Harnessing protein folding neural networks for peptide–protein docking

Abstract

Highly accurate protein structure predictions by deep neural networks such as AlphaFold2 and RoseTTAFold have tremendous impact on structural biology and beyond. Here, we show that, although these deep learning approaches have originally been developed for the in silico folding of protein monomers, AlphaFold2 also enables quick and accurate modeling of peptide–protein interactions. Our simple implementation of AlphaFold2 generates peptide–protein complex models without requiring multiple sequence alignment information for the peptide partner, and can handle binding-induced conformational changes of the receptor. We explore what AlphaFold2 has memorized and learned, and describe specific examples that highlight differences compared to state-of-the-art peptide docking protocol PIPER-FlexPepDock. These results show that AlphaFold2 holds great promise for providing structural insight into a wide range of peptide–protein complexes, serving as a starting point for the detailed characterization and manipulation of these interactions.

Brachypodium antifreeze protein gene products inhibit ice recrystallization, attenuate ice nucleation, and reduce immune response

“…Protein-protein docking predictions between various AFPs and INPs were performed using FRODOCK 18,19 . LRRs and flg22 binding predictions were modelled using AlphaFold with a 30 glycine linker connecting the amino terminal of the LRR to the flg22 peptide sequence as described 34 with the LRR carboxyl apoplast localization signal sequences omitted to allow for more representative binding events. The LRR-linker-flg22 models were opened in PyMOL (version 2.4.1) and each was independently selected as a separate object and the linker was hidden.…”

Bacterial Transcriptional Regulators: A Road Map for Functional, Structural, and Biophysical Characterization

“…Thus, to predict the holo form of a TR with its cognate ligand, it may be more insightful to use traditional homology modeling (template-based) using templates that contain the ligand coordinates since more accurate deep learning algorithm models of the apo form would be missing the ligand information entirely. This complication, currently present in the most recent models released by AlphaFold [7,91,92], will likely be solved once user selection of the appropriate ligand-bound template is allowed [93,94] or docking tools are incorporated into deep learning algorithm models [95]. Additionally, sequence similarity networks are a useful tool to provide multiple sequence alignments that would inform better structural predictions, as they have been shown to define isofunctional clusters in TR families [96].…”

Accurate Protein-Ligand Complex Structure Prediction using Geometric Deep Learning

“…These methods provided a novel computational approach to generate protein structures directly from their chemical sequences, rethinking the structure prediction paradigm in a data-driven perspective without following the conventional force-őeld assumption [19]. Recent progresses [20,21,22,23] demonstrated that using these methods as the protein/peptide structure estimator can predict precise structures for amino acid-based complexes such as the protein-protein and protein-peptide complexes, showing their generalizability to the downstream tasks. However, these methods are not speciőcally designed for SOM ligands, which inevitably restricted the related applications such as proteinligand complex structure prediction and affinity estimation.…”