Lncrna subcellular localization prediction method and system for imbalanced data

Abstract

The application belongs to the field of subcellular localization prediction, and provides an lncRNA subcellular localization prediction method and system for unbalanced data, in order to fully utilize the sequence information of lncRNA, physical and chemical mode features and distributed representation features of nucleic acids are extracted, and two basic classifiers including a convolutional neural network and a gated recurrent unit are integrated, a subcellular classification result is obtained through adaptive weighted averaging of the two models, and subcellular localization is carried out based on the classification result, so that the precision of subcellular localization prediction on an unbalanced data set can be improved. In order to solve the problem that the model performs poorly on the unbalanced data set, a label distribution margin perception loss function is used in the training process. Compared with traditional machine learning models and existing predictors, the method has stronger tolerance to class imbalance and robustness.

Description

Technical Field

[0001] This invention belongs to the field of subcellular localization prediction, and particularly relates to a method and system for predicting lncRNA subcellular localization for imbalanced data. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Determining subcellular localization is crucial for understanding the interaction patterns, post-transcriptional or co-transcriptional regulatory modifications, and diverse roles of lncRNAs in cellular homeostasis. lncRNAs regulate transcriptional processes in the nucleus through chromatin interactions and provide scaffolding for the spatial organization of the nuclear compartments. Recent studies have revealed a greater abundance of cytoplasmic lncRNAs than previously thought, which play a coordinating and constructive role in the cytoplasmic environment. Therefore, lncRNAs located in the cytoplasm or nucleus play important roles in cellular processes, and determining their subcellular localization is essential for understanding their functions. Increasing research focuses on using computational methods to determine the subcellular localization of lncRNAs.

[0004] However, existing technologies use the GloVe method as the nucleotide sequence encoding method, and then input the encoding into a classifier composed of CNN, LSTM and MLP for classification and prediction. The drawback of this method is that it is not optimized for imbalanced data, and it only uses a single feature encoding, resulting in poor performance and poor robustness on cell lines with imbalanced data. Summary of the Invention

[0005] To address at least one of the technical problems mentioned above, this invention provides a method and system for predicting subcellular localization of lncRNAs for imbalanced data. In order to fully utilize the sequence information of lncRNAs, it extracts two types of features, including physicochemical pattern features and distributed representation features of nucleic acids, and integrates two basic classifiers, including convolutional neural networks (CNN) and gated recurrent units (GRU), thus solving the problem of poor model performance on imbalanced datasets.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The first aspect of the present invention provides a method for predicting subcellular localization of lncRNAs for imbalanced data, comprising the following steps:

[0008] Obtain long non-coding RNA sequences;

[0009] Physicochemical pattern features were obtained by encoding long non-coding RNA sequences using the normalized Moreau-Broto autocross correlation method.

[0010] A corpus was obtained by processing long non-coding RNA sequences. Based on the corpus, the Word2Vec sequence word embedding algorithm based on the hopscotch model was used to encode two-dimensional nucleic acid distributed representation features.

[0011] The first classification result is obtained by combining the first dimension of nucleic acid distributed representation features and the first classification model;

[0012] The second classification result is obtained by fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features with the second classification model.

[0013] The subcellular classification result is obtained by adaptively weighting the first and second classification results, and the subcellular localization is performed based on the classification result.

[0014] A second aspect of the present invention provides a lncRNA subcellular localization prediction system for imbalanced data, comprising:

[0015] The data acquisition module is used to acquire long non-coding RNA sequences;

[0016] The feature extraction module uses the normalized Moreau-Broto automatic cross-correlation method to encode long non-coding RNA sequences to obtain physicochemical pattern features;

[0017] A corpus was obtained by processing long non-coding RNA sequences. Based on the corpus, the Word2Vec sequence word embedding algorithm based on the hopscotch model was used to encode two-dimensional nucleic acid distributed representation features.

[0018] The classification module is used to combine the first dimension of nucleic acid distributed representation features and the first classification model to obtain the first classification result;

[0019] The second classification result is obtained by fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features with the second classification model.

[0020] The localization module is used to adaptively weight and average the first and second classification results to obtain the subcellular classification result, and to perform subcellular localization based on the classification result.

[0021] A third aspect of the present invention provides a computer-readable storage medium.

[0022] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the lncRNA subcellular localization prediction method for imbalanced data as described in the first aspect above.

[0023] A fourth aspect of the present invention provides a computer device.

[0024] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps in the lncRNA subcellular localization prediction method for imbalanced data as described in the first aspect above.

[0025] Compared with the prior art, the beneficial effects of the present invention are:

[0026] To fully utilize lncRNA sequence information, physicochemical pattern features and distributed representation features of nucleic acids were extracted. Two basic classifiers, a convolutional neural network (CNN) and a gated recurrent unit (GRU), were integrated. Subcellular classification results were obtained through adaptive weighted averaging of the two models. Subcellular localization was then performed based on these classification results. During training, the marginal awareness of label distribution (LDAM) loss function was utilized to improve the accuracy of subcellular localization prediction on imbalanced datasets. Compared with traditional machine learning models, this approach provides more detailed classification and stronger robustness on imbalanced cell lines.

[0027] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0028] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0029] Figure 1 This is a flowchart of the lncRNA subcellular localization prediction method for imbalanced data according to Embodiment 1 of the present invention;

[0030] Figure 2 This is a block diagram of lncRNA subcellular localization prediction for imbalanced data according to Embodiment 1 of the present invention;

[0031] Figures 3(a)-3(d) This is a comparison of localization performance on different cell line datasets in Embodiment 1 of the present invention. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0034] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0035] Terminology Explanation

[0036] Word2vec is a group of models used to generate word vectors. These models are shallow, two-layer neural networks trained to reconstruct linguistic word text. The network represents words and needs to guess the input words in adjacent positions. Under the bag-of-words assumption in word2vec, the order of words is unimportant. After training, the word2vec model can be used to map each word to a vector, which can be used to represent the relationship between words. This vector is the hidden layer of the neural network.

[0037] Skip-gram model:

[0038] The Skip-gram model is a simple yet highly practical model. In natural language processing, the selection of corpus is a crucial issue: firstly, the corpus must be sufficient.

[0039] On the one hand, the dictionary must have a large enough vocabulary, and on the other hand, it must contain as many sentences as possible that reflect the relationships between words. For example, only when there are as many sentences as possible in the corpus such as "fish swim in the water" can the model learn the semantic and grammatical relationships in the sentence. This is the same principle as humans learning natural language. The more times it is repeated, the more it will imitate.

[0040] Second, the corpus must be accurate. That is, the selected corpus must accurately reflect the semantic and grammatical relationships of the language. For example, in Chinese, the corpus of the *People's Daily* is relatively accurate. However, more often than not, the concern about accuracy arises not from the selection of the corpus, but from the processing method.

[0041] In n-gram models, the limited window size prevents the relationship between words outside the window and the current word from being accurately reflected in the model. Simply increasing the window size would increase the training complexity. The Skip-gram model effectively solves these problems.

[0042] As the name suggests, Skip-gram means "skipping certain symbols". For example, the sentence "This song is sung so well" has four ternary phrases: "This song is sung well", "The song is sung really well", "The song is sung really well", and "It's really good". However, we find that the original meaning of this sentence is "This song is sung well", but the above four ternary phrases do not reflect this information.

[0043] The Skip-gram model, however, allows certain words to be skipped, thus forming the 3-gram phrase "this song is sung well". If two words are allowed to be skipped, it becomes a 2-Skip-gram.

[0044] Example 1

[0045] Reference Figures 1-2 This embodiment provides a method for predicting subcellular localization of lncRNAs in imbalanced data, including the following steps:

[0046] Step 1: Obtain long non-coding RNA sequences;

[0047] Step 2: The physicochemical pattern features of long non-coding RNA sequences were obtained by encoding them using the normalized Moreau-Broto autocross correlation method;

[0048] Step 3: Process the long non-coding RNA sequences to obtain a corpus. Based on the corpus, use the Word2Vec sequence word embedding algorithm based on the hopscotch model to encode and obtain two-dimensional nucleic acid distributed representation features.

[0049] Step 4: Combine the first dimension of nucleic acid distributed representation features with the first classification model to obtain the first classification result;

[0050] Step 5: After fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features, combine them with the second classification model to obtain the second classification result;

[0051] Step 6: Adaptively weight the first classification result and the second classification result to obtain the subcellular classification result, and perform subcellular localization based on the classification result.

[0052] In step 2, the process of encoding long non-coding RNA sequences using the Normalized Moreau-Broto automatic cross correlation (NMBACC) to obtain physicochemical pattern features includes:

[0053] Step 201: Divide the lncRNA sequence into several dinucleotide sets and represent them as follows:

[0054] lncRNA = S1, S2, S 3, …,S n-1 ,S n ,S n ∈(AA,AC,……,TT) (1)

[0055] Among them, S n For the nth dinucleotide, S n The Cartesian product of the nucleotide set AGCT and itself, i.e., all dinucleotides.

[0056] Step 202: The NMBACC formula combined with the dinucleotide physicochemical property matrix is ​​used to extract sequence features, as described below:

[0057]

[0058]

[0059] Where λ is the autocorrelation lag of formula (1), i is the i-th attribute in the nucleotide physicochemical property matrix, j is the lag value, which is equal to the value of λ, and L is the length of the dinucleotide group. This means that a dinucleotide is equal to S. l The value of the i-th attribute. V NMBACC The meaning is the NMBACC eigenvector, and d(i,j) means the lag value of the physicochemical properties of the i-th nucleotide.

[0060] Using only normalized Moreau-Broto autocross correlation feature descriptors leads to the loss of lncRNA sequence order information. To overcome this limitation and effectively utilize sequence context information, sequence embedding is used to combine the distributed representation of nucleic acids with NMBACC features to compensate for the loss of lncRNA sequence order information.

[0061] This embodiment uses the word2vec framework, which is an efficient distributed representation sequence word embedding algorithm used in the field of natural language processing. In recent years, it has been widely used for biological sequences, such as proteins, mRNA, and non-coding RNA.

[0062] In step 3, the process of processing long non-coding RNA sequences to obtain a corpus, and then encoding the distributed representation features of nucleic acids using the Word2Vec sequence word embedding algorithm based on the hopscotch model, specifically includes:

[0063] Step 301: Process the data of the long non-coding RNA sequence training set to obtain a corpus. Specifically, use a word segmenter with a window size of 3 and a span of 1 to decompose all lncRNA sequences into sentences containing multiple trinucleotides, and integrate all sentences to obtain the corpus.

[0064] Step 302: Using the corpus as training data, train two different word vector embedding models using Word2Vec based on the hopscotch model to obtain two-dimensional nucleic acid distributed representation features;

[0065] The first word embedding model outputs a 16-dimensional nucleic acid distributed representation feature, which is then fed into the first classification model to process the word embedding in order to more effectively obtain the contextual information of the RNA sequence.

[0066] The second word vector embedding model outputs a 64-dimensional nucleic acid distributed representation feature, which is then combined with the second classification model for classification.

[0067] To ensure that the distributed representation length of all sequences is consistent and to reduce dimensionality to save computational resources, an adaptive average pooling technique was used to reduce the feature representation dimension of each lncRNA sequence from 64*(L-3+1) dimensions to 64 dimensions.

[0068] Step 4: Combine the first dimension of nucleic acid distributed representation features with the first classification model to obtain the first classification result;

[0069] GRU is a type of recurrent neural network. Both GRU and LSTM were proposed to solve the problems of gradient vanishing and exploding, as well as long-term memory problems in backpropagation.

[0070] GRU achieved results comparable to LSTM, and is easier to train than LSTM, which can significantly improve training efficiency.

[0071] Therefore, the first classification model in this embodiment uses a gated recurrent unit (GRU) neural network.

[0072] The 16-dimensional nucleic acid distributed representation features are fed into a gate-controlled recurrent unit (GRU) neural network, and the first classification result is obtained after training and testing.

[0073] Step 5: After fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features, combine them with the second classification model to obtain the second classification result;

[0074] The 64-dimensional Word2Vec features and the physicochemical pattern features extracted from long non-coding RNA sequences in step 2 using the normalized Moreau-Broto automatic cross-correlation method are concatenated into a vector and fed into a convolutional neural network (CNN). After training and testing, the second classification result is obtained.

[0075] The new model integrating convolutional neural networks based on convolutional layers and recurrent neural networks based on gated recurrent units mainly consists of convolutional layers, pooling layers, gated recurrent units, and fully connected layers.

[0076] Convolutional layers are used to extract latent information from input features, while pooling layers can effectively reduce the size of feature vectors and the number of parameters in the final connected layers, thereby accelerating computation and preventing overfitting.

[0077] Step 6: Adaptively weighted average the first and second classification results to obtain the subcellular classification results.

[0078] To predict the subcellular localization of LncRNAs, fully connected layers were used in both the first and second classification models, and the final result was obtained by weighted averaging of the results from the two sub-models.

[0079] When a dataset has class imbalance, it can lead to poor prediction performance or even the inability to predict classes with fewer samples, because deep learning models often do not learn enough feature representations for the minority of classes.

[0080] Therefore, this embodiment utilizes the Label Distribution Margin Awareness (LDAM) loss function during training for both the convolutional neural network based on convolutional layers and the recurrent neural network based on gated recurrent units. This function encourages a larger margin for the minority class, resulting in a more balanced model performance. The equation for the LDAM loss function is as follows:

[0081]

[0082] Where (x,y) is a sample, f is the model, and z y z is the output value of class y obtained after building a model for a sample x with label y. j C is the output value of class j obtained from this sample, and C is a hyperparameter that needs to be adjusted.

[0083] To facilitate comparison with other methods, the method of this invention is named lncLocator-imb. It is compared with three existing methods, including lncLocator, DeepLncLoc, and lncLocator 2.0, on each cell line. All predictors for comparison are performed and evaluated using an independent test set.

[0084] Detailed results of the predictive performance of the comparative predictors are shown in Table 1.

[0085] Table 1 Performance comparison on different cell line datasets

[0086]

[0087]

[0088] The formulas for calculating the Matthews correlation coefficient (MCC), mean accuracy (BACC), and macro average F-score used in the performance comparison are as follows:

[0089]

[0090]

[0091]

[0092]

[0093]

[0094] Among them, TP, FP, TN, and FN represent the number of true positives, false positives, true negatives, and false negatives, respectively. (i) FP (i) and FN () These represent the number of true positives, false positives, and false negatives for the i-th class, respectively.

[0095] lncLocator feeds features into an ensemble machine learning model to identify five distinct localizations. DeepLncLoc uses Word2Vec to encode nucleotide sequences into embeddings and feeds these embeddings into a convolutional neural network.

[0096] Both lncLocator and DeepLncLoc originally identified five subcellular localizations, so their network structures were modified to output only two subcellular localizations. lncLocator uses GloVe to encode nucleotide sequences and further feeds them into a regressor composed of CNN, LSTM, and MLP.

[0097] from Figures 3(a)-3(d)As shown in Table 1, lncLocator-imb achieves better performance than lncLocator, DeepLncLoc, and lncLocator 2.0. lncLocator-imb performs best, followed by DeepLncLoc and lncLocator 2.0, while lncLocator ranks last. Compared to DeepLncLoc, lncLocator-imb's MCC is increased by an average of 4.38%. Furthermore, lncLocator-imb's average macroscopic F-measure, BACC, and AUC are 2.11%, 2.34%, and 5.69% higher than DeepLncLoc, respectively. This demonstrates the effectiveness of the proposed deep learning framework on imbalanced datasets.

[0098] Example 2

[0099] This embodiment provides a lncRNA subcellular localization prediction system for imbalanced data, including:

[0100] The data acquisition module is used to acquire long non-coding RNA sequences;

[0101] The feature extraction module uses the normalized Moreau-Broto automatic cross-correlation method to encode long non-coding RNA sequences to obtain physicochemical pattern features;

[0102] A corpus was obtained by processing long non-coding RNA sequences. Based on the corpus, the Word2Vec sequence word embedding algorithm based on the hopscotch model was used to encode two-dimensional nucleic acid distributed representation features.

[0103] The classification module is used to combine the first dimension of nucleic acid distributed representation features and the first classification model to obtain the first classification result;

[0104] The second classification result is obtained by fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features with the second classification model.

[0105] The localization module is used to adaptively weight and average the first and second classification results to obtain the subcellular classification result, and to perform subcellular localization based on the classification result.

[0106] Example 3

[0107] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the lncRNA subcellular localization prediction method for imbalanced data as described in Embodiment 1 above.

[0108] Example 4

[0109] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the lncRNA subcellular localization prediction method for imbalanced data as described in Embodiment 1 above.

[0110] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0111] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0115] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting subcellular localization of lncRNAs in imbalanced data, characterized in that, Includes the following steps: Obtain long non-coding RNA sequences; Physicochemical pattern features were obtained by encoding long non-coding RNA sequences; A corpus was obtained by processing long non-coding RNA sequences. Based on the corpus, a Word2Vec sequence word embedding algorithm based on a hopscotch model was used to encode two-dimensional distributed nucleic acid representation features: The data from the long non-coding RNA sequence training set were processed to obtain a corpus; Using a corpus as training data, two different word embedding models were trained using Word2Vec based on a hopscotch model. Based on the first word embedding model, a 16-dimensional nucleic acid distributed representation feature is obtained as the first dimension of the nucleic acid distributed representation feature. Based on the second word embedding model, a 64-dimensional nucleic acid distributed representation feature is obtained as the second dimension of the nucleic acid distributed representation feature. The first classification result is obtained by combining the first dimension of nucleic acid distributed representation features and the first classification model; The second classification result is obtained by fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features with the second classification model. The subcellular classification result is obtained by adaptively weighting the first and second classification results, and the subcellular localization is performed based on the classification result.

2. The lncRNA subcellular localization prediction method for imbalanced data as described in claim 1, characterized in that, The normalized Moreau-Broto autocross correlation method was used to encode the physicochemical pattern features of the long non-coding RNA sequence.

3. The lncRNA subcellular localization prediction method for imbalanced data as described in claim 1, characterized in that, The method for processing long non-coding RNA sequences to obtain a corpus involves using a word segmenter with a window size of 3 and a span of 1 to decompose all long non-coding RNA sequences into sentences containing multiple trinucleotides, and then integrating all the sentences to obtain the corpus.

4. The lncRNA subcellular localization prediction method for imbalanced data as described in claim 2, characterized in that, The physicochemical pattern characteristics obtained by encoding long non-coding RNA sequences using the normalized Moreau-Broto automatic cross-correlation method are as follows: in, It is the [number]th [item] in the nucleotide physicochemical property matrix. One attribute, It is a lag value, which is equal to The value of , where L is the length of the dinucleotide sequence. This means that a dinucleotide is equal to Time The value of each attribute.

5. The lncRNA subcellular localization prediction method for imbalanced data as described in claim 1, characterized in that, The first classification model uses a gated recurrent unit neural network, and the second classification model uses a convolutional neural network.

6. The lncRNA subcellular localization prediction method for imbalanced data as described in claim 1, characterized in that, During the training of the first and second classification models, the marginal perception loss function of label distribution is used.

7. A lncRNA subcellular localization prediction system for imbalanced data, characterized in that, include: The data acquisition module is used to acquire long non-coding RNA sequences; The feature extraction module encodes long non-coding RNA sequences to obtain physicochemical pattern features; A corpus was obtained by processing long non-coding RNA sequences. Based on the corpus, a Word2Vec sequence word embedding algorithm based on a hopscotch model was used to encode two-dimensional distributed nucleic acid representation features: The data from the long non-coding RNA sequence training set were processed to obtain a corpus; Using a corpus as training data, two different word embedding models were trained using Word2Vec based on a hopscotch model. A 16-dimensional nucleic acid distributed representation feature was obtained based on the first word embedding model, and a 64-dimensional nucleic acid distributed representation feature was obtained based on the second word embedding model. The classification module is used to combine the first dimension of nucleic acid distributed representation features and the first classification model to obtain the first classification result; The second classification result is obtained by fusing the physicochemical pattern features and the second-dimensional nucleic acid distributed representation features with the second classification model. The localization module is used to adaptively weight and average the first and second classification results to obtain the subcellular classification result, and to perform subcellular localization based on the classification result.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the steps in the lncRNA subcellular localization prediction method for imbalanced data as described in any one of claims 1-6.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the lncRNA subcellular localization prediction method for imbalanced data as described in any one of claims 1-6.

Images