BRIDGE: A Coarse-Grained Architecture to Embed Protein-Protein Interactions for Therapeutic Applications
Khade, P. M.; Watkins, A. M.
Show abstract
In this work, we present Biophysical Representation of Interfaces via Delaunay-based Graph Embeddings (BRIDGE), a coarse-grained graph neural network that captures embeddings containing meaningful information about protein-protein interactions. The BRIDGE model takes as input graphs defined by the Delaunay tesselation of the individual chains and is pre-trained to predict the Delaunay adjacency at the protein-protein interface. The model achieves state-of-the-art performance in this task. The biophysical information captured by the BRIDGE embedding layer due to this pre-training task can further be used for downstream tasks, including for therapeutically relevant property prediction. We demonstrate the use of these embeddings by training models to predict antibody-antigen (Ab-Ag) affinity and antibody viscosity. These predictors, which require only the structures of the individual molecules in isolation, are especially well suited to large molecule therapeutic applications where complex structures are rarely determined until the very latest stages of a project.
Matching journals
The top 4 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- nanoBERT: A deep learning model for gene agnostic navigation of the nanobody mutational space 96%
- Estimating Protein Complex Model Accuracy Using Graph Transformers and Pairwise Similarity Graphs 95%
- SAINT-Angle: self-attention augmented inception-inside-inception network and transfer learning improve protein backbone torsion angle prediction 95%
Similar papers in this journal
- EGRET: Edge Aggregated Graph Attention Networks and Transfer Learning Improve Protein-Protein Interaction Site Prediction 96%
- CASTER-DTA: Equivariant Graph Neural Networks for Predicting Drug-Target Affinity 96%
- To pack or not to pack: revisiting protein side-chain packing in the post-AlphaFold era 96%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.