Hallucinating protein assemblies

Abstract

Deep learning generative approaches provide an opportunity to broadly explore protein structure space beyond the sequences and structures of natural proteins. Here we use deep network hallucination to generate a wide range of symmetric protein homo-oligomers given only a specification of the number of protomers and the protomer length. Crystal structures of 7 designs are very close to the computational models (median RMSD: 0.6 Å), as are 3 cryoEM structures of giant rings with up to 1550 residues, C33 symmetry, and 10 nanometer in diameter; all differ considerably from previously solved structures. Our results highlight the rich diversity of new protein structures that can be created using deep learning, and pave the way for the design of increasingly complex nanomachines and biomaterials.

Evolutionary-scale prediction of atomic level protein structure with a language model

“…When evaluating AlphaFold2 and RosettaFold on single sequences by ablating the multiple sequence alignment, performance degrades substantially, and falls well below that of ESMFold. Note that this is an artificial setting as AlphaFold2 has not been explicitly trained for single sequences, however it has recently emerged as important in protein design, where these models have been used with single sequence inputs for de novo protein design (42)(43)(44).…”

Now What Sequence? Pre-trained Ensembles for Bayesian Optimization of Protein Sequences

“…84 There are also other approaches to finding peptide binders that do not involve BO. 85,86 Figure 6 shows the BO of Equation 11 averaged across 5 runs with variable sequence length. We compared against results from Anupam Patgiri [81] , which experimentally found the peptide FEGIYRLELLKAEEAN to bind well to Ras GTPase.…”

“…80 There are also other approaches to finding peptide binders that do not involve BO. 81,82 Figure 6 shows the BO of Equation 11 averaged across 10 runs with variable sequence length. The algorithm found better binders than the known binder from Anupam Patgiri [77] (FEGIYRLELLKAEEAN) based on wild type (WT) SoS protein.…”

Building blocks of non-Euclidean ribbons: size-controlled self-assembly via discrete frustrated particles

“…[27][28][29] In principle, recent progress in the synthesis of colloidal-scale particles with programmed shape can allow for tunable shape frustration that can more fully test the continuum theory description of size control. 30,31 Notably, advances in DNA nanotechnology 32,33 as well as synthetic protein engineering [34][35][36][37] allow for both careful design and control of the shape frustration of self-assembling nanoscale units as well as new opportunities for programming the interactions to separately tune the strength of cohesion and costs associated with distinct modes of assembly deformation. However, due to the primary reliance on continuum descriptions of GFA, several basic challenges remain to relate emergent thermodynamic behaviors in a particular system of self-assembling frustrated subunits.…”