Biophysical Deconvolution of Polyclonal Adaptive Immunity from Deep Mutational Scanning Landscapes
Faculty: Prof. Rahul Roy (DCE) and Prof. Debnath Pal (CDS)
Abstract
Deep mutational scanning (DMS) of viral surface proteins has become a standard tool for mapping antibody escape. When the viral mutant library is held constant and patient serum is varied longitudinally, the resulting escape profiles cease to report on the virus and instead become high-resolution projections of the evolving antibody repertoire onto mutational space. This project develops the computational and biophysical framework needed to invert those projections — to recover, from a single polyclonal escape map, the weighted clonal composition of the underlying humoral response and its trajectory over time.
The central methodological contribution is a physics-informed deconvolution architecture. Coarse-grained molecular dynamics (CG-MD, MARTINI force field) simulations of antibody–RBD complexes across all four major structural classes (Classes 1–4) are used to compute per-mutation binding free energy perturbations (∆∆G), yielding a class-specific escape fingerprint for each of ~3,800 RBD single-point mutants. These fingerprints serve as physically constrained basis vectors in a linear mixing model: an observed polyclonal escape map e is decomposed as e ≈ Σ k w k f k + ε, where f is the CG-MD fingerprint of antibody class k, wk ≥ 0 are the class weights reflecting clonal abundance, and ε captures epistatic corrections and experimental noise. The weights are inferred via a machine learning framework trained on monoclonal ground-truth data and validated against published molecular-level polyclonal decompositions.
A second computational axis addresses epistasis. For the top 50–100 mutation pairs showing non-additive escape in experimental DMS data, double-mutant CG-MD simulations distinguish cooperative destabilisation of a single antibody interface (structural epistasis, ε = ∆∆GAB − ∆∆GA − ∆∆GB < 0 with both residues contacting one paratope) from apparent synergy arising when two mutations independently affect distinct antibodies in the polyclonal mixture (compositional epistasis). This distinction is inaccessible to experimental escape scores alone and determines whether synergistic escape reflects a vulnerability in one antibody or a gap in repertoire coverage.
Applied longitudinally to serum from an Indian cohort (20–30 patients, 6 timepoints over 24 months, stratified by vaccine platform and immunological status), the deconvolution framework tracks clonal composition dynamics: which antibody classes expand through affinity maturation, which undergo competitive displacement, and which persist as immunological imprints. Original Antigenic Sin (OAS) is modelled thermodynamically as a kinetic trap in CDR sequence space, where the energetic barrier to remodelling recall-response antibodies for a new variant epitope exceeds the selection pressure available in the germinal centre. The framework yields a per-patient imprinting strength index and, ultimately, computational designs for variant immunogens that redirect responses away from imprint-dominant epitopes toward conserved, subdominant targets.
The project thus establishes DMS not as a virology assay but as a quantitative immunological instrument — one whose readout, when coupled with biophysical priors, can resolve the clonal architecture of adaptive immunity at a resolution currently available only through far more laborious single-cell approaches.
Keywords: deep mutational scanning · polyclonal deconvolution · coarse-grained molecular dynamics · antibody escape · affinity maturation · Original Antigenic Sin · epistasis · SARS-CoV-2 RBD