Method and system for detecting activity of ACE inhibitory peptide on scallop skirt based on deep learning

Abstract

The invention discloses a scallop skirt ACE inhibitory peptide activity method and system based on deep learning, and relates to the technical field of bioengineering.The scallop skirt ACE inhibitory peptide activity method comprises the steps that a comprehensive data set is obtained based on simulated in-vivo test data and peptide level feature data of different peptides, key features are screened, and a training set and a test set are divided; initializing a random forest model, an SVM model and a logistic regression model, performing independent training by using the training set, and determining an optimal prediction model of the peptide antihypertensive activity based on a prediction result of the test set; extracting importance data of key features based on the optimal prediction model for the antihypertensive activity of the peptide, sorting the importance data, and visually analyzing the antihypertensive activity features, feature importance sorting and model performance of different peptides; and all key results in the analysis process are integrated and systematically stored. According to the method, normal form reference with prediction accuracy and mechanism depth can be provided for clinical transformation of marine bioactive peptides, and AI-driven novel peptide drug research and development are promoted to enter a new stage.

Description

Technical Field

[0001] This invention relates to the field of bioengineering technology, and more specifically to a method and system for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning. Background Technology

[0002] Bioactive peptides, as functional fragments of proteins, have become core targets for the development of functional foods and drugs due to their high selectivity, low toxicity, and diverse physiological regulatory functions (such as antihypertensive, antioxidant, and antibacterial effects). However, traditional research models relying on experimental screening face three major efficiency bottlenecks: ① The discovery cycle for bioactive peptides is long, requiring several months of laboratory work to screen 10,000 peptide fragments with a success rate of less than 0.1%; ② Structure-activity mechanism analysis is difficult, as the transient interactions between short peptides and target proteins (such as ACE and Keap1) are difficult to capture experimentally; ③ In vivo activity verification is lagging, with significant differences often existing between in vitro activity and in vivo effects, resulting in low clinical translation rates. These bottlenecks severely restrict the transformation of bioactive peptides from basic research to industrial applications, necessitating the development of new research paradigms to overcome existing limitations.

[0003] In recent years, breakthroughs in computational simulation technology have provided new solutions to the aforementioned challenges. Deep learning (such as XGBoost and Random Forest) can quickly identify high-potential peptides from massive sequences through feature weight analysis, improving efficiency by hundreds of times compared to traditional methods. For example, the DeepRT method developed by Ma et al. uses deep learning to predict peptide retention times in liquid chromatography. This method combines CNN (a 4-layer convolutional network for processing peptide sequences) and RNN (treating amino acids as 20-length vectors for processing sequences) to extract features. After dimensionality reduction using principal component analysis, it employs three machine learning methods—support vector machine, random forest, and gradient boosting—to predict retention times. Because neural networks can automatically learn and process data, reducing reliance on human experience, they can extract features more effectively, resulting in higher prediction accuracy. After training on a training, validation, and test set divided in an 8:1:1 ratio, the correlation between the theoretical prediction and the actual value is close to 0.99. Compared to software such as ELUDE and GPTime, the retention time prediction is more accurate, forming a scientific closed-loop model establishment method of "virtual prediction → in vivo empirical verification".

[0004] However, existing computational simulation techniques still have many shortcomings in the study of ACE inhibitory peptides from scallop skirts. How to develop a deep learning research method that integrates computational simulation and experimental verification for ACE inhibitory peptides from scallop skirts, and solve the problems of low efficiency, unclear structure-activity mechanism, and disconnect between in vivo and in vitro activity of traditional methods, so as to provide technical support for the efficient research and development and industrial application of ACE inhibitory peptides from scallop skirts, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, which solves the problems existing in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A deep learning-based method for detecting the activity of ACE inhibitory peptides in scallop skirts includes the following steps: S1. Collect simulated in vivo experimental data and peptide level characteristic data of different peptides and merge them to obtain a comprehensive dataset; S2. Select key features from the comprehensive dataset and divide it into training and test sets; S3. Initialize three deep learning models: random forest, SVM, and logistic regression. Train each of the three deep learning models independently using the training set. Based on the prediction results of the test set, confirm the optimal prediction model for peptide antihypertensive activity. S4. Based on the optimal prediction model for peptide antihypertensive activity, extract the importance data of key features and sort them, and visualize and analyze the antihypertensive activity characteristics, feature importance ranking, and model performance of different peptides. S5. Integrate all key results from the analysis process and store them systematically.

[0007] Optionally, in S1, the simulated in vivo experimental data includes information on systolic blood pressure changes, time information, and significance information for RW peptide, RF peptide, KF peptide, VF peptide, GL peptide, and GIL peptide; the peptide level characteristic data includes the basic physicochemical characteristics of RW peptide, RF peptide, KF peptide, VF peptide, AF peptide, NF peptide, VW peptide, GF peptide, GL peptide, and GIL peptide.

[0008] Optionally, the specific steps of S1 are as follows: using peptide sequences as associated keywords, the simulated in vivo experimental data and peptide level feature data are merged through data linking to form a comprehensive dataset containing multi-dimensional information.

[0009] Optionally, the specific steps of S2 are as follows: Five key features were selected from the comprehensive dataset: docking energy, activity score, molecular weight, net charge, and water solubility. These features were then standardized using the StandardScaler tool to eliminate dimensional differences. Stratified sampling is used to divide the standardized feature data and corresponding target variables into training and test sets.

[0010] Optionally, in S3, if the training set has a single class, only the random forest model will be trained.

[0011] Optionally, in S3, the optimal prediction model for peptide antihypertensive activity is confirmed based on the prediction results of the test set, specifically as follows: Input the test set into each trained deep learning model to obtain the prediction results; By combining the real labels of the test set, the classification accuracy and AUC value of each deep learning model are calculated. The model with the highest classification accuracy and the largest AUC value is selected as the final optimal prediction model for peptide antihypertensive activity.

[0012] Optionally, the specific steps of S4 are as follows: Using the optimal prediction model for peptide antihypertensive activity, the importance data of key features were extracted and ranked to clarify the degree of influence of key features on the prediction results of peptide antihypertensive activity. Line graphs are used to present the depressurization effects of different peptides over time series, bar graphs are used to show the probability distribution of peptide activity and the importance of features, scatter plots are used to show the relationship between molecular docking energy and activity score, heatmaps are used to present the confusion matrix of random forest, and ROC curves are used to reflect the model's discriminative ability. Based on computational analysis and visualization results, the antihypertensive activity characteristics of different peptides, the ranking of feature importance, and the model performance were analyzed.

[0013] A system for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, for performing any of the above-described methods for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, comprising: The data collection module is used to collect and merge simulated in vivo experimental data and peptide level characteristic data of different peptides to obtain a comprehensive dataset. The data splitting module is used to filter key features from the comprehensive dataset and divide it into training and test sets; The model building and training module is used to initialize three deep learning models: random forest, SVM, and logistic regression. The training set is used to train the three deep learning models independently. Based on the prediction results of the test set, the optimal prediction model for peptide antihypertensive activity is confirmed. The computational analysis and visualization module is used to extract and rank the importance data of key features through the optimal prediction model of peptide antihypertensive activity, and to visualize and analyze the antihypertensive activity characteristics of different peptides, the ranking of feature importance, and the performance of the model. The results output and storage module is used to integrate all key results from the analysis process and store them systematically.

[0014] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses a method and system for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, which has the following beneficial effects: This invention develops a deep learning research method that integrates computational simulation and experimental verification for ACE inhibitory peptides in scallop skirts. It solves the problems of low efficiency, unclear structure-activity mechanism, and disconnect between in vivo and in vitro activity of traditional methods. It can provide a paradigm reference with both predictive accuracy and mechanistic depth for the clinical translation of marine bioactive peptides, and promote the development of new peptide drugs driven by AI into a new stage. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 Flowchart of the deep learning-based method for inhibiting the activity of ACE inhibitory peptides in scallop skirts provided by this invention; Figure 2 The peptide activity probability distribution diagram provided by this invention; Figure 3 Feature importance distribution map provided for this invention; Figure 4 The molecular docking energy and activity score diagram provided by this invention; Figure 5 The random forest confusion matrix analysis diagram provided by this invention; Figure 6 The ROC curve provided for this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Addressing the technological bottlenecks identified in the background section, this invention focuses on scallop-derived ACE-inhibiting peptides, constructing a comprehensive research framework encompassing "computational simulation-directed optimization - in vitro and in vivo activity verification - in-depth analysis of structure-activity mechanism." In the computationally guided peptide design stage, BIOPEP virtual enzymatic digestion, AlphaFold2 structure prediction, and molecular docking energy gradient analysis are integrated to screen for Zn targeting the ACE active site. 2+Highly efficient peptides with charge-active domains. A three-tiered system for cross-scale activity validation was established, comprising a Caco-2 cell uptake model, SHR blood pressure time-course monitoring, and dynamic tracking of molecular interactions, bridging the gap between computation and experiment. Particularly in exploring drug resistance reversal mechanisms, umbrella sampling was used to simulate the conformational stability of the peptide-ACE complex under blood pressure fluctuations, elucidating the contributions of Trp indole ring π-π stacking and the Ala356 double hydrogen bond network to long-acting antihypertensive effects. This study aims to provide a paradigmatic reference with both predictive accuracy and mechanistic depth for the clinical translation of marine bioactive peptides, propelling AI-driven novel peptide drug development into a new stage.

[0019] Therefore, this invention discloses a method for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, such as... Figure 1 As shown, it includes the following steps: S1. Collect simulated in vivo experimental data and peptide level characteristic data of different peptides and merge them to obtain a comprehensive dataset; S2. Select key features from the comprehensive dataset and divide it into training and test sets; S3. Initialize three deep learning models: random forest, SVM, and logistic regression. Train each of the three deep learning models independently using the training set. Based on the prediction results of the test set, confirm the optimal prediction model for peptide antihypertensive activity. S4. Based on the optimal prediction model for peptide antihypertensive activity, extract the importance data of key features and sort them, and visualize and analyze the antihypertensive activity characteristics, feature importance ranking, and model performance of different peptides. S5. Integrate all key results from the analysis process and store them systematically.

[0020] Next, for Figure 1 The process shown will be described in detail to further understand the technical solution to be protected by this invention.

[0021] I. Experimental Methods

[0022] This experiment used Windows 11 as the system and PyCharm (student version) and Anconda (free version) as the development tools. It was completed in a food laboratory at a university. The specific configuration is shown in Table 1.

[0023] Table 1 System Development Hardware Configuration

[0024] 1. Combine in vivo experimental data and simulation data

[0025] Merging in vivo experimental data is the process of organically integrating previously generated simulated in vivo experimental data with peptide level characteristic data. In S1 of this embodiment, the simulated in vivo experimental data includes information on systolic pressure changes, time information, and significance information for RW peptides, RF peptides, KF peptides, VF peptides, GL peptides, and GIL peptides; the peptide level characteristic data includes basic physicochemical characteristics such as docking energy, activity score, molecular weight, net charge, and water solubility for RW peptides, RF peptides, KF peptides, VF peptides, AF peptides, NF peptides, VW peptides, GF peptides, GL peptides, and GIL peptides.

[0026] Furthermore, the specific steps in S1 are as follows: using peptide sequences as associated keywords, simulated in vivo experimental data and peptide level feature data are merged through data linking to form a comprehensive dataset containing multi-dimensional information. This process achieves the fusion of dynamic experimental data and static feature data, preserving dynamic information such as the antihypertensive effect of different peptides in vivo over time, while also supplementing their inherent physicochemical properties. This provides a comprehensive and coherent data foundation for subsequent feature engineering and model construction, and helps to more accurately analyze the antihypertensive activity of peptides and the correlation between various features.

[0027] 2. Data Preprocessing

[0028] In this embodiment, the specific steps of S2 are as follows: Five key features—docking energy, activity score, molecular weight, net charge, and water solubility—are selected from the comprehensive dataset. These features are then standardized using the StandardScaler tool to eliminate dimensional differences and ensure the features are on the same scale. A stratified sampling method is then used to divide the standardized feature data and the corresponding target variable into training and testing sets. This method ensures that the proportion of significantly lowering blood pressure peptides to non-significantly lowering blood pressure peptides is consistent between the training and testing sets, laying the foundation for fair model evaluation.

[0029] 3. Model Building and Training

[0030] The model training phase is a crucial transformation process from data to an effective model. In this embodiment, the specific steps of S3 are as follows: initialize three deep learning models—random forest, SVM, and logistic regression—and train each model independently using the training set. Considering practical considerations, if the training set contains only one peptide category, targeted measures will be taken, such as training only the random forest model or adjusting the training process, to ensure the feasibility and effectiveness of model training. Ultimately, a model capable of predicting whether a peptide is a significantly blood pressure-lowering peptide will be constructed.

[0031] Furthermore, in S3, the optimal prediction model for peptide antihypertensive activity is confirmed based on the prediction results of the test set. Specifically, the test set is input into each trained deep learning model to obtain the prediction results; combined with the real labels of the test set, the classification accuracy and AUC value of each deep learning model are calculated, and the model with the highest classification accuracy and the largest AUC value is selected as the final optimal prediction model for peptide antihypertensive activity.

[0032] 4. Computational Analysis and Visualization

[0033] Computational analysis and visualization work closely together to help uncover the value of data. In this embodiment, the specific steps of S4 are as follows: Using the optimal prediction model for peptide antihypertensive activity, we extracted and ranked the importance data of key features to clarify the degree of influence of key features on the prediction results of peptide antihypertensive activity and identified key influencing factors. Line charts are used to present the blood pressure-lowering effects of different peptides over time, visually demonstrating the trend of systolic blood pressure changes over time; bar charts are used to display the probability distribution of peptide activity and the importance of features, making the probability and feature influence clear at a glance; scatter plots are used to present the relationship between molecular docking energy and activity score, exploring the correlation between the two; heatmaps are used to present the random forest confusion matrix to help evaluate the model's classification performance; and ROC curves are used to reflect the model's discriminative ability. Based on computational analysis and visualization results, the antihypertensive activity characteristics of different peptides, the ranking of feature importance, and the model performance were analyzed.

[0034] 5. Results Output and Saving

[0035] This stage is a crucial step in systematically storing the key results generated during the analysis process. Its core purpose is to achieve traceability, reusability, and sharing of the results. In this stage, key results such as peptide antihypertensive activity prediction results tables, classification model performance evaluation reports, and feature importance analysis tables will be output first. Various generated visualization charts, such as time-series antihypertensive effect line graphs and peptide activity probability distribution bar graphs, will be saved in 300dpi resolution PNG format as "peptide antihypertensive classification results.png".

[0036] After integrating and processing the structured data, it was exported along with the raw data containing in vivo experimental data and peptide characteristics as CSV files, such as "Peptide Blood Pressure Classification Results.csv" and "Simulated In vivo Data.csv," using UTF-8 SIG encoding to ensure proper display of Chinese content. These operations not only fully preserved the raw data, processing results, and visualizations from the analysis process but also incorporated the final research conclusions and recommendations, providing strong data support and decision-making basis for subsequent peptide-related research and applications.

[0037] II. Results Analysis

[0038] 1. Probability distribution of peptide activity

[0039] peptide activity probability distribution as follows Figure 2 As shown in the horizontal bar chart, the predicted probabilities of the antihypertensive activity of the six peptides are visually presented (RW: 0.857 > KF: 0.836 > RF: 0.833 > VF: 0.610 > GIL: 0.569 > GL: 0.517). The probabilities of RW, KF, and RF all exceed 0.83, which is highly consistent with the experimentally measured significant antihypertensive effect (systolic blood pressure reduction ≥14 mmHg), confirming the reliability of the model in identifying highly effective peptides. It is noteworthy that although VF peptide was labeled as an effective peptide (minimum reduction in systolic blood pressure of -10.51 mmHg), its probability value (0.610) is close to the critical region of the classification threshold of 0.6, suggesting that its effect may be unique: on the one hand, the molecular docking energy of VF (54.29 kcal / mol) is significantly lower than that of other effective peptides (RW: 71.97 kcal / mol), suggesting that its binding mode may be atypical; on the other hand, its water solubility characteristic value is 0 (hydrophobic), which may lead to reduced oral absorption efficiency and thus weaken its in vivo activity. The predicted probabilities of GL and GIL are both below 0.57, consistent with the labeling of actual ineffective peptides, but the minimum reduction in systolic blood pressure of GL (-5.24 mmHg) is close to the effectiveness threshold (-5 mmHg), reflecting the model's conservative judgment tendency for weak activity signals.

[0040] 2. Feature Importance Analysis

[0041] The feature importance analysis obtained from the random forest model revealed key structural determinants of the activity of antihypertensive peptides. For example... Figure 3 As shown, molecular docking energy (importance 0.250) is listed as the highest contributing characteristic, which is directly related to the mechanism of action of ACE inhibitory peptides—the larger the negative docking energy, the stronger the binding strength between the peptide and the ACE active site (e.g., RW peptides are π-π stacked with ACE via the Trp indole ring and Ala356 double hydrogen bonds). Water solubility and net charge (both importance 0.206) are tied for second place, highlighting the regulatory role of bioavailability on in vivo activity; peptides with positive net charges (e.g., RW, RF net charge = 1) are more likely to penetrate the intestinal mucosal barrier; while peptides with poor water solubility (e.g., VF water solubility = 0), although they may have increased affinity for the hydrophobic pocket of ACE due to enhanced hydrophobicity, will significantly reduce oral bioavailability. The effect of molecular weight (importance 0.162) is relatively weak, but still conforms to the rule that small molecule peptides (<1000 Da) are more easily absorbed (e.g., GL molecular weight 188.22 Da, although small, is still ineffective due to the lack of key active groups).

[0042] 3. Molecular docking energy and activity score analysis

[0043] Molecular docking energy and activity score, as follows Figure 4 As shown, this scatter plot reveals the core rules of structure-activity relationship through three-dimensional visualization (horizontal axis: docking energy; vertical axis: activity score; dot size: minimum reduction in systolic blood pressure; color: predicted probability). Highly efficient peptides cluster in the upper left quadrant (docking energy ≤ -60 kcal / mol and activity score ≥ 0.9). For example, RW and RF appear as large dark dots in this region, confirming the positive correlation between strong binding force and high activity. VF peptide, as a key outlier (docking energy -54.29 kcal / mol, activity score 0.815, probability 0.610), reveals the limitations of the model due to its anomalous location: although its docking energy is weak, experiments show that it still has a certain blood pressure-lowering effect (systolic blood pressure reduction of -10.51 mmHg), which may be due to the active fragments generated after it is dissociated by digestive enzymes (similar to the highly active tripeptide GRP released by the degradation of soybean peptide WGPRL). GL peptides (upper right quadrant) typically exhibit the characteristics of "high docking energy - low activity" as ineffective peptides, as they cannot stably bind to the ACE catalytic domain due to the lack of key C-terminal aromatic amino acids (such as Phe and Trp).

[0044] 4. Confusion Matrix Analysis

[0045] The confusion matrix shown in the heatmap quantifies the model's classification performance (4 effective peptides and 2 invalid peptides, e.g.) Figure 5 (As shown). The model successfully identified all invalid peptides (GL, GIL, 100% true negative rate), but missed one valid peptide (VF was misclassified as invalid). This error stemmed from the boundary characteristics of VF; its predicted probability (0.610) was slightly higher than the threshold but lower than other valid peptides, and the experimentally measured minimum decrease in systolic blood pressure (-10.51 mmHg), although exceeding the threshold, was significantly lower than RW (-18.24 mmHg) and RF (-14.22 mmHg). Notably, this error did not lead to a false positive (invalid peptide misclassified as valid), ensuring the model's practicality in priority screening scenarios. High-probability peptides (>0.8) are reliable, but critical value peptides require in vitro validation.

[0046] 5. ROC curve analysis

[0047] The ROC curve (AUC=0.83) confirms that the model possesses excellent classification and discriminative abilities. Figure 6As shown, the steep rise in the upper left of the curve indicates that when the specificity is set to 80%, the model sensitivity reaches 90%, demonstrating its extremely high capture efficiency for true positive peptides (RW, KF, RF). The area under the curve is significantly higher than the random baseline (diagonal) and close to the excellent threshold of 0.85, which is consistent with the accuracy on the independent test set (83%). The strategy of choosing the optimal threshold of 0.6 has also been validated as reasonable: if the threshold is increased to 0.65, VF will be correctly classified but may introduce false positives; if it is decreased to 0.55, although VF can be included, GL (probability 0.517) may be misjudged. In this trade-off, maintaining 0.6 can maximize the Matthews correlation coefficient (MCC=0.86).

[0048] and Figure 1 Corresponding to the method described above, this embodiment of the invention also provides a system for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, for use in... Figure 1 The specific implementation of the method, a system for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning provided in this embodiment of the invention, can be applied to computer terminals or various mobile devices, specifically including: The data collection module is used to collect and merge simulated in vivo experimental data and peptide level characteristic data of different peptides to obtain a comprehensive dataset. The data splitting module is used to filter key features from the comprehensive dataset and divide it into training and test sets; The model building and training module is used to initialize three deep learning models: random forest, SVM, and logistic regression. The training set is used to train the three deep learning models independently. Based on the prediction results of the test set, the optimal prediction model for peptide antihypertensive activity is confirmed. The computational analysis and visualization module is used to extract and rank the importance data of key features through the optimal prediction model of peptide antihypertensive activity, and to visualize and analyze the antihypertensive activity characteristics of different peptides, the ranking of feature importance, and the performance of the model. The results output and storage module is used to integrate all key results from the analysis process and store them systematically.

[0049] In summary, this embodiment effectively overcomes the efficiency bottleneck of traditional active peptide screening by constructing a closed-loop research framework of "computational simulation → in vivo validation → mechanism analysis". Based on the random forest model (AUC=0.83), the ACE inhibitory activity prediction of six scallop peptides showed that RW, KF, and RF had high activity probabilities (>0.83), and their significant in vivo hypotensive effect (systolic blood pressure reduction ≥14 mmHg) verified the reliability of the model in screening highly efficient peptides. Feature importance analysis revealed that molecular docking energy (importance 0.250) is the core determining factor, and its strong negative value (e.g., RW peptide score of -71.97 kcal / mol) reflects the high binding strength between the peptide and the ACE active center (e.g., the π-π stacking and hydrogen bond network formed by the Trp indole ring and Ala356). Water solubility (importance 0.206) and net charge (importance 0.206) jointly regulate bioavailability. Positively charged peptides are more likely to penetrate the intestinal barrier, while hydrophobic peptides (such as VF) may enhance their affinity for the hydrophobic pocket of ACE, but their in vivo activity is weakened due to poor oral absorption.

[0050] The case of the critical peptide VF (predicted probability 0.610, experimental blood pressure reduction of -10.51 mmHg) highlights the current model's insufficient coverage of atypical mechanisms of action, suggesting that it may exert its effects indirectly through pathways such as the release of active fragments after enzymatic hydrolysis. Meanwhile, the model accurately identified the ineffective peptides GL and GIL (probability <0.57), but the weak blood pressure reduction effect of GL (-5.24 mmHg) also suggests that there are still potential weakly active fragments in the turbot peptide library worth exploring. In summary, this embodiment not only confirms the feasibility of scallops as a source of high-quality ACE inhibitory peptides, but also lays the foundation for future optimization of the predictive model (such as integrating enzymatic hydrolysis stability parameters) and in-depth exploration of multiple blood pressure reduction mechanisms.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0052] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, characterized in that, Includes the following steps: S1. Collect simulated in vivo experimental data and peptide level characteristic data of different peptides and merge them to obtain a comprehensive dataset; S2. Select key features from the comprehensive dataset and divide it into training and test sets; S3. Initialize three deep learning models: random forest, SVM, and logistic regression. Train each of the three deep learning models independently using the training set. Based on the prediction results of the test set, confirm the optimal prediction model for peptide antihypertensive activity. S4. Based on the optimal prediction model for peptide antihypertensive activity, extract the importance data of key features and sort them, and visualize and analyze the antihypertensive activity characteristics, feature importance ranking, and model performance of different peptides. S5. Integrate all key results from the analysis process and store them systematically.

2. The method for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning according to claim 1, characterized in that, In S1, the simulated in vivo experimental data includes information on systolic blood pressure changes, time information, and significance information for RW peptide, RF peptide, KF peptide, VF peptide, GL peptide, and GIL peptide; peptide level characteristic data includes the basic physicochemical characteristics of RW peptide, RF peptide, KF peptide, VF peptide, AF peptide, NF peptide, VW peptide, GF peptide, GL peptide, and GIL peptide.

3. The method for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning according to claim 1, characterized in that, The specific steps of S1 are as follows: using peptide sequences as associated keywords, the simulated in vivo experimental data and peptide level feature data are merged through data linking to form a comprehensive dataset containing multi-dimensional information.

4. The method for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning according to claim 1, characterized in that, The specific steps of S2 are as follows: Five key features were selected from the comprehensive dataset: docking energy, activity score, molecular weight, net charge, and water solubility. These features were then standardized using the StandardScaler tool to eliminate dimensional differences. Stratified sampling is used to divide the standardized feature data and corresponding target variables into training and test sets.

5. The method for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning according to claim 1, characterized in that, In S3, if the training set contains only one class, then only the random forest model is trained.

6. The method for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning according to claim 1, characterized in that, In S3, the optimal prediction model for peptide antihypertensive activity was confirmed based on the prediction results of the test set, specifically as follows: Input the test set into each trained deep learning model to obtain the prediction results; By combining the real labels of the test set, the classification accuracy and AUC value of each deep learning model are calculated. The model with the highest classification accuracy and the largest AUC value is selected as the final optimal prediction model for peptide antihypertensive activity.

7. The method for detecting ACE inhibitory peptide activity in scallop skirts based on deep learning according to claim 4, characterized in that, The specific steps of S4 are as follows: Using the optimal prediction model for peptide antihypertensive activity, the importance data of key features were extracted and ranked to clarify the degree of influence of key features on the prediction results of peptide antihypertensive activity. Line graphs are used to present the depressurization effects of different peptides over time series, bar graphs are used to show the probability distribution of peptide activity and the importance of features, scatter plots are used to show the relationship between molecular docking energy and activity score, heatmaps are used to present the confusion matrix of random forest, and ROC curves are used to reflect the model's discriminative ability. Based on computational analysis and visualization results, the antihypertensive activity characteristics of different peptides, the ranking of feature importance, and the model performance were analyzed.

8. A system for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning, for performing the method for detecting the activity of ACE inhibitory peptides in scallop skirts based on deep learning as described in any one of claims 1-7, characterized in that, include: The data collection module is used to collect and merge simulated in vivo experimental data and peptide level characteristic data of different peptides to obtain a comprehensive dataset. The data splitting module is used to filter key features from the comprehensive dataset and divide it into training and test sets; The model building and training module is used to initialize three deep learning models: random forest, SVM, and logistic regression. The training set is used to train the three deep learning models independently. Based on the prediction results of the test set, the optimal prediction model for peptide antihypertensive activity is confirmed. The computational analysis and visualization module is used to extract and rank the importance data of key features through the optimal prediction model of peptide antihypertensive activity, and to visualize and analyze the antihypertensive activity characteristics of different peptides, the ranking of feature importance, and the performance of the model. The results output and storage module is used to integrate all key results from the analysis process and store them systematically.

Images