Abstract:Abstract A designable protein backbone is one for which amino acid sequences that stably fold into it exist. To design such backbones, a general method is much needed for continuous sampling and optimization in the backbone conformational space without specific amino acid sequence information. The energy functions driving such sampling and optimization must faithfully recapitulate the characteristically coupled distributions of multiplexes of local and non-local conformational variables in designable backbones. It is also desired that the energy surfaces are continuous and smooth, with easily computable gradients. We combine statistical and neural network (NN) approaches to derive a model named SCUBA, standing for Side-Chain-Unspecialized-Backbone-Arrangement. In this approach, high-dimensional statistical energy surfaces learned from known protein structures are analytically represented as NNs. SCUBA is composed as a sum of NN terms describing local and non-local conformational energies, each NN term derived by first estimating the statistical energies in the corresponding multi-variable space via neighbor-counting (NC) with adaptive cutoffs, and then training the NN with the NC-estimated energies. To determine the relative weights of different energy terms, SCUBA-driven stochastic dynamics (SD) simulations of natural proteins are considered. As initial computational tests of SCUBA, we apply SD simulated annealing to automatically optimize artificially constructed polypeptide backbones of different fold classes. For a majority of the resulting backbones, structurally matching native backbones can be found with Dali Z-scores above 6 and less than 2 Å displacements of main chain atoms in aligned secondary structures. The results suggest that SCUBA-driven sampling and optimization can be a general tool for protein backbone design with complete conformational flexibility. In addition, the NC-NN approach can be generally applied to develop continuous, noise-filtered multi-variable statistical models from structural data. Linux executables to setup and run SCUBA SD simulations are publicly available ( http://biocomp.ustc.edu.cn/servers/download_scuba.php ). Interested readers may contact the authors for source code availability.

Scaffold-Lab: Critical Evaluation and Ranking of Protein Backbone Generation Methods in A Unified Framework

ProteinInvBench: Benchmarking Protein Inverse Folding on Diverse Tasks, Models, and Metrics.

Global-Context Aware Generative Protein Design

Generative De Novo Protein Design with Global Context

PDB-Struct: A Comprehensive Benchmark for Structure-based Protein Design

ProteinWeaver: A Divide-and-Assembly Approach for Protein Backbone Design

Out of Many, One: Designing and Scaffolding Proteins at the Scale of the Structural Universe with Genie 2

Diffusion probabilistic modeling of protein backbones in 3D for the motif-scaffolding problem

A backbone-centred energy function of neural networks for protein design

Improved motif-scaffolding with SE(3) flow matching

Building Confidence in Deep Generative Protein Design

Protein Language Model Supervised Scalable Approach for Diverse and Designable Protein Motif-Scaffolding with GPDL

Protein Language Model Supervised Precise and Efficient Protein Backbone Design Method

ProteinBench: A Holistic Evaluation of Protein Foundation Models

Generating Novel, Designable, and Diverse Protein Structures by Equivariantly Diffusing Oriented Residue Clouds

Amalga: Designable Protein Backbone Generation with Folding and Inverse Folding Guidance

Combining statistical and neural network approaches to derive energy functions for completely flexible protein backbone design

Protein Design with A Comprehensive Statistical Energy Function and Boosted by Experimental Selection for Foldability

Functional Geometry Guided Protein Sequence and Backbone Structure Co-Design

Energy Functions in De Novo Protein Design: Current Challenges and Future Prospects

Proteus: exploring protein structure generation for enhanced designability and efficiency