EGR: Equivariant Graph Refinement and Assessment of 3D Protein Complex Structures

Abstract

Protein complexes are macromolecules essential to the functioning and well-being of all living organisms. As the structure of a protein complex, in particular its region of interaction between multiple protein subunits (i.e., chains), has a notable influence on the biological function of the complex, computational methods that can quickly and effectively be used to refine and assess the quality of a protein complex's 3D structure can directly be used within a drug discovery pipeline to accelerate the development of new therapeutics and improve the efficacy of future vaccines. In this work, we introduce the Equivariant Graph Refiner (EGR), a novel E(3)-equivariant graph neural network (GNN) for multi-task structure refinement and assessment of protein complexes. Our experiments on new, diverse protein complex datasets, all of which we make publicly available in this work, demonstrate the state-of-the-art effectiveness of EGR for atomistic refinement and assessment of protein complexes and outline directions for future work in the field. In doing so, we establish a baseline for future studies in macromolecular refinement and structure analysis. 1 Preprint. Under review.

EGG: Accuracy Estimation of Individual Multimeric Protein Models Using Deep Energy-Based Models and Graph Neural Networks

“…The group names associated with our presented methods are denoted as EBM–{layer} and Regression–{layer}, with the layer being either “Transformer” or “MetaLayer”. Our models were trained as single-model predictors [ 8 , 17 , 18 , 19 ] and, therefore, we filtered all consensus-model predictors.…”

Improving Protein-Ligand Interaction Modeling with cryo-EM Data, Templates, and Deep Learning in 2021 Ligand Model Challenge

“…In the near future, we plan to develop advanced end-to-end deep learning architectures similar to some components in AlphaFold to predict better protein structures from cryo-EM maps and reference structures. Moreover, we plan to design 3D-equivariant deep learning architectures like SE(3)-equivariant Transformer network [23][24][25][26] to tackle the problem of geometric constraints which are not addressed by current methods. Finally, an end-to-end direct deep learning prediction of the structure of protein-ligand complexes from cryo-EM density maps, reference structures and ligand information to fully automate all the steps of the entire pipeline in this work is also an interesting direction to pursue.…”

A Survey of Deep Learning Methods for Estimating the Accuracy of Protein Quaternary Structure Models

“…More recently, several methods [20,24,27,28,30,31,[43][44][45][46][47][48][49] have used graphs to represent protein complex structures. Compared to other forms, graph structures offer several advantages: 1) they can effectively represent residue-residue or atom-atom interactions, with the capacity to easily assign chemical, physical, biological, and artificial features to the nodes and edges in the graph; 2) graph representations can scale to match protein structures of various complexity and size; and 3) they are particularly suitable for some deep learning algorithms such as graph neural networks while requiring fewer computational resources than 2D and 3D grid representations.…”

“…Statistical potential-based methods convert the distribution of distance-relevant or irrelevant pairwise contacts at atom or residue level into statistical potentials [8]. Machine learning or deep learning-based methods [11,13,[17][18][19][20][21][22][23][24][25][26][27][28][29][30][31] generally use features to represent a protein quaternary structure, which are then used to predict the structures quality scores. Similar to the protein tertiary structure EMA task, quaternary structure EMA methods can also be classified into multi-model [23,32] and single-model approaches [24,25].…”