Design, build, test, repeat—the essence of design thinking and a guiding principle of synthetic biology. Given the complexities of biological systems, there are bottlenecks that slow this virtuous design cycle for many of its potential applications. Yet, in the realm of small protein design and testing, the floodgates have just been thrust open.The recent study by David Baker’s group creates a platform for the design, testing, and refinement of small therapeutic proteins (Chevalier et al., 2017xChevalier, A., Silva, D.A., Rocklin, G.J., Hicks, D.R., Vergara, R., Murapa, P., Bernard, S.M., Zhang, L., Lam, K.H., Yao, G. et al. Nature. 2017; 550: 74–79PubMedSee all ReferencesChevalier et al., 2017) and uses this to create de novo designed protein inhibitors of influenza and botulinum toxin.A designed small protein (Bot.0671.2) bound the receptor binding domain (HcB) of botulinum neurotoxin. White is the design prediction; rainbow is the X-ray structure bound to the target (gray surface). Image courtesy of G. Rocklin.View Large Image | View Hi-Res Image | Download PowerPoint SlideThe key element to their work is that it is “massively parallel” in most of its facets. In the initial stage, they design and screen hundreds of thousands of peptides computationally for likelihood of high-affinity binding to a target using Rosetta, a program developed in the Baker lab that predicts protein structure (https://www.rosettacommons.org/). Although protein structures have long been used to computationally aid in the design of small molecules, chemical space is fantastically vast, and our ability to build and test in that space is constrained by the limits of chemical synthesis or the availability of compounds in existing libraries. These limits are much less pronounced in protein space and the effort by Chevalier et al. illustrates that making and testing designed proteins should be far easier to scale.The authors point to the convergence of two technical advances that enable this massively parallel pipeline. First is the ability to accurately model the structure of small proteins in the size range of 40 amino acids, and second, the capacity to synthesize oligonucleotides large enough to encode them. After computationally identifying approximately 10,000 peptides with high-predicted binding and stability for the two targets, they are able to directly make the oligonucleotides encoding them. These are then assembled in yeast libraries from which cells expressing potential binders are sorted and the proteins identified using deep sequencing. This pipeline ultimately results in the identification of thousands of high-affinity binders. Further analysis of which predicted binders do bind and which don’t is used to refine the computational model, thus completing the design cycle.These landmark efforts build off the group’s long-term goal of figuring out how sequence impacts folding and stability and their use of iterative cycles to improve computational-based design (e.g., Rocklin et al., 2017xRocklin, G.J., Chidyausiku, T.M., Goreshnik, I., Ford, A., Houliston, S., Lemak, A., Carter, L., Ravichandran, R., Mulligan, V.K., Chevalier, A. et al. Science. 2017; 357: 168–175Crossref | PubMed | Scopus (6)See all ReferencesRocklin et al., 2017).To cap off their latest study, the authors choose individual peptides for additional proof-of-principle testing, including confirmatory crystal structures and evidence of in vitro and in vivo activity. Specifically, the influenza inhibitor is shown to have impressive efficacy in mouse models of infection, and the toxin inhibitor protects cultured rat cortical neurons.It is also very important to note that repeated administration of the influenza-targeting small proteins does not elicit a strong immune response. This portends well for de novo designed small proteins as potential therapeutics, although further work is needed to fully explore their immunogenicity and determine whether additional steps in the design stages will be needed to filter out peptides with high immunogenic potential. It is also an open question whether computational analysis can be used to filter out small proteins that have greater potential for off-target effects.Time to take out the crystal ball. What could this mean for a future bench researcher, who for example, might want to make a new inhibitor for a target protein? The starting point for such a project might one day be as simple as pulling up the structure, running a computer program, and ordering a few thousand oligonucleotides—perhaps an afternoon’s work. What could a shift to massively parallel peptide design mean for the larger scientific community? The number of clinical trials based on small peptide therapeutics has been increasing in recent years. This should only accelerate given the speed and ease of protein synthesis methods coupled with these new computational tools. Perhaps in a decade, the majority of new filings for clinical trials will be for small designer proteins and not small molecules or antibodies.