| To design proteins of a specific function is an important objective of protein design.The function of a protein often involves the binding with another molecule,such as a small molecule,another protein,or a nucleic acid.Thus an important problem for designing proteins with functions is to design proteins that can bind a given small molecule ligand.One approach to designing small molecule-binding proteins is based on the "inside-out" strategy,in which the overall design process is divided into three steps,which correspond respectively to solving three different sub-problems.The first sub-problem is the design of a pocket model that comprises disconnected amino acid residues arranged around the target small molecule to provide a favorable environment for binding.The second sub-problem is to construct designable backbone conformations of the host protein,including the conformations of the loop regions surrounding the pocket.These backbone conformations provide the candidate scaffolds to connect/support the residues of a pocket model.The third sub-problem is to assemble the pocket residues onto the backbone scaffolds.This involves finding on the candidate backbones those backbone positions that can match the spatial arrangement of the amino acid residues defined by the pocket model.In the subsequent sequence design of the host protein,the residues types at these sites can be fixed so the final designed sequence can form the desired binding pocket through autonomous protein folding.In this work,we follow the overall "inside-out" strategy and develop computational methods to address the different sub-problems.We develop a data-driven DEPACT(design pocket as a cluster based on templates)method for the de novo design of pocket models.In this method,disconnected amino acid residues or short peptide fragments that favorably interacting with sub-structures of the target ligand are extracted from natural complex structures in Protein Data Bank(PDB),and then sampled and combined to form intact pockets wrapping around the target small molecule.Compared with existing methods,DEPACT can design pockets with diverse binding environments which are not restricted by artificial design rules or approximately defined energy functions.Moreover,pockets with metal ion or water molecule-mediated ligand-protein interactions can also be designed by DEPACT.We compare two recent methods,SCUBA and SCUBA-D,for generating backbone conformations of loops for potential use in designing backbones surrounding binding pockets.Both methods can be used to design backbone without pre-specifying the amino acid sequence.SCUBA(sidechain-unknown backbone arrangement)uses a neural network-based statistical energy,while SCUBA-D uses an artificial intelligence approach of diffusion generation.Results of conformation sampling of loops in various native structures suggest that both SCUBA and SCUBA-D can generate diverse,reasonable loop conformations,while SCUBA-D is computationally much more efficient than SCUBA.We develop PACMatch to solve the pocket model-protein scaffold matching problem.PACMatch uses key atoms of pocket residues as representative features of the binding environment provided by the pocket.By using a multi-step strategy,pockets of more than a few residues can be efficiently processed.Compared with the existing method RosettaMatch for the same benchmark set,PACMatch achieves a higher rate of recovering native matches between pocket residues and backbone positions.Moreover,PACMatch can handle pocket models containing as many as 10 amino acid residues.Hence it is suitable for processing the multi-residue pocket models produced by DEPACT.Based on the above developments of computational method,we explore the design of an artificial urate oxidase.The process demonstrates the overall computational workflow and some remaining challenges emerging from the practical applications of computational design.In summary,we investigate and develop multiple computational tools for the’inside-out" design of small molecule-binding proteins.Our methods form a chain of tools that provide a basis for the practical construction of artificial proteins with small molecule-binding functions based on computational protein design. |
