Prediction method and system of relative abundance of proteins on protein crown based on machine learning

Abstract

The present invention relates to a prediction method and system for the relative abundance of proteins on a protein corona based on machine learning. The method comprises: training and evaluating each machine learning model to determine a primary predictor, and saving the trained hyperparameters and decision boundaries; performing secondary training based on the hyperparameters and decision boundaries and a protein regression data set to obtain a secondary regression prediction model; inputting the characteristic information of the protein to be predicted into the primary predictor to obtain a classification prediction result; when there is protein adsorbed on nanoparticles in the classification prediction result, the protein to be predicted is input into the secondary regression prediction model to obtain a protein relative abundance prediction result. The secondary regression task is performed on the basis of the primary classification task, which can reduce the training time; combining classification with regression can realize a comprehensive prediction of the relative abundance of proteins on the protein corona, and improve the prediction performance of the relative abundance of proteins.

Description

Technical Field

[0001] The present invention relates to the field of biological detection technology, and in particular to a method and system for predicting the relative abundance of proteins on a protein crown based on machine learning. Background Art

[0002] Nanoparticles, with their small size and multiple unique properties, have had a profound impact on the development of many technological fields, including medicine. These nanoparticles are so small that they can enter almost all parts of the human body. This has led to the development of a new medical approach, namely nanomedicine. Once nanoparticles enter biological fluids, they interact with a range of biopolymers, including proteins, which bind to the nanoparticles to form a "protein corona" associated with the nanoparticles. The formation of the protein corona can be regarded as an interaction between nanoparticles and proteins. This surface "biotransformation" of nanomaterials can change their overall pharmacological and toxicological properties and their potential therapeutic or diagnostic functions in an unpredictable way. A thorough understanding of the composition of the protein corona and its interaction with nanoparticles is essential for evaluating their biological effects and potential applications in biomedicine. The protein corona composition refers to the relative protein abundance (RPA) of the total amount of proteins in the corona and is an important parameter for describing the protein corona. The use of artificial intelligence to establish predictive models can quickly characterize the adsorption behavior of proteins and the formation of the protein corona. To date, many factors (such as the physicochemical properties of nanoparticles, incubation and separation conditions) have been shown to affect the biological response and composition of the protein corona, and machine learning algorithms are very suitable for extracting key information from these factors to accurately predict the composition of the protein corona. The application of machine learning can improve the breadth and depth of protein corona research and make the composition analysis of the protein corona more convenient. It can be seen that machine learning has great potential in the field of protein corona-related predictions.

[0003] Currently, most studies on the composition of the protein corona are based on liquid chromatography-mass spectrometry (LC-MS / MS) to obtain information on the composition of the protein corona, and a small number of them use the random forest algorithm in machine learning to perform classification or regression analysis to predict the relative abundance of proteins in the protein corona. However, the analysis technology based on liquid chromatography-mass spectrometry (LC-MS / MS) is expensive and time-consuming; the prediction based on random forest is relatively simple and does not dig out deeper important feature information. At present, the prediction of RPA using machine learning is relatively simple. Predicting RPA through simple regression analysis or classification prediction cannot give a comprehensive prediction result of RPA, and its prediction performance needs to be improved. Summary of the invention

[0004] Based on this, in order to solve the above technical problems, a method and system for predicting the relative abundance of proteins on the protein crown based on machine learning is provided, which can improve the prediction performance of protein relative abundance.

[0005] A method for predicting the relative abundance of proteins on a protein crown based on machine learning, the method comprising:

[0006] Acquire a protein classification data set, perform training and evaluation on various machine learning models according to the protein classification data set, and determine a target machine learning algorithm according to the evaluation results;

[0007] Using the target machine learning algorithm as a first-level predictor, and saving trained hyperparameters and decision boundaries based on the first-level predictor;

[0008] Perform secondary training based on the hyperparameters, decision boundaries and protein regression data set to obtain a secondary regression prediction model;

[0009] Determining a protein to be predicted, and inputting feature information of the protein to be predicted into the primary predictor to obtain a classification prediction result;

[0010] When there is protein adsorbed on the nanoparticles in the classification prediction result, the protein to be predicted is input into the secondary regression prediction model to obtain the protein relative abundance prediction result.

[0011] In one embodiment, the training and evaluation of each machine learning model according to the protein classification dataset includes:

[0012] Dividing the protein classification data set into a protein classification training set and a protein classification test set;

[0013] Based on the protein classification training set and the protein classification test set, the extreme random tree, random forest, gradient boosting decision tree, XGBoost, Lightgbm, and neural network are trained and evaluated respectively, and each evaluation result is obtained.

[0014] In one embodiment, determining a target machine learning algorithm according to the evaluation result includes:

[0015] According to each of the evaluation results, the extremely random tree is determined as the target machine learning algorithm from the extremely random tree, random forest, gradient boosting decision tree, XGBoost, Lightgbm, and neural network.

[0016] In one embodiment, the secondary training is performed based on the hyperparameters, decision boundaries and protein regression dataset to obtain a secondary regression prediction model, including:

[0017] Acquire a protein regression data set, and divide the protein regression data set into a protein regression training set and a protein regression test set;

[0018] Based on the protein regression training set and the protein regression test set, the hyperparameters and decision boundaries are transferred to the protein regression data set for secondary training to obtain a secondary regression prediction model.

[0019] In one embodiment, the step of determining the protein to be predicted comprises:

[0020] Acquire a number of protein monomers, and select a protein monomer from the protein monomers as the protein to be predicted;

[0021] The qualitative features and quantitative features of the protein to be predicted are extracted, and the qualitative features and quantitative features are used as feature information of the protein to be predicted.

[0022] In one embodiment, the method further comprises:

[0023] The characteristic information of the protein to be predicted is normalized to obtain processed characteristic information.

[0024] In one embodiment, the step of inputting the characteristic information of the protein to be predicted into the primary predictor to obtain a classification prediction result comprises:

[0025] Inputting the processed feature information into the primary predictor, and predicting whether the protein to be predicted is adsorbed on the nanoparticles by the primary predictor;

[0026] Adsorbing the protein to be predicted on the nanoparticles as a first classification prediction result;

[0027] The fact that the protein to be predicted is not adsorbed on the nanoparticles is taken as the second classification prediction result.

[0028] In one embodiment, the method further comprises:

[0029] When the classification prediction result is the second classification prediction result, there is no need to input the protein to be predicted into the secondary regression prediction model, thereby ending the prediction of the relative abundance of the protein on the protein crown.

[0030] A prediction system for relative abundance of proteins on a protein crown based on machine learning, the system comprising:

[0031] An algorithm evaluation and determination module is used to obtain a protein classification data set, perform training and evaluation on various machine learning models according to the protein classification data set, and determine a target machine learning algorithm according to the evaluation results;

[0032] A data saving module, used to use the target machine learning algorithm as a primary predictor and save the trained hyperparameters and decision boundaries based on the primary predictor;

[0033] A secondary training module, used for performing secondary training based on the hyperparameters and decision boundaries and the protein regression data set to obtain a secondary regression prediction model;

[0034] A classification prediction module, used to determine the protein to be predicted, and input the characteristic information of the protein to be predicted into the primary predictor to obtain a classification prediction result;

[0035] The protein relative abundance prediction module is used to input the protein to be predicted into the secondary regression prediction model to obtain the protein relative abundance prediction result when there is protein adsorbed on the nanoparticles in the classification prediction result.

[0036] The above-mentioned machine learning-based prediction method and system for the relative abundance of proteins on the protein corona selects the optimal first-level predictor by evaluating multiple machine learning algorithms, and then based on the migration idea, migrates the optimal hyperparameters and optimal decision boundaries obtained in the first-level classification training to the regression data set for secondary training to construct a complete second-level regression prediction model. The second-level regression task is performed on the basis of the first-level classification task, which can reduce a certain amount of training time; combining classification with regression can realize a comprehensive prediction of the relative abundance of proteins on the protein corona. On the basis of predicting whether adsorption occurs or not, the specific value of the relative abundance of proteins on the protein corona can be further predicted, thereby improving the prediction performance of protein relative abundance. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is an application environment diagram of a method for predicting relative abundance of proteins on a protein crown based on machine learning in one embodiment;

[0038] Figure 2 Schematic diagram of a method for predicting the relative abundance of proteins on a protein crown based on machine learning in one embodiment;

[0039] Figure 3 A schematic diagram of a framework for predicting the relative abundance of proteins on a protein crown based on machine learning in one embodiment;

[0040] Figure 4 The structure diagram of a prediction system for relative abundance of proteins on protein crowns based on machine learning in one embodiment;

[0041] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0043] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe classification prediction results, but these classification prediction results are not limited by these terms. These terms are only used to distinguish the first classification prediction result from another classification prediction result. For example, without departing from the scope of this application, the first classification prediction result may be referred to as the second classification prediction result, and similarly, the second classification prediction result may be referred to as the first classification prediction result. Both the first classification prediction result and the second classification prediction result are classification prediction results, but they are not the same classification prediction result.

[0044] The prediction method of protein relative abundance on the protein crown based on machine learning provided in the embodiments of the present application can be applied to Figure 1 In the application environment shown. Figure 1 As shown, the application environment includes a computer device 110. The computer device 110 can obtain a protein classification data set, train and evaluate each machine learning model according to the protein classification data set, and determine the target machine learning algorithm according to the evaluation result; the computer device 110 can use the target machine learning algorithm as a primary predictor, and save the trained hyperparameters and decision boundaries based on the primary predictor; the computer device 110 can perform secondary training based on the hyperparameters and decision boundaries and the protein regression data set to obtain a secondary regression prediction model; the computer device 110 can determine the protein to be predicted, and input the feature information of the protein to be predicted into the primary predictor to obtain a classification prediction result; when there is a protein adsorbed on the nanoparticle in the classification prediction result, the computer device 110 can input the protein to be predicted into the secondary regression prediction model to obtain a protein relative abundance prediction result. Among them, the computer device 110 can be, but is not limited to, various personal computers, laptops, smart phones, robots, unmanned aerial vehicles, tablet computers and other devices.

[0045] In one embodiment, Figure 2 As shown, a method for predicting the relative abundance of proteins on the protein crown based on machine learning is provided, comprising the following steps:

[0046] Step 202, obtain a protein classification data set, train and evaluate various machine learning models based on the protein classification data set, and determine a target machine learning algorithm based on the evaluation results.

[0047] The protein corona composition refers to the relative protein abundance (RPA) of the total amount of proteins in the corona, which is an important parameter for describing the protein corona. In this embodiment, the computer device can obtain a protein classification data set, and then use various machine learning models to make preliminary predictions on the protein classification data set to determine whether multiple proteins are adsorbed on the surface of the nanoparticles.

[0048] In one embodiment, a method for predicting the relative abundance of proteins on a protein crown based on machine learning is provided, which may also include a process for training and evaluating each machine learning model. The specific process includes: dividing the protein classification data set into a protein classification training set and a protein classification test set; based on the protein classification training set and the protein classification test set, respectively, training and evaluating the extreme random tree, random forest, gradient boosting decision tree, XGBoost, Lightgbm, and neural network, and obtaining each evaluation result.

[0049] Specifically, the computer device can train and evaluate each machine learning model in turn based on the protein classification training set and the protein classification test set, and record the training and evaluation results of each machine learning model respectively, so as to select the machine learning algorithm with the best performance, namely the target machine learning algorithm, as the first-level predictor.

[0050] In this embodiment, six machine learning models are used to preliminarily predict whether multiple proteins are adsorbed on the surface of nanoparticles. Specific machine learning models may include: Extremely Random Tree (ERT), Random Forest (RF), Gradient Boosting Decision Tree (GBDT), XGBoost (XGB), Lightgbm (LGBM) and Neural Network (NN). The machine learning model is used to preliminarily predict whether multiple proteins are adsorbed on the surface of nanoparticles. The purpose is to predict whether the RPA value is greater than zero, which is a binary classification prediction task.

[0051] In one embodiment, the computer device can determine the extremely random tree as the target machine learning algorithm from extremely random trees, random forests, gradient boosting decision trees, XGBoost, Lightgbm, and neural networks based on various evaluation results.

[0052] That is, these six machine learning algorithms were trained and evaluated in 178 independent protein data sets, and the machine learning algorithm with the best performance was selected: Extremely Randomized Trees (ERT), which was used as the first-level predictor for classification prediction.

[0053] Step 204, using the target machine learning algorithm as a first-level predictor, and saving the trained hyperparameters and decision boundaries based on the first-level predictor.

[0054] Specifically, the first-level predictor can perform the first-level classification task. According to the performance evaluation, the extremely random tree (ERT) is selected as the first-level predictor, where the first-level predictor can be calculated according to the formula To obtain the prediction results. Among them, represents the final classification result, y represents the classification label (for binary classification problems, it can be defined as y∈{0,1}), K represents the number of basic classifiers in RF, represents the indicator function, represents the kth basic classifier in ERT, x represents the input data, Θ k Represents a random attribute subset of the kth basic classifier. After training the first-level classification predictor, save the hyperparameters and decision boundaries trained on ERT in the first-level classification task. On this basis, secondary training can be performed on the regression dataset to obtain a secondary regression prediction model.

[0055] Step 206 , performing secondary training based on the hyperparameters, decision boundaries and protein regression dataset to obtain a secondary regression prediction model.

[0056] Specifically, based on the idea of ​​migration, a secondary predictor is constructed on the basis of the primary predictor ERT to predict the specific value of RPA. The decision tree constructed by the primary predictor on the classification data set and the decision boundary learned on the decision tree are used as the basis to construct a secondary regression prediction model. Based on the primary predictor, secondary training is performed on the regression data set to obtain a secondary predictor to predict the specific value of RPA.

[0057] In one embodiment, a method for predicting the relative abundance of proteins on a protein crown based on machine learning is provided, which may also include a secondary training process, the specific process including: obtaining a protein regression data set, and dividing the protein regression data set into a protein regression training set and a protein regression test set; based on the protein regression training set and the protein regression test set, migrating the hyperparameters and decision boundaries to the protein regression data set for secondary training to obtain a secondary regression prediction model.

[0058] Specifically, for the secondary regression prediction task, secondary regression training can be performed based on the results of the primary classification training, which can be done according to the formula To obtain the regression prediction results, represents the final regression result, K represents the number of basic regressors in RF, represents the kth base regressor in ERT, x represents the input data, Θ k represents a random subset of attributes for the k-th base regressor.

[0059] Among them, regression prediction is based on classification prediction, which is also carried out around 178 independent proteins. At the same time, in order to ensure the reliability of the regression prediction results, it is also evaluated based on the performance of the baseline model corresponding to the 178 proteins (such as the R2 evaluation indicator).

[0060] In one embodiment, the performance of the first-level classification prediction model can be evaluated using six indicators, namely, accuracy rate (ACC), precision, recall rate, F1 score, and Matthews correlation coefficient (MCC). The performance of the second-level regression prediction model is evaluated using the coefficient of determination (R2) and root mean square error (RMSE). The calculation formulas are respectively expressed as: accuracy rate Accuracy Recall F1 score Matthews correlation coefficient Coefficient of determination Root mean square error Among them, TP is true positive, TN is true negative, FP is false positive, FN is false negative, y is the true label value, To predict the results, is the label mean.

[0061] Step 208, determine the protein to be predicted, and input the characteristic information of the protein to be predicted into the first-level predictor to obtain the classification prediction result.

[0062] In one embodiment, a method for predicting the relative abundance of proteins on a protein crown based on machine learning is provided, which may also include a process for determining the protein to be predicted, the specific process including: obtaining a number of protein monomers, and selecting a protein monomer from the number of protein monomers as the protein to be predicted; extracting qualitative features and quantitative features from the protein to be predicted, and using the qualitative features and quantitative features as feature information of the protein to be predicted.

[0063] Specifically, for the comprehensive prediction of protein RPA on the protein crown, it is necessary to first determine the protein to be predicted. Through the 178 related protein monomers provided, first select a protein monomer as the protein to be predicted. The protein monomer corresponds to the ID number in the PDB database.

[0064] The characteristic information in the protein to be predicted can include 8 qualitative features and 13 quantitative features. The 8 qualitative features are: nanoparticles, nanoparticle surface modification, nanoparticle type, nanoparticle shape, surface modification type, dispersion medium, incubation protein source, incubation culture; the 13 quantitative features are: size measured by transmission electron microscopy, size measured by dynamic light scattering, dispersion medium pH, Zeta point, polydispersity index, incubation plasma concentration, incubation nanoparticle concentration, incubation time, incubation temperature, centrifugation speed, centrifugation time, centrifugation temperature, and number of centrifugation repetitions.

[0065] Next, a method for predicting the relative abundance of proteins on a protein crown based on machine learning is provided, which may also include a normalization process, and the specific process includes: normalizing the characteristic information of the protein to be predicted to obtain the processed characteristic information.

[0066] Specifically, the input data, i.e., the characteristic information of the protein to be predicted, can be compressed to between 0 and 1 through maximum and minimum normalization.

[0067] In one embodiment, a method for predicting the relative abundance of proteins on a protein crown based on machine learning is provided, which may also include a classification prediction process, the specific process including: inputting the processed feature information into a first-level predictor, predicting by the first-level predictor whether the protein to be predicted is adsorbed on the nanoparticles; taking the adsorption of the protein to be predicted on the nanoparticles as a first classification prediction result; taking the non-adsorption of the protein to be predicted on the nanoparticles as a second classification prediction result.

[0068] In one embodiment, when the classification prediction result is the second classification prediction result, there is no need to input the protein to be predicted into the secondary regression prediction model, and the prediction of the relative abundance of the protein on the protein crown is ended.

[0069] That is to say, the trained first-level classification prediction model, i.e., the first-level predictor, is used to predict whether the selected protein is adsorbed on the nanoparticles under the condition of the input relevant protein corona information. If the prediction result is adsorption, the next step of regression prediction is performed. If the prediction result is no adsorption, no regression prediction is performed.

[0070] Step 210: When there is protein adsorbed on the nanoparticles in the classification prediction result, the protein to be predicted is input into the secondary regression prediction model to obtain the protein relative abundance prediction result.

[0071] Specifically, if the prediction result in the classification prediction is that the protein is adsorbed on the nanoparticles, the trained secondary regression prediction model is used to predict the specific value of RPA of the selected protein on the protein corona to obtain the protein relative abundance prediction result, that is, the relative size of the adsorption amount.

[0072] In this embodiment, the comprehensive prediction of protein RPA on the protein corona mainly includes the following steps: selecting protein, inputting the corresponding information of the protein corona as features, normalizing the data, first performing classification prediction through a primary predictor, and when the classification prediction result RPA>0, performing regression prediction to predict the specific value of RPA.

[0073] By combining classification and regression to construct a secondary prediction model, a comprehensive prediction of the RPA on the protein corona of the selected protein under the condition of the corresponding input protein corona information is achieved. That is, first, it is predicted whether the protein is adsorbed on the corresponding nanoparticles. If adsorbed, the relative size of the adsorption amount, that is, the specific value of RPA, is predicted. The secondary prediction model can provide a reference for the preliminary design of the protein corona.

[0074] Since it is proposed to combine classification with regression to construct a secondary prediction model to comprehensively predict the RPA of multiple proteins on the protein corona, while predicting whether the RPA is greater than zero, the specific value of the RPA when the RPA is greater than zero is given. Specifically, the optimal hyperparameters and decision boundaries learned from the primary classification are used as the basis for secondary training in the secondary regression data to obtain a secondary prediction model, which greatly reduces the search range of hyperparameters and decision boundaries while ensuring model performance. Most of the existing technologies only use random forests for simple classification or regression prediction, and do not combine these two analysis methods to predict the RPA of multiple proteins on the protein corona. Combining classification with regression to construct a secondary prediction model to comprehensively predict the RPA of multiple proteins on the protein corona can provide a reference for the design of the protein corona, and greatly facilitate related research on the analysis of the components of the protein corona.

[0075] In one embodiment, a method for predicting the relative abundance of proteins on a protein crown based on machine learning is provided and applied to Figure 3 The secondary prediction model framework shown in the figure is shown in the figure. Figure 3 As shown, the collected classification data set is divided into 90% training set and 10% test set; 178 independent protein data sets are input into the first-level predictor ERT; the first-level predictor ERT is used to predict whether RPA is greater than 0. If greater than 0, the hyperparameters and decision boundaries in the first-level predictor training process are migrated, and secondary training is performed to obtain a second-level regression prediction model; the regression data set is divided into 80% training set and 20% test set; 178 independent protein data sets are input into the second-level regression prediction model; the specific value of RPA is predicted by the second-level regression prediction model to complete the prediction of the relative abundance of proteins on the protein crown.

[0076] It should be understood that, although the various steps in the above-mentioned flow chart are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the above-mentioned flow chart may include a plurality of sub-steps or a plurality of stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0077] In one embodiment, Figure 4 As shown, a prediction system for the relative abundance of proteins on the protein crown based on machine learning is provided, comprising: an algorithm evaluation and determination module 410, a data storage module 420, a secondary training module 430, a classification prediction module 440 and a protein relative abundance prediction module 450, wherein:

[0078] An algorithm evaluation and determination module 410 is used to obtain a protein classification data set, perform training and evaluation on each machine learning model according to the protein classification data set, and determine a target machine learning algorithm according to the evaluation results;

[0079] A data saving module 420, configured to use the target machine learning algorithm as a primary predictor and save the trained hyperparameters and decision boundaries based on the primary predictor;

[0080] A secondary training module 430 is used to perform secondary training based on the hyperparameters, the decision boundary and the protein regression data set to obtain a secondary regression prediction model;

[0081] The classification prediction module 440 is used to determine the protein to be predicted and input the characteristic information of the protein to be predicted into the primary predictor to obtain the classification prediction result;

[0082] The protein relative abundance prediction module 450 is used to input the protein to be predicted into the secondary regression prediction model to obtain the protein relative abundance prediction result when there is protein adsorbed on the nanoparticles in the classification prediction result.

[0083] In one embodiment, the algorithm evaluation and determination module 410 is also used to divide the protein classification data set into a protein classification training set and a protein classification test set; based on the protein classification training set and the protein classification test set, the extremely random tree, random forest, gradient boosting decision tree, XGBoost, Lightgbm, and neural network are trained and evaluated respectively, and each evaluation result is obtained.

[0084] In one embodiment, the algorithm evaluation determination module 410 is also used to determine the extremely random tree as the target machine learning algorithm from extremely random trees, random forests, gradient boosting decision trees, XGBoost, Lightgbm, and neural networks based on various evaluation results.

[0085] In one embodiment, the secondary training module 430 is also used to obtain a protein regression data set, and divide the protein regression data set into a protein regression training set and a protein regression test set; based on the protein regression training set and the protein regression test set, the hyperparameters and decision boundaries are migrated to the protein regression data set for secondary training to obtain a secondary regression prediction model.

[0086] In one embodiment, the classification prediction module 440 is also used to obtain several protein monomers, and select a protein monomer from the several protein monomers as the protein to be predicted; extract qualitative features and quantitative features from the protein to be predicted, and use the qualitative features and quantitative features as feature information of the protein to be predicted.

[0087] In one embodiment, the classification prediction module 440 is further used to perform normalization processing on the feature information of the protein to be predicted to obtain processed feature information.

[0088] In one embodiment, the classification prediction module 440 is also used to input the processed feature information into a first-level predictor, and predict whether the protein to be predicted is adsorbed on the nanoparticles through the first-level predictor; the adsorption of the protein to be predicted on the nanoparticles is used as the first classification prediction result; and the non-adsorption of the protein to be predicted on the nanoparticles is used as the second classification prediction result.

[0089] In one embodiment, the protein relative abundance prediction module 450 is used to terminate the prediction of the relative abundance of the protein on the protein crown without inputting the protein to be predicted into the secondary regression prediction model when the classification prediction result is the second classification prediction result.

[0090] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 5As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a prediction method for the relative abundance of proteins on the protein crown based on machine learning is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a button, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0091] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0092] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of a method for predicting the relative abundance of proteins on a protein crown based on machine learning are implemented.

[0093] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of a method for predicting the relative abundance of proteins on a protein crown based on machine learning are implemented.

[0094] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0095] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0096] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for predicting the relative abundance of proteins on a protein crown based on machine learning, characterized in that: The method comprises: Acquire a protein classification data set, perform training and evaluation on various machine learning models according to the protein classification data set, and determine a target machine learning algorithm according to the evaluation results; Using the target machine learning algorithm as a first-level predictor, and saving trained hyperparameters and decision boundaries based on the first-level predictor; Perform secondary training based on the hyperparameters, decision boundaries and protein regression data set to obtain a secondary regression prediction model; Determining the protein to be predicted, including: extracting qualitative features and quantitative features from the protein to be predicted, and using the qualitative features and quantitative features as feature information of the protein to be predicted; inputting the feature information of the protein to be predicted into the primary predictor to obtain a classification prediction result, including: normalizing the feature information of the protein to be predicted to obtain processed feature information; inputting the processed feature information into the primary predictor, and predicting whether the protein to be predicted is adsorbed on nanoparticles by the primary predictor; using the adsorption of the protein to be predicted on nanoparticles as a first classification prediction result; using the non-adsorption of the protein to be predicted on nanoparticles as a second classification prediction result; When there is protein adsorbed on the nanoparticles in the classification prediction result, the protein to be predicted is input into the secondary regression prediction model to obtain the protein relative abundance prediction result.

2. The method for predicting the relative abundance of proteins on the protein crown based on machine learning according to claim 1, characterized in that: The training and evaluation of each machine learning model according to the protein classification data set includes: Dividing the protein classification data set into a protein classification training set and a protein classification test set; Based on the protein classification training set and the protein classification test set, the extreme random tree, random forest, gradient boosting decision tree, XGBoost, Lightgbm, and neural network are trained and evaluated respectively, and each evaluation result is obtained.

3. The method for predicting the relative abundance of proteins on the protein crown based on machine learning according to claim 2, characterized in that: Determine the target machine learning algorithm based on the evaluation results, including: According to each of the evaluation results, the extremely random tree is determined as the target machine learning algorithm from the extremely random tree, random forest, gradient boosting decision tree, XGBoost, Lightgbm, and neural network.

4. The method for predicting the relative abundance of proteins on the protein crown based on machine learning according to claim 1, characterized in that: The secondary training is performed based on the hyperparameters, the decision boundary and the protein regression data set to obtain a secondary regression prediction model, including: Acquire a protein regression data set, and divide the protein regression data set into a protein regression training set and a protein regression test set; Based on the protein regression training set and the protein regression test set, the hyperparameters and decision boundaries are transferred to the protein regression data set for secondary training to obtain a secondary regression prediction model.

5. The method for predicting the relative abundance of proteins on the protein crown based on machine learning according to claim 1, characterized in that: The step of determining the protein to be predicted comprises: A plurality of protein monomers are obtained, and a protein monomer is selected from the plurality of protein monomers as the protein to be predicted.

6. The method for predicting the relative abundance of proteins on the protein crown based on machine learning according to claim 1, characterized in that: The method further comprises: When the classification prediction result is the second classification prediction result, there is no need to input the protein to be predicted into the secondary regression prediction model, thereby ending the prediction of the relative abundance of the protein on the protein crown.

7. A prediction system for the relative abundance of proteins on the protein crown based on machine learning, characterized in that: The system comprises: An algorithm evaluation and determination module is used to obtain a protein classification data set, perform training and evaluation on various machine learning models according to the protein classification data set, and determine a target machine learning algorithm according to the evaluation results; A data saving module, used to use the target machine learning algorithm as a primary predictor and save the trained hyperparameters and decision boundaries based on the primary predictor; A secondary training module, used for performing secondary training based on the hyperparameters and decision boundaries and the protein regression data set to obtain a secondary regression prediction model; A classification prediction module is used to determine the protein to be predicted, including: extracting qualitative features and quantitative features from the protein to be predicted, and using the qualitative features and quantitative features as feature information of the protein to be predicted; inputting the feature information of the protein to be predicted into the primary predictor to obtain a classification prediction result, including: normalizing the feature information of the protein to be predicted to obtain processed feature information; inputting the processed feature information into the primary predictor, and predicting whether the protein to be predicted is adsorbed on nanoparticles by the primary predictor; using the adsorption of the protein to be predicted on nanoparticles as a first classification prediction result; using the non-adsorption of the protein to be predicted on nanoparticles as a second classification prediction result; The protein relative abundance prediction module is used to input the protein to be predicted into the secondary regression prediction model to obtain the protein relative abundance prediction result when there is protein adsorbed on the nanoparticles in the classification prediction result.

8. The prediction system for relative abundance of proteins on the protein crown based on machine learning according to claim 7, characterized in that: The algorithm evaluation and determination module is also used to: divide the protein classification data set into a protein classification training set and a protein classification test set; based on the protein classification training set and the protein classification test set, respectively, perform training evaluation on extreme random trees, random forests, gradient boosting decision trees, XGBoost, Lightgbm, and neural networks, and obtain various evaluation results.

Images