Exploring the ability of the MD+FoldX method to predict SARS-CoV-2 antibody escape mutations using large-scale data

Figures

• 

  Fig. 1. SARS-CoV-2 mutations could lead to Ab escape. (a) A wild-type viral antigen (Ag) can mutate into a variant that may either retain Ab-neutralization hence preventing infection, or lead to Ab-escape hence maintaining infection. (b) RBD-directed neutralizing Abs can be divided into four main classes based on the epitopes they target11. Abs tested in this study are listed according to their class. For each class, one representative Ab bound to the RBD is shown: Class 1, REGN10933; Class 2, LY-CoV555; Class 3, REGN10987; Class 4, S2X259. Protein images on the left were generated with Protein Imager software16.

• 

  Fig. 2. The methodology pipeline of the study is structured as follows: (a) systems are selected based on the availability of data; (b) binding affinity predictions are generated for each system using the MD+FoldX approach; and (c) predictions are compared with experimental escape fraction data, leading to the optimization of the energy cutoff and the precision evaluation, followed by the creation of an Ab-escape mutation list using the optimized cutoff.

• 

  Fig. 3. Binding interface mutation for Ab within a specific class. Interface residues highlighted in red represent the union set of RBD residues located within 10 Å of any antibody chains in this class. A total of 85 sites were mutated in Class 1, 91 in Class 2, 81 in Class 3, and 141 in Class 4. Mutations are mapped onto the 3D structure of one representative Ab/Ag complex per class: Class 1 (AZD8895), Class 2 (LY-CoV555), Class 3 (REGN10987), and Class 4 (S2X259).

• 

  Fig. 4. Heat map of predicted \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta \Delta G_{bind}$$\end{document}ΔΔGbind values using MD+FoldX for mutations in the RBD when in complex with LY-CoV555 Class 2 Ab. The horizontal axis displays the amino acid positions in the RBD along with the corresponding native amino acids. The vertical axis represents the mutation of amino acids colored according to physicochemical properties: small non-polar, hydrophobic, polar, negatively charged and positively charged. White squares represent the amino acids of the wild-type. For clarity in visualization, sites where all mutations led to changes in binding free energy of less than 0.5 kcal/mol were omitted from the heat map.

• 

  Fig. 5. Pearson and Spearman correlation coefficients between predicted \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta G_{\text {bind}}$$\end{document}ΔGbind values and experimental escape fractions (f), represented by \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {F}(f)$$\end{document}F(f) as defined in Eq. (2). We show all the Ab/Ag complexes spanning the 4 Ab classes: Class 1 (purple), Class 2 (blue), Class 3 (green), and Class 4 (yellow).

• 

  Fig. 6. Example of a cutoff optimization using the S2H14 Class 1 system. (a) Precision, recall, and F-Score values across a range of cutoffs. (b) Scatter of predicted \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta G_{\text {bind}}$$\end{document}ΔGbind values against \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\mathcal {F}(f)$$\end{document}F(f); the green dashed line represents the optimized cutoff, the orange dashed line indicates the classic cutoff, the grey dashed line represents the escape fraction value at which the 50% of the cells expressing a specific variant escape Ab binding, and the solid grey line indicates the value of the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek}…

• 

  Fig. 7. Optimized cutoff\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_{\Delta \Delta G}$$\end{document}ΔΔG for each system. The orange dashed line represents the classic cutoff and the grey dashed line is the median optimized cutoff\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$_{\Delta \Delta G}$$\end{document}ΔΔG. Asterisks indicate each system within the cases discussed in the “Discussion” section: Case 1 (*), Case 2 (**), and Case 3 (***).

• 

  Fig. 8. Precision (expressed in %) comparison across systems as defined by Eq. (3) using optimized cutoffs for each system. Asterisks indicate each system within the cases discussed in the Discussion section: Case 1 (*), Case 2 (**), and Case 3 (***).