Overview

Computational protein design is the inverse of structure prediction: rather than predicting a structure from a sequence, it searches for sequences that will fold into a predetermined target structure with desired functional properties. The design process simultaneously optimizes a large number of sequence and conformational variables to maximize folding stability, binding affinity, catalytic activity, or other targeted properties. Recent advances in deep learning and Monte Carlo optimization have dramatically expanded the scope of de novo protein design, enabling the creation of novel folds, functional enzymes, and protein-based therapeutics.

Methods

Design proceeds through iterative cycles of sequence selection and structure evaluation. The rotamer library enumerates discrete side-chain conformations, and a potential energy function (typically a molecular mechanics force field with solvation and statistical terms) scores each candidate. The dead-end elimination algorithm provably removes rotamers that cannot participate in the global minimum energy conformation. Protein Language Models and diffusion-based generative models now enable sequence design conditioned on backbone coordinates, producing diverse and experimentally expressible proteins.

Applications

Computational design has produced novel enzymes for industrial biocatalysis, stable protein scaffolds for vaccine display, and high-affinity binders for diagnostic or therapeutic applications. Designed proteins are expressed and characterized using site-directed mutagenesis to validate and refine computational predictions. The approach relies on detailed knowledge of protein structure and amino acids to specify the target backbone and to model sequence-structure compatibilities. Insights from protein folding and chaperones further guide the design of proteins that fold reliably and maintain stability in cellular environments.