Alphafold2 refinement improves designability of large de novo proteins

Abstract

Recent advances in computational protein design have enabled the creation of novel proteins for a variety of purposes. The capability for producing custom-shape high-quality backbones for very large proteins would further expand the scope of protein design. To this end, here we introduce the AlphaFold2 (AF2) cycler (af2cycler) design pipeline. AF2cycler is used to refine draft protein backbone geometries of arbitrary shape created by the Chroma diffusion model or other methods, using a combination of AF2 and ProteinMPNN to achieve high designability with minimal deviations from the initial target structure. In silico testing on multiple protein designs (100-1000 amino acids) demonstrated improvements in structural integrity and designability after af2cycling confirmed by enhanced ESMFold repredictions. Experimental wet lab validation through the design of a variety of test proteins with distinct shapes, each comprising 1000 amino acids, showed structural agreement between in silico predictions and transmission electron microscopy (TEM) imaging, establishing the af2cyclers efficacy in translating designs into real-world results. Af2cycler provides a convenient and reliable protein design workflow, particularly for large proteins, with potential for expanding to applications in areas such as design of higher-order protein complexes and multi-state backbone optimization.