5hmC prediction method and system based on multi-branch deep learning and adaptive feature screening

Abstract

The invention discloses a 5hmC prediction method based on multi-branch deep learning and adaptive feature screening. The 5hmC prediction method comprises the following steps: S1, constructing a positive and negative sample set; s2, sequence type feature coding, physical and chemical attribute feature coding and statistical class feature coding are carried out on the positive and negative samples respectively, and a multi-dimensional feature set is generated; s3, respectively inputting the features of different coding types into corresponding neural network branches for processing; s4, screening a plurality of feature subsets with the maximum contribution to 5hmC prediction; and S5, training a prediction model comprising a plurality of neural network branches, a self-attention mechanism and a classifier for predicting the 5hmC modification site. According to the scheme, by combining different types of features such as sequence feature coding, physical and chemical attribute feature coding and statistical feature coding, information of multiple layers such as a local mode, a global mode, physical and chemical attributes and base coding of the DNA sequence is comprehensively depicted, through multi-feature fusion, single-surface description of sequence information by an existing method is avoided, and the method is convenient to implement and high in practicability. Therefore, the identification capability of the model on the 5hmC modification site is greatly improved.

Description

Technical Field

[0001] The present invention belongs to the field of computer bioinformatics, and particularly relates to a 5hmC prediction method and system based on multi-branch deep learning and adaptive feature screening. Background Art

[0002] 5-hydroxymethylcytosine (5hmC) is an important form of DNA methylation and plays a key role in gene expression regulation, cell differentiation, and diseases such as cancer and neurodegenerative diseases. The change in the modification level of 5hmC is closely related to various pathological states, so its research occupies an important position in the field of epigenetics. Accurately identifying and analyzing the distribution and regulatory mechanism of 5hmC can not only provide a new perspective on the genomic regulatory network but also provide a theoretical basis for disease diagnosis and treatment. However, due to the weak signal of 5hmC and its similarity to other DNA modifications (such as 5mC), the detection of 5hmC still faces great challenges.

[0003] The current technologies for 5hmC detection are divided into two categories: experimental methods and computational methods. Experimental methods mainly include high-throughput sequencing technologies (such as TAB-Seq, oxBS-Seq) and biomarker methods. High-throughput sequencing efficiently distinguishes 5hmC from other modifications through the combination of chemical modification and sequencing, but has problems such as high cost, cumbersome steps, and high requirements for sample quality, which limit its large-scale application. The biomarker method relies on enzymatic reactions and antibody labeling technologies and also faces challenges such as high cost and low throughput, making it difficult to meet the needs of large-scale data analysis.

[0004] To overcome the limitations of experimental methods, computational methods have gradually become a research hotspot, especially driven by machine learning and deep learning. These methods do not rely on expensive experimental equipment and can show obvious advantages in large-scale screening. Traditional machine learning methods, such as support vector machines (SVM) and XGBoost, combine the features of DNA sequences for prediction, with relatively high computational efficiency and can meet certain research needs. However, these methods are difficult to capture complex patterns in the sequences and have limited ability to process high-dimensional features, resulting in limited improvement in model performance.

[0005] Deep learning methods, through technologies such as convolutional neural networks (CNN) and recurrent neural networks (RNN), have significantly improved the ability to extract sequence features, especially outstanding in capturing local patterns and long-range dependencies. However, these methods are prone to overfitting on small sample datasets and have weak ability to fuse multi-source features, still facing certain challenges.

[0006] Despite the advantages and disadvantages of experimental and computational methods, there are still some key problems in current 5hmC detection methods: experimental methods are unable to achieve large-scale detection due to high costs and low throughput; traditional computational methods are insufficient in sequence complexity modeling and difficult to adapt to the increasing data requirements; while deep learning methods suffer from overfitting problems and insufficient feature fusion. Therefore, developing an efficient and robust 5hmC prediction method that can fully integrate multi-source features and improve the generalization ability of the model is a technical challenge that urgently needs to be solved. Summary of the Invention

[0007] The object of the present invention is to propose a 5hmC prediction method and system based on multi-branch deep learning and adaptive feature screening for the problems of the existing technology.

[0008] To achieve the above object, the present invention adopts the following technical solutions: A 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, the method comprising: S1. Screening common 5hmC modification sites from various sequencing data to construct positive and negative sample sets; S2. Performing sequence-type feature encoding, physicochemical property feature encoding, and statistical-type feature encoding on the positive and negative samples respectively to generate a multi-dimensional feature set; S3. Inputting features of different encoding types into corresponding neural network branches for processing; S4. Dynamically weighting and fusing the features output by each neural network branch through a self-attention mechanism, and combining dynamic SHAP value calculation and multi-stage LASSO regression to screen several feature subsets that contribute the most to 5hmC prediction; S5. Adopting an adaptive regularization and adaptive resampling strategy to train a prediction model including multiple neural network branches, a self-attention mechanism, and a classifier through a class imbalance loss function for predicting 5hmC modification sites.

[0009] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, the sequence-type feature encoding includes any one or more of One-Hot encoding, Binary encoding, Kmer encoding, PSTNPsS encoding, Mismatch encoding, and ENAC encoding; The physicochemical property feature encoding includes any one or more of PCP encoding and SCPseKNC encoding; The statistical-type feature encoding includes any one or more of DAC encoding, ANF encoding, and ASDC encoding; The neural network branches include a CNN branch, an LSTM branch, and a DNN branch.

[0010] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, in step S2, Kmer frequency encoding, PSTNPsS encoding, One-Hot encoding, Binary encoding, SCPseKNC encoding, and DAC encoding are respectively performed on the positive and negative samples to generate a multi-dimensional feature set; The neural network branches include two CNN branches that respectively process Kmer frequency encoding features and PSTNPsS encoding features, two LSTM branches that respectively process SCPseKNC encoding features and Binary encoding features, and two DNN branches that respectively process One-Hot encoding features and DAC encoding features.

[0011] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, the CNN branch includes multiple convolutional kernels for extracting local sequence-type features of Kmer frequency encoding and PSTNPsS encoding; The LSTM branch adopts a bidirectional structure for processing the physicochemical property time-series features encoded by SCPseKNC and the binary sequence dependency encoded by Binary; The DNN branch includes a fully connected layer and a non-linear activation function for extracting the global non-linear associations of One-Hot encoding features and DAC encoding.

[0012] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, in step S3, each neural network branch outputs a feature vector with a fixed and identical dimension, and the features output by each neural network branch are dynamically weighted and fused by the self-attention mechanism to capture the complementary information between different features.

[0013] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, step S4 specifically includes: Stack the outputs of each neural network branch into a tensor with a specific shape; Calculate the importance of different features through the multi-head attention mechanism and perform weight normalization; After calculating the weighted features, introduce residual connections and layer normalization to keep the model stable during deep training; Perform average pooling on the fused features to form a candidate feature vector with a fixed length; Screen out the final feature vector from the candidate feature vector through SHAP value calculation and multi-stage LASSO regression as the input of the classifier.

[0014] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, in step S4, the SHAP value calculation adopts a dynamic update mechanism. After each training round, SHAP is recalculated. Through multiple rounds of calculation, the change of SHAP value with training is monitored, and the time decay weight is used to smooth the SHAP change. Finally, several features with the highest SHAP values are selected, such as the top 50%.

[0015] After several features are selected through SHAP value calculation, they enter the multi-stage LASSO regression screening stage. Finally, a part of the features with the greatest impact on 5hmC prediction are retained from the selected several features, such as the top 50%.

[0016] In the above 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, in step S4, the process of screening features by SHAP value also includes multi-model cross-comparison: Use XGBoost, LightGBM, and random forest to train models, and calculate SHAP values respectively; Select a part of the features that rank in the top set percentage, such as 50%, in multiple models and whose importance consistency coefficient is higher than the set value.

[0017] A 5hmC prediction system based on multi-branch deep learning and adaptive feature screening includes: A data preprocessing module for screening common 5hmC modification sites from various sequencing data and constructing positive and negative sample sets according to the relevant characteristic parameters of the modification sites; A multi-source feature encoding module for performing sequence-type feature encoding, physicochemical property feature encoding, and statistical-type feature encoding on the positive and negative sample sets to generate a multi-dimensional feature set; A multi-branch deep learning module includes multiple neural network branches and a self-attention fusion unit; the multiple neural network branches are used to process the features of the corresponding encoding types respectively, and the self-attention fusion unit is used to dynamically weight and fuse the features output by each neural network branch; A dynamic feature screening module includes SHAP calculation and LASSO regression, and is used to combine dynamic SHAP value calculation and multi-stage LASSO regression to screen several feature subsets that contribute the most to 5hmC prediction from the features output by the self-attention fusion unit, such as screening the top 50% or top 30% of the feature subsets with the greatest contribution.

[0018] A training module uses an adaptive regularization and adaptive resampling strategy to train a prediction model including multiple neural network branches, a self-attention mechanism, and a classifier through a class imbalance loss function for predicting 5hmC modification sites.

[0019] A 5hmC prediction system based on multi-branch deep learning and adaptive feature screening includes: An interaction module, configured to receive a DNA context sequence to be predicted and output a prediction result; A multi-source feature encoding module, configured to perform sequence-based feature encoding, physicochemical property feature encoding, and statistical feature encoding on the DNA context sequence to be predicted to generate a multi-dimensional feature set; A multi-branch deep learning module, including multiple neural network branches and a self-attention fusion unit; the multiple neural network branches are configured to process features of corresponding encoding types respectively, and the self-attention fusion unit dynamically weights and fuses the features output by each neural network branch; A feature screening module, pre-storing feature types obtained through dynamic SHAP value calculation and multi-stage LASSO regression screening during the training process, and configured to extract the pre-stored feature types from the features output by the neural network branches as a feature subset for output; A classification module, configured to output a prediction result based on the feature subset output by the feature screening module.

[0020] The advantages of the present invention are as follows: 1. By combining different types of features such as Kmer, One-Hot, PSTNPsS, and PCP, this solution comprehensively depicts information at multiple levels such as the local pattern, global pattern, physicochemical properties, and base encoding of the DNA sequence. Through multi-feature fusion, it avoids the single-sided description of sequence information in existing methods, thereby greatly improving the model's ability to identify 5hmC modification sites; 2. For different feature forms (such as sequence-based features, physicochemical property features, numerical features, etc.), this solution uses different deep learning network branches (CNN, LSTM, DNN) for modeling, giving full play to the advantages of various features and improving the model's ability to capture different features. In addition, the introduction of a self-attention fusion module can dynamically weight the outputs of each branch, improve the efficiency of feature fusion, and further improve the model performance; 3. In the feature screening stage, this solution adopts strategies such as dynamic SHAP calculation, time decay, multi-model cross-comparison, and multi-stage LASSO screening to ensure the stability and interpretability of feature screening, avoid the overfitting problem caused by redundant high-dimensional features, and thus improve the model's generalization ability; 4. This solution has a more flexible and intelligent feature screening method compared with existing site prediction methods, and adopts a branch strategy for different forms of features, having higher efficiency and effect in dealing with complex features compared with the prior art. Description of the Drawings

[0021] Figure 1 is a method flow chart of the 5hmC prediction method based on multi-branch deep learning and adaptive feature screening of the present invention; Figure 2It is the model architecture diagram of the 5hmC prediction method based on multi-branch deep learning and adaptive feature screening of the present invention; Figure 3 It is the accuracy test result diagram of 12 encoding methods based on the traditional machine learning model XGBoost in the 5hmC recognition task; Figure 4 It is the model performance comparison diagram of the embodiment of the present invention; Figure 5 It is the performance of the accuracy performance index under different encoded feature data; Figure 6 It is the performance of the area under the ROC curve performance index under different encoded feature data; Figure 7 It is the performance of the precision performance index under different encoded feature data; Figure 8 It is the performance of the recall performance index under different encoded feature data; Figure 9 It is the performance of the Matthews correlation coefficient performance index under different encoded feature data; Figure 10 It is the performance of the F1 score performance index under different encoded feature data; Figure 11 It is the system composition in the model construction stage of the 5hmC prediction method based on multi-branch deep learning and adaptive feature screening of the present invention; Figure 12 It is the system composition in the model application / test stage of the 5hmC prediction method based on multi-branch deep learning and adaptive feature screening of the present invention.

[0022] Reference numerals: interaction module 1; multi-source feature encoding module 2; multi-branch deep learning module 3; feature screening module 4; classifier 5; data preprocessing module 6; dynamic feature screening module 7; training module 8. Detailed implementation manners

[0023] This embodiment provides a 5hmC prediction method based on multi-branch deep learning and adaptive feature screening. As Figure 1 shown, the overall process of the prediction method is divided into the following main steps: data acquisition and cleaning, feature extraction and fusion, multi-branch deep learning modeling, feature fusion, feature screening and optimization, and final prediction, specifically as follows: S1. Data acquisition and cleaning First, four types of 5hmC-related sequencing data are obtained, including WGBS, TAPS, TAPSbeta, and CAPS. These data respectively detect the positions of 5hmC through different modification methods. Specifically: WGBS: Unmodified cytosine is converted to T, while 5mC and 5hmC remain as C.

[0024] TAPS: Unmodified cytosine remains as C, while 5mC and 5hmC are converted to T.

[0025] TAPSbeta: Only detect 5mC (converted to T).

[0026] CAPS: Only detect 5hmC (converted to T).

[0027] The fields included in the data file are information such as chromosome, start position, end position, modification level, unmodified reads, and modified reads. By converting these fields into numerical data and cleaning them, the integrity of the data is ensured, especially in handling missing values of modification levels and reads counts.

[0028] Furthermore, extract the CpG positions, modification levels, and related metrics (such as unmodified and modified reads) of 5hmC modification sites from the four sequencing files, and ensure that all data is formatted with complete fields and missing values filled. Next, filter out the 5hmC sites common to these four sequencing methods through an intersection operation to ensure data consistency.

[0029] Based on this, according to the distribution of modification levels, select negative and positive samples: Negative sample selection: Determine the modification inflection point through histogram analysis, and select sites with lower modification levels as negative samples. Specifically, we select samples with lower modification according to the rate of change of modification levels. At the same time, special attention is also paid to the sites that are "close to unmodified but have signals" to ensure the diversity of negative samples.

[0030] Positive sample selection: By selecting sites with higher modification levels, ensure that the selected positive samples have higher modification levels and can accurately reflect the actual distribution of 5hmC.

[0031] Since 5hmC sites are relatively scarce, the downsampling technique (SMOTE) is used to balance the sample ratio and solve the problem of data imbalance.

[0032] Finally, in step S1, extract the context sequence of each CpG site from the reference genome. Centered on each site, intercept a 41bp window (including 20bp prefix, the current site, and 20bp suffix). Filter the sequences containing "N", save the processed context sequences in FASTA format, and label the sample tags (positive samples as "pos" and negative samples as "neg") for subsequent feature extraction.

[0033] S2. Feature Extraction and Fusion As Figure 2As shown, in the feature extraction stage, multiple encoding methods are used to extract different types of features from the context sequence to ensure that the local patterns, global features, physicochemical properties, and statistical information of the DNA sequence can be comprehensively captured. The encoding methods used include: Sequence-based encoded features: such as One-Hot, Binary, Kmer, PSTNPsS, etc., which are designed to describe the local or global patterns of the base sequence or consider the regularity of the position distribution.

[0034] Physicochemical property features: such as PCP, SCPseKNC, etc., which mainly reflect the physicochemical properties of DNA such as melting temperature, polarizability, hydration, and energy stacking.

[0035] Statistical / correlation-based features: such as DAC, ANF, ASDC, etc., which are mainly used to characterize the correlation relationships and statistical distributions between dinucleotides or higher-order nucleotides.

[0036] In this embodiment, 12 different feature encoding methods are used. The following is a brief description of several of these feature encoding methods: Kmer frequency encoding: 5hmC modification is related to specific K-mer combinations, and specific short sequence fragment patterns may determine the occurrence probability of 5hmC modification. The frequency of nucleotide substrings with a fixed length of k (such as k = 3) is statistically calculated to provide statistical pattern information and generate local pattern features of the sequence. The calculation formula for the feature value is: Mismatch encoding: Calculate the occurrence of adjacent nucleic acids of length k with at most m mismatches (m < k). Based on the Kmer frequency encoding, a certain matching tolerance is introduced to capture sequence types with minor mutations.

[0037] L is the total length of the input sequence; m is the maximum number of allowed mismatches, satisfying m < k; represents the reference k-mer When appears, represents the number of times the k-mer with a Hamming distance of exactly j from it appears in the sequence; is the i-th type of k-mer; represents the Hamming distance between k-mer u and v, that is, the number of non-matching nucleotides at the corresponding positions of the two k-mers; is the indicator function, which takes 1 when the condition in the parentheses holds, otherwise takes 0.

[0038] PSTNPsS (Position-Specific Trinucleotide Propensity) Encoding: Analyze the trinucleotides and their position-specific propensities in the sequence. By calculating all possible trinucleotide combinations and using a feature extraction method with position information, capture the spatial distribution patterns of nucleotides and reveal possible functional or structural patterns in the sequence.

[0039] Where: and represent the frequencies of the trinucleotide at position j in the positive and negative samples, respectively.

[0040] SCPseKNC Encoding: Extract k-mers by sliding window on the sequence, combine them with physicochemical properties such as entropy values, and calculate the squared difference nucleotide composition, which can capture more subtle nucleotide association changes in the sequence and has a certain robustness to mutations or sequencing noise.

[0041] One-Hot Encoding: Represent each base (A, T, C, G) in the sequence as a one-hot vector to reflect the local structure of the base sequence and completely retain the original sequence information.

[0042] Binary Encoding: Represent each base as a binary vector, which is simple and efficient and can express the type distribution of the sequence in a low dimension.

[0043] DAC (Dinucleotide Auto-Covariance) Encoding: Evaluate the association between two dinucleotides in the sequence, considering their physicochemical properties.

[0044] is the physicochemical property value of the dinucleotide and is the average value of this property.

[0045] ENAC (Enhanced Nucleic Acid Composition): Statistically calculate the frequencies of each base in a given window in a sliding window manner to capture local base distribution information.

[0046] ANF (Accumulated Nucleotide Frequency) Encoding: Encode the nucleotide distribution of the entire RNA sequence. It provides a density value for each nucleotide according to the occurrence of each nucleotide at each position in the sequence.

[0047] where f(si) is the nucleotide frequency at the i-th position and l is the length of the sequence.

[0048] ASDC (Adaptive Skip Dipeptide Composition) encoding: Considering the relationship between adjacent and spaced bases in the sequence, it goes beyond simple dipeptide composition and models by considering the distance of nucleotides with the maximum skip in the sequence as the distance of nucleotides: represent nucleotides, and capture their occurrence frequencies in the sequence.

[0049] PCP (Physical-Chemical Property) encoding: Physical-chemical properties such as the melting temperature, polarizability, and hydration of DNA have important effects on 5hmc formation and can be used to globally analyze the influence of DNA structural characteristics on the modification pattern. A feature matrix is constructed based on the physical and chemical properties of dinucleotides (such as stacking energy, melting temperature) to quantify the physical and chemical properties of the sequence. The mean value of all dinucleotides in the sequence is calculated to obtain an overall feature representation, and autocovariance is used to calculate the characteristic changes between different dinucleotides.

[0050] ACGT2Vec encoding: Drawing on the ideas of natural language processing methods such as Word2Vec, DNA / RNA sequences are converted into vectors. The embedding technique is used to transform each k-mer in the sequence into a low-dimensional vector, thereby obtaining a higher-level feature representation. The embedding vectors are trained according to the context of bases in the DNA sequence, which can better represent the structural information in the DNA sequence.

[0051] The above 12 feature encoding methods can comprehensively characterize the local patterns, physical and chemical properties, statistical distributions, and other potential information of DNA sequences.

[0052] In the initial stage, pre-experiments are carried out on the above 12 encoding methods respectively to obtain a candidate set with better performance and feature dimensions. Subsequently, as Figure 3 shown, 6 most valuable feature types (Kmer, PSTNPsS, One-Hot, Binary, SCPseKNC, DAC) are roughly screened from the 12 encoding methods as input features for subsequent modeling to further reduce the high-dimensional redundancy problem.

[0053] S3. Multi-branch Deep Learning Modeling To improve the accuracy and robustness of the model, this scheme designs a multi-branch deep learning network, which processes different types of features using different network branches. Each feature is extracted through an independent network branch to capture the key pattern information of each feature. This embodiment mainly provides the following three networks for different features: CNN branch: It includes multiple convolutional kernels and pooling layers. The size of the convolutional kernel is 3×1 or 5×1, which is used to capture the local patterns of adjacent bases or short sequences.

[0054] LSTM Branch: Adopts a bidirectional structure with a hidden layer dimension of 64 - 128, used to process features with temporal dependencies.

[0055] DNN Branch: Includes fully connected layers with a hidden layer dimension of 128 - 256, used to process features without obvious temporal dependencies and capture the non - linear relationships between features.

[0056] The specific encoding of each type of feature uses different network branches as follows: SCPseKNC Encoding: SCPseKNC encoding describes the spatial distribution of nucleotides in a sequence. In this embodiment, the LSTM branch is used for processing, and the LSTM processes this temporal dependence to capture the long - range dependence relationships in the sequence.

[0057] Binary Encoding: Binary encoding represents each base in binary. In this embodiment, the LSTM branch is used for processing to capture the temporal dependence in this sequence feature.

[0058] PSTNPsS Encoding: PSTNPsS encoding reflects the combined characteristics of trinucleotides (short sequence fragments), and extracts features by analyzing the position - specificity of trinucleotides in a longer DNA sequence. In this embodiment, the CNN branch is used for processing to extract local features from short sequences.

[0059] One - Hot Encoding: One - Hot encoding represents the local structure of the base sequence through one - hot vectors. In this embodiment, the DNN branch is used to process it to capture the non - linear relationships of long sequences.

[0060] DAC Encoding: DAC encoding evaluates the correlation of dinucleotides. In this embodiment, the DNN is used to process this type of global features and extract their complex relationships.

[0061] Kmer Encoding: Kmer encoding generates encoded features by statistically counting the frequencies of nucleotide substrings of a fixed length k (k = 3). In this embodiment, the CNN is used for processing to extract local sequence - type information.

[0062] S4. Feature Fusion Since there may be information redundancy or complementary relationships among different types of features, simple feature concatenation may lead to too high feature dimensions, thus affecting the learning ability of the model. Therefore, this solution introduces a multi - layer self - attention mechanism in the feature fusion stage. By dynamically calculating the weights of features from different branches, the model can automatically select the most informative feature components.

[0063] Each branch in S3 will ultimately output a feature vector with a fixed dimension. In the feature fusion stage, a self-attention mechanism is adopted to further adjust the weights of the outputs of each branch to capture the complementary information between different features, ensuring that the model can better adapt to the special requirements of the 5hmC prediction task. The fusion process of the self-attention mechanism: Feature stacking: Concatenate the outputs of each branch (Concatenation, abbreviated as cat) into a tensor of shape (B, 6, 64), where B is the batch size, 6 represents 6 branches, and each branch outputs a 64-dimensional feature vector.

[0064] Multi-head self-attention calculation: Calculate the importance of different features through the multi-head attention mechanism and perform weight normalization: where Q, K, and V are the query, key, and value matrices respectively, and d k is the scaling factor.

[0065] Residual connection and layer normalization: After calculating the weighted features, introduce residual connection and layer normalization to keep the model stable during deep training.

[0066] Global pooling: Perform average pooling on the fused features to form a final feature vector of a fixed length.

[0067] The fused feature vector will be used as the input for SHAP calculation, LASSO screening, and final classification.

[0068] S5. Feature screening and optimization After feature fusion, this solution further introduces SHAP value calculation, multi-model cross-comparison, and multi-stage LASSO screening to ensure that the finally used features have the optimal contribution to 5hmC prediction.

[0069] Dynamic SHAP calculation: The SHAP value is an important method for evaluating feature importance. The core idea is to calculate the marginal contribution of each feature to the model prediction result. During the model training process, the SHAP value will be calculated dynamically to reflect the importance of features in real time. In this embodiment, through multiple rounds of calculations, monitor the change of the SHAP value with training, and use time decay weights to smooth the noise to ensure that the stable feature contributions are concerned throughout the training process. Its calculation formula is: where f(s) is the model prediction value using any feature set S, and N is the set of all features.

[0070] The SHAP calculation of this scheme adopts a dynamic update mechanism: after each epoch, the SHAP value is recalculated to ensure that the feature importance is continuously adjusted as the training progresses. The SHAP changes are smoothed using time decay weighting to avoid noise affecting feature selection.

[0071] Furthermore, since the calculation of SHAP value depends on the model structure, different models may give different evaluations of feature importance. Therefore, in order to improve the stability of feature screening, this scheme introduces multi-model cross comparison, namely: 1. Use XGBoost, LightGBM, and Random Forest to train the model and calculate the SHAP value respectively.

[0072] 2. Select the features that rank in the top 50% in multiple models and have high consistency in importance, and eliminate the features that fluctuate greatly between different models. In this embodiment, the standard deviation is used as the consistency coefficient.

[0073] 3. Through cross-comparison, ensure that the final selected features have a high contribution under different modeling methods, and improve the robustness of feature screening.

[0074] Only features that have stable contributions in multiple models will enter the LASSO screening stage.

[0075] Multi-stage LASSO screening: LASSO is a feature selection method based on L1 regularization, which can automatically remove features with small weights. This embodiment uses multi-stage LASSO regression for feature screening based on dynamic SHAP sorting. Through multiple rounds of LASSO regression, redundant features are further eliminated, and finally a set of streamlined and efficient feature subsets is formed. The optimization objectives of LASSO are as follows: Among them, λ is the regularization coefficient, w is the feature weight vector, y is the label, and X is the feature matrix. This scheme uses multi-stage LASSO screening: First-stage LASSO: LASSO screening is performed based on the initial feature set to remove the least important 20% features.

[0076] Second stage LASSO (based on SHAP ranking adjustment): further screening, only retaining features with high SHAP importance, and performing LASSO dimensionality reduction again.

[0077] Final feature set: Only the 50% fused feature vectors with the greatest impact on 5hmC prediction are retained to improve the generalization ability of the model and reduce the computational overhead caused by redundant features.

[0078] S4. Final prediction and training optimization After feature screening, the obtained final feature set is input into the fully connected layer FC to predict 5hmc sites. The fully connected layer FC is the classifier used in this model. During training, BCEWithLogitsLoss (binary cross-entropy loss) or FocalLoss (focal loss) is used to handle imbalanced data. The AdamW optimizer is adopted and combined with CosineAnnealingLR (cosine annealing learning rate adjustment) to dynamically adjust the learning rate. To avoid overfitting, this solution adds regularization means such as Dropout (dropout layer) and BatchNormalization (batch normalization). At the same time, an early stopping mechanism (patience) is used to monitor the validation set loss, and when the loss does not improve in multiple iterations, the training is stopped in advance. In addition, through the adaptive regularization and adaptive resampling strategy, the robustness of the model is further improved.

[0079] To verify the performance and robustness of this solution, in this embodiment, indicators such as ROC AUC (area under the ROC curve), Precision (precision), SN (sensitivity / true positive rate), SP (specificity), MCC (Matthews correlation coefficient), F1_score (F1 score), Recall (recall rate), and Accuracy (accuracy rate) are used to compare and evaluate the classification effects of the model implemented by this solution and traditional methods. Figure 4 It is a radar performance comparison chart of this solution model and the traditional model iRhm5CNN regarding indicators such as ROC AUC, Precision, MCC, F1_score, Recall, and Accuracy. It can be seen that the model of this solution is superior to the traditional model in any indicator.

[0080] In addition, this embodiment also compares the performance of different encoded feature data in the 5hmc recognition task, such as Figure 5 、 Figure 6 、 Figure 7 、 Figure 8 、 Figure 9 、 Figure 10 As shown, in the figure, our_features are the fusion features obtained in this application, and the rest are the corresponding features. It can be seen that the fusion features are basically superior to single features in each indicator. For recall, although the fusion features are slightly lower than kmer, combined with other aspects such as Precision, the comprehensive performance is obvious, which can verify the effectiveness of this solution in multi-source data integration.

[0081] As Figure 11 shown, this solution also proposes a 5hmC prediction system based on multi-branch deep learning and adaptive feature screening to implement the above steps to construct a prediction model for 5hmC modification site prediction, including: A data preprocessing module 6, which is used to clean the sequencing data, extract the context sequences, screen the common modification sites, and construct positive and negative sample sets; A multi-source feature encoding module 2, which is used to perform Kmer frequency encoding, PSTNPsS encoding, One-Hot encoding, Binary encoding, SCPseKNC encoding, and DAC encoding on the positive and negative sample sets to generate a multi-dimensional feature set; A multi-branch deep learning module 3, including a CNN branch, an LSTM branch, a DNN branch, and a self-attention fusion unit; the CNN branch processes the Kmer frequency encoding features and PSTNPsS encoding features, the LSTM branch processes the SCPseKNC encoding features and Binary encoding features, and the DNN branch processes the One-Hot encoding features and DAC encoding features respectively; the self-attention fusion unit dynamically weights and fuses the features output by each neural network branch; A dynamic feature screening module 7, including SHAP calculation and LASSO regression, which is used to combine dynamic SHAP value calculation and multi-stage LASSO regression to screen several feature subsets that contribute the most to the 5hmC prediction from the features output by the self-attention fusion unit; A training module 8, which uses an adaptive regularization and adaptive resampling strategy to train a prediction model including multiple neural network branches, a self-attention mechanism, and a classifier 5 through a class imbalance loss function for predicting 5hmC modification sites.

[0082] Furthermore, as Figure 12 shown, this solution also proposes a 5hmC prediction system in which a prediction model can be used for predicting 5hmC modification sites after training, including an interaction module 1, which is used to receive the DNA context sequence to be predicted and output the prediction result; A multi-source feature encoding module 2, which is used to perform Kmer frequency encoding, PSTNPsS encoding, One-Hot encoding, Binary encoding, SCPseKNC encoding, and DAC encoding on the DNA context sequence to be predicted to obtain a multi-dimensional feature set; A multi-branch deep learning module 3, including a CNN branch, an LSTM branch, a DNN branch, and a self-attention fusion unit; the CNN branch processes the Kmer frequency encoding features and PSTNPsS encoding features, the LSTM branch processes the SCPseKNC encoding features and Binary encoding features, and the DNN branch processes the One-Hot encoding features and DAC encoding features respectively; the self-attention fusion unit weights and fuses the features output by each neural network branch; A feature screening module 4, which pre-stores the feature types screened during the training process and is used to extract the pre-stored feature types from the features output by the neural network branches as a feature subset for output; A classifier 5, configured to output a prediction result based on the feature subset output by the feature screening module 4.

[0083] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the described specific embodiments or use similar means for substitution, without departing from the spirit of the present invention or exceeding the scope defined by the appended claims.

Claims

1. A 5hmC prediction method based on multi-branch deep learning and adaptive feature screening, characterized in that: The method includes: S1. Screen common 5hmC modification sites from various sequencing data and construct positive and negative sample sets; S2. Perform sequence feature encoding, physical and chemical property feature encoding and statistical feature encoding on positive and negative samples respectively to generate a multidimensional feature set; S3. Input the features of different encoding types into the corresponding neural network branches for processing; S4. Dynamically weighted fusion of features output by each neural network branch through self-attention mechanism, combined with dynamic SHAP value calculation and multi-stage LASSO regression, to screen several feature subsets that contribute most to 5hmC prediction; S5. Adaptive regularization and adaptive resampling strategies are used to train a prediction model including multiple neural network branches, self-attention mechanism and classifier through class imbalance loss function to predict 5hmC modification sites.

2. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 1, characterized in that: The sequence feature coding includes any one or more of One-Hot coding, Binary coding, Kmer coding, PSTNPsS coding, Mismatch coding, and ENAC coding; The physical and chemical property characteristic coding includes any one or more of PCP coding and SCPseKNC coding; The statistical feature coding includes any one or more of DAC coding, ANF coding, and ASDC coding; The neural network branches include CNN branches, LSTM branches and DNN branches.

3. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 2, characterized in that: In step S2, Kmer frequency encoding, PSTNPsS encoding, One-Hot encoding, Binary encoding, SCPseKNC encoding and DAC encoding are performed on the positive and negative samples to generate a multidimensional feature set; The neural network branches include two CNN branches that process Kmer frequency coding features and PSTNPsS coding features respectively, two LSTM branches that process SCPseKNC coding features and Binary coding features respectively, and two DNN branches that process One-Hot coding features and DAC coding features respectively.

4. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 2 or 3, characterized in that: The CNN branch includes a multi-layer convolution kernel for extracting Kmer frequency coding features and PSTNPsS coded local sequence features; The LSTM branch adopts a bidirectional structure to process the temporal characteristics of the physical and chemical properties encoded by SCPseKNC and the binary sequence dependencies encoded by Binary; The DNN branch includes a fully connected layer and a nonlinear activation function, which is used to extract the global nonlinear association of One-Hot coding features and DAC coding.

5. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 1, characterized in that: In step S3, each neural network branch outputs a fixed feature vector of the same dimension, and the self-attention mechanism dynamically weights and fuses the features output by each neural network branch to capture the complementary information between different features.

6. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 5, characterized in that: Step S4 specifically includes: Stack the outputs of each neural network branch into a tensor of a specific shape; The importance of different features is calculated through a multi-head attention mechanism, and the weights are normalized; After calculating the weighted features, residual connections and layer normalization are introduced; Perform average pooling on the fused features to form a candidate feature vector of fixed length; The final feature vector is selected from the candidate feature vectors through SHAP value calculation and multi-stage LASSO regression as the input of the classifier.

7. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 6, characterized in that: In step S4, the SHAP value calculation adopts a dynamic update mechanism. After each training round, SHAP is recalculated. Through multiple rounds of calculation, the change of SHAP value with training is monitored, and the time decay weight is used to smooth the SHAP change, and finally several features with the highest SHAP value are selected; After several features were screened out through SHAP value calculation, the multi-stage LASSO regression screening stage was entered, and finally some features with the greatest impact on 5hmC prediction were retained from the screened features.

8. The 5hmC prediction method based on multi-branch deep learning and adaptive feature screening according to claim 7, characterized in that: In step S4, the SHAP value screening feature process also includes multi-model cross-comparison: Use XGBoost, LightGBM, and Random Forest to train the model and calculate the SHAP value respectively; Select some features that rank in the top set percentage in multiple models and whose importance consistency coefficient is higher than the set value.

9. A 5hmC prediction system based on multi-branch deep learning and adaptive feature screening, characterized in that: include: Data preprocessing module (6), used to screen common 5hmC modification sites from various sequencing data to construct positive and negative sample sets; A multi-source feature encoding module (2) is used to perform sequence feature encoding, physical and chemical property feature encoding and statistical feature encoding on the positive and negative sample sets to generate a multi-dimensional feature set; A multi-branch deep learning module (3), comprising a plurality of neural network branches and a self-attention fusion unit; the plurality of neural network branches are used to process features of corresponding encoding types respectively, and the self-attention fusion unit is used to dynamically weight and fuse features output by each neural network branch; Dynamic feature screening module (7), used to combine dynamic SHAP value calculation with multi-stage LASSO regression to screen several feature subsets that contribute most to 5hmC prediction from the features output by the self-attention fusion unit; The training module (8) uses adaptive regularization and adaptive resampling strategies to train a prediction model including multiple neural network branches, self-attention mechanism and classifier (5) through a class imbalance loss function to predict 5hmC modification sites.

10. A 5hmC prediction system based on multi-branch deep learning and adaptive feature screening, characterized in that: include: An interactive module (1) is used to receive a DNA context sequence to be predicted and output a prediction result; A multi-source feature encoding module (2) is used to perform sequence feature encoding, physical and chemical property feature encoding and statistical feature encoding on the DNA context sequence to be predicted to generate a multi-dimensional feature set; A multi-branch deep learning module (3) includes a plurality of neural network branches and a self-attention fusion unit; the plurality of neural network branches are used to process features of corresponding encoding types respectively, and the self-attention fusion unit dynamically weights and fuses features output by each neural network branch; A feature screening module (4) pre-stores feature types obtained by dynamic SHAP value calculation and multi-stage LASSO regression screening during training, and is used to extract the pre-stored feature types from the features output by the neural network branch and output them as feature subsets; The classifier (5) is used to output a prediction result based on the feature subset output by the feature screening module (4).

Images