Robust de novo design of protein binding proteins from target structural information alone

Abstract

The design of proteins that bind to a specific site on the surface of a target protein using no information other than the three-dimensional structure of the target remains an outstanding challenge. We describe a general solution to this problem which starts with a broad exploration of the very large space of possible binding modes and interactions, and then intensifies the search in the most promising regions. We demonstrate its very broad applicability by de novo design of binding proteins to 12 diverse protein targets with very different shapes and surface properties. Biophysical characterization shows that the binders, which are all smaller than 65 amino acids, are hyperstable and bind their targets with nanomolar to picomolar affinities. We succeeded in solving crystal structures of four of the binder-target complexes, and all four are very close to the corresponding computational design models. Experimental data on nearly half a million computational designs and hundreds of thousands of point mutants provide detailed feedback on the strengths and limitations of the method and of our current understanding of protein-protein interactions, and should guide improvement of both. Our approach now enables targeted design of binders to sites of interest on a wide variety of proteins for therapeutic and diagnostic applications.

Randomized gates eliminate bias in sort-seq assays

“…Likelihood-based inference is a common strategy used with the intent to circumvent the resolution limitation of the histogram approach 5;8;9;15 . However, this approach can fail on real data.…”

“…Quantitative, multiplexed assays relying on fluorescence activated cell sorting (FACS) followed by high-throughput sequencing are critical to modern biology and molecular engineering because they enable construction of large scale datasets connecting sequence to function. For example, these “sort-seq” assays are widely used to profile the strength of protein-protein binding interactions via yeast display 4;5;11;17 . In particular one (i) synthesizes a library of 10 4 to 10 5 DNA sequences encoding proteins that may bind to a target of interest; (ii) transforms the library into yeast such that each putative binder is expressed on the surface of a population of cells; (iii) incubates cells with fluorescently labeled target protein; (iv) physically separates 10 6 to 10 8 cells based on binding affinity by FACS; and finally, (v) quantifies the prevalence, and thereby binding affinity, of each library member by high throughput sequencing.…”

“…This challenge has spurred much work on how to effectively reduce histogram bias 1;8;13;14 . One common approach seeks to overcome the resolution limits of histograms by assuming fluorescence is log-normally distributed for each population and using maximum likelihood estimation to estimate moments 5;8;9;15 . However, on real data, this assumption is violated and the resulting estimates can have greater bias than the naive approach (Figure S1).…”

Next-Generation Molecular Discovery: From Bottom-Up In Vivo and In Vitro Approaches to In Silico Top-Down Approaches for Therapeutics Neogenesis

“…The David Baker lab at EMBL-EBI is one of the first groups claiming to be able to perform de novo design for more transient interactions of protein binders and targets. In this work, the group has moved beyond using a native or homologous protein scaffold and previously identified hot-spot residues to identify and predict proteins with favourable and more transient interactions [ 133 ]. This approach begins by performing target docking of a library of “disembodied” amino acids and creating what is termed a “rotamer interaction field” (RIF).…”

Deep learning of Protein Sequence Design of Protein-protein Interactions

“…The success rates for the de novo generation of protein-protein interactions archived by existing methods is very low, with only a few examples demonstrating it is possible (Cao et al, 2020;Fleishman et al, 2011;Strauch et al, 2014). Even most recent work illustrates an average success rate of 0.25% while requiring substantial computational and laboratory resources (Cao et al, 2021) underlining that it is still highly challenging. With DeepMind using deep learning algorithms and entering the field of protein structure prediction, much has changed about how machine learning approaches can be used for protein folding and also design.…”