An Atomistic Statistically Effective Energy Function for Computational Protein Design

Topham, Christopher M.; Barbe, Sophie; André, Isabelle

doi:10.1021/acs.jctc.6b00090

Cited by 10 publications

(11 citation statements)

References 171 publications

(519 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As of July 2017, there are ~132,000 structures in the protein data bank (PDB) 29 with a yearly increase of ~10,000, but the number of unique folds has not changed in the past few years, suggesting more data are accumulated on each fold, and therefore statistical learning and utilizing the existing structures are likely able to improve the design methods. 30,31 Recently, two statistical potentials for protein design have been developed, 32,33 and the ABACUS potential 34 has been successfully used in designing proteins. 33,35 While these statistical potentials have a physical basis, machine learning especially deep-learning neural network has recently become a popular method to analyze big data sets, extract complex features, and make accurate predictions.…”

Section: Introductionmentioning

confidence: 99%

Computational Protein Design with Deep Learning Neural Networks

Wang

Cao

Zhang

et al. 2018

Sci Rep

141

142

View full text Add to dashboard Cite

Computational protein design has a wide variety of applications. Despite its remarkable success, designing a protein for a given structure and function is still a challenging task. On the other hand, the number of solved protein structures is rapidly increasing while the number of unique protein folds has reached a steady number, suggesting more structural information is being accumulated on each fold. Deep learning neural network is a powerful method to learn such big data set and has shown superior performance in many machine learning fields. In this study, we applied the deep learning neural network approach to computational protein design for predicting the probability of 20 natural amino acids on each residue in a protein. A large set of protein structures was collected and a multi-layer neural network was constructed. A number of structural properties were extracted as input features and the best network achieved an accuracy of 38.3%. Using the network output as residue type restraints improves the average sequence identity in designing three natural proteins using Rosetta. Moreover, the predictions from our network show ~3% higher sequence identity than a previous method. Results from this study may benefit further development of computational protein design methods.

show abstract

Section: Introductionmentioning

confidence: 99%

Computational Protein Design with Deep Learning Neural Networks

Wang

Cao

Zhang

et al. 2018

Sci Rep

141

142

View full text Add to dashboard Cite

show abstract

“…That protein structure can be designed computationally was established a long time ago (1) and has been demonstrated numerous times since (23,26). It is also true that reliance on prior structural data has been broadly explored, both in terms of various statistics-based methods (46,49,52,54) as well as in the creation of chimeras that fused domains or larger fragments from existing structures (64)(65)(66)(67). What is new and exciting about our results here is the marriage between the generality of our approach (i.e., its ability to design sequences for arbitrarily-defined structures) and its reliance on motif-based structural data.…”

Section: Discussionmentioning

confidence: 99%

“…PANENERGY is an example of atomistic statistical energy function built for CPD and related applications (49). It builds upon a traditional atom-pair PDB-derived statistical interaction energy, but applies a novel "connectivity scaling factor" to modulate inter-atomic interactions.…”

Section: Introductionmentioning

confidence: 99%

“…The factor is based on the degree of chemical connectivity of interacting atoms, which effectively emulates course-graining (thus introducing some manybody character to the underlying atomic pairwise interaction model) without loss of atomistic detail. The function is further supplemented with an amino-acid environment statistical energy and a non-polar surface areabased cavitation term (49). PANENERGY has been shown to perform well on predicting mutational protein stability changes and protein-ligand binding free energies, decoy discrimination, and native sequence recovery (49).…”

Section: Introductionmentioning

confidence: 99%

“…The function is further supplemented with an amino-acid environment statistical energy and a non-polar surface areabased cavitation term (49). PANENERGY has been shown to perform well on predicting mutational protein stability changes and protein-ligand binding free energies, decoy discrimination, and native sequence recovery (49).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A general-purpose protein design framework based on mining sequence-structure relationships in known protein structures

Zhou

Panaitiu

Grigoryan

2018

Preprint

View full text Add to dashboard Cite

The ability to routinely design functional proteins, in a targeted manner, would have enormous implications for biomedical research and therapeutic development. Computational protein design (CPD) offers the potential to fulfill this need, and though recent years have brought considerable progress in the field, major limitations remain. Current state-of-the-art approaches to CPD aim to capture the determinants of structure from physical principles. While this has led to many successful designs, it does have strong limitations associated with inaccuracies in physical modeling, such that a robust general solution to CPD has yet to be found. Here we propose a fundamentally novel design framework-one based on identifying and applying patterns of sequence-structure compatibility found in known proteins, rather than approximating them from models of inter-atomic interactions. Specifically, we systematically decompose the target structure to be designed into structural building blocks we call TERMs (tertiary motifs) and use rapid structure search against the Protein Data Bank (PDB) to identify sequence patterns associated with each TERM from known protein structures that contain it. These results are then combined to produce a sequence-level pseudo-energy model that can score any sequence for compatibility with the target structure. This model can then be used to extract the optimal-scoring sequence via combinatorial optimization or otherwise sample the sequence space predicted to be well compatible with folding to the target. Here we carry out extensive computational analyses, showing that our method, which we dub dTERMen (design with TERM energies): 1) produces native-like sequences given native crystallographic or NMR backbones, 2) produces sequence-structure compatibility scores that correlate with thermodynamic stability, and 3) is able to predict experimental success of designed sequences generated with other methods, and 4) designs sequences that are found to fold to the desired target by structure prediction more frequently than sequences designed with an atomistic method. As an experimental validation of dTERMen, we perform a total surface redesign of Red Fluorescent Protein mCherry, marking a total of 64 residues as variable. The single sequence identified as optimal by dTERMen harbors 48 mutations relative to mCherry, but nevertheless folds, is monomeric in solution, exhibits similar stability to chemical denaturation as mCherry, and even preserves the fluorescence property. Our results strongly argue that the PDB is now sufficiently large to enable proteins to be designed by using only examples of structural motifs from unrelated proteins. This is highly significant, given that the structural database will only continue to grow, and signals the possibility of a whole host of novel data-driven CPD methods. Because such methods are likely to have orthogonal strengths relative to existing techniques, they could represent an important step towards removing remaining barriers to robust CPD.

show abstract

Computational Protein Design

Shifman¹,

Singh²

2018

Encyclopedia of Biophysics

View full text Add to dashboard Cite

An Atomistic Statistically Effective Energy Function for Computational Protein Design

Cited by 10 publications

References 171 publications

Computational Protein Design with Deep Learning Neural Networks

Computational Protein Design with Deep Learning Neural Networks

A general-purpose protein design framework based on mining sequence-structure relationships in known protein structures

Computational Protein Design

Contact Info

Product

Resources

About