Deep learning approaches for conformational flexibility and switching properties in protein design

Rudden, Lucas S. P.; Hijazi, Mahdi; Barth, Patrick

doi:10.3389/fmolb.2022.928534

Cited by 8 publications

(9 citation statements)

References 73 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the design of proteins with emergent functions of core relevance to life, such as spatiotemporal self-organization upon energy dissipation, is still in its infancy. Such complex biological functions only arise from finely tuned interactions with the cellular environment, such as other proteins and lipid membranes, and the ability to switch between different conformational states, which is not yet possible to design de novo by ML-based design ( 14 ).…”

Section: Mainmentioning

confidence: 99%

See 1 more Smart Citation

Designing a protein with emergent function by combinedin silico, in vitroandin vivoscreening

Kohyama¹,

Frohn

Babl

et al. 2023

Preprint

View full text Add to dashboard Cite

Recently, utilization of machine learning (ML)-based methods has led to astonishing progress in protein design and, thus, the design of new biological functionality. However, emergent functions that require higher-order molecular interactions, such as the ability to self-organize, are still extremely challenging to implement. Here, we describe a comprehensivein silico,in vitro, andin vivoscreening pipeline (i3-screening) to develop and validate ML-designed artificial homologs of a bacterial protein that confers its role in cell division through the emergent function of spatiotemporal pattern formation. Moreover, we present complete substitution of a wildtype gene by an ML-designed artificial homolog inEscherichia coli. These results raise great hopes for the next level of synthetic biology, where ML-designed synthetic proteins will be used to engineer cellular functions.

show abstract

Section: Mainmentioning

confidence: 99%

“…the ability to switch between different conformational states, which is not yet possible to design de novo by ML-based design (14).…”

mentioning

confidence: 99%

Designing a protein with emergent function by combinedin silico, in vitroandin vivoscreening

Kohyama¹,

Frohn

Babl

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…1−3 A central feature of protein dynamics is internal protein mobility. 4,5 Internal/local motions are necessarily spatially restricted/ordered by their immediate protein surroundings. They are characterized by a structural element (the local restrictions), a kinetic element (the type of motion and the pertinent rate constants), and related features of local geometry.…”

Section: Introductionmentioning

confidence: 99%

Local Structures in Proteins from Microsecond Molecular Dynamics Simulations: A Symmetry-Based Perspective

Pshetitsky,

Mendelman,

Buck

et al. 2024

J. Phys. Chem. B

View full text Add to dashboard Cite

We report on a new method for the characterization of local structures in proteins based on extensive molecular dynamics (MD) simulations, here, 1 μs in length. The N−H bond of the Rho GTPase binding domain of plexin-B1 (RBD) serves as a probe and the potential, u(MD), which restricts its internal motion, as a qualifier of the local dynamic structure. u(MD) is derived from the MD trajectory as a function of the polar angles, (θ, φ), which specify the N−H orientation in the protein. u(MD) is statistical in character yielding empirical descriptions. To establish more insightful methodical descriptions, we develop a comprehensive method which approximates u(MD) by combinations of analytical Wigner functions that belong to the D 2h point group. These combinations, called u(simulated), make it possible to gain a new perspective of local dynamic structures in proteins based on explicit potentials/free energy surfaces and associated probability densities, entropy, and ordering. A simpler method was developed previously using 100 ns MD simulations. In that case, the traditional "perpendicular N−H ordering" setting centered at C α −C α with (θ, φ) = (90, 90) and generally, featuring positive φ, prevailed. u(MD) derived from 1 μs MD simulations is considerably more complex requiring substantial model enhancement. The enhanced method applies to the well-structured sections of the RBD. It only applies partly to its loops where u(MD) extends into the negative-φ region where we detect nonperpendicular N−H ordering. This arrangement requires devising new reference structures and making substantial algorithmic changes, to be performed in future work. Here, we focus on developing the comprehensive method and using it to investigate perpendicular ordering settings. We find that secondary structures (loops) exhibit varying (virtually invariant) potentials with A g , B 2u , and B 1u (A g and B 2u ) D 2h symmetry. Application to RBD dimerization and RBD binding to the GTPase Rac1 is described in the subsequent article. Applications to other probes, proteins, and biological functions, based on explicit local potentials, probability densities, entropy, and ordering, are possible. 2 D 2,2 , where D 2,0 and D 2,2 are the Wigner functions with D 2h symmetry, D L,K , with L = 2 and K = 0 and 2 (cf. the Supporting Information (SI)). The D L,K functions, hence u, are given in terms of the polar angles, θ and φ, which specify the orientation of the N−H bond in the protein. 12 More enhanced potential forms comprising additional D L,K functions could not be used in view

show abstract

“…The benefit is particularly prominent in protein-relevant tasks, such as protein representation [zhang2022protein, ingraham2019generative, 29], protein folding [35], protein design [36,37], and protein-protein interaction [38][39][40][41][42][43]. Second, various research groups have developed new techniques to improve the interpretability of GNNs [24,44,45].…”

Section: Introductionmentioning

confidence: 99%

G-PLIP: Knowledge graph neural network for structure-free protein-ligand bioactivity prediction

Crouzet,

Lieberherr,

Atz

et al. 2023

Preprint

View full text Add to dashboard Cite

Protein-ligand interaction (PLI) shapes efficacy and safety profiles of small-molecule drugs. Most existing methods rely on resource-intensive computation to predict PLI based on structural information. Here we show that a light-weight graph neural network (GNN), trained with quantitative PLIs of a small number of proteins and ligands, is able to predict the strength of unseen PLIs. The model has no direct access to structural information of protein-ligand complexes. Instead, the predictive power is provided by encoded knowledge of proteins and ligands, including primary protein sequence, gene expression, protein-protein interaction network, and structural similarities between ligands. Our novel model performs competitively with or better than structure-aware models. Our results suggest that existing PLI-prediction methods may be further improved by using omics data and representation learning techniques that embed biological and chemical knowledge.

show abstract

Deep learning approaches for conformational flexibility and switching properties in protein design

Cited by 8 publications

References 73 publications

Designing a protein with emergent function by combinedin silico, in vitroandin vivoscreening

Designing a protein with emergent function by combinedin silico, in vitroandin vivoscreening

Local Structures in Proteins from Microsecond Molecular Dynamics Simulations: A Symmetry-Based Perspective

G-PLIP: Knowledge graph neural network for structure-free protein-ligand bioactivity prediction

Contact Info

Product

Resources

About