Computational design of a predicted high-affinity humanized nanobody against KRAS Switch-I using spectral sequence optimization and deep learning validation

Abstract The KRAS oncoprotein remains a critical pharmacological challenge due to the plasticity of its active site and the emergence of resistance to covalent inhibitors (G12C). This study presents the rational in silico design of PIA-KRASv2-Nb, a humanized nanobody targeting the epitope DEYDPTIEDS in the Switch-I region, generated using a Spectral Sequence Optimization algorithm. Unlike directed evolution methods, our approach utilizes physicochemical signal analysis (Fourier transform of amino acid properties) to maximize resonant complementarity with the target. Structural prediction using AlphaFold-Multimer v3 identified a complex with high structural confidence (ipTM = 0.78). To determine whether this topological confidence correlates with physical affinity, we performed orthogonal thermodynamic evaluation using the PRODIGY estimator, predicting a favorable dissociation constant (Kd) of 6.7 nM. Critically, the candidate achieved a perfect humanization score (H-score = 1.0), indicating intrinsic human compatibility. Furthermore, all-atom molecular dynamics simulations verified the immediate steric viability of the complex, maintaining a highly robust interfacial network averaging 29.17 residue–residue contacts over 10 ns. This dynamic persistence provides predictive support for short-timescale structural stability, circumventing the steric clash limitations often encountered when classical end-point free energy calculations are applied to unrelaxed AI-derived models. These results position PIA-KRASv2-Nb as a promising in silico candidate, evaluated via supporting computational metrics and a favorable theoretical humanization profile, warranting future experimental characterization.