Abstract

Background: We aimed to establish a novel diagnostic model for kidney diseases by combining artificial intelligence with complete mass spectrum information from urinary proteomics.
Methods: We enrolled 134 patients (IgA nephropathy, membranous nephropathy, and diabetic kidney disease) and 68 healthy participants as controls, with a total of 610,102 mass spectra from their urinary proteomic profiles. The training data set (80%) was used to create a diagnostic model using XGBoost, random forest (RF), a support vector machine (SVM), and artificial neural networks (ANNs). The diagnostic accuracy was evaluated using a confusion matrix with a test dataset (20%). We also constructed receiver operating-characteristic, Lorenz, and gain curves to evaluate the diagnostic model.
Results: Compared with the RF, SVM, and ANNs, the modified XGBoost model, called Kidney Disease Classifier (KDClassifier), showed the best performance. The accuracy of the XGBoost diagnostic model was 96.03%. The area under the curve of the extreme gradient boosting (XGBoost) model was 0.952 (95% confidence interval, 0.9307–0.9733). The Kolmogorov-Smirnov (KS) value of the Lorenz curve was 0.8514. The Lorenz and gain curves showed the strong robustness of the developed model.
Conclusion:The KDClassifier achieved high accuracy and robustness and thus provides a potential tool for the classification of kidney diseases.

Keywords

Kidney disease classification, urinary proteomics, machine learning algorithm, diagnosis, artificial intelligence

Study population

All patients and health controls were recruited from Sichuan Provincial People’s Hospital from September 2019 to May 2020. Written consent was obtained from the participants prior to the physical examination or biopsy procedure. All the participants were recruited on a voluntary basis. Our study samples were considered representative of the Chinese population. The study protocol was approved by the medical ethics committee of Sichuan Provincial People’s Hospital and West China Hospital.
In this study, 202 urine samples from patients with IgAN (n = 50), MN (n = 50), and DKD (n = 34) and HCs (n = 68) were collected in tubes in accordance with the standard hospital operating procedures. All the patients with kidney diseases were examined through a renal biopsy, and those with secondary types of IgAN or MN were excluded. The urine samples were collected within 1 week before the renal biopsy. Briefly, the midstream urine from the second morning void was collected in appropriate containers and centrifuged at 1,000×g for 20 min. The precipitate was discarded, and 500 μL of the supernatant (including the soluble proteins, exosomes, and other membrane elements) was collected in a 1.5-mL tube and stored at −80°C until use.

Urinary protein digestion

Human urinary protein digestion was performed using a filter-aided sample preparation. Each 100-μL urine supernatant was loaded onto a 30-kDa ultrafiltration device. After centrifugation at 13,000×g for 15 min, a 100-μL uranyl acetate (UA) solution with 20 mM dithiothreitol was added and reacted for 4 h at 37°C. An alkylation reaction was then achieved by adding a 100-μL UA solution with 50 mM iodoacetamide and incubated in the dark for 1 h at room temperature. The buffer was replaced with 50 mM ammonium bicarbonate. Finally, 10-μg trypsin was added to each filter tube, and the reaction was maintained for 16h at 37°C. The digestion was collected, and the measured concentration was 480 nm. The urinary protein digestions were freeze-dried and stored at −80°C.

Mass spectrometric analysis

A urinary peptide analysis was performed using an Orbitrap Fusion Lumos mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA). Briefly, the peptides were dissolved in 0.1% fulvic acid (FA) and separated into a column with a 75-μm inner diameter and 15-cm length over a 78-min gradient (buffer A, 0.1% FA in water; buffer B, 0.1% FA in 80% ACN) at a flow rate of 300 nL/min. MS1 was analyzed with a scan mass range of 300–1,400 at a resolution of 120,000 at 200 m/z. The radiofrequency lens, automatic gain control (AGC), maximum injection time (MIT), and exclusion duration were set at 30%, 5.0 e5, 50 ms, and 18 s, respectively. MS2 was analyzed in data-dependent mode for the 20 most intense ions. The isolation window (m/z), collision energy, AGC, and MIT were set at 1.6, 35%, 5.0 e3, and 35 ms, respectively.

Spectral establishment

The raw mass spectrometric data were converted into mascot generic format (MGF) files, with each file containing thousands of pieces of mass spectrum information. The x-coordinate was the mass-to-charge ratio (m/z), and the y-coordinate was the relative peak intensity. The mass spectra from each file were used to profile the urinary proteome of each patient. The mass spectrometric data from the urinary proteomics were deposited in the ProteomeXchange Consortium through the PRIDE partner repository with the dataset identifier PXD018996.

Data preprocessing

The mass spectra contained four classes of CKD urinary proteomic information (IgAN, MN, DKD, and HC). The MGF files were processed using an illumination normalization method. The data of all the original urinary proteomic mass spectra were transformed into double-column arrays of indefinite lengths (with the abscissa and ordinate values of the peaks in the spectrum). Owing to the unequal lengths of the arrays, we set an array with a length of 50 rows (the maximum value). We then merged each double-column array into a single feature data line with a length of 100. Data of insufficient length were considered missing values. Finally, a data set with four different data labels (IgAN, MN, DKD, and HC) was created and imported into the XGBoost model.

XGBoost model

XGBoost, developed by Chen et al., is a machine learning technique that assembles weak prediction models [18]. It generates a series of decision trees in a gradientboosting manner, which means that it generates the next decision tree based on the current tree to better predict the outcome. After training, a classification prediction system composed of a series of decision trees is achieved. This is an extendible and cutting-edge application of a gradient boosting machine and has been proven to push the limits of computing power for boosted tree algorithms. Gradient boosting is an algorithm in which new models are created for predicting the residuals of the prior models and then adding them together to make the final prediction. A gradient descent algorithm is used to minimize the loss during the addition of the new models. XGBoost with a multi-core central processing unit reduces the lookup times of the individual trees created. With this algorithm, the K additive function ensemble model (K trees) is defined as follows:

where indicates the sample, F is the space containing all trees, and refers to the function in the functional space F. To train the ensemble model, the objective is minimized as follows:

Here, is a loss function that measures the difference between the target and the prediction . In addition, penalizes the complexity and is defined as

The number of leaves in a tree is defined as T; in addition, indicates the minimum loss reduction, is the weight of the regularization, and represents the corresponding score of the leaves.
The XGBoost algorithm can handle missing data automatically by adding a default direction for the missing values in each tree node. The default direction is learned during the training procedure. When a value is missing in the validation data, the instance is classified into the default direction. This means inputting only a reduced number of important variables while leaving the others as null values during the application stage.
We maintained 20% of the data as the validation set and used the remaining 80% to train our diagnosis XGBoost model. The hyperparameters used in our analysis were as follows: learning rate = 0.01, minimum loss reduction = 10, maximum tree depth = 10, number of subsamples = 0.8, number of trees = 300, and number of rounds = 100. A simultaneous grid search over gamma, reg lambda, and the subsample was used to reexamine the model and check for differences between the optimum values.

Other machine learning models

Random forest (RF) is a type of classifier that uses randomly generated samples from existing situations and consists of multiples trees [19]. To classify a sample, each tree in the forest is given an input vector, and a result is produced for each tree. The tree with the most votes is chosen as the result. RF divides each node into branches by using the best randomly selected variables on each node.
A SVM is a controlled classification algorithm based on the statistical learning theory [20]. The working principle of a SVM is based on the prediction of the most appropriate decision function that separates the two classes; in other words, on the basis of the definition of a hyperplane, it can distinguish two classes from each other in the most appropriate manner possible. Similar to a classification, kernel functions are used to process nonlinear states during the regression. In cases in which the data cannot be separated linearly, nonlinear classifiers can be used instead of linear classifiers. A SVM transforms into a high dimensional feature space, which can be easily classified linearly from the original input space by means of a nonlinear mapping function. Thus, instead of finding values by repeatedly multiplying them using kernel functions, the value is directly substituted in the kernel function, and its counterpart is found in the feature space. Thus, this does not require dealing with a space with a very high dimensional quality. A SVM has four widely used kernel functions, namely linear, polynomial, sigmoid, and radial basis functions.
Artificial neural networks (ANNs) compose an information processing system inspired by biological neural networks and include some performance characteristics similar to those of biological neural networks [21]. The simplest artificial neuron consists of five main components as follows: inputs, weights, transfer function, activation function, and output. In an ANN, neurons are organized in layers. The layer between the input and output layers is called the hidden layer. The network is regulated by minimizing the error function. The connection weights are recalculated and updated to minimize the error. Thus, it is aimed at bringing output values that are closest to the ground truth values of the network.

Performance evaluation and statistical analysis

We divided all mass spectrum data from the CKD urinary proteomics into a training data set (80%) and a validation data set (20%). The training data set was directly used to train the framework and create a diagnostic model using XGBoost, RF, a SVM, and an ANN. The validation data set was used to calculate the diagnostic accuracy. We compared the accuracy of the four machine learning models and constructed a confusion matrix to calculate the sensitivity, specificity, positive predictive value, and negative predictive value of the XGBoost diagnostic model. We also constructed ROC curves for the CKD diagnostic model. We calculated the area under the curve (AUC) of the ROC curves to evaluate the prediction capabilities of the diagnostic model. Lorenz and gain curves were then constructed to evaluate the goodness of fit of the XGBoost diagnostic model.
The Lorenz and gain curves were established as graphical representations of the econometric distribution. These have been proven to be valuable analytic tools in other fields as well, including the evaluation of classifier models. Kendall and Stuart introduced a Lorenz curve arranged in ascending order according to the probability returned by the classification model. The divided points from dividing 0-1 equally into N parts are the threshold (abscissa), and the true positivity rate (TPR) and false positivity rate (FPR) are calculated. By taking the TPR and FPR as ordinates, two curves, both Lorenz curves (or KS curves), are drawn. The cutoff point (KS value) is the position where the distance between the TPR and FPR curves is the largest. A KS value of >0.2 is considered to indicate good prediction accuracy. The gain plot is an index used to describe the overall accuracy of the classifier models. With an increase in depth, the gain rate of the classifier model is compared with the natural random classification model. The steeper the curve and the larger the slope, the better the TPR obtained by the model.
Continuous variables are expressed as the mean ± standard deviation and compared using a T test. Categorical variables are expressed as percentages, and a chi-square test or Fisher exact test was used to compare the differences in the variables. The SPSS version 22.0 software (IBM Corp) was used for the comparative analysis of the basic characteristics. The machine learning models were developed using Python 3.4.3 (using the XGBoost, DF, SVM, and ANN libraries). In the evaluation and analysis method for determining the performance of the XGBoost model (KDClassifier), R version 3.5.2 was applied (using the pROC, dplyr, caret, lattice, and ggplot2 packages). The 95% confidence intervals (CIs) were then calculated. All P values were two-tailed, and a P value < 0.05 was considered statistically significant.

CKD represents a major public health issue in terms of its substantial financial burden and consumption of healthcare resources [1]. In addition, it is a risk factor of hypertension and cardiovascular diseases, which together constitute a substantial cause of death in most societies [22]. Accurate identification and early screening of CKD in the population have long been important topics. The development of a noninvasive and accurate early diagnostic method is needed. The diagnostic ability of a single biomarker is slightly weak, and renal biopsy is invasive, with a risk of major bleeding. With the development of mass spectrometry, urinary proteomes can now be both quantitatively and qualitatively detected [17]. Our study was focused on artificial intelligence-assisted noninvasive diagnostic methods for different types of kidney disease based on mass spectra information from urinary proteomics.
Previous studies showed that measurement of the level of a single protein marker in the clinical diagnosis of CKD does not take advantage of the overall value and macro efficiency of proteomics [17]. In addition, the feasibility of using a single protein marker in the clinical diagnosis of CKD requires further research and validation. The use of several or even dozens of protein panels can improve the diagnostic accuracy. The existing mass spectrometry applied in proteomics is used in the identification of differential proteins and selection of individual proteins for further differential studies. In fact, the overall data from a mass spectrometric analysis is not applied. Moreover, the efficacy of its clinical application requires further evaluation. By using big data, machine learning can be applied by integrating all information from a mass spectrometric analysis. We analyzed all data to take full advantage of the overall efficiency of proteomic mass spectrometry. Therefore, for CKD classification, considering the comprehensiveness of a mass spectrum analysis, as the feature data of our AI algorithm, we applied a first-order mass spectrum analysis of the proteomics without further processing. Artificial intelligence algorithms such as ANNs, SVMs, DTs, and XGBoost combined with medical or biological experience have obtained remarkable results [23, 24]. Through the training of big data sets, a machine learning model can predict classifications. Machine learning outperforms the conventional statistical methods with its ability to better identify variables, achieve a better predictive performance and modeling of complex relationships, and learn from multiple modules of data, and its robustness against data noise. It has therefore been applied in the diagnosis of certain diseases such as lung cancer [25], cardiovascular disease [26], and chronic kidney disease [27]. Machine and deep learning algorithms can not only impute missing data in the training sets but also identify existing characteristics that are otherwise unrecognizable. Most existing diagnostic machine models for CKD are based on records and detection indicators that are currently used in clinical practice [28]. However, the training data types of these models vary, and the accuracy of artificial collection is relatively low, with poor clinical application. To date, no studies have been conducted on machine learning models for diagnosis based on the full spectra of CKD urinary proteomics. In addition, of the many existing machine learning models, XGBoost achieved an outstanding classification performance without a high computation time and is a practical approach. XGBoost is a type of tree-structured model, the basic idea of which is to design an ensemble approach for several rule-based binary trees. XGBoost is derived from the most famous tree ensemble method, called gradient boosting decision tree. XGBoost has gained popularity by winning numerous machine learning competitions since its initial development [18]. Advances in big data and artificial intelligence have enabled clinicians to process information more efficiently and make diagnosis and treatment decisions more accurately [29]. It is unquestionable that big data and artificial intelligence are transforming medicine from various perspectives, including precision medicine and clinical intelligence. On the basis of the big data applied in urinary proteomic mass spectra, the strategy of using artificial intelligence and machine learning algorithms has been used to provide a new direction for the classification of kidney diseases. To the best of our knowledge, our study is the first to combine artificial intelligence and urinary proteomic mass spectra information in the diagnosis and classification of kidney diseases.
Compared with RF, SVM, and ANNs, the XGBoost model with mass spectra information for urinary proteomics showed a perfect performance in the diagnosis of kidney diseases. This is consistent with the classification ability of XGBoost models when applied for other clinical diseases. Therefore, compared with other machine learning algorithms, the advantages of the XGBoost algorithm are as follows [30]: First, XGBoost adds a regularization term to the objective function, which reduces the variance of the model, simplifying the model while preventing an overfitting. Second, XGBoost used not only the first derivative but also the second derivative to make the loss more accurate. Third, when the training data are sparse, the default direction of the branch can be specified for a missing or specified value, which can significantly improve the efficiency of XGBoost. Fourth, XGBoost supports column sampling and parallel optimization, thereby reducing the number of computations and improving efficiency. The peak value of the urine proteome mass spectrometry data is presented in the form of a set of numbers in abscissa and ordinate coordinates, which is used in the construction of the XGBoost model to maximize such advantages. In our study, the overall accuracy of the diagnostic XGBoost model for the four disease groups was 96.03%, which is basically consistent with the accuracy of renal biopsy. Therefore, our study highlights the advantages of using a noninvasive diagnostic method as an artificial intelligence model in proteomics. In addition, we conducted a detailed assessment of the modelling accuracy for each type of kidney disease and the HCs. The specificity of the diagnostic model for the four disease types was >95% (97.07%–99.91%); thus, its misdiagnosis rate is extremely low, and its ability to distinguish each type of disease shows excellent stability. Although the sensitivity for the four disease types was approximately 90% (89.81%–97.63%), the sensitivity for the three types of kidney disease, excluding the HC group, was lower than the specificity. Therefore, the missed diagnosis rate of this model is higher than its misdiagnosis rate, which indicates that this model may be more suitable for accurate disease diagnosis than for disease screening. The next steps of this research will focus on improving the prediction sensitivity of the model. For all four disease types, the sensitivity of the model regarding DKD diagnosis was the worst (89.81%), probably because of the smaller number of patients with DKD included, smaller mean number of spectra, or significant differences with the other disease groups. The low mean number of urinary proteomic mass spectra is a response to the state of the real disease, which cannot be avoided. We hope to include more samples in our future study to reduce the problems caused by data imbalance. In addition, through analyses of the ROC curve, gain plot, and Lorenz curve, this study showed that the model achieved strong robustness and high accuracy.
At present, a few existing XGBoost models for kidney disease diagnosis have been constructed using data on the clinical characteristics and individual laboratory test indicators. Ogunleye et al. [7] enrolled 250 CKD cases and 150 HCs to train and validate the XGBoost model with 22 clinical features. The accuracy, sensitivity, and specificity of the XGBoost model were all 100%. Xiao et al. [27] also constructed a XGBoost model for the prediction of CKD progression in 551 patients with proteinuria. A total of 13 blood-derived tests and 5 demographic features were used as variables to train the model. The accuracy of this progression model was 0.87. By applying 36 characteristics of 2,047 Chinese patients from 18 renal centers, Chen et al. [31] used a XGBoost model for the prediction of end-stage CKD. The C statistical value of this XGBoost model was 0.84. As all of these reports were constructed using clinical information and outcome indicators were inconsistent, poor comparability with our diagnostic XGBoost model was achieved. However, the accuracy of our model is high.
The KDClassifier classified the characteristic differences of the pathological types of CKD at the level of the integrated information of mass spectrometry proteomics in urine. No specific proteins or laboratory indicators of clinical concern have been identified, such as GFR, urine protein, or creatinine. This is significantly different from our normal assumption. The information captured by a machine learning model is more abundant than that obtained using comparative analysis of differential proteins. To explain the specific content of the information captured by machine learning with human logic requires further research and discussion.
Overall, the KDClassifier, an XGBoost diagnostic model, established in this study showed its feasibility and superiority for clinical application. However, in terms of economics, the current cost of mass spectrometry analysis of proteomics is relatively high, and realization of its clinical application will still take a long time. With further innovations in science and technology, however, we expect the cost of mass spectrometric analysis to inevitably decline. The KDClassifier is not only suitable for the classification of the three types of kidney disease considered but also has the potential to be extended to all types of kidney disease. The diagnostic advantages of this model will be fully demonstrated.
Our study also has certain limitations. First, the cohort used was not from a prospective trial, and selective bias was inevitable. Second, only three common types of kidney disease were included. Whether this learning machine diagnostic method is suitable for other types of kidney disease needs further research and validation using a larger sample size. Third, owing to the relatively small sample size, we did not include more clinical parameters for an artificial intelligence-assisted analysis. Including more clinical data will further improve the diagnostic power of our model. Fourth, this study only compared four mainstream machine learning methods with certain limitations. Fifth, only the mass spectra of urinary proteomic information were used, and the clinical information of the patients was omitted. If both types of information are combined, patients can be better diagnosed. We expect to develop more suitable artificial intelligence algorithms for a noninvasive and accurate diagnosis of kidney diseases.

In conclusion, the KDClassifier, a machine learning classification model that applies information on mass spectra from urinary proteomics, showed high accuracy in the diagnosis of different types of CKD. This study provides new insights into the application of artificial intelligence in the accurate and noninvasive diagnosis of kidney diseases. In addition, the KDClassifier provides a potential tool for the classification of all types of kidney disease.

Funding

This work was funded by grants from the National Natural Science Foundation of China (grant no. 31901038), the China Postdoctoral Science Foundation (2019M653438), the Post-Doctoral Research Foundation of West China Hospital of Sichuan University (2018HXBH062/2018HXBH074).

Conflicts of interest

The authors declare no conflict of interest.

Availability of data and material

The mass spectra data on the urinary proteomics were deposited into the ProteomeXchange Consortium through the PRIDE partner repository using the dataset identifier PXD018996.

Code availability

The codes that support the findings of this study are available on request from the corresponding author. The codes are not publicly available due to privacy or ethical restrictions.

Statement

We confirm that all methods were carried out in accordance with relevant guidelines and regulations in the manuscript.

Authors’ contributions

G.L., H.Y., J.C., J.Z. and F.L. directed and designed research;
Y.Z. Y.M.and W.Z. directed and performed analyses of mass spectrometry data;
W. Z., X.L. and Y.M. adapted algorithms and software for data analysis;
C.W. and L.Z. coordinated acquisition, distribution and quality evaluation of samples;
W.Z. and Y.Z. wrote the manuscript.

1. Coresh J. Update on the Burden of CKD. Journal of the American Society of Nephrology, 2017, 28(4): 1020- 1022.

2. Xie Y, Bowe B, Mokdad AH, et al. Analysis of the Global Burden of Disease study highlights the global, regional, and national trends of chronic kidney disease epidemiology from 1990 to 2016. Kidney International, 2018, 94(3): 567-581.

3. Glassock RJ, Warnock DG, Delanaye P. The global burden of chronic kidney disease: estimates, variability and pitfalls. Nature Reviews Nephrology, 2017, 13(2): 104-114.

4. Okparavero A, Foster MC, Tighiouart H, et al. Prevalence and complications of chronic kidney disease in a representative elderly population in Iceland. Nephrology Dialysis Transplantation, 2016, 31(3): 439-447.

5. Marcussen N, Olsen S, Larsen S, et al. Reproducibility of the WHO classification of glomerulonephritis. Clinical Nephrology, 1995, 44(4): 220-224.

6. Sethi S, Haas M, Markowitz GS, et al. Mayo Clinic/Renal Pathology Society Consensus Report on Pathologic Classification, Diagnosis, and Reporting of GN. Journal of the American Society of Nephrology, 2016, 27(5): 1278- 1287.

7. Fogo AB. Morphology expands understanding of lesions. Kidney International, 2020, 97(4): 627-360.

8. Lemley KV. Machine Learning Comes to Nephrology. Journal of the American Society of Nephrology, 2019, 30(10): 1780-1781.

9. Hermsen M, de Bel T, den Boer M, et al. Deep Learning- Based Histopathologic Assessment of Kidney Tissue. Journal of the American Society of Nephrology, 2019, 30(10): 1968-1979.

10. Sekulic M, Crary GS. Kidney Biopsy Yield: An Examination of Influencing Factors. American Journal Of Surgical Pathology, 2017, 41(7): 961-972.

11. Ito S. Aneurysmal dilatation associated with arteriovenous fistula in a transplanted kidney after renal biopsies. Pediatric Transplantation, 2014, 18(7): E216-E219.

12. Lin RC. Lipidomics: new insight into kidney disease. Advances in Clinical Chemistry, 2015, 68: 153-175.

13. Zhao YY. Metabolomics in chronic kidney disease. Clinica Chimica Acta, 2013, 422: 59-69.

14. Wu J, Chen YD, Gu W. Urinary proteomics as a novel tool for biomarker discovery in kidney diseases. Journal of Zhejiang University-Science B, 2010, 11(4): 227-237.

15. Beck LH, Bonegio RGB, Lambeau G, et al. M-Type Phospholipase A(sub 2) Receptor as Target Antigen in Idiopathic Membranous Nephropathy. New England Journal of Medicine, 2009, 361(1): 11-21.

16. Scheubert K, Hufsky F, Petras D, et al. Significance estimation for large scale metabolomics annotations by spectral matching. Nature Communications, 2017, 8(1): 1-10.

17. Mischak H, Delles C, Vlahou A, et al. Proteomic biomarkers in kidney disease: issues in development and implementation. Nature Reviews Nephrology, 2015, 11(4): 221-232.

18. Chen TQ, Guestrin C. XGBoost: A Scalable Tree Boosting System. Kdd’16: Proceedings of the 22nd Acm Sigkdd International Conference on Knowledge Discovery and Data Mining. 2016: 785-794.

19. Xie N, Chu CL, Tian XY, et al. An Endogenous Project Performance Evaluation Approach Based on Random Forests and IN-PROMETHEE II Methods. Mathematical Problems in Engineering, 2014.

20. Sang YS, Zhang HX, Zuo L. Least Squares Support Vector Machine Classifiers Using PCNNs. 2008 Ieee Conference on Cybernetics and Intelligent Systems, Vols 1 and 2, 2008: 828-833.

21. Goodacre R, Kell DB. Correction of mass spectral drift using artificial neural networks. Analytical Chemistry, 1996, 68(2): 271-280.

22. Ene-Iordache B, Perico N, Bikbov B, et al. Chronic kidney disease and cardiovascular risk in six regions of the world (ISN-KDDC): a cross-sectional study. Lancet Global Health. 2016, 4(5): e307-319.

23. Delahunt CB, Mehanian C, Hu LM, et al. Automated Microscopy and Machine Learning for Expert-Level Malaria Field Diagnosis. Proceedings of the Fifth Ieee Global Humanitarian Technology Conference Ghtc 2015, 2015: 393-399.

24. Mandal S. A Survey of Adaptive Fuzzy Controllers: Nonlinearities and Classifications. IEEE Transactions on Fuzzy Systems, 2015, 24(5): 1095-1107.

25. Baxi V, Beck A, Pandya D, et al. Artificial intelligencepowered retrospective analysis of PD-L1 expression in nivolumab trials of advanced non-small cell lung cancer. Journal for Immunotherapy of Cancer, 2019, 7.

26. Tsipouras MG, Voglis C, Fotiadis DI. A framework for fuzzy expert system creation - Application to cardiovascular diseases. Ieee Transactions on Biomedical Engineering, 2007, 54(11): 2089-2105.

27. Xiao J, Ding RF, Xu XL, et al. Comparison and development of machine learning tools in the prediction of chronic kidney disease progression. Journal of Translational Medicine, 2019, 17(1): 1-13.

28. Ogunleye AA, Qing-Guo W. XGBoost Model for Chronic Kidney Disease Diagnosis. IEEE-ACM Transactions on Computational Biology and Bioinformatics, 2019, 17(6): 2131-2140.

29. Yang C, Kong GL, Wang LW, et al. Big data in nephrology: Are we ready for the change? Nephrology, 2019, 24(11): 1097-1102.

30. Li CB, Zheng XS, Yang ZK, et al. Predicting Short-Term Electricity Demand by Combining the Advantages of ARMA and XGBoost in Fog Computing Environment. Wireless Communications & Mobile Computing, 2018.

31. Chen TY, Li X, Li YX, et al. Prediction and Risk Stratification of Kidney Outcomes in Iga Nephropathy. American Journal of Kidney Diseases, 2019, 74(3): 300-309.