includes the following steps: (s1) Provide a high-throughput screening aptamer sequence dataset, the dataset including nucleic acid sequence information and copy number information of aptamers; and randomly divide the aptamer sequence dataset into a training set and a test set according to a certain proportion; (s2) Data processing: Each nucleotide character in the nucleic acid sequence in the dataset is converted into a corresponding numerical index to obtain the numerical form of each nucleic acid sequence; and the numerical index is expanded into a vector to obtain a nucleic acid sequence representation composed of vectors; In this context, the vector representation of nucleotides in the nucleic acid sequences of the training set is called the first vector, and the vector representation of nucleotides in the nucleic acid sequences of the test set is called the second vector. (s3) Encoder-decoder training: The nucleic acid sequence representation formed by the first vector is input into the Gaussian mixture variational encoder (GM-VAE) for training, so that the trained encoder can capture the latent structural features and aptamer family distribution in the aptamer nucleic acid sequence and map these features to a low-dimensional latent space; and enable the decoder to generate the original input sequence; and then use the nucleic acid sequence representation formed by the second vector for testing; (s4) Data dimensionality reduction and clustering: The encoder trained in (s3) is used to perform embedded dimensionality reduction on the nucleic acid sequence of the target round number, thereby mapping the nucleic acid sequence to a low-dimensional latent space, capturing the potential similarity and structural features between nucleic acid sequences; and based on the potential similarity and structural features, different clusters or families of nucleic acid sequences are obtained; (s5) Selection of active sequence regions and deep implicit evolution: select clusters corresponding to high copy number nucleic acid sequences and define them as active sequence regions; and perform implicit spatial genetic evolution strategy on the active sequence regions; (s6) Obtain the optimized aptamer sequence: After the evolution in (s5), the designed and optimized aptamer sequence is obtained; and (s7) Decoder output: The designed and optimized aptamer sequence obtained using the decoder output (s6).

2. The method as described in claim 1, characterized in that, The different clusters of nucleic acid sequences refer to the family distribution of the aptamers corresponding to the nucleic acid sequences.

3. The method as described in claim 1, characterized in that, The method further includes the following steps: (s0) Provides single-round and / or multi-round nucleic acid aptamer sequence information and related experimental data obtained by SELEX technology screening, wherein the experimental data includes copy number information; and preprocesses the data to obtain a high-throughput screened aptamer sequence dataset; The preprocessing includes removing redundancy and measurement errors from the data.

4. The method as described in claim 1, characterized in that, In step (s1), the aptamer sequence dataset is randomly divided into a training set and a test set in a 9:1 ratio using the "train_test_split" function in the "sklearn.model_selection" module.

5. The method as described in claim 1, characterized in that, Step (s2) includes the following sub-steps: (s2a) Convert each character in the nucleic acid sequence in the dataset into a corresponding numerical index, that is, map nucleotides A, T / U, G, and C to 0, 1, 2, and 3 respectively; (s2b) Using the "nn.Embedding" module in the PyTorch framework version 1.5.0, each numerical index is expanded into a 32-dimensional vector to obtain a vector representation of each nucleotide; thus, a nucleic acid sequence representation composed of vectors is obtained; wherein, the vector representation of nucleotides in the nucleic acid sequences in the training set is the first vector, and the vector representation of nucleotides in the nucleic acid sequences in the test set is the second vector.

6. The method as described in claim 1, characterized in that, In step (s3), the data information input to the Gaussian hybrid variable encoder includes: the number of nucleic acid sequences, the dimension of the nucleic acid sequences, and the length of the nucleic acid sequences.

7. The method as described in claim 1, characterized in that, The Gaussian hybrid variable self-dividing encoder (GM-VAE) includes an encoder and a decoder.

8. The method as described in claim 7, characterized in that, The encoder is a convolutional neural network (CNN) architecture containing multiple skip connection layers.

9. The method as described in claim 8, characterized in that, Each skip connection layer consists of a convolutional layer, a batch normalization layer, and a Leaky ReLU nonlinear activation function.

10. The method as described in claim 1, characterized in that, The Gaussian Mixture Variational Encoder (GM-VAE) also includes a Gaussian Mixture Model (GMM).

11. The method as described in claim 1, characterized in that, In step (s5), the genetic evolution strategies include elitism, crossover, and mutation.

12. The method as described in claim 11, characterized in that, The elitism mentioned includes the following two algorithms: Gaussian Mixture Model (GMM) and K-means clustering algorithm.

13. An apparatus for aptamer design and optimization, characterized in that, include: (a) A data input module, configured to input a high-throughput filtered aptamer sequence dataset, the dataset including nucleic acid sequence information and copy number information of the aptamers; and to randomly divide the aptamer sequence dataset into a training set and a test set according to a certain proportion; (b) A data processing module configured to convert each nucleotide character in the nucleic acid sequence in the dataset into a corresponding numerical index, thereby obtaining a numerical form of each nucleic acid sequence; and to expand the numerical index into a first vector, thereby obtaining a nucleic acid sequence representation composed of the first vector; In this context, the vector representation of nucleotides in the nucleic acid sequences of the training set is called the first vector, and the vector representation of nucleotides in the nucleic acid sequences of the test set is called the second vector. (c) An analysis module, which is configured to perform the following operations: (i) Encoder-decoder training: The nucleic acid sequence representation composed of the first vector is input into the Gaussian mixture variational encoder (GM-VAE) for training, so that the trained encoder can capture the latent structural features and aptamer family distribution in the aptamer nucleic acid sequence and map these features to a low-dimensional latent space; and enable the decoder to generate the original input sequence; and then use the nucleic acid sequence representation composed of the second vector for testing; (ii) Data dimensionality reduction and clustering: The trained encoder is used to perform embedded dimensionality reduction on the nucleic acid sequence of the target round number, thereby mapping the nucleic acid sequence to a low-dimensional latent space, capturing the potential similarity and structural features between nucleic acid sequences; and different clusters or families of nucleic acid sequences are obtained based on the potential similarity and structural features. (iii) Selection of active sequence regions and deep implicit evolution: select clusters corresponding to high copy number nucleic acid sequences and define them as active sequence regions; and perform implicit spatial genetic evolution strategy on the active sequence regions; (iv) Obtain the optimized aptamer sequence: After the evolution in (iii), the designed and optimized aptamer sequence is obtained; (d) Output module: The output module is configured to output the designed and optimized aptamer sequence.

14. The apparatus as claimed in claim 13, characterized in that, The device further includes (a0) a data preprocessing module, which is configured to: preprocess the nucleic acid aptamer sequence information and related experimental data obtained by SELEX technology screening in one round and / or multiple rounds for a specific target, thereby obtaining a high-throughput screened aptamer sequence dataset; The experimental data includes copy number information; the preprocessing includes removing redundancy and sequences with measurement errors from the data.

15. The apparatus as claimed in claim 13, characterized in that, Execute the following procedure in the data processing module: (b1) Convert each character in the nucleic acid sequence in the dataset into a corresponding numerical index, that is, map nucleotides A, T / U, G, and C to 0, 1, 2, and 3 respectively; (b2) Using the "nn.Embedding" module in the PyTorch framework version 1.5.0, each numerical index is expanded into a 32-dimensional vector to obtain a vector representation of each nucleotide; thus obtaining a nucleic acid sequence representation composed of vectors; wherein, the vector representation of nucleotides in the nucleic acid sequences in the training set is the first vector, and the vector representation of nucleotides in the nucleic acid sequences in the test set is the second vector.

16. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the method of claim 1.

Images