Discrimination of Etiologically Different Cholestasis by Modeling Proteomics Datasets

Abstract

Cholestasis is characterized by disrupted bile flow from the liver to the small intestine. Although etiologically different cholestasis displays similar symptoms, diverse factors can contribute to the progression of the disease and determine the appropriate therapeutic option. Therefore, stratifying cholestatic patients is essential for the development of tailor-made treatment strategies. Here, we have analyzed the liver proteome from cholestatic patients of different etiology. In total, 7161 proteins were identified and quantified, of which 263 were differentially expressed between control and cholestasis groups. These differential proteins point to deregulated cellular processes that explain part of the molecular framework of cholestasis progression. However, the clustering of different cholestasis types was limited. Therefore, a machine learning pipeline was designed to identify a panel of 20 differential proteins that segregate different cholestasis groups with high accuracy and sensitivity. In summary, proteomics combined with machine learning algorithms provides valuable insights into the molecular mechanisms of cholestasis progression and a panel of proteins to discriminate across different types of cholestasis. This strategy may prove useful in developing precision medicine approaches for patient care.

Keywords: cholestasis; liver; machine learning; quantitative proteomics.

Figure 1

Quantitative proteomics results. (a) Table showing a summary of the number of PSMs, peptides, and proteins identified and differentially expressed in cholestasis (FDR < 1%). (b) PCA analysis of a proteomics dataset showing the segregation of cholestasis and control samples. IS corresponds to the internal standard containing equal amounts of each sample (ANOVA adj. p-value < 0.05). (c) Heatmap and clustering of differentially expressed proteins in cholestasis (ANOVA adj. p-value < 0.05).

Figure 2

Functional analysis of the cholestatic proteome. (a) Volcano plot of differential proteins when control and cholestasis samples were compared (t-test adj. p-value < 0.05). (b) PCA analysis of proteomics dataset showing the segregation of cholestasis and control samples. IS corresponds to the internal standard containing equal amounts of each sample (t-test adj. p-value < 0.05). (c) Cellular processes altered in cholestasis using the list of differentially expressed proteins in cholestasis regardless of the etiology (KEGG pathways enrichment using KOBAS tool, adj. p-value < 0.05).

Figure 3

Machine learning analysis graphical workflow. (a) Workflow for iterative random forest-based feature selection and posterior model building and evaluation. The workflow involved dividing the original dataset into a 30–70% split for training and testing, ensuring an equal representation of samples from each class. We repeated this process 100 times. We conducted iterative random forest-based feature selection within each partition on the training set. We then used a binomial test p-value to select the top 20 features consistently appearing in the highest-ranked features across a significant portion of the data partitions. Using these selected features, we constructed a final model in a new dataset partition using various classifiers. We then assessed the performance of these classifiers (e.g., accuracy, area under the curve) using the test set from that partition. (b) Detailed view of recursive random forest importance-based feature selection. In the recursive random forest importance-based feature selection process, the training set was employed for each data partition to select features based on their importance, which was determined through multiple iterations of the random forest algorithm.

Figure 4

Machine learning analysis. (a) Performance of different algorithms: random forest, extreme gradient boosting, linear discriminant analysis, k-nearest neighbors, naïve Bayes, support vector machine, and logistic regression. (b) Confusion matrix of SVM algorithm; (c) ROC curves; (d) MDS (multidimensional scaling) 3D plot of samples.

Figure 5

Classification of cholestasis of different etiologies using the described machine learning pipeline. (a) Heatmap representation of cholestasis subtypes clustering using the 20 proteins panel selected by the algorithm. (b) Contribution of each protein to the classification of each cholestasis group based on their estimated importance. (c) Boxplots representing the abundance values of each protein from the classifier in the different cholestasis groups.