GDockScore: a graph-based protein-protein docking scoring function

Abstract

Protein complexes play vital roles in a variety of biological processes such as mediating biochemical reactions, the immune response, and cell signalling, with three-dimensional structure specifying function. Computational docking methods provide a means to determine the interface between two complexed polypeptide chains without using time-consuming experimental techniques. The docking process requires the optimal solution to be selected with a scoring function. Here we propose a novel graph-based deep learning model that utilizes mathematical graph representations of proteins to learn a scoring function (GDockScore). GDockScore was pre-trained on docking outputs generated with the Protein Data Bank (PDB) biounits and the RosettaDock protocol, and then fine-tuned on HADDOCK decoys generated on the ZDOCK Protein Docking Benchmark. GDockScore performs similarly to the Rosetta scoring function on docking decoys generated using the RosettaDock protocol. Furthermore, state-of-the-art is achieved on the CAPRI score set, a challenging dataset for developing docking scoring functions.

EuDockScore: euclidean graph neural networks for scoring protein-protein interfaces

“…To test the models ability to classify near-native from incorrect antibody-antigen complexes we used a non-redundant set of antibody structures generated by docking with RosettaDock (Gray et al ., 2003) and relaxation with the Rosetta protocol as well (Tyka et al ., 2011). Local and global docking was performed as in GDockScore (McFee and Kim, 2023) to ensure we captured incorrect and near-native decoys in the set. Using a hold out, non-redundant test set we see that EuDockScore-Ab can differentiate near-native from incorrect docks in unseen data (Fig.…”

“…Here we have presented several EuDockScore scoring functions that can be used to accurately score protein-protein docks as well as antibody-antigen complexes. We iterate upon our previous work (McFee and Kim, 2023) by using modern, increasingly expressive Euclidean graph networks that can operate on raw coordinates (Liao and Smidt, 2023; Liao et al ., 2023) and supplemented the inputs with NLP embeddings from a BERT-style model (Geffen et al ., 2022).…”

“…Due to our previous success with GDockScore and the pre-training regime we established (McFee and Kim, 2023). We decided to replicate this training scheme for an antibody-antigen complex dataset.…”

“…Furthermore, groups have shown that combining scoring functions with AlphaFold-Multimer quality prediction output metrics may actually improve the filtering of protein-protein complex models. In particular, our previous model GDockScore (McFee and Kim, 2023) was used to screen AlphaFold-Multimer models for COVID-19 related interfaces (Tomezsko et al ., 2024). Thus, scoring and traditional docking remain relevant in the current realm of protein-protein interactions and structural bioinformatics.…”

“…EuDockScore outperforms state-of-the-art models on the challenging CAPRI score set for protein-protein docking As in our previous work GDockScore (McFee and Kim, 2023), we first pre-trained the model using docking decoys generated from solved structures in the PDB. Training the model resulted in lower training and validation losses in comparison to GDockScore and training converged in roughly half the number of previously required training iterations.…”