Predicting cell-penetrating peptides using machine learning algorithms and navigating in their chemical space

Abstract

Cell-penetrating peptides (CPPs) are naturally able to cross the lipid bilayer membrane that protects cells. These peptides share common structural and physicochemical properties and show different pharmaceutical applications, among which drug delivery is the most important. Due to their ability to cross the membranes by pulling high-molecular-weight polar molecules, they are termed Trojan horses. In this study, we proposed a machine learning (ML)-based framework named BChemRF-CPPred (beyond chemical rules-based framework for CPP prediction) that uses an artificial neural network, a support vector machine, and a Gaussian process classifier to differentiate CPPs from non-CPPs, using structure- and sequence-based descriptors extracted from PDB and FASTA formats. The performance of our algorithm was evaluated by tenfold cross-validation and compared with those of previously reported prediction tools using an independent dataset. The BChemRF-CPPred satisfactorily identified CPP-like structures using natural and synthetic modified peptide libraries and also obtained better performance than those of previously reported ML-based algorithms, reaching the independent test accuracy of 90.66% (AUC = 0.9365) for PDB, and an accuracy of 86.5% (AUC = 0.9216) for FASTA input. Moreover, our analyses of the CPP chemical space demonstrated that these peptides break some molecular rules related to the prediction of permeability of therapeutic molecules in cell membranes. This is the first comprehensive analysis to predict synthetic and natural CPP structures and to evaluate their chemical space using an ML-based framework. Our algorithm is freely available for academic use at http://comptools.linc.ufpa.br/BChemRF-CPPred .

Figure 1

Boxplot of accuracy from tenfold cross-validation of ANN (red), GPC (blue), SVM (green), and BChemRF-CPPred (orange).

Figure 2

(A) Accuracy of ANN (red), GPC (blue), SVM (green), and BChemRF-CPPred (orange) by FCs evaluated in the independent test. (B) ROC curves and AUC of ML-based frameworks using the FC-1, FC-2, FC-3, and FC-4 in the independent test.

Figure 3

Boxplot of accuracy from tenfold cross-validation of ANN (red), GPC (blue), SVM (green), and BChemRF-CPPred (orange) using FASTA input.

Figure 4

Accuracy of ANN (red), GPC (blue), SVM (green), and BChemRF-CPPred (orange) by FCs evaluated in the independent test, using FASTA input.

Figure 5

Normalized cumulative information entropy (CIE) provided by structure-based, AAC, DPC, and PseAAC descriptors, and calculated by ERT algorithm. (A) Training dataset; (B) independent test dataset.

Figure 6

Analysis of 3D dimensionality reduction using PCA of the sequence- and structure-based descriptors present in FC-1 to FC-4. Panel (A) 3D PCA of FC-1 showing a contribution of explained variance ratio of 10.93% (PC1), 7.26% (PC2), and 6% (PC3), and cumulative explained variance ratio (CEVR) of 24.19%. (B) 3D PCA of FC-2 showing a contribution of explained variance ratio of 48.9% (PC1), 21.94% (PC2), and 14.34%(PC3), and CEVR = 85.19%. (C) 3D PCA of FC-3 showing a contribution of explained variance ratio of 16.31% (PC1), 12.03% (PC2), and 7.22% (PC3) and CEVR = 35.58%. (D) 3D PCA of FC-4 showing a contribution of explained variance ratio of 17.81% (PC1), 12.48% (PC2), and 8.93% (PC3), and CEVR = 39.29%.

Figure 7

General structure of BChemRF-CPPred framework with ANN, GPC and SVM machine learning algorithms.

Figure 8

Process of hyper-parameters tuning applied for ANN, GPC, and SVM by FC using Grid Search method. The best models obtained in x-th feature composition (ANN_bFC-X, GPC_bFC-X, SVM_bFC-X) were used to compose the respective framework.