Protein sequence design by conformational landscape optimization

By Christoffer H Norn, B. Wicky, David Juergens, Sirui Liu, David E. Kim, D. Tischer, B. Koepnick, Ivan V. Anishchenko, David Baker, Sergey Ovchinnikov. 2021

142 citations recorded by the source corpus.

Significance Almost all proteins fold to their lowest free energy state, which is determined by their amino acid sequence. Computational protein design has primarily focused on finding sequences that have very low energy in the target designed structure. However, what is most relevant during folding is not the absolute energy of the folded state but the energy difference between the folded state and the lowest-lying alternative states. We describe a deep learning approach that captures aspects of the folding landscape, in particular the presence of structures in alternative energy minima, and show that it can enhance current protein design methods. The protein design problem is to identify an amino acid sequence that folds to a desired structure. Given Anfinsen’s thermodynamic hypothesis of folding, this can be recast as finding an amino acid sequence for which the desired structure is the lowest energy state. As this calculation involves not only all possible amino acid sequences but also, all possible structures, most current approaches focus instead on the more tractable problem of finding the lowest-energy amino acid sequence for the desired structure, often checking by protein structure prediction in a second step that the desired structure is indeed the lowest-energy conformation for the designed sequence, and typically discarding a large fraction of designed sequences for which this is not the case. Here, we show that by backpropagating gradients through the transform-restrained Rosetta (trRosetta) structure prediction network from the desired structure to the input amino acid sequence, we can directly optimize over all possible amino acid sequences and all possible structures in a single calculation. We find that trRosetta calculations, which consider the full conformational landscape, can be more effective than Rosetta single-point energy estimations in predicting folding and stability of de novo designed proteins. We compare sequence design by conformational landscape optimization with the standard energy-based sequence design methodology in Rosetta and show that the former can result in energy landscapes with fewer alternative energy minima. We show further that more funneled energy landscapes can be designed by combining the strengths of the two approaches: the low-resolution trRosetta model serves to disfavor alternative states, and the high-resolution Rosetta model serves to create a deep energy minimum at the design target structure.

Subfields: Materials Chemistry, Molecular Biology, Computational Biology & Bioinformatics, Computational Chemistry & Drug Discovery

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Materials Chemistry

In Molecular Biology

In Computational Biology & Bioinformatics

In Computational Chemistry & Drug Discovery

Citations in the corpus

AlphaFold2 and its applications in the fields of biology and medicine (2023, methodology): “111,112 The integrity of the structures predicted is validated and con ﬁ rmed by standard ab initio folding, protein structure analysis methods, and rigorous all-atom molecular dynamics simulations.”
De novo protein design by deep network hallucination (2020, methodology): “On the sampling side, the Monte Carlo approach can be made more efficient by direct gradient-based minimization by tracing the gradients back to the inputs 25 .”
Protein sequence design with a learned potential (2022, methodology): “…learning methods for design thus far have focused on mutation prediction 25 – 30 , rotamer repacking of native sequences 31 , or amino acid sequence design without modeling side-chain conformers 32 – 35 , with some experimental validation including circular dichroism data 33 and ﬂ uorescence 26 .”
Scaffolding protein functional sites using deep learning (2022, methodology): “…fold to the hallucinated structures (12). trRosetta can also be used to design sequences that fold into a target backbone structure by carrying out sequence optimization using a structure recapitulation loss function that rewards similarity of the predicted structure to the target structure (13).”
Masked Inverse Folding with Sequence Transfer for Protein Representation Learning (2023, methodology): “More recent methods attempt to invert deep-learning protein structure prediction models (Jendrusch et al., 2021; Moffat et al., 2021; 2022; Anishchenko et al., 2021; Norn et al., 2021; Wang et al., 2021).”
Improving de novo protein binder design with deep learning (2022, unknown): “As observed previously 32 , such consideration of the overall folding landscape enables considerably more accurate assessment of the likelihood a design will fold and bind as intended compared to evaluation of only thedepth of the designed energy well.”
Using AlphaFold for Rapid and Accurate Fixed Backbone Protein Design (2021, methodology): “A gradient descent approach is likely to also be susceptible to becoming trapped in local minima but it would hopefully reduce the the number of design iterations currently required down to a handful of forward passes (Norn et al., 2021).”
Mega-scale experimental analysis of protein folding stability in biology and design (2023, methodology): “Beyond studying stability, cDNA display proteolysis will have other applications, including assaying designed proteins on a massive scale to systematically improve design methods 18,30,31,49 and to dissect the relationships between folding stability and function 50 .”
Deep learning methods for designing proteins scaffolding functional sites (2021, methodology): “design sequences that fold into a target backbone structure by carrying out sequence optimization using a structure recapitulation loss function that rewards similarity of the predicted structure to the target structure (13).”
Guided Generative Protein Design using Regularized Transformers (2022, methodology): “Additional approaches to protein design have included structure-based design Rohl et al. [2004], Norn et al. [2021] wherein ideal structures are chosen a priori and the task is to ﬁt a sequence to the design.”
De novo design of luciferases using deep learning (2023, methodology): “…of ideal NTF2-like structures, we developed a deep-learning-based ‘family-wide hallu-cination’ approach that integrates unconstrained de novo design 17,18 and Rosetta sequence-design approaches 19 to enable the generation of an essentially unlimited number of proteins that have a desired fold (Fig.”
AlphaDesign: A de novo protein design framework based on AlphaFold (2021, methodology): “[71, 72] use both Markov-Chain Monte Carlo and gradient descent as their means for optimisation.”
Fast, accurate antibody structure prediction from deep learning on massive set of natural antibodies (2022, methodology): “Meanwhile, methods like AlphaFold and RoseTTAFold have been adapted for 343 gradient-based design of novel protein structures and even sca olding binding loops (52, 53).”
Efficient generative modeling of protein sequences using simple autoregressive models (2021, methodology): “via a VAE [26] or a Graph Neural Network [51], or by inverting structural prediction models [4, 21, 37, 40].”