Efficient siRNA effectiveness prediction method and system based on XGBOOST algorithm

Abstract

The invention relates to the field of bioinformatics and drug development, and discloses an efficient siRNA effectiveness prediction method and system based on an XGBOOST algorithm. The method comprises the following steps: acquiring a training data set containing siRNA sequence information, chemical modification information and corresponding validity data; carrying out cleaning, feature extraction and data balance processing on the data; training the prediction model by using an XGBOOST algorithm, and optimizing hyper-parameters; performing model evaluation and optimization through cross validation; a new siRNA sequence is input for effectiveness prediction; explaining a prediction result by using an SHAP value analysis technology; the actual interference efficiency of the efficient siRNA sequence is verified through an in-vitro cell experiment, and the model is further optimized based on an experiment result. According to the method, the accuracy and efficiency of siRNA effectiveness prediction are improved, the influence of chemical modification is comprehensively considered, the problem of data imbalance is solved, the model interpretation is enhanced, and the calculation efficiency is remarkably improved.

Description

Technical Field

[0001] The present application relates to the fields of bioinformatics and drug development, and in particular to a small interfering RNA (siRNA) effectiveness prediction technology based on machine learning. Background Art

[0002] Small nucleic acid drugs, especially small interfering RNA (siRNA) drugs, have attracted much attention in the medical field in recent years. siRNA can specifically silence target genes through the RNA interference (RNAi) mechanism, thereby achieving disease treatment. Since the first siRNA drug was approved for marketing in 2018, as of 2023, the US FDA has approved 6 small nucleic acid drugs, showing the huge potential and rapid development trend in this field.

[0003] However, there are still many challenges in the development of siRNA drugs, one of the most critical bottlenecks is how to quickly and accurately identify and screen efficient siRNA sequences. Traditional experimental methods are time-consuming and costly, and it is difficult to meet the needs of rapid drug development. Therefore, researchers began to turn to computer-assisted prediction methods.

[0004] At present, siRNA effectiveness prediction methods are mainly divided into three categories: rule-based empirical methods, traditional machine learning algorithms, and deep learning methods. These methods have exposed a series of problems in practice:

[0005] Insufficient prediction accuracy: Existing machine learning prediction methods, such as random forest (RF) and support vector machine (SVM), have limited prediction accuracy when dealing with complex siRNA biological data. They often fail to fully capture the complex relationship between siRNA sequences and their effectiveness, resulting in inaccurate prediction results.

[0006] The influence of chemical modification is not fully considered: Chemical modification of siRNA has a significant impact on its effectiveness, but traditional prediction methods often ignore this factor. With the continuous emergence of new chemical modifications, how to integrate this information into the prediction model has become an urgent problem to be solved.

[0007] Data imbalance problem: In siRNA research, effective sequences are often in the minority, resulting in a serious imbalance in the training data set. Existing methods perform poorly in dealing with such imbalanced data, especially in predicting the minority class (i.e., highly effective siRNA).

[0008] Models lack interpretability: Although some machine learning models perform well in predictive performance, their complexity often leads to a "black box" effect, making it difficult to explain the biological significance behind the predicted results. This limits researchers' in-depth understanding of the mechanism of action of siRNA.

[0009] Low computational efficiency: As the scale of siRNA data continues to expand, the computational efficiency of existing prediction methods is difficult to meet the needs of high-throughput screening, which leads to a longer new drug development cycle and increased R&D costs.

[0010] Difficult to adapt to new chemical modifications: Pharmaceutical companies are constantly developing new siRNA chemical modification patents, but existing prediction models are difficult to quickly adapt to these new modifications, affecting the conversion efficiency from patents to practical applications.

[0011] Personalized treatment needs are difficult to meet: With the development of precision medicine, how to design siRNA drugs for specific patients or subgroups has become a new challenge. Existing methods are difficult to provide sufficiently accurate guidance for personalized siRNA design.

[0012] These technical problems seriously restrict the development efficiency and success rate of siRNA drugs. Therefore, there is an urgent need for a new siRNA effectiveness prediction method that can comprehensively solve the above problems, improve prediction accuracy, consider the effects of chemical modifications, handle data imbalance, provide interpretable results, improve computational efficiency, and quickly adapt to new chemical modifications and personalized treatment needs. This will not only greatly accelerate the development of siRNA drugs, but also provide strong technical support for the development of a new generation of RNA interference therapies. Summary of the invention

[0013] The purpose of this application is to provide an efficient siRNA effectiveness prediction method and system based on the XGBOOST algorithm to solve the problems raised in the above background technology.

[0014] The present application discloses a highly efficient siRNA effectiveness prediction method based on the XGBOOST algorithm, comprising the following steps:

[0015] Step A: obtaining a training data set including siRNA sequence information, chemical modification information and corresponding validity data; preprocessing the training data set, including data cleaning, feature extraction and data balancing, to obtain balanced training data;

[0016] Step B: Based on the balanced training data, the XGBOOST algorithm is used to train the siRNA effectiveness prediction model and optimize the algorithm hyperparameters; multi-index performance evaluation and optimization are performed on the trained prediction model to obtain the optimized prediction model;

[0017] Step C: Input the new siRNA sequence and its chemical modification information into the optimized prediction model to perform effectiveness prediction and obtain prediction results; perform interpretability analysis on the prediction results and generate a biologically meaningful interpretation report;

[0018] Step D: verifying the actual interference efficiency of the predicted high-efficiency siRNA sequence through in vitro cell experiments; based on the interpretation results and experimental verification results, further optimizing and iterating the prediction model, including updating the feature extraction method and training data set;

[0019] Step E: Repeat steps B to D until the model performance reaches a predetermined standard, thereby obtaining a final siRNA effectiveness prediction model.

[0020] In a preferred embodiment, the preprocessing in step A specifically includes: cleaning and standardizing the data; extracting siRNA sequence features and chemical modification features to generate a feature matrix; and using data balancing technology to handle data imbalance in the feature matrix.

[0021] In a preferred example, the data balancing technology includes one or more of oversampling, undersampling or generative adversarial networks (GANs).

[0022] In a preferred example, the XGBOOST algorithm used in step B adopts parallel computing technology to improve computing efficiency.

[0023] In a preferred example, the method for optimizing the hyperparameters of the XGBOOST algorithm in step B includes one or more of grid search, random search or Bayesian optimization, and the hyperparameters include learning rate, maximum tree depth and regularization parameter.

[0024] In a preferred example, the multi-index performance evaluation in step B adopts a cross-validation method, and the evaluation indicators include two or more of accuracy, precision, recall, F1 score and ROC-AUC.

[0025] In a preferred embodiment, the interpretability analysis in step C adopts SHAP value analysis technology.

[0026] In a preferred example, the optimization and iteration in step D include adjusting the feature extraction method, updating the training data set and re-optimizing hyperparameters.

[0027] In a preferred embodiment, the in vitro cell experiment verification in step D includes chemical synthesis of siRNA, optimization of transfection conditions and evaluation of RNA interference efficiency.

[0028] In a preferred embodiment, the feature extraction step includes extracting information such as the nucleotide composition, thermodynamic properties, and type, position, and quantity of chemical modifications of the siRNA sequence.

[0029] In a preferred embodiment, the SHAP value analysis technique is used to identify the most influential features for predicting the effectiveness of siRNA, and compare these features with known siRNA design rules, thereby providing a biologically explainable prediction basis.

[0030] The present application also discloses an efficient siRNA effectiveness prediction device based on the XGBOOST algorithm, comprising:

[0031] Data acquisition and preprocessing module: used to acquire a training data set containing siRNA sequence information, chemical modification information and corresponding validity data, and preprocess the training data set, including data cleaning, feature extraction and data balancing, to obtain balanced training data;

[0032] Model training and optimization module: used to train the siRNA effectiveness prediction model based on the balanced training data using the XGBOOST algorithm and optimize the algorithm hyperparameters; perform multi-index performance evaluation and optimization on the trained prediction model to obtain the optimized prediction model;

[0033] Prediction and interpretation module: used to input new siRNA sequences and their chemical modification information into the optimized prediction model, perform validity prediction, obtain prediction results, and perform interpretability analysis on the prediction results to generate a biologically meaningful interpretation report;

[0034] Experimental verification and model optimization module: used to verify the actual interference efficiency of the predicted high-efficiency siRNA sequence through in vitro cell experiments, and further optimize and iterate the prediction model based on the interpretation results and experimental verification results, including updating the feature extraction method and training data set;

[0035] Loop execution module: used to control the repeated execution of the model training and optimization module, the prediction and interpretation module, the experimental verification and model optimization module until the model performance reaches the predetermined standard and the final siRNA effectiveness prediction model is obtained.

[0036] Compared with the prior art, the embodiments of the present application have the following technical effects:

[0037] First, by introducing the XGBOOST algorithm, the gradient boosting technology is used to more accurately capture the complex relationship between siRNA sequences and their effectiveness. Compared with traditional random forest (RF) and support vector machine (SVM) algorithms, the prediction model of this application significantly improves the accuracy and robustness of prediction results when processing complex biological data.

[0038] Furthermore, the influence of chemical modification is fully considered. Traditional siRNA prediction methods often ignore the influence of chemical modification on its effectiveness. This application integrates data of siRNAs with various chemical modifications, fully considers the influence of chemical modification on the effectiveness of siRNA, and improves the applicability and reliability of the prediction model for siRNAs with different modification types.

[0039] Furthermore, the problem of data imbalance is solved. Existing methods have poor prediction results when processing unbalanced data, especially the prediction accuracy on minority class data is low. This application adopts advanced data balancing technologies such as random oversampling, undersampling and generative adversarial networks (GANs), which significantly improves the generalization ability of the model on unbalanced data sets.

[0040] Furthermore, the model interpretability is enhanced. Although some machine learning models have advantages in predictive performance, their complexity may lead to a lack of interpretability of the model. This application introduces explanatory analysis techniques such as SHAP values ​​(Shapley Additive Explanations) to provide biological interpretation of the results of the siRNA prediction model, helping researchers understand the mechanism behind the prediction results and enhancing the interpretability and application value of the model.

[0041] Furthermore, the computational efficiency is improved. The computational efficiency of existing prediction methods is low, and it is difficult to quickly process large-scale siRNA data. This application significantly improves the computational speed and efficiency by adopting the XGBOOST algorithm and parallel computing technology, which can meet the needs of high-throughput siRNA screening, greatly shorten the R&D cycle, and reduce R&D costs.

[0042] In summary, this application solves the problems of accuracy, reliability, data imbalance, interpretability and computational efficiency in the existing technology for siRNA effectiveness prediction by introducing the advanced XGBOOST algorithm and comprehensively processing a variety of chemical modification data, and provides a new platform for efficient, accurate and interpretable siRNA design and prediction, which has significant technical advantages and application prospects.

[0043] A large number of technical features are recorded in the specification of this application, which are distributed in various technical solutions. If all possible combinations of technical features of this application (i.e., technical solutions) are to be listed, the specification will be too long. In order to avoid this problem, the various technical features disclosed in the above-mentioned invention content of this application, the various technical features disclosed in the various embodiments and examples below, and the various technical features disclosed in the accompanying drawings can be freely combined with each other to form various new technical solutions (these technical solutions are all deemed to have been recorded in this specification), unless the combination of such technical features is technically infeasible. For example, in one example, feature A+B+C is disclosed, and in another example, feature A+B+D+E is disclosed, and features C and D are equivalent technical means that play the same role. Technically, only one can be used, and it is impossible to use them at the same time. Feature E can be combined with feature C technically. Then, the solution of A+B+C+D should not be deemed to have been recorded because it is technically infeasible, and the solution of A+B+C+E should be deemed to have been recorded. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a schematic diagram of an overview of the process of a highly efficient siRNA effectiveness prediction method based on the XGBOOST algorithm according to the first embodiment of the present application.

[0045] Figure 2 It is the overall process and main steps of the machine learning or data analysis project in the efficient siRNA effectiveness prediction method based on the XGBOOST algorithm according to the first embodiment of the present application.

[0046] Figure 3 It is a schematic diagram of the process of the highly efficient siRNA effectiveness prediction method based on the XGBOOST algorithm according to the first embodiment of the present application.

[0047] Figure 4 It is a schematic diagram of the structure of a highly efficient siRNA effectiveness prediction system based on the XGBOOST algorithm according to the second embodiment of the present application. DETAILED DESCRIPTION

[0048] In the following description, many technical details are provided to help readers better understand the present application. However, those skilled in the art can understand that the technical solution claimed in the present application can be implemented even without these technical details and various changes and modifications based on the following embodiments.

[0049] Description of some concepts:

[0050] siRNA (small interfering RNA): Small interfering RNA is a double-stranded RNA molecule, usually 20-25 nucleotides in length, which can specifically inhibit gene expression through the RNA interference mechanism.

[0051] RNA interference (RNAi): A double-stranded RNA-induced, sequence-specific gene silencing mechanism that can inhibit the expression of target genes.

[0052] XGBOOST (eXtreme Gradient Boosting): A decision tree-based ensemble learning algorithm that improves prediction accuracy and computational efficiency through gradient boosting technology.

[0053] Chemical modification: chemical structural changes to RNA molecules, such as 2'-O-methylation, phosphorothioation, etc., are used to improve the stability, specificity and cellular uptake rate of siRNA.

[0054] Data imbalance: In machine learning, it refers to the situation where there is a significant difference in the number of samples of different categories in the training data set. In siRNA research, it is usually manifested as the number of efficient siRNA sequences is far less than that of inefficient sequences.

[0055] Oversampling: A technique for dealing with data imbalance by duplicating or generating minority class samples to increase their proportion in the dataset.

[0056] Undersampling: Another technique to deal with data imbalance is to balance the dataset by reducing the number of majority class samples.

[0057] Generative Adversarial Networks (GANs): A deep learning model consisting of two networks, a generator and a discriminator, that can be used to generate new and balanced data samples.

[0058] SHAP value (SHapley Additive exPlanations): A method for explaining the prediction results of machine learning models, based on the concept of Shapley value in game theory, which is used to evaluate the contribution of each feature to the prediction results.

[0059] ROC-AUC (Receiver Operating Characteristic-Area Under the Curve): The area under the receiver operating characteristic curve is used to evaluate the performance of the binary classification model. The larger the value, the better the model performance.

[0060] Cross-validation: A model evaluation method that divides the dataset into multiple subsets and repeatedly trains and validates the model to evaluate the generalization ability of the model.

[0061] Hyperparameters: In machine learning, hyperparameters are parameters that are set before the learning process begins, such as learning rate, tree depth, etc., which need to be optimized to determine the optimal value.

[0062] Feature extraction: Extracting meaningful information for model training from raw data, which may include sequence features, thermodynamic features, etc. in siRNA prediction.

[0063] The specific application fields and application scenarios of this application are introduced below.

[0064] The siRNA effectiveness prediction method of the present application is particularly suitable for the field of precision medicine, including the field of personalized tumor treatment and the field of complex diseases.

[0065] 1. Personalized cancer treatment

[0066] In precision medicine for tumors, designing targeted siRNA drugs for specific gene mutations of patients is an important trend. However, since the gene mutations of each patient may be different, traditional siRNA design methods are difficult to quickly respond to personalized needs. The prediction method of this application can quickly generate targeted siRNA sequence suggestions based on the patient's genetic test results.

[0067] For example, for patients with non-small cell lung cancer carrying KRAS G12C mutation, this system can generate dozens of highly effective siRNA candidate sequences targeting the mutation within a few minutes, and provide a predicted effectiveness score and possible mechanism of action explanation for each sequence. This information can directly guide the laboratory to conduct rapid verification, greatly shortening the time from gene detection to personalized siRNA drug design.

[0068] 2. Complex disease areas

[0069] With the continuous improvement and growth of genomic technology and data, new targets for complex diseases such as cardiovascular disease and metabolic disease are constantly being discovered, which provides a wealth of candidate targets for the development of small nucleic acid drugs and greatly promotes the scope of indications of small nucleic acid drugs.

[0070] For example, for specific mutations or abnormal expression of cardiovascular disease-related genes (such as the PCSK9 gene), this system can design efficient siRNA sequences to reduce blood lipid levels and reduce the risk of cardiovascular events. Specifically, for patients with dyslipidemia, the overexpression of the PCSK9 gene is determined through genetic testing, and this platform can quickly generate efficient siRNA sequences targeting the PCSK9 gene. After the laboratory verifies these sequences, the efficient siRNA is applied to clinical trials to evaluate its effectiveness in lowering blood lipid levels and preventing cardiovascular disease.

[0071] In addition, the system can also recommend the most suitable siRNA chemical modification scheme based on the patient's metabolic characteristics to improve the in vivo stability and targeting of the drug. For example, according to the physiological conditions of different patients (such as liver function and kidney function), the platform can recommend different siRNA modification types (such as 2'-O-methyl and 2'-fluoro modifications) to ensure that siRNA has the best stability and activity in vivo. This highly personalized siRNA design method not only improves the accuracy of treatment, but also significantly reduces the risk of side effects.

[0072] Through its application in relevant treatments, this application demonstrates its unique advantages and innovative value in complex biomedical scenarios, and provides strong technical support for the development of precision medicine.

[0073] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0074] The first embodiment of the present application relates to an efficient siRNA effectiveness prediction method based on the XGBOOST algorithm, the process of which is as follows: Figure 1 , Figure 2 and Figure 3 As shown, the method comprises the following steps:

[0075] Step 100: Obtain a training data set including siRNA sequence information, chemical modification information and corresponding validity data.

[0076] Specifically, in this embodiment, a comprehensive and reliable training data set needs to be established first. This data set is the cornerstone of the entire prediction method, and its quality directly affects the performance of subsequent models. Optionally, siRNA-related data can be collected in a multi-channel manner. For example:

[0077] First, a large amount of verified siRNA data is extracted from public databases such as siRNAdb and siRecords. These databases bring together experimental results from research institutions around the world and are widely representative. At the same time, we deeply explore published research papers and extract valuable siRNA experimental data from them. These literature data often contain detailed experimental conditions and results, which are of great significance for understanding the mechanism of action of siRNA.

[0078] In addition to public data, original experimental data can also be obtained through independent experiments or cooperation with other research institutions. These data have high controllability and consistency and can serve as an important supplement to model validation. In addition, you can pay attention to published patent literature and collect relevant siRNA design and effect data from them, which often represent the latest technological advances.

[0079] For each siRNA data, its nucleotide sequence information is recorded in detail, including the double-stranded RNA sequence of 19-25 nucleotides in length, the start and end positions on the target mRNA, and the distinction between the guide strand and the passenger strand. This basic sequence information is the basis of siRNA function and is crucial to predicting its effect.

[0080] It is particularly noteworthy that chemical modification information of siRNA can also be collected. This includes the type of modification (such as 2'-O-methylation, 2'-F modification, phosphorothioation, etc.), the specific position of the modification in the siRNA sequence, and the number or ratio of modifications on a single siRNA molecule. This chemical modification information has an important impact on the stability, specificity and efficacy of siRNA, and is a key innovation of this prediction method.

[0081] For each siRNA, its effectiveness data can also be collected. This usually includes the percentage of reduction in the mRNA level or protein level of the target gene, the siRNA concentration required to achieve a 50% inhibitory effect (IC50 value), and the effect data at different time points and in different cell lines. These multi-dimensional effectiveness data enable us to comprehensively evaluate the performance of siRNA.

[0082] After data collection is completed, strict data processing and quality control are carried out. First, data from different sources are converted into a consistent format, and validity data under different experimental conditions are converted into comparable standard scores. Then, duplicate entries in the data set are removed, missing information is reasonably filled or eliminated, and possible abnormal data points are identified and processed. These steps ensure the quality and reliability of the data set.

[0083] Finally, the acquired data set is initially divided into training set, validation set and test set to prepare for subsequent model training and evaluation. At the same time, the source of the data, acquisition time, processing method and other meta-information are recorded in detail to ensure the traceability and repeatability of the data.

[0084] Through this comprehensive and rigorous data acquisition and processing process, a high-quality siRNA training data set was established. This data set not only contains the basic sequence information of siRNA, but also covers detailed information on various chemical modifications, as well as effectiveness data under different conditions. This provides rich and reliable training materials for the prediction model, enabling the model to more comprehensively consider various factors that affect the effect of siRNA, thus laying a solid foundation for subsequent prediction of efficient siRNA.

[0085] Step 200: preprocessing the training data set, including data cleaning, feature extraction and data balancing, to obtain balanced training data.

[0086] Optionally, the preprocessing in step 200 specifically includes:

[0087] a) Clean and standardize the data;

[0088] b) extracting siRNA sequence features and chemical modification features to generate a feature matrix;

[0089] c) Use data balancing techniques to deal with the data imbalance problem in the feature matrix.

[0090] Optionally, the data balancing technique includes one or more of oversampling, undersampling or generative adversarial networks (GANs).

[0091] Optionally, 200, the feature extraction step includes extracting information such as the nucleotide composition, thermodynamic properties, and type, position, and quantity of chemical modifications of the siRNA sequence.

[0092] Specifically, in this step, after obtaining the original training data set, it needs to be fully preprocessed to ensure data quality and extract meaningful features. This preprocessing step is crucial to improving the performance of subsequent machine learning models. This step mainly includes three key links: data cleaning, feature extraction, and data balancing. For example:

[0093] First, the data was cleaned and standardized. During this process, each siRNA record was carefully checked and obviously erroneous or incomplete data entries were deleted. For numerical data, such as the inhibition efficiency of siRNA, the Z-score standardization method was used to convert it to a uniform scale to eliminate data differences under different experimental conditions. For categorical data, such as chemical modification type, the one-hot encoding method was used for standardization. This step ensures the consistency and comparability of the data, laying the foundation for subsequent feature extraction and model training.

[0094] Next, feature extraction is performed, which is one of the core innovations of this method. This step not only extracts traditional siRNA sequence features, but also pays special attention to chemical modification features to comprehensively capture factors that affect siRNA effectiveness. For sequence features, the nucleotide composition ratio is calculated, including GC content, the frequency of occurrence of various dinucleotides and trinucleotides, etc. At the same time, the thermodynamic properties of the sequence are analyzed, such as melting energy (ΔG), melting point (Tm), etc. These parameters have an important influence on the stability and activity of siRNA.

[0095] In terms of chemical modification features, information such as the type, location, and quantity of modifications is extracted. For example, the specific locations of modifications such as 2'-O-methylation, 2'-F modification, and phosphorothioation in the siRNA sequence are recorded, and the total number of each modification and its distribution pattern in the sequence are calculated. These chemical modification features have a significant impact on the stability, targeting, and off-target effects of siRNA, and are important factors often overlooked by traditional prediction methods.

[0096] Through these feature extraction steps, a high-dimensional feature vector is generated for each siRNA record, and these vectors together constitute a feature matrix. This feature matrix comprehensively reflects the sequence and chemical modification characteristics of siRNA, providing rich input information for subsequent machine learning models.

[0097] Finally, the feature matrix is ​​subjected to data balancing. In siRNA research, highly efficient siRNAs are usually in the minority, which leads to serious imbalance in the data set. In order to solve this problem, a variety of data balancing techniques are used.

[0098] Oversampling techniques, such as SMOTE (Synthetic Minority Over-sampling Technique), increase the number of highly effective siRNAs by generating new minority class samples in the feature space. This method can increase the model's learning of highly effective siRNAs, but care should be taken to avoid overfitting.

[0099] Undersampling techniques, such as random undersampling or cluster undersampling, balance the dataset by reducing the number of majority class samples. This approach can reduce computational complexity but may lose some useful information.

[0100] In addition, the latest generative adversarial network (GANs) technology was also tried. GANs can generate high-quality synthetic minority class samples, which can not only balance the dataset but also increase the diversity of the data and improve the generalization ability of the model.

[0101] In practical applications, these techniques are usually used in combination, and the best balance strategy is selected through cross-validation. For example, SMOTE may be used to moderately oversample the minority class, and then cluster undersampling methods may be used to reduce the majority class samples, and finally GANs may be used to generate some high-quality synthetic samples to further enhance the representativeness of the dataset.

[0102] Through this series of preprocessing steps, we finally obtained a cleaned, feature-rich and balanced training dataset. This dataset not only retains the key information of the original data, but also enhances its representativeness and learning value through feature extraction and data balancing. This lays a solid foundation for the subsequent XGBOOST model training and helps to improve the model's prediction accuracy for highly effective siRNA, especially in terms of predictive ability considering the effects of chemical modifications.

[0103] Step 300: Based on the balanced training data obtained in step 200, the XGBOOST algorithm is used to train the siRNA effectiveness prediction model and optimize the algorithm hyperparameters.

[0104] Optionally, the XGBOOST algorithm used in step 300 adopts parallel computing technology to improve computing efficiency.

[0105] Optionally, the method for optimizing the hyperparameters of the XGBOOST algorithm in step 300 includes one or more of grid search, random search or Bayesian optimization, and the hyperparameters include learning rate, maximum tree depth and regularization parameter.

[0106] Specifically, this step is the core of the entire prediction method, mainly involving model training and optimization. XGBOOST (eXtreme Gradient Boosting) algorithm, as an efficient ensemble learning method, was selected to build a siRNA effectiveness prediction model. This algorithm has powerful feature processing capabilities and excellent prediction performance, and is particularly suitable for processing complex siRNA feature data in this application. For example:

[0107] First, the balanced training data obtained in step 200 is input into the XGBOOST algorithm. These data contain rich siRNA sequence features and chemical modification features, providing comprehensive learning materials for the model. The XGBOOST algorithm constructs multiple decision trees and weights the prediction results of these trees to form a powerful prediction model.

[0108] During the training process, the XGBOOST algorithm uses parallel computing technology to improve computational efficiency. Specifically, when constructing each decision tree, the algorithm can search for split points for different features in parallel. This parallelization strategy significantly reduces the time required for model training, especially when processing large-scale siRNA datasets.

[0109] In order to further improve the performance of the model, this step also includes the optimization of the hyperparameters of the XGBOOST algorithm. Hyperparameters refer to the parameters that need to be set before model training, which directly affect the learning process and final performance of the model. In this method, the key hyperparameters optimized include learning rate, maximum tree depth and regularization parameter.

[0110] The learning rate controls the contribution of each tree to the final prediction result. A smaller learning rate usually improves the generalization ability of the model, but may require more iterations. The maximum tree depth limits the complexity of each decision tree. The appropriate tree depth can strike a balance between the model's expressiveness and the risk of overfitting. The regularization parameter is used to control the complexity of the model and prevent overfitting.

[0111] In order to find the optimal hyperparameter combination, this method uses a variety of optimization strategies, for example:

[0112] Grid search: This method systematically tries all possible combinations in a predefined parameter space. Although computationally intensive, it can fully explore the parameter space and is suitable for situations where the parameter range is relatively small.

[0113] Random search: Randomly sampling from a predefined parameter distribution, this method can explore a larger parameter space than grid search and is more computationally efficient.

[0114] Bayesian optimization: This is a more advanced method that intelligently selects the next set of hyperparameters to be evaluated by building a probabilistic model between hyperparameters and model performance. This method can converge to the optimal parameter combination faster and is particularly suitable for situations where computing resources are limited.

[0115] In practical applications, these optimization methods can be selected or combined according to specific circumstances. For example, random search can be used to quickly locate potential high-quality parameter areas in a large range, and then a more detailed grid search or Bayesian optimization can be performed in this area.

[0116] The hyperparameter optimization process is usually carried out in a cross-validation manner to ensure that the found parameter combination has good generalization ability. Specifically, k-fold cross-validation can be used to divide the training data into k parts, use k-1 parts for training each time, and the remaining 1 part for validation. This process is repeated k times, and finally the parameter combination with the best average performance is selected.

[0117] Through this rigorous training and optimization process, a high-performance XGBOOST model was finally obtained. This model can fully utilize the sequence characteristics and chemical modification information of siRNA to accurately predict the effectiveness of siRNA. Because the influence of chemical modification is taken into account, this model has stronger prediction ability than traditional methods, especially for chemically modified siRNA molecules.

[0118] This optimized XGBOOST model provides a powerful tool for subsequent siRNA design and screening, and is expected to significantly improve the efficiency and success rate of siRNA drug development.

[0119] In order to better understand the innovation and technical advantages of this method, the specific principles and methods of the XGBOOST algorithm in siRNA effectiveness prediction are discussed below. Unlike traditional machine learning methods, this method cleverly incorporates the complex characteristics and chemical modification information of siRNA sequences into the objective function and tree structure optimization of XGBOOST, achieving accurate modeling of the siRNA mechanism of action. By analyzing the core components of the algorithm, including objective function design, loss function selection, regularization strategy, and model optimization process, we can more clearly see how this method overcomes many challenges in the field of siRNA prediction.

[0120] Derivation of XGBoost model related formulas

[0121] In this example, we use the XGBoost algorithm to predict the effectiveness of siRNA. Specifically, we divide the siRNA sequences into effective and ineffective categories by setting a threshold (cut-off), thereby converting the problem into a binary classification task. As a powerful ensemble learning method, the XGBoost algorithm is particularly suitable for handling such complex classification problems.

[0122] When dealing with classification problems, XGBoost usually uses binary logistic regression as its base learner. The core of the algorithm lies in the design of its objective function, which mainly includes two key components: log loss function and regularization term. The log loss function is used to measure the difference between the model prediction and the actual label, while the regularization term is used to control the model complexity and prevent overfitting.

[0123] Below, we will introduce in detail the application of the XGBoost algorithm in siRNA effectiveness prediction, including the construction of the objective function, the selection of the loss function, the regularization method, and the specific steps of model optimization. These mathematical formulas and derivation processes will help to deeply understand the working principle of the XGBoost algorithm in this embodiment.

[0124] The core of the XGBoost algorithm lies in the design of its objective function. In the siRNA effectiveness prediction task, the objective function consists of two parts: the loss function and the regularization term. We will introduce these components one by one and their role in the model.

[0125] 1. Objective Function

[0126] The objective function of XGBoost can be expressed as:

[0127]

[0128] in:

[0129] n is the number of training samples;

[0130] is the loss function of the i-th sample;

[0131] K is the number of trees;

[0132] Ω(f k ) is the regularization term of the K-th tree.

[0133] 2. Loss Function

[0134] In the binary classification problem of siRNA effectiveness, we use the logarithmic loss function (Log Loss):

[0135]

[0136] in:

[0137] y i is the true label of the i-th siRNA sample (0 means invalid, 1 means valid);

[0138] is the predicted probability of the model for the i-th sample;

[0139] This loss function can effectively measure the difference between the model prediction and the true label, which helps the model learn the effectiveness characteristics of the siRNA sequence.

[0140] 3. Regularization term

[0141] In order to control model complexity and prevent overfitting, we introduced a regularization term:

[0142]

[0143] in:

[0144] T is the number of leaf nodes of the decision tree;

[0145] γ is the penalty coefficient that controls the number of leaf nodes;

[0146] λ is the penalty coefficient that controls the weight of the leaf node;

[0147] w j is the weight of the jth leaf node.

[0148] The introduction of regularization terms helps to balance model complexity and prediction accuracy in the siRNA effectiveness prediction model. By adjusting γ and λ, the depth and weight distribution of the model can be controlled, thereby obtaining a more robust and generalizable prediction model.

[0149] This regularization method is particularly suitable for the task of predicting siRNA effectiveness because it helps the model capture key features in siRNA sequences while avoiding overfitting to the noise in the training data.

[0150] In the process of optimizing the siRNA effectiveness prediction model using the XGBoost algorithm, we use the second-order derivative approximation method of Taylor expansion to approximate the objective function. This method can improve computational efficiency while maintaining high prediction accuracy. Specifically, we use the first-order derivative (gradient) and second-order derivative (Hessian matrix) of the objective function to approximate the original objective function.

[0151] For the siRNA effectiveness prediction task, the first and second order derivatives are calculated as follows:

[0152] First-order derivative (gradient):

[0153] Second-order derivative (Hessian):

[0154] in:

[0155] is the predicted probability of the model for the i-th siRNA sample;

[0156] y i is the true label of the i-th siRNA sample (0 means invalid, 1 means valid).

[0157] These derivatives play a key role in the XGBoost algorithm:

[0158] 1) The first-order derivative (g_i) represents the difference between the current model prediction value and the true label. It guides the model to optimize in the direction of reducing the prediction error, which helps to improve the accuracy of siRNA effectiveness prediction.

[0159] 2) The second-order derivative (h_i) reflects the curvature of the loss function, that is, the rate at which the prediction error changes. In siRNA effectiveness prediction, it helps the algorithm to more accurately estimate the adjustment that should be made in each iteration, thereby accelerating convergence and improving model stability.

[0160] Using this second-order approximation method, the XGBoost algorithm is able to more accurately estimate the importance of siRNA sequence features in each iteration and adjust the model parameters accordingly. This allows the algorithm to quickly and accurately capture key factors that affect siRNA effectiveness, such as sequence composition, thermodynamic properties, and chemical modifications.

[0161] By combining first-order and second-order information, XGBoost can achieve faster convergence speed and higher prediction accuracy in siRNA effectiveness prediction tasks, which has significant advantages over traditional gradient boosting methods.

[0162] The XGBoost algorithm uses an incremental tree model to optimize the siRNA effectiveness prediction. This method allows the model to gradually learn the complex features of the siRNA sequence, thereby continuously improving the prediction accuracy. Below we explain the optimization process of the incremental tree model in detail:

[0163] Optimization of incremental tree model

[0164] In the XGBoost algorithm, we optimize the model by gradually adding new decision trees. Suppose we have built the first t-1 trees and now want to add the tth tree. Our goal is to minimize the following objective function:

[0165]

[0166] in:

[0167] n is the number of siRNA samples in the training set;

[0168] g i and h iare the first-order and second-order gradient statistics of the i-th sample respectively;

[0169] f t (x i ) is the predicted value of the t-th tree for the i-th siRNA sample;

[0170] Ω(f t ) is the regularization term for the t-th tree.

[0171] The regularization term Ω(f t ) Expand and substitute, we get:

[0172]

[0173] in:

[0174] γ is the penalty coefficient that controls the number of leaf nodes in the tree;

[0175] T is the number of leaf nodes in the tree;

[0176] λ is the penalty coefficient that controls the weight of the leaf node;

[0177] is the weight of the jth leaf node.

[0178] The design of this objective function has several important implications for the siRNA effectiveness prediction task:

[0179] 1) Gradient information utilization: g i and h i Contains the error information of the current model in predicting the effectiveness of siRNA. By minimizing these gradients, the newly added trees can focus on correcting the errors of the existing model, thereby improving the overall prediction accuracy.

[0180] 2) Tree structure optimization: The γT term controls the complexity of the tree. In the siRNA prediction task, this helps prevent the model from overfitting to the noise in the training data and instead captures the key features that truly affect the effectiveness of siRNA.

[0181] 3) Leaf node weight adjustment: The term penalizes the weight of the leaf node. This ensures that the model does not rely too much on certain specific features when predicting the effectiveness of siRNA, but takes multiple factors into consideration.

[0182] 4) Balance and flexibility: By adjusting γ and λ, the best balance between model complexity and prediction accuracy can be found. This is particularly important for processing different types of siRNA sequence datasets, as it allows the model to be adaptively adjusted according to specific circumstances.

[0183] By minimizing this objective function, the XGBoost algorithm can gradually build a model that can accurately predict the effectiveness of siRNA without overfitting. This method is particularly suitable for dealing with complex nonlinear relationships and high-dimensional feature spaces that may exist in siRNA sequence data.

[0184] In the process of XGBoost algorithm building a decision tree to predict the effectiveness of siRNA, selecting the best split node is a key step. To achieve this goal, the algorithm calculates the gain of each potential split and selects the split with the largest gain. This method can effectively capture the key features of siRNA sequences that affect their effectiveness.

[0185] Gain calculation for split nodes

[0186] Assume that feature k is considered for splitting at node j, and the splitting point is s. The gain is calculated as follows:

[0187]

[0188] in:

[0189] I L and I R Respectively represent the sample sets of the left child node and the right child node after splitting

[0190] I represents the sample set of nodes before splitting

[0191] g i and h i are the first-order and second-order gradient statistics of the i-th sample, respectively.

[0192] λ is the L2 regularization term coefficient

[0193] γ is a parameter used to control the complexity of the tree

[0194] This gain formula is of great significance in predicting the effectiveness of siRNA:

[0195] 1) Feature importance assessment: By calculating the gains of different features and split points, XGBoost is able to assess the importance of various siRNA sequence features (such as nucleotide composition, sequence position, thermodynamic properties, etc.) to the predictive validity.

[0196] 2) Capturing nonlinear relationships: The gain calculation takes into account the first-order and second-order gradient information, enabling the model to capture the complex nonlinear relationship between siRNA sequence features and its effectiveness.

[0197] 3) Overfitting control: By introducing the regularization term λ and the complexity penalty γ, the algorithm not only improves the prediction accuracy but also controls the complexity of the model to prevent overfitting to the noise in the training data.

[0198] 4) Balanced split effect: The formula takes into account the situation of the left and right child nodes after the split, ensuring that each split can maximize the model's ability to predict the effectiveness of siRNA.

[0199] 5) Computational efficiency: This gain calculation method allows the XGBoost algorithm to efficiently evaluate a large number of possible splitting schemes, maintaining a fast training speed even when dealing with high-dimensional siRNA feature spaces.

[0200] Based on this gain formula, the XGBoost algorithm can efficiently select the best split features and split points to build a powerful siRNA effectiveness prediction model. This process is repeated until the preset stop condition (such as the maximum tree depth or the minimum number of leaf node samples) is reached.

[0201] The decision tree ensemble model constructed in this way can comprehensively consider various factors that affect the effectiveness of siRNA, including sequence features, structural features, thermodynamic properties, etc. This makes the final prediction model not only highly accurate, but also provides interpretable feature importance information, which helps researchers to deeply understand the key factors affecting the effectiveness of siRNA.

[0202] Step 400: Perform multi-index performance evaluation and optimization on the prediction model trained in step 300 to obtain an optimized prediction model.

[0203] Optionally, the multi-index performance evaluation in step 400 adopts a cross-validation method, and the evaluation indicators include two or more of accuracy, precision, recall, F1 score and ROC-AUC.

[0204] Specifically, this step aims to comprehensively evaluate the performance of the XGBOOST model and further optimize it based on the evaluation results to ensure the reliability and effectiveness of the model in practical applications. Through multi-angle and multi-indicator evaluation, we can gain an in-depth understanding of the advantages and disadvantages of the model and provide a clear direction for subsequent optimization. For example:

[0205] First, the cross-validation method is used to evaluate the model. Cross-validation is a robust evaluation technique that can effectively reduce the risk of overfitting and provide more reliable performance estimates. In this method, k-fold cross-validation is used, and k is usually set to 5 or 10. The specific steps are as follows:

[0206] The dataset is randomly divided into k subsets of similar size.

[0207] Use k-1 subsets as training sets and the remaining 1 subset as validation set.

[0208] Train the model and evaluate the performance on the validation set.

[0209] Repeat step 2-3k times, using a different subset as validation set each time.

[0210] The average of k evaluation results is calculated as the final performance indicator.

[0211] This approach can make full use of limited data and provide more robust performance estimates.

[0212] During the evaluation process, a variety of evaluation indicators are used to comprehensively measure the model performance. For example, these indicators include but are not limited to:

[0213] Accuracy: The ratio of correctly predicted samples to the total number of samples. This is an intuitive overall performance indicator, but it may be misleading in the case of class imbalance.

[0214] Precision: The proportion of siRNAs predicted to be effective that are actually effective. This indicator reflects the reliability of the model in predicting effective siRNAs.

[0215] Recall: The proportion of correctly predicted siRNAs among the effective siRNAs. This indicator reflects the ability of the model to capture effective siRNAs.

[0216] F1 score: The harmonic mean of precision and recall. The F1 score balances precision and recall and is particularly useful in cases of class imbalance.

[0217] ROC-AUC (Receiver Operating Characteristic-Area Under Curve): The area under the ROC curve. This indicator reflects the ability of the model to distinguish between effective and ineffective siRNAs at different thresholds, and is very useful for evaluating the overall discrimination ability of the model.

[0218] By calculating these indicators, the performance of the model can be evaluated from different perspectives. For example, a high accuracy rate indicates that the overall prediction of the model is accurate, a high precision rate means that the effective siRNA predicted by the model is more reliable, and a high recall rate indicates that the model can capture most of the actual effective siRNAs. The F1 score provides a balanced perspective of precision and recall, while the ROC-AUC reflects the overall discriminative ability of the model.

[0219] Based on these evaluation indicators, targeted model optimization can be performed. Optimization strategies may include:

[0220] Feature engineering: Analyze feature importance, remove unimportant features or add new meaningful features.

[0221] Model structure adjustment: For example, adjust the number of XGBOOST trees, the depth of each tree and other parameters.

[0222] Regularization strategy: Adjust the L1 and L2 regularization parameters to control model complexity and prevent overfitting.

[0223] Sampling strategy: Adjust the column sampling and row sampling parameters of XGBOOST to improve the generalization ability of the model.

[0224] Integration strategy: Consider integrating XGBOOST with other models (such as random forests, neural networks, etc.) to further improve performance.

[0225] The optimization process is an iterative process. After each adjustment, cross-validation and multi-indicator evaluation need to be re-performed to compare the performance changes before and after optimization. This process may need to be repeated many times until a satisfactory performance level is achieved.

[0226] It is worth noting that during the optimization process, it is necessary to carefully balance the performance of the model on the training set and the validation set. If the model performs very well on the training set but performs poorly on the validation set, it may mean overfitting; conversely, if the model performs poorly on both sets, there may be an underfitting problem.

[0227] Through this rigorous and careful evaluation and optimization process, we finally obtained a siRNA effectiveness prediction model with balanced performance and strong generalization ability. This optimized model can not only accurately predict the effectiveness of siRNA, but also adapt to different types of siRNA data, including siRNAs with various chemical modifications. This provides a reliable computational tool for subsequent siRNA design and screening, and is expected to significantly improve the efficiency and success rate of siRNA drug development.

[0228] Step 500: Input the new siRNA sequence and its chemical modification information into the prediction model optimized in step 400 to perform effectiveness prediction and obtain prediction results.

[0229] Specifically, in step 500, the newly designed or to-be-evaluated siRNA sequence and its chemical modification information are input into the previously optimized prediction model. Optionally, the specific process includes:

[0230] Data preparation: First, the new siRNA data is preprocessed to ensure that its format is consistent with the training data. This includes extracting sequence features (such as GC content, nucleotide composition, etc.) and chemical modification features (such as modification type, position, etc.).

[0231] Feature generation: Generate feature vectors for new siRNAs using the same feature extraction method used for the training data. This ensures that the new data is in the same format as the model’s input.

[0232] Prediction: The generated feature vector is input into the optimized XGBOOST model. The model outputs a prediction effectiveness score for each siRNA, usually a value between 0 and 1, indicating the probability that the siRNA is predicted to be effective.

[0233] Results collation: The prediction results are integrated with the original siRNA information to generate a comprehensive report containing siRNA sequence, chemical modification information and prediction effectiveness score.

[0234] This step provides an important basis for subsequent siRNA screening and optimization, and can quickly evaluate the potential effectiveness of a large number of candidate siRNAs, thereby improving experimental efficiency.

[0235] Step 600: Perform an interpretability analysis on the prediction results of step 500 to generate an explanation report with biological significance.

[0236] Optionally, the interpretability analysis in step 600 adopts SHAP (SHapley Additive ex Planations) value analysis technology.

[0237] Optionally, 600, the SHAP value analysis technology is used to identify the features that have the greatest impact on the prediction of siRNA effectiveness, and compare these features with known siRNA design rules, thereby providing a biologically interpretable basis for prediction.

[0238] Step 600 aims to deeply understand the prediction mechanism of the model and transform the complex machine learning model into an understandable biological explanation. This embodiment uses SHAP (SHapley Additional explanations) value analysis technology to perform interpretability analysis. SHAP is a method based on game theory that can quantify the contribution of each feature to model prediction.

[0239] Optionally, the specific steps of SHAP value analysis are as follows:

[0240] Calculate SHAP value: For each prediction sample, calculate the SHAP value of all features. The SHAP value reflects the positive or negative contribution of the feature to the prediction result.

[0241] Global importance analysis: Summarize the SHAP values ​​of all samples to get the global importance ranking of features. This helps to identify the most influential features for predicting siRNA effectiveness.

[0242] Local interpretation: For a single siRNA prediction result, analyze its SHAP value distribution to understand which features play a decisive role in this specific prediction.

[0243] Feature interaction analysis: Use SHAP interaction values ​​to explore how interactions between features affect prediction results.

[0244] Biological interpretation: SHAP analysis results are compared with known siRNA design rules to provide biologically interpretable predictions.

[0245] Based on the SHAP analysis, a detailed interpretation report can be generated, optionally including the following:

[0246] Key feature identification: List the most important features for predicting siRNA effectiveness, such as nucleotide composition at specific positions, chemical modification type, etc.

[0247] Feature influence direction: describes how each important feature affects the prediction results, for example, the presence of certain features will increase or decrease the predictive effectiveness of siRNA.

[0248] Design rule verification: Compare the important features identified by the model with traditional siRNA design rules (such as Tuschl rule, Reynolds rule, etc.) to verify the validity of known rules and possibly discover new design principles.

[0249] Effects of chemical modification: Detailed analysis of the effects of chemical modification on siRNA effectiveness, including the effects of different modification types and locations.

[0250] Sequence pattern analysis: Identification of specific sequence patterns or structural features associated with highly effective siRNAs.

[0251] Individual siRNA interpretation: Case studies of specific siRNA samples predicted to be highly or inefficient are performed, explaining the key factors that led to this prediction.

[0252] Summary of New Findings: Summarizes the novel design principles discovered through SHAP analysis, which may provide new directions for siRNA design.

[0253] Through this detailed interpretability analysis, researchers can:

[0254] Gain a deeper understanding of the model’s prediction mechanism and increase confidence in the prediction results.

[0255] Validate and improve existing siRNA design principles.

[0256] Discover new factors that influence siRNA potency, especially new knowledge about chemical modifications.

[0257] Provide guidance for customized design of siRNA and select the most suitable sequence and modification strategy according to the characteristics of specific target genes.

[0258] Promote basic research on the mechanism of action of siRNA and provide new research directions and hypotheses.

[0259] This analytical method that combines machine learning and biological knowledge not only improves the accuracy and efficiency of siRNA design, but also provides valuable scientific observations for the development of RNA interference technology. By continuously accumulating and analyzing data, this method has the potential to promote the continuous advancement of siRNA technology and bring new breakthroughs to the fields of gene therapy and drug development.

[0260] Step 700: verifying the actual interference efficiency of the highly effective siRNA sequence predicted in step 500 through in vitro cell experiments.

[0261] Optionally, the in vitro cell experiment verification in step 700 includes chemical synthesis of siRNA, optimization of transfection conditions and evaluation of RNA interference efficiency.

[0262] Specifically, step 700 is to verify the actual interference efficiency of the high-efficiency siRNA sequence predicted in step 500 through in vitro cell experiments. This step is crucial for evaluating the accuracy and practicality of the prediction model. Optionally, the in vitro cell experiment verification mainly includes three key links: chemical synthesis of siRNA, optimization of transfection conditions and evaluation of RNA interference efficiency.

[0263] First, chemical synthesis of siRNA is performed. According to the highly efficient siRNA sequence predicted in step 500, a professional nucleic acid synthesis company is commissioned to perform chemical synthesis. During synthesis, chemical modifications included in the predicted model need to be considered, such as 2'-O-methyl modification, phosphorothioate modification, etc. After the synthesis is completed, quality control is performed, including mass spectrometry analysis and high performance liquid chromatography (HPLC) purity detection.

[0264] Secondly, optimize the transfection conditions. This step is very important to ensure that siRNA can effectively enter the cell and play a role. It mainly includes selecting the appropriate cell line, screening the best transfection reagent, optimizing the siRNA concentration and transfection reagent dosage, and determining the best post-transfection observation time point.

[0265] Finally, evaluate the RNA interference efficiency. This is the core link to verify the actual effect of siRNA, including detecting the expression changes of the target gene at the mRNA and protein levels. Real-time quantitative PCR (qPCR) technology is usually used to measure the mRNA expression level, and Western Blot technology is used to detect the protein expression level. In addition, corresponding functional experiments can be performed according to the function of the target gene.

[0266] In the experimental design, it is necessary to set up appropriate control groups, including negative control, positive control, transfection reagent control and blank control. The setting of these control groups helps to eliminate non-specific effects and ensure the reliability of experimental results.

[0267] Through this series of in vitro cell experiments, the accuracy and practicality of the prediction model can be comprehensively evaluated, the actual interference efficiency of siRNA can be verified, and an important basis can be provided for further optimization of siRNA design strategies.

[0268] Step 800: Based on the interpretation results of step 600 and the experimental verification results of step 700, further optimize and iterate the prediction model, including updating the feature extraction method and the training data set.

[0269] Optionally, the optimization and iteration in step 800 include adjusting the feature extraction method, updating the training data set, and re-optimizing hyperparameters.

[0270] Specifically, step 800 is to further optimize and iterate the prediction model based on the interpretation results of step 600 and the experimental verification results of step 700. The purpose of this step is to continuously improve the performance and accuracy of the model through a feedback mechanism.

[0271] First, the features that the model values ​​and possible biases during prediction can be identified based on the interpretation results of step 600. This information can be used to adjust the feature extraction method, such as increasing or decreasing the weight of certain features, or introducing new feature types.

[0272] Secondly, the experimental verification results of step 700 provide real siRNA interference efficiency data. These new experimental data can be used to update and expand the training data set. By adding the latest, experimentally verified data, the generalization ability and prediction accuracy of the model can be improved.

[0273] Finally, based on the adjusted feature extraction method and the updated training dataset, the model's hyperparameters can be re-optimized. This may include adjusting the learning rate, changing the network structure, modifying regularization parameters, etc. Hyperparameter optimization can be performed through methods such as grid search, random search, or Bayesian optimization.

[0274] Through this iterative process, the prediction model can continuously learn new information and adapt to new data patterns, thereby continuously improving its ability to predict siRNA interference efficiency. This optimization process is dynamic and can continue as new data is accumulated and new knowledge is discovered, ensuring that the model always maintains its optimal performance.

[0275] Step 900: Repeat steps 300 to 800 until the model performance reaches a predetermined standard, thereby obtaining a final siRNA effectiveness prediction model.

[0276] Specifically, step 900 describes an iterative optimization process, which aims to continuously improve the siRNA effectiveness prediction model by repeatedly executing the previous steps until the model performance reaches a predetermined standard.

[0277] For example, this process may include the following main steps:

[0278] Repeat: Starting from step 300 (feature extraction and data preprocessing) to step 800 (model optimization and iteration), this series of steps is repeated.

[0279] Continuous improvement: Each iteration is adjusted and optimized based on the results of the previous round, including feature selection, model training, model interpretation, experimental verification, and model updating.

[0280] Performance evaluation: After each iteration, the performance of the model is evaluated. This may include the evaluation of prediction accuracy, sensitivity, specificity, etc.

[0281] Predetermined standard: Set a performance goal or standard in advance as a condition for stopping iteration. This standard should be determined based on actual application requirements and technical feasibility.

[0282] Termination condition: When the model performance reaches the predetermined standard, the iteration process is stopped.

[0283] Final model: The model obtained after meeting the predetermined criteria was considered the final siRNA effectiveness prediction model.

[0284] Through this iterative approach, the predictive ability and reliability of the model can be gradually improved, and finally a siRNA effectiveness prediction model that meets the needs of practical applications can be obtained. This process embodies the concept of continuous learning and optimization in machine learning, ensuring that the model can adapt to new data and knowledge and maintain its predictive performance.

[0285] The following summarizes some key technical means in the above embodiments, which are as follows:

[0286] 1) Efficient siRNA prediction model based on XGBOOST algorithm

[0287] The above embodiment uses the XGBOOST (eXtreme Gradient Boosting) algorithm to build a high-precision siRNA effectiveness prediction model. XGBOOST is an advanced ensemble learning algorithm that makes predictions by constructing multiple decision trees and combining them. In this model, the above embodiment performs feature engineering on the characteristics of the siRNA sequence and extracts multidimensional features including nucleotide composition, thermodynamic properties, etc. The hyperparameters of the model are optimized by grid search and other methods, such as adjusting the learning rate, maximum tree depth, subsampling rate, etc., to achieve the best prediction effect. This method significantly improves the accuracy and robustness of the prediction results and can better adapt to different types of siRNA sequences and chemical modifications.

[0288] 2) Comprehensive data processing method of siRNA with multiple chemical modifications

[0289] The above embodiment has developed a comprehensive data processing method for processing and integrating siRNA data containing multiple chemical modifications. First, a large amount of siRNA sequence data, including its chemical modification information and experimentally verified validity data, was collected from multiple public databases and laboratories. Then, data cleaning was performed to remove duplicate, erroneous or incomplete data. Next, a feature extraction method was designed that not only took into account the sequence information of siRNA, but also encoded the type, position and quantity of chemical modifications as features. Finally, data from different sources were made comparable through standardized processing. This method ensures the quality and consistency of the data and provides high-quality input for subsequent model training.

[0290] 3) In vitro cell experiment verification method

[0291] The above embodiment combines computational prediction and experimental verification to form a closed-loop optimization system. First, the efficient siRNA sequence screened out according to the prediction model is chemically synthesized. Then, a set of in vitro cell experimental protocols is designed, including selecting a suitable cell line (such as HEK293T), optimizing transfection conditions (such as transfection reagents, siRNA concentration, transfection time, etc.), and using methods such as qRT-PCR or Western blot to evaluate RNA interference efficiency. The experimental results are not only used to verify the accuracy of the prediction, but also used to update the training data set, thereby optimizing the parameters of the prediction model. This method establishes a complete process from prediction to verification, and then to model optimization, and continuously improves the accuracy of the prediction.

[0292] 4) Data imbalance processing technology

[0293] In order to solve the problem that there are usually few efficient sequences in siRNA datasets, the above embodiments use a variety of advanced data balancing technologies. Specifically, they include: (1) oversampling technology, such as SMOTE (Synthetic Minority Over-sampling Technique), which generates new synthetic samples between minority class samples to increase their number; (2) undersampling technology, such as ENN (Edited Nearest Neighbors), which reduces the number of majority class samples by deleting some noisy majority class samples; (3) Generative Adversarial Networks (GANs), which generates high-quality synthetic minority class samples by training generators and discriminators. The combined application of these technologies significantly improves the model's ability to identify efficient siRNA sequences (minority class samples), while avoiding the bias that may be caused by a single technology.

[0294] 5) Model interpretation analysis

[0295] In order to enhance the interpretability of the model, the above embodiment introduces SHAP (SHapley AdditiveexPlanations) value analysis technology. SHAP value is based on the concept of Shapley value in game theory, which can quantify the contribution of each feature to the prediction result. In the above embodiment, the SHAP value of each siRNA sequence feature (such as nucleotides at specific positions, GC content, chemical modification, etc.) is calculated to identify the most influential factors for siRNA effectiveness prediction. These results are used to generate detailed explanation reports, including feature importance ranking, feature positive and negative effects on prediction, etc. By comparing these analysis results with known siRNA design rules, the above embodiment can provide biologically meaningful explanations, help researchers understand the prediction mechanism of the model, and provide guidance for the design and optimization of siRNA.

[0296] In order to better understand the technical solution of the present application, a specific example is provided below for illustration. The details listed in the example are mainly for ease of understanding and are not intended to limit the scope of protection of the present application.

[0297] This example proposes a siRNA effectiveness prediction platform based on the XGBOOST algorithm, which improves the prediction accuracy and efficiency of siRNA sequences through advanced machine learning technology and data processing methods. Figure 1 The overall process of this example is demonstrated, including key steps such as data collection and processing, model training and optimization, model evaluation and interpretation, experimental verification and result feedback.

[0298] Specifically, this example includes the following steps:

[0299] Step 10: Data processing, including:

[0300] Data collection: siRNA sequences containing various chemical modifications and their experimental verification data were collected from public databases and laboratories.

[0301] Data cleaning and preprocessing: Clean and standardize the raw data, remove redundant and inconsistent data, fill in missing values, and perform necessary standardization. Extract relevant features, such as sequence features, secondary structure features, and chemical modification information. One-hot encoding is used for RNA sequences, and one-hot encoding is performed for chemical modification types according to the corresponding RNA base positions.

[0302] Data balancing processing: To address the problem of imbalanced data sets, techniques such as oversampling (Random Oversampling), undersampling (Undersampling) and generative adversarial networks (GANs) are used to generate balanced data sets.

[0303] Step 20: Model training and optimization, including:

[0304] Model selection: The XGBOOST algorithm was used to establish a siRNA effectiveness prediction model.

[0305] Hyperparameter tuning: Use K-fold validation and grid search techniques to optimize the hyperparameters of the XGBOOST model (such as learning rate, maximum depth, subsampling rate, etc.) to ensure the efficiency and robustness of the model.

[0306] Model training: Model training is performed on the processed dataset. XGBOOST gradually optimizes model parameters through continuous iteration to capture the complex relationship between siRNA sequences and their effectiveness.

[0307] Step 30: Model evaluation and interpretation, including:

[0308] Model evaluation: The performance of the model is evaluated through indicators such as accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve (ROC-AUC)*1.

[0309] Result interpretation: Use explanatory analysis techniques such as SHAP values ​​(Shapley Additive Explanations) to explain the model prediction results and provide biologically meaningful explanations.

[0310] Step 40: Experimental verification, including:

[0311] siRNA synthesis: Chemical synthesis is performed based on the efficient siRNA sequences screened out by the model.

[0312] In vitro cell experiments: Select cell lines with high transfection efficiency (such as HEK293T cells), optimize transfection conditions, and evaluate the RNA interference efficiency of siRNA.

[0313] Result feedback: Feedback the experimental results to the model to further optimize model performance and hyperparameter settings.

[0314] Optionally, in this example, the following metrics are used to evaluate model performance:

[0315] Accuracy: The ratio of the number of correctly predicted samples to the total number of samples.

[0316] Precision: The proportion of samples predicted to be positive that are actually positive.

[0317] Recall: The proportion of samples that are predicted to be positive among samples that are actually positive.

[0318] F1 Score: The harmonic mean of precision and recall, which comprehensively evaluates model performance.

[0319] ROC-AUC: The area under the receiver operating characteristic curve reflects the ability of the model to distinguish between positive and negative samples.

[0320] These indicators reflect the accuracy and efficiency of siRNA prediction as follows:

[0321] Accuracy reflects the overall predictive ability of the model, but it may be distorted when the data is imbalanced.

[0322] A high precision indicates that most of the predicted positive samples are correct, and a high recall indicates that most of the actual positive samples are correctly predicted.

[0323] The F1 score comprehensively evaluates precision and recall and is suitable for data imbalance situations.

[0324] ROC-AUC reflects the classification performance of the model at different thresholds and more comprehensively evaluates the predictive ability of the model.

[0325] Furthermore, the key hyperparameter settings and their effects of the XGBOOST algorithm in this example are as follows (all can be adjusted according to experimental results):

[0326] Learning rate (eta): determines the size of each iteration step, affecting the model convergence speed and prediction accuracy. It is generally set between 0.01 and 0.3.

[0327] Maximum depth (max_depth): controls the maximum depth of the tree. A larger depth can capture more features, but may lead to overfitting. It is generally set between 3 and 10.

[0328] Subsample rate: determines the sample ratio used in each iteration to prevent overfitting. It is usually set between 0.5 and 1.

[0329] Regularization parameter (lambda, alpha): controls the complexity of the model and reduces overfitting through penalty terms. It is usually set between 1 and 10.

[0330] See also Figure 2 , the whole process of this example can be divided into four units:

[0331] The data processing unit includes data collection, data cleaning, data balancing and feature extraction modules. High-performance servers are used for data processing to ensure data quality and consistency.

[0332] The model training and optimization unit includes model selection, hyperparameter tuning, and model training modules. The high-performance computing platform is used to run the XGBOOST algorithm for model training and optimization.

[0333] The model evaluation and interpretation unit includes model evaluation and result interpretation modules. It uses explanatory analysis techniques to provide biological explanations for model prediction results.

[0334] The experimental verification unit includes siRNA synthesis and in vitro cell experiment modules. Laboratory equipment is used to perform chemical synthesis and cell experiment verification of siRNA to evaluate the RNA interference efficiency of siRNA.

[0335] The above embodiments and specific examples have the following technical effects:

[0336] 1) Improved prediction accuracy: The XGBOOST algorithm can more accurately capture the complex relationship between siRNA sequences and their effectiveness, significantly improving the accuracy and robustness of prediction results. Compared with traditional algorithms such as random forests, XGBOOST shows higher prediction accuracy when processing complex biological data.

[0337] 2) Comprehensive consideration of the impact of chemical modifications: By processing and integrating data on siRNAs with various chemical modifications, the applicability and reliability of the prediction model for siRNAs with different modified types are improved. This enables the model to better adapt to new chemical modifications and provide more accurate guidance for drug development.

[0338] 3) Effectively solve the problem of data imbalance: The use of advanced data balancing techniques, such as oversampling, undersampling, and generative adversarial networks (GANs), enhances the model's ability to identify minority class samples. This solves the shortcomings of existing methods in dealing with unbalanced siRNA data and improves the comprehensiveness of predictions.

[0339] 4) Enhanced model interpretability: By introducing explanatory analysis techniques such as SHAP values, biologically meaningful explanations are provided for the prediction results. This helps researchers better understand the mechanisms behind the prediction results and improves the credibility and practicality of the model.

[0340] 5) Improve computational efficiency: The XGBOOST algorithm uses parallel processing and incremental training to significantly improve computational speed. This enables the platform to quickly process large-scale siRNA data and meet the needs of high-throughput screening.

[0341] 6) Shorten the drug development cycle: By quickly and accurately screening highly effective siRNA sequences, the initial stage of new drug development is greatly shortened. This is of great significance for pharmaceutical companies to quickly respond to market demand and seize opportunities.

[0342] 7) Reduce R&D costs: High-precision prediction reduces the number of candidate sequences that need to be experimentally verified, thereby reducing experimental costs. At the same time, it improves the success rate of siRNA drug development and reduces ineffective R&D investment.

[0343] 8) Improve the application efficiency of chemical modification patents: For newly obtained chemical modification patents, the platform can quickly evaluate their application potential in siRNA design, accelerating the transformation process from patents to practical applications.

[0344] 9) Enhanced application potential in multiple fields: Although initially targeted at cardiovascular diseases, the flexibility of the platform makes it easy to expand to the development of siRNA drugs in other disease areas, such as tumors, neurological diseases, etc.

[0345] 10) Promote the development of personalized medicine: By accurately predicting the effectiveness of different siRNA sequences, it lays the foundation for the development of personalized siRNA drugs for specific patients or subpopulations.

[0346] In summary, this efficient siRNA effectiveness prediction platform based on the XGBOOST algorithm significantly improves the efficiency and success rate of siRNA drug development, and provides strong technical support for the rapid development of a new generation of RNA interference therapies.

[0347] The second embodiment of the present application relates to an efficient siRNA effectiveness prediction system based on the XGBOOST algorithm, and its structure is as follows: Figure 4As shown, the efficient siRNA effectiveness prediction system based on the XGBOOST algorithm includes:

[0348] Data acquisition and preprocessing module: used to acquire a training data set containing siRNA sequence information, chemical modification information and corresponding validity data, and preprocess the training data set, including data cleaning, feature extraction and data balancing, to obtain balanced training data;

[0349] Model training and optimization module: used to train the siRNA effectiveness prediction model based on the balanced training data using the XGBOOST algorithm and optimize the algorithm hyperparameters; perform multi-index performance evaluation and optimization on the trained prediction model to obtain the optimized prediction model;

[0350] Prediction and interpretation module: used to input new siRNA sequences and their chemical modification information into the optimized prediction model, perform validity prediction, obtain prediction results, and perform interpretability analysis on the prediction results to generate a biologically meaningful interpretation report;

[0351] Experimental verification and model optimization module: used to verify the actual interference efficiency of the predicted high-efficiency siRNA sequence through in vitro cell experiments, and further optimize and iterate the prediction model based on the interpretation results and experimental verification results, including updating the feature extraction method and training data set;

[0352] Loop execution module: used to control the repeated execution of the model training and optimization module, the prediction and interpretation module, the experimental verification and model optimization module until the model performance reaches the predetermined standard and the final siRNA effectiveness prediction model is obtained.

[0353] The first implementation manner is a method implementation manner corresponding to the present implementation manner. The technical details in the first implementation manner can be applied to the present implementation manner, and the technical details in the present implementation manner can also be applied to the first implementation manner.

[0354] It should be noted that those skilled in the art should understand that the implementation functions of each module shown in the above-mentioned embodiment of the efficient siRNA effectiveness prediction system based on the XGBOOST algorithm can be understood by referring to the relevant description of the efficient siRNA effectiveness prediction method based on the XGBOOST algorithm. The functions of each module shown in the above-mentioned embodiment of the efficient siRNA effectiveness prediction system based on the XGBOOST algorithm can be realized by a program (executable instruction) running on a processor, or by a specific logic circuit. If the efficient siRNA effectiveness prediction method based on the XGBOOST algorithm in the embodiment of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM, Read Only Memory), a disk or an optical disk. Thus, embodiments of the present application are not limited to any specific combination of hardware and software.

[0355] Accordingly, an embodiment of the present application further provides a computer storage medium, in which computer executable instructions are stored. When the computer executable instructions are executed by a processor, the various method embodiments of the present application are implemented.

[0356] In addition, the embodiments of the present application also provide an efficient siRNA effectiveness prediction system based on the XGBOOST algorithm, which includes a memory for storing computer executable instructions, and a processor; the processor is used to implement the steps in the above-mentioned method implementations when executing the computer executable instructions in the memory. Among them, the processor can be a central processing unit (Central Processing Unit, referred to as "CPU"), or other general-purpose processors, digital signal processors (Digital Signal Processor, referred to as "DSP"), Application Specific Integrated Circuit (Application Specific Integrated Circuit, referred to as "ASIC"), etc. The aforementioned memory can be a read-only memory (read-only memory, referred to as "ROM"), a random access memory (random access memory, referred to as "RAM"), a flash memory (Flash), a hard disk or a solid-state hard disk, etc. The steps of the method disclosed in each embodiment of the present application can be directly embodied as a hardware processor execution, or a combination of hardware and software modules in the processor.

[0357] It should be noted that in the application documents of this patent, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one" do not exclude the existence of other identical elements in the process, method, article or device including the elements. In the application documents of this patent, if it is mentioned that an action is performed according to an element, it means that the action is performed at least according to the element, which includes two situations: performing the action only according to the element, and performing the action according to the element and other elements. Expressions such as multiple, multiple, and multiple include 2, 2 times, 2 kinds, and more than 2, more than 2 times, and more than 2 kinds.

[0358] All documents mentioned in this application are considered to be included in the disclosure of this application as a whole, so that they can be used as the basis for modification when necessary. In addition, it should be understood that after reading the above disclosure of this application, those skilled in the art can make various changes or modifications to this application, and these equivalent forms also fall within the scope of protection claimed in this application.

Claims

1. A highly efficient siRNA effectiveness prediction method based on XGBOOST algorithm, characterized in that: The following steps are involved: Step A: obtaining a training data set containing siRNA sequence information, chemical modification information and corresponding effectiveness data; Preprocessing the training data set, including data cleaning, feature extraction and data balancing, to obtain balanced training data; Step B: Based on the balanced training data, the XGBOOST algorithm is used to train the siRNA effectiveness prediction model and optimize the algorithm hyperparameters; multi-index performance evaluation and optimization are performed on the trained prediction model to obtain the optimized prediction model; Step C: Input the new siRNA sequence and its chemical modification information into the optimized prediction model to perform effectiveness prediction and obtain prediction results; perform interpretability analysis on the prediction results and generate a biologically meaningful interpretation report; Step D: verifying the actual interference efficiency of the predicted high-efficiency siRNA sequence through in vitro cell experiments; based on the interpretation results and experimental verification results, further optimizing and iterating the prediction model, including updating the feature extraction method and training data set; Step E: Repeat steps B to D until the model performance reaches a predetermined standard, thereby obtaining a final siRNA effectiveness prediction model.

2. The method according to claim 1, characterized in that The preprocessing in step A specifically includes: cleaning and standardizing the data; extracting siRNA sequence features and chemical modification features to generate a feature matrix; and using data balancing technology to handle data imbalance problems in the feature matrix.

3. The method according to claim 2, characterized in that The data balancing technique includes one or more of oversampling, undersampling or generative adversarial networks (GANs).

4. The method according to claim 1, characterized in that: The XGBOOST algorithm used in step B adopts parallel computing technology to improve computing efficiency.

5. The method according to claim 1, characterized in that: The method for optimizing the hyperparameters of the XGBOOST algorithm in step B includes one or more of grid search, random search or Bayesian optimization, and the hyperparameters include learning rate, maximum tree depth and regularization parameter.

6. The method according to claim 1, characterized in that The multi-index performance evaluation in step B adopts a cross-validation method, and the evaluation indicators include two or more of accuracy, precision, recall, F1 score and ROC-AUC.

7. The method according to claim 1, characterized in that The interpretability analysis in step C adopts the SHAP value analysis technique.

8. The method according to claim 1, characterized in that The optimization and iteration in step D include adjusting the feature extraction method, updating the training data set, and re-optimizing the hyperparameters.

9. The method according to claim 1, characterized in that: The in vitro cell experiment verification in step D includes chemical synthesis of siRNA, optimization of transfection conditions and evaluation of RNA interference efficiency.

10. The method according to claim 2, characterized in that The feature extraction step includes extracting information such as the nucleotide composition, thermodynamic properties, and type, position, and quantity of chemical modifications of the siRNA sequence.

11. The method according to claim 7, characterized in that The SHAP value analysis technique is used to identify the most influential features for predicting siRNA effectiveness, and compare these features with known siRNA design rules, thereby providing a biologically interpretable basis for prediction.

12. An efficient siRNA effectiveness prediction device based on XGBOOST algorithm, characterized in that: include: Data acquisition and preprocessing module: used to acquire a training data set containing siRNA sequence information, chemical modification information and corresponding validity data, and preprocess the training data set, including data cleaning, feature extraction and data balancing, to obtain balanced training data; Model training and optimization module: used to train the siRNA effectiveness prediction model based on the balanced training data using the XGBOOST algorithm and optimize the algorithm hyperparameters; perform multi-index performance evaluation and optimization on the trained prediction model to obtain the optimized prediction model; Prediction and interpretation module: used to input new siRNA sequences and their chemical modification information into the optimized prediction model, perform effectiveness prediction, obtain prediction results, and perform interpretability analysis on the prediction results to generate a biologically meaningful interpretation report; Experimental verification and model optimization module: used to verify the actual interference efficiency of the predicted high-efficiency siRNA sequence through in vitro cell experiments, and further optimize and iterate the prediction model based on the interpretation results and experimental verification results, including updating the feature extraction method and training data set; Loop execution module: used to control the repeated execution of the model training and optimization module, the prediction and interpretation module, the experimental verification and model optimization module until the model performance reaches the predetermined standard and the final siRNA effectiveness prediction model is obtained.

Images