Multiscale affinity maturation simulations to elicit broadly neutralizing antibodies against HIV

Abstract

The design of vaccines against highly mutable pathogens, such as HIV and influenza, requires a detailed understanding of how the adaptive immune system responds to encountering multiple variant antigens (Ags). Here, we describe a multiscale model of B cell receptor (BCR) affinity maturation that employs actual BCR nucleotide sequences and treats BCR/Ag interactions in atomistic detail. We apply the model to simulate the maturation of a broadly neutralizing Ab (bnAb) against HIV. Starting from a germline precursor sequence of the VRC01 anti-HIV Ab, we simulate BCR evolution in response to different vaccination protocols and different Ags, which were previously designed by us. The simulation results provide qualitative guidelines for future vaccine design and reveal unique insights into bnAb evolution against the CD4 binding site of HIV. Our model makes possible direct comparisons of simulated BCR populations with results of deep sequencing data, which will be explored in future applications.

On the Rapid Calculation of Binding Affinities for Antigen and Antibody Design and Affinity Maturation Simulations

“…Because our primary interests are antibody and antigen design, and affinity maturation simulations for modeling HIV [ 31 ], as the experimental dataset we chose binding affinities between the HIV envelope glycoprotein (gp120) and the VRC01 broadly neutralizing antibody (bnAb), obtained from two studies. First, Clark et al [ 32 ] published a dataset of 86 binding affinities for three different antibody–antigen complexes.…”

ppdx: Automated modeling of protein–protein interaction descriptors for use with machine learning

“…Then there are two examples related to modeling HIV/Antibody interactions: the first is a benchmark of scoring functions for modeling the binding of the VRC01 antibody to HIV antigens 5,64 ; the second is to compute the breadth of broadly neutralizing antibodies against HIV. 11,15 The last example consists of a subset of protein-protein interactions from the curated SKEMPI database. 65 We selected only protein-protein complexes with isothermal titration calorimetry (ITC) data.…”

“…Our efforts began with the adjustment of coefficients in a linear regression model for the specific application of predicting binding affinities between anti-HIV antibodies and their cognate antigens. 11 However, we soon realized that improved agreement with experiments could be obtained by using different descriptors, a different form of regression, or both. With the advent of easily accessible machine learning libraries, 12,13 such as scikit-learn, 14 we focused our efforts on developing method for the calculation of a range of molecular properties and descriptors, which could be used in a machine learning model using only a few lines of code.…”

“…The simple standalone tools we had used initially 11,15 thus evolved into a more complete software package, called ppdx, for Protein-Protein ΔX, where X can, in principle, be any property of interest: a binding affinity, an inhibitory concentration (IC 50 ), or even rates of binding and unbinding. The package has three main components: (i) an interface to structure modeling tools for creating the 3D coordinates of the protein-protein complexes starting with the protein sequences and a template for the structure; (ii) computation of a variety of descriptors, which can be used to train regression or machine learning models; (iii) a simple interface to use the trained models for predictions of new protein-protein complexes.…”

“…The present software represents an approach to improve the second class of methods to create a general tool that achieves a compromise between speed and accuracy appropriate for computational protein design applications. Our efforts began with the adjustment of coefficients in a linear regression model for the specific application of predicting binding affinities between anti‐HIV antibodies and their cognate antigens 11 . However, we soon realized that improved agreement with experiments could be obtained by using different descriptors, a different form of regression, or both.…”