Protein Docking Model Evaluation by Graph Neural Networks

Abstract

Physical interactions of proteins play key functional roles in many important cellular processes. To understand molecular mechanisms of such functions, it is crucial to determine the structure of protein complexes. To complement experimental approaches, which usually take a considerable amount of time and resources, various computational methods have been developed for predicting the structures of protein complexes. In computational modeling, one of the challenges is to identify near-native structures from a large pool of generated models. Here, we developed a deep learning–based approach named Graph Neural Network–based DOcking decoy eValuation scorE (GNN-DOVE). To evaluate a protein docking model, GNN-DOVE extracts the interface area and represents it as a graph. The chemical properties of atoms and the inter-atom distances are used as features of nodes and edges in the graph, respectively. GNN-DOVE was trained, validated, and tested on docking models in the Dockground database and further tested on a combined dataset of Dockground and ZDOCK benchmark as well as a CAPRI scoring dataset. GNN-DOVE performed better than existing methods, including DOVE, which is our previous development that uses a convolutional neural network on voxelized structure models.

Quality Assessment of Protein Docking Models Based on Graph Neural Network

“…Furthermore, a top-k-pooling strategy was employed to select a fixed size of residue (node) embeddings from protein docking interfaces to form graph-level representations for prediction. Our GNN model is significantly different from an earlier GNN model for docking evaluation (Wang et al, 2021). In the definition of our graph, we treated the residues at docking models as nodes instead of atoms in that work, which reduced the graph complexity and computational time cost.…”

Protein design via deep learning

“…Many researches of this field focused on algorithm development and in silico evaluation with barely few experimental verifications and practical applications. Taking pharmacy and therapeutics as an example, although conventional drug discovery methodologies concentrated on molecular dynamics simulations and molecular docking [ 40 ] have made great achievements, protein design approaches are gradually showing their impressive capability and promising future. There are many roadmaps involving protein design in this field, which aim at various diseases afflicting human beings.…”

Deep Local Analysis evaluates protein docking conformations with Locally oriented Cubes

“…This database has 61 target complexes with on average 108 candidate conformations per target generated by the Fast Fourier Transform-based method GRAMM-X [45]. For comparison purpose we followed the experimental setups of GNN-DOVE [48]: in summary, 59 target complexes were chosen and divided into 4 non-redundant groups with respect to the sequence identity (less than 30%) and TM-score [52] (less than 0.5). On average each of these complexes has 9.83 acceptable conformations (L-RMSD ≤ 5.0 Å) and 98.5 incorrect ones.…”

“…In [13], SE(3)-equivariant hierarchical convolutions were applied to a point-cloud representation of the whole conformation. Finally, graph-based representations, as those used in GNN-DOVE [48] and DeepRank-GNN [41], are invariant to 3D rotations, but at the expense of losing information about the orientations of the atoms with respect to each other.…”