Predicting Protein-Peptide Interactions: Benchmarking Deep Learning Techniques and a Comparison with Focused Docking

Main methods and tools for peptide development based on protein-protein interactions (PPIs)

Protein-protein interactions (PPIs) regulate essential physiological and pathological processes. Due to their large and shallow binding surfaces, PPIs are often considered challenging drug targets for small molecules. Peptides offer a viable alternative, as they can bind these targets, acting as regulators or mimicking interaction partners. This review focuses on competitive peptides, a class of orthosteric modulators that disrupt PPI formation. We provide a concise yet comprehensive overview of recent advancements in in-silico peptide design, highlighting computational strategies that have improved the efficiency and accuracy of PPI-targeting peptides. Additionally, we examine cutting-edge experimental methods for evaluating PPI-based peptides. By exploring the interplay between computational design and experimental validation, this review presents a structured framework for developing effective peptide therapeutics targeting PPIs in various diseases.

Protein-Peptide Docking with ESMFold Language Model

Designing peptide therapeutics requires precise peptide docking, which remains a challenge. We assessed the ESMFold language model, originally designed for protein structure prediction, for its effectiveness in protein-peptide docking. Various docking strategies, including polyglycine linkers and sampling-enhancing modifications, were explored. The number of acceptable-quality models among top-ranking results is comparable to traditional methods and generally lower than AlphaFold-Multimer or Alphafold 3, though ESMFold surpasses it in some cases. The combination of result quality and computational efficiency underscores ESMFold's potential value as a component in a consensus approach for high-throughput peptide design.

Neutrophil elastase binds at the central domain of extracellular Toll-like receptor 4: AI prediction, docking, and validation in disease model

The interaction between Neutrophil Elastase (NE) and Toll-like receptor 4 (TLR4) has attracted substantial scientific attention, particularly regarding its potential role in cardiovascular diseases. Employing AlphaFold2, biomolecular docking, and MMGBSA calculation we aimed to predict their binding and validated the results through a co-immunoprecipitation study in a rat model with isoproterenol (ISO) -induced cardiac hypertrophy. Our findings strongly suggest a specific and plausible interaction between rat NE and rat TLR4, distinct from other neutrophil-derived serine proteases. Notably, AlphaFold2's precision was confirmed through cross-validation with known protein crystal structures, while Consurf analysis emphasized the evolutionary variable to conserve the rat NE - rat TLR4 binding site. HADDOCK, RosettaDock, ZDOCK, MD simulation, MMGBSA calculations, and superimposition with the stabilized structure complex all predicted strong binding between rat NE and rat TLR4. Our animal experiments revealed elevated NE and TLR4 expression in the hypertrophied myocardium following ISO infusion, with data confirming the physical interaction between NE and TLR4. Overall, this study sheds light on the intricate molecular association between NE and TLR4, underlining their potential significance in cardiovascular pathophysiology. Furthermore, it underscores AlphaFold2's reliability as a robust tool for predicting protein-protein interactions and complex structures.

Peptide design to control protein-protein interactions

Targeting of protein-protein interactions has become of huge interest in every aspect of medicinal and biological sciences. The control of protein interactions selectively offers the opportunity to control biological processes while limiting off target effects. This interest has massively increased with the development of cryo-EM and protein structure prediction with tools such as RosettaFold and AlphaFold. When designing molecules to control protein interactions, either inhibition or stabilisation, a starting point is commonly peptide design. This tutorial review describes that process, highlighting the selection of an initial sequence with and without structural information. Subsequently, methods for how the sequence can be analysed for key residues and how this information can be used to optimise the ligand efficiency are highlighted. Finally a discussion on how peptides can be further modified to increase their affinity and cell permeability, improving their drug-like properties, is presented.

Accurate sequence-to-affinity models for SH2 domains from multi-round peptide binding assays coupled with free-energy regression

Short linear peptide motifs play important roles in cell signaling. They can act as modification sites for enzymes and as recognition sites for peptide binding domains. SH2 domains bind specifically to tyrosine-phosphorylated proteins, with the affinity of the interaction depending strongly on the flanking sequence. Quantifying this sequence specificity is critical for deciphering phosphotyrosine-dependent signaling networks. In recent years, protein display technologies and deep sequencing have allowed researchers to profile SH2 domain binding across thousands of candidate ligands. Here, we present a concerted experimental and computational strategy that improves the predictive power of SH2 specificity profiling. Through multi-round affinity selection and deep sequencing with large randomized phosphopeptide libraries, we produce suitable data to train an additive binding free energy model that covers the full theoretical ligand sequence space. Our models can be used to predict signaling network connectivity and the impact of missense variants in phosphoproteins on SH2 binding.

Leveraging coevolutionary insights and AI-based structural modeling to unravel receptor-peptide ligand-binding mechanisms

Secreted signaling peptides are central regulators of growth, development, and stress responses, but specific steps in the evolution of these peptides and their receptors are not well understood. Also, the molecular mechanisms of peptide-receptor binding are only known for a few examples, primarily owing to the limited availability of protein structural determination capabilities to few laboratories worldwide. Plants have evolved a multitude of secreted signaling peptides and corresponding transmembrane receptors. Stress-responsive SERINE RICH ENDOGENOUS PEPTIDES (SCOOPs) were recently identified. Bioactive SCOOPs are proteolytically processed by subtilases and are perceived by the leucine-rich repeat receptor kinase MALE DISCOVERER 1-INTERACTING RECEPTOR-LIKE KINASE 2 (MIK2) in the model plant Arabidopsis thaliana. How SCOOPs and MIK2 have (co)evolved, and how SCOOPs bind to MIK2 are unknown. Using in silico analysis of 350 plant genomes and subsequent functional testing, we revealed the conservation of MIK2 as SCOOP receptor within the plant order Brassicales. We then leveraged AI-based structural modeling and comparative genomics to identify two conserved putative SCOOP-MIK2 binding pockets across Brassicales MIK2 homologues predicted to interact with the "SxS" motif of otherwise sequence-divergent SCOOPs. Mutagenesis of both predicted binding pockets compromised SCOOP binding to MIK2, SCOOP-induced complex formation between MIK2 and its coreceptor BRASSINOSTEROID INSENSITIVE 1-ASSOCIATED KINASE 1, and SCOOP-induced reactive oxygen species production, thus, confirming our in silico predictions. Collectively, in addition to revealing the elusive SCOOP-MIK2 binding mechanism, our analytic pipeline combining phylogenomics, AI-based structural predictions, and experimental biochemical and physiological validation provides a blueprint for the elucidation of peptide ligand-receptor perception mechanisms.

Insight into the Binding Interaction between PEDCs and hERRγ Utilizing Molecular Docking and Molecular Dynamics Simulations

Phenolic environmental endocrine-disrupting chemicals (PEDCs) are persistent EDCs that are widely found in food packaging materials and environmental media and seriously threaten human health and ecological security. Human estrogen-related receptor γ (hERRγ) has been proposed as a mediator for the low-dose effects of many environmental PEDCs; however, the atomic-level descriptions of dynamical structural features and interactions of hERRγ and PEDCs are still unclarified. Herein, how three PEDCs, 4-(1-methylpropyl)phenol (4-sec-butylphenol), 5,6,7,8-tetrahydro-2-naphthol (tetrahydro-2-napthol), and 2,2-bis(4-hydroxy-3,5-dimethoxyphenyl)propane (BP(2,2)(Me)), interact with hERRγ to produce its estrogenic disruption effects was studied. Molecular docking and multiple molecular dynamics (MD) simulations were first conducted to distinguish the detailed interaction pattern of hERRγ with PEDCs. These binding structures revealed that residues around Leu271, Leu309, Leu345, and Phe435 are important when binding with PEDCs. Furthermore, the binding energies of PEDCs with hERRγ were also characterized using the molecular mechanics/Poisson Boltzmann surface area (MM-PBSA) and solvated interaction energy (SIE) methods, and the results showed that the interactions of CH-π, π-π, and hydrogen bonds are the major contributors for hERRγ binding to these three PEDCs. What is striking is that the methoxide groups of BP(2,2)(Me), as hydrophobic groups, can help to reduce the binding energy of PEDCs binding with hERRγ. These results provide important guidance for further understanding the influence of PEDCs on human health problems.

Glutamine-derived peptides: Current progress and future directions

Glutamine, the most abundant amino acid in the body, plays a critical role in preserving immune function, nitrogen balance, intestinal integrity, and resistance to infection. However, its limited solubility and instability present challenges for its use a functional nutrient. Consequently, there is a preference for utilizing glutamine-derived peptides as an alternative to achieve enhanced functionality. This article aims to review the applications of glutamine monomers in clinical, sports, and enteral nutrition. It compares the functional effectiveness of monomers and glutamine-derived peptides and provides a comprehensive assessment of glutamine-derived peptides in terms of their classification, preparation, mechanism of absorption, and biological activity. Furthermore, this study explores the potential integration of artificial intelligence (AI)-based peptidomics and synthetic biology in the de novo design and large-scale production of these peptides. The findings reveal that glutamine-derived peptides possess significant structure-related bioactivities, with the smaller molecular weight fraction serving as the primary active ingredient. These peptides possess the ability to promote intestinal homeostasis, exert hypotensive and hypoglycemic effects, and display antioxidant properties. However, our understanding of the structure-function relationships of glutamine-derived peptides remains largely exploratory at current stage. The combination of AI based peptidomics and synthetic biology presents an opportunity to explore the untapped resources of glutamine-derived peptides as functional food ingredients. Additionally, the utilization and bioavailability of these peptides can be enhanced through the use of delivery systems in vivo. This review serves as a valuable reference for future investigations of and developments in the discovery, functional validation, and biomanufacturing of glutamine-derived peptides in food science.

Leveraging machine learning models for peptide-protein interaction prediction

Peptides play a pivotal role in a wide range of biological activities through participating in up to 40% protein-protein interactions in cellular processes. They also demonstrate remarkable specificity and efficacy, making them promising candidates for drug development. However, predicting peptide-protein complexes by traditional computational approaches, such as docking and molecular dynamics simulations, still remains a challenge due to high computational cost, flexible nature of peptides, and limited structural information of peptide-protein complexes. In recent years, the surge of available biological data has given rise to the development of an increasing number of machine learning models for predicting peptide-protein interactions. These models offer efficient solutions to address the challenges associated with traditional computational approaches. Furthermore, they offer enhanced accuracy, robustness, and interpretability in their predictive outcomes. This review presents a comprehensive overview of machine learning and deep learning models that have emerged in recent years for the prediction of peptide-protein interactions.

Thyroid Hormone Receptor Agonistic and Antagonistic Activity of Newly Synthesized Dihydroxylated Polybrominated Diphenyl Ethers: An In Vitro and In Silico Coactivator Recruitment Study

Dihydroxylated polybrominated diphenyl ethers (DiOH-PBDEs) could be the metabolites of PBDEs of some organisms or the natural products of certain marine bacteria and algae. OH-PBDEs may demonstrate binding affinity to thyroid hormone receptors (TRs) and can disrupt the functioning of the systems modulated by TRs. However, the thyroid hormone disruption mechanism of diOH-PBDEs remains elusive due to the absence of diOH-PBDEs standards. This investigation explores the potential disruptive effects of OH/diOH-PBDEs on thyroid hormones via competitive binding and coactivator recruitment with TRα and TRβ. At levels of 5000 nM and 25,000 nM, 6-OH-BDE-47 demonstrated significant recruitment of steroid receptor coactivator (SRC), whereas none of the diOH-PBDEs exhibited SRC recruitment within the range of 0.32-25,000 nM. AutoDock CrankPep (ADCP) simulations suggest that the conformation of SRC and TR-ligand complexes, particularly their interaction with Helix 12, rather than binding affinity, plays a pivotal role in ligand agonistic activity. 6,6'-diOH-BDE-47 displayed antagonistic activity towards both TRα and TRβ, while the antagonism of 3,5-diOH-BDE-100 for TRα and TRβ was concentration-dependent. 3,5-diOH-BDE-17 and 3,5-diOH-BDE-51 exhibited no discernible agonistic or antagonistic activities. Molecular docking analysis revealed that the binding energy of 3,3',5-triiodo-L-thyronine (T3) surpassed that of OH/diOH-PBDEs. 3,5-diOH-BDE-100 exhibited the highest binding energy, whereas 6,6'-diOH-BDE-47 displayed the lowest. These findings suggest that the structural determinants influencing the agonistic and antagonistic activities of halogen phenols may be more intricate than previously proposed, involving factors beyond high-brominated PBDEs or hydroxyl group and bromine substitutions. It is likely that the agonistic or antagonistic propensities of OH/diOH-PBDEs are instigated by protein conformational changes rather than considerations of binding energy.

Deep learning for advancing peptide drug development: Tools and methods in structure prediction and design

Peptides can bind challenging disease targets with high affinity and specificity, offering enormous opportunities for addressing unmet medical needs. However, peptides' unique features, including smaller size, increased structural flexibility, and limited data availability, pose additional challenges to the design process compared to proteins. This review explores the dynamic field of peptide therapeutics, leveraging deep learning to enhance structure prediction and design. Our exploration encompasses various facets of peptide research, ranging from dataset curation handling to model development. As deep learning technologies become more refined, we channel our efforts into peptide structure prediction and design, aligning with the fundamental principles of structure-activity relationships in drug development. To guide researchers in harnessing the potential of deep learning to advance peptide drug development, our insights comprehensively explore current challenges and future directions of peptide therapeutics.

From interaction networks to interfaces, scanning intrinsically disordered regions using AlphaFold2

The revolution brought about by AlphaFold2 opens promising perspectives to unravel the complexity of protein-protein interaction networks. The analysis of interaction networks obtained from proteomics experiments does not systematically provide the delimitations of the interaction regions. This is of particular concern in the case of interactions mediated by intrinsically disordered regions, in which the interaction site is generally small. Using a dataset of protein-peptide complexes involving intrinsically disordered regions that are non-redundant with the structures used in AlphaFold2 training, we show that when using the full sequences of the proteins, AlphaFold2-Multimer only achieves 40% success rate in identifying the correct site and structure of the interface. By delineating the interaction region into fragments of decreasing size and combining different strategies for integrating evolutionary information, we manage to raise this success rate up to 90%. We obtain similar success rates using a much larger dataset of protein complexes taken from the ELM database. Beyond the correct identification of the interaction site, our study also explores specificity issues. We show the advantages and limitations of using the AlphaFold2 confidence score to discriminate between alternative binding partners, a task that can be particularly challenging in the case of small interaction motifs.

OmeDDG: Improved Protein Mutation Stability Prediction Based on Predicted 3D Structures

Determining changes in the protein's thermal stability following mutations is critical in protein engineering and understanding pathogenic missense mutations. Despite the development of various computational methods to predict the effects of single-point mutations, their accuracy remains limited. In this study, we propose a new computational method, OmeDDG, that more accurately predicts mutation-induced Gibbs free energy changes in protein folding (ΔΔG). OmeDDG takes the sequences of wild-type and mutant proteins as input, utilizes OmegaFold to obtain the 3D structure, employs a convolutional neural network to extract structural features, and combines them with protein mutation features and pretraining features to predict the stability of single-point mutations in proteins. We performed a comprehensive comparison between OmeDDG and other available prediction methods on four blind test datasets, confirming that OmeDDG can effectively enhance protein mutation prediction performance. Notably, on the antisymmetric dataset Ssym, OmeDDG achieves the best performance, demonstrating favorable antisymmetry with PCC = 0.79 and RMSE = 0.96 for forward mutations and PCC = 0.77 and RMSE = 0.97 for reverse mutant types.

A Perspective on the Prospective Use of AI in Protein Structure Prediction

AlphaFold2 (AF2) and RoseTTaFold (RF) have revolutionized structural biology, serving as highly reliable and effective methods for predicting protein structures. This article explores their impact and limitations, focusing on their integration into experimental pipelines and their application in diverse protein classes, including membrane proteins, intrinsically disordered proteins (IDPs), and oligomers. In experimental pipelines, AF2 models help X-ray crystallography in resolving the phase problem, while complementarity with mass spectrometry and NMR data enhances structure determination and protein flexibility prediction. Predicting the structure of membrane proteins remains challenging for both AF2 and RF due to difficulties in capturing conformational ensembles and interactions with the membrane. Improvements in incorporating membrane-specific features and predicting the structural effect of mutations are crucial. For intrinsically disordered proteins, AF2's confidence score (pLDDT) serves as a competitive disorder predictor, but integrative approaches including molecular dynamics (MD) simulations or hydrophobic cluster analyses are advocated for accurate dynamics representation. AF2 and RF show promising results for oligomeric models, outperforming traditional docking methods, with AlphaFold-Multimer showing improved performance. However, some caveats remain in particular for membrane proteins. Real-life examples demonstrate AF2's predictive capabilities in unknown protein structures, but models should be evaluated for their agreement with experimental data. Furthermore, AF2 models can be used complementarily with MD simulations. In this Perspective, we propose a "wish list" for improving deep-learning-based protein folding prediction models, including using experimental data as constraints and modifying models with binding partners or post-translational modifications. Additionally, a meta-tool for ranking and suggesting composite models is suggested, driving future advancements in this rapidly evolving field.

Revolutionizing Peptide-Based Drug Discovery: Advances in the Post-AlphaFold Era

Peptide-based drugs offer high specificity, potency, and selectivity. However, their inherent flexibility and differences in conformational preferences between their free and bound states create unique challenges that have hindered progress in effective drug discovery pipelines. The emergence of AlphaFold (AF) and Artificial Intelligence (AI) presents new opportunities for enhancing peptide-based drug discovery. We explore recent advancements that facilitate a successful peptide drug discovery pipeline, considering peptides' attractive therapeutic properties and strategies to enhance their stability and bioavailability. AF enables efficient and accurate prediction of peptide-protein structures, addressing a critical requirement in computational drug discovery pipelines. In the post-AF era, we are witnessing rapid progress with the potential to revolutionize peptide-based drug discovery such as the ability to rank peptide binders or classify them as binders/non-binders and the ability to design novel peptide sequences. However, AI-based methods are struggling due to the lack of well-curated datasets, for example to accommodate modified amino acids or unconventional cyclization. Thus, physics-based methods, such as docking or molecular dynamics simulations, continue to hold a complementary role in peptide drug discovery pipelines. Moreover, MD-based tools offer valuable insights into binding mechanisms, as well as the thermodynamic and kinetic properties of complexes. As we navigate this evolving landscape, a synergistic integration of AI and physics-based methods holds the promise of reshaping the landscape of peptide-based drug discovery.

Growing ecosystem of deep learning methods for modeling protein-protein interactions

Numerous cellular functions rely on protein-protein interactions. Efforts to comprehensively characterize them remain challenged however by the diversity of molecular recognition mechanisms employed within the proteome. Deep learning has emerged as a promising approach for tackling this problem by exploiting both experimental data and basic biophysical knowledge about protein interactions. Here, we review the growing ecosystem of deep learning methods for modeling protein interactions, highlighting the diversity of these biophysically informed models and their respective trade-offs. We discuss recent successes in using representation learning to capture complex features pertinent to predicting protein interactions and interaction sites, geometric deep learning to reason over protein structures and predict complex structures, and generative modeling to design de novo protein assemblies. We also outline some of the outstanding challenges and promising new directions. Opportunities abound to discover novel interactions, elucidate their physical mechanisms, and engineer binders to modulate their functions using deep learning and, ultimately, unravel how protein interactions orchestrate complex cellular behaviors.