A deep-learning framework for multi-level peptide–protein interaction prediction

Abstract

Peptide-protein interactions are involved in various fundamental cellular functions and their identification is crucial for designing efficacious peptide therapeutics. Recently, a number of computational methods have been developed to predict peptide-protein interactions. However, most of the existing prediction approaches heavily depend on high-resolution structure data. Here, we present a deep learning framework for multi-level peptide-protein interaction prediction, called CAMP, including binary peptide-protein interaction prediction and corresponding peptide binding residue identification. Comprehensive evaluation demonstrated that CAMP can successfully capture the binary interactions between peptides and proteins and identify the binding residues along the peptides involved in the interactions. In addition, CAMP outperformed other state-of-the-art methods on binary peptide-protein interaction prediction. CAMP can serve as a useful tool in peptide-protein interaction prediction and identification of important binding residues in the peptides, which can thus facilitate the peptide drug discovery process.

Improving Peptide-Protein Docking with AlphaFold-Multimer using Forced Sampling

“…The samples of observed peptide-protein interaction complexes in the test set were based on the state-of-the art paper for peptide-protein interaction prediction by Lei et al (2021), which consisted of 262 peptide-protein complexes with experimentally solved structures in the PDB (Berman et al, 2000). After selecting one complex representative per ECOD (Schaeffer et al, 2017) family, the final redundancy reduced set consisted of 112 peptide-protein complexes.…”

“…For peptide-protein interaction prediction, a negative set was created in the same manner as in Lei et al (2021); by randomly pairing 560 (so ×5 as many negatives as positives) peptides and protein receptors from the positive set. Although randomly pairing proteins is no guarantee for lack of interaction, the resulting false negative rate will be statistically insignificant, especially considering the proteins hail from different species, and is as such common practice for constructing negative sets for protein-protein interaction prediction (Wang et al, 2019;Guo et al, 2008).…”

“…These are interactions between a protein and a smaller peptide fragment, sometimes referred to as short linear motif or SLiM, which can be part of a disordered region of a larger protein. Because peptide fragments display a high degree of flexibility and are often disordered when unbound, investigating the molecular details of such interactions and indeed even identifying direct interaction at all has been difficult both from an experimental and computational point of view, and specialized solutions have been developed to investigate peptide-protein interactions specifically (Petsalaki and Russell, 2008; Alam et al ., 2017; Lei et al ., 2021; Öztürk et al ., 2018; Johansson-Åkhe et al ., 2021).…”

Structure-based prediction of HDAC6 substrates validated by enzymatic assay reveals determinants of promiscuity and detects new potential substrates

“…Machine Learning-based approaches can be used for these aims (e.g. Hidden Markov Models 17 and naive Bayes 18 ), but such approaches depend on considerable amounts of data 19 – 21 . Moreover, enzyme substrate patterns may not always adequately be depicted by a sequence-based description, like in the case of O-glycosylation 22 or HIV-1 protease substrates 23 .…”

Computational design of dynamic receptor—peptide signaling complexes applied to chemotaxis

“…Unbound peptide ligands are often partially disordered in solution, which challenges structure determination, and computational sampling of the vast conformational space. In contrast to rigid protein binders and small-molecule ligands, structural information on peptide binding is scarce and limits supervised training and validation of deep-learning [5][6][7] and physics-based 8 protein:peptide complex structure prediction approaches. The specific receptor:peptide engineering problem is further complicated by the high flexibility of both receptor and peptide ligand which through mutual induced fit often adopt a new conformation together to reach the active state and initiate signal transduction.…”