Rotamer Density Estimator is an Unsupervised Learner of the Effect of Mutations on Protein-Protein Interaction

Abstract

Protein-protein interactions are crucial to many biological processes, and predicting the effect of amino acid mutations on binding is important for protein engineering. While data-driven approaches using deep learning have shown promise, the scarcity of annotated experimental data remains a major challenge. In this work, we propose a new approach that predicts mutational effects on binding using the change in conformational flexibility of the protein-protein interface. Our approach, named Rotamer Density Estimator (RDE), employs a flow-based generative model to estimate the probability distribution of protein side-chain conformations and uses entropy to measure flexibility. RDE is trained solely on protein structures and does not require the supervision of experimental values of changes in binding affinities. Furthermore, the unsupervised representations extracted by RDE can be used for downstream neural network predictions with even greater accuracy. Our method outperforms empirical energy functions and other machine learning-based approaches.

PPFlow: Target-Aware Peptide Design with Torsional Flow Matching

“…Stability evaluates binding scores of the original pose of the generated peptides, which directly reflects the quality of generated peptides without re-docking. The stability scores are calculated by F old X (Schymkowitz et al, 2005) since it performs fast and accurate stability calculation in protein binding tasks compared with other energy-based methods (Luo et al, 2023). We give IMP%-S as the percentages of the designed ones with better stability than reference rather than average, because some extremely unstable structures will make the comparison on average values meaningless.…”

“…Recently, protein backbone generation methods employ flow matching techniques, to explore the applicability and effectiveness (Yim et al, 2023a;Bose et al, 2023). For protein side chains, methods usually focus on protein-protein complexes, such as RED-PPI (Luo et al, 2023) and DiffPack (Zhang et al, 2023a). Our side-chain packing methods follow RDE-PPI since it is pre-trained on a large amount of protein data, and achieves a good generalization performance on peptides.…”

DSMBind: SE(3) denoising score matching for unsupervised binding energy prediction and nanobody design

“…Baselines. Luo et al [24] reported a comprehensive list of baselines on the SKEMPI test set. They can be categorized into three groups: physics-based models, protein-language models, and supervised models.…”

“…• Supervised models. We consider two state-of-the-art baselines (MIFNet [45] and RDENet [24]) that have the best performance on SKEMPI. MIFNet is a masked inverse folding model similar to ESM-IF, but fine-tuned on the SKEMPI dataset with three-fold cross validation.…”

“…We validate DSMBind on three computational benchmarks related to protein-ligand binding prediction, protein-protein binding mutation effect prediction, and antibody-antigen binding prediction. We compare DSMBind with state-of-the-art binding prediction models including protein language models (ESM) [10,27], physics-based models [1,6], and supervised deep learning models [23,24,47] trained on experimental binding affinity data in the public domain. We also explore alternative EBM training objectives (e.g.…”