Sage Journals: Discover world-class research

Abstract

A key step in most viral infections is the binding of a viral protein to a host receptor, leading to the virus entering the host cell. Disrupting this protein-protein interaction is an effective strategy for preventing infection and subsequent disease. Building on recent advances in computational tools for structural biology, we introduce Virus Inhibition via Peptide Engineering and Receptor Mimicry (VIPER), a novel approach for the automatic derivation and optimization of biomimetic decoy peptides that mimic binding sites of human proteins. VIPER leverages structural data from human-pathogen protein complexes, yielding peptides that can competitively inhibit viral entry by mimicking the natural receptor. We computationally validated VIPER using molecular dynamics simulations and showcased its applicability on three clinically relevant viruses, highlighting its potential to accelerate therapeutic development. With a focus on reproducibility and extensibility, VIPER can facilitate the rapid development of antiviral inhibitors by automating the design and optimization of biomimetic compounds.

1. INTRODUCTION

Recently, significant progress has been made in designing antiviral peptides, with major breakthroughs occurring during the Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic (Basit et al., 2021; Chatterjee et al., 2020; Chitsike et al., 2021; Curreli et al., 2020; Huang et al., 2020; Jaiswal et al., 2022; Karoyan et al., 2021; Larue et al., 2021; Odolczyk et al., 2021; Panda et al., 2021; Rajpoot et al., 2021; Renzi et al., 2023). The key idea behind these novel inhibitors is to mimic the binding partner that the viral surface protein (VSP) uses to dock to the human receptor. These biomimetic peptides serve as “decoys” that bind directly to the VSP and prevent it from attaching to and entering human cells. While most work has focused on basic research and preclinical efforts, where antiviral peptides so far have shown promise (Jackman, 2022), a biomimetic antiviral peptide inhibitor, Enfuvirtide, has already been approved for clinical use by the FDA (Matthews et al., 2004).

Improvements in in silico approaches (Grisoni et al., 2018; Müller et al., 2018; Plisson et al., 2020; Yoshida et al., 2018; Zakharova et al., 2022) have contributed to the progress seen in peptide engineering in the past few years (Fosgerau and Hoffmann, 2015; Mustafa et al., 2018; Naeimi et al., 2022; Pountos et al., 2016), by reducing the time and cost needed for wet lab experiments. Recently, Google’s DeepMind released AlphaProteo, a machine-learning model designed to generate proteins that bind to target proteins (Zambaldi et al., 2024). However, the model appears to focus on generating larger proteins (>50 amino acids) rather than small peptides. Moreover, AlphaProteo generates these proteins de novo, without mimicking existing human proteins, which could potentially affect the immunogenicity of the resulting binding proteins. Therefore, to the best of our knowledge, there is no dedicated computational method that facilitates the design of biomimetic antiviral entry inhibitor peptides. To fill this gap, we developed Virus Inhibition via Peptide Engineering and Receptor Mimicry (VIPER), a computational approach to automatically design and optimize decoy peptides using the structural data of the human and pathogen proteins in a complex. VIPER is released as an extensible open-source Python program, with emphasis on reproducibility and customizability, and an extensive manual documenting the software is provided in the GitHub repository.

2. METHODS

VIPER is a computational approach that automatically derives potential entry inhibitor peptides from the structural data of a viral protein bound to the human receptor. Its modular design allows for customizations and extensions of the core components. The workflow can be divided into four components: (1) input, (2) complex analysis, (3) residue selection, and (4) iterative improvement. An overview of the VIPER control flow is given in Figure 1.

FIG. 1.

An overview of VIPER’s control flow. VIPER, Virus Inhibition via Peptide Engineering and Receptor Mimicry.

2.1. Input and complex analysis steps

The only necessary input files for VIPER are: (1) a Protein Data Bank (PDB) structure of a VSP bound to a receptor and (2) a configuration file that specifies all settings and options. VIPER makes extensive use of the RosettaCommons software suite (Leaver-Fay et al., 2011) to relax the structures for downstream analysis, with the goal of producing structures that are likely to be more energetically favorable (Conway et al., 2014; Khatib et al., 2011; Nivón et al., 2013; Tyka et al., 2011). All relaxations are performed using the standard relax application, except for additional restraints to the heavy atom positions to avoid deviating too much from the crystal structure. However, these restraints are dropped during the iterative improvement step of the peptide, as there is no experimentally solved structure for the peptide-VSP complex. Further details and all default parameters can be found in Supplementary Data (Supplementary Fig. S1 and Fig. S2).

The residue_energy_breakdown1 (part of the RosettaCommons suite) is used to identify residues strongly involved in binding between the receptor and VSP, as measured by the internal RosettaCommons score function (Alford et al., 2017). Running this application generates a score file that specifies the different types of energy terms such as electrostatic or van der Waals at a per-residue level. This information can then be used to rank residues based on their contribution to the total binding energy score.

2.2. Residue selection step

After relaxing the structures and computing the per-residue binding energy, the next step involves identifying a subset of candidate residues (implemented in VIPER in the FragmentJoiner module). FragmentJoiner performs a linear scan through the residues on the receptor chain(s) and starts a fragment when it encounters a residue with a negative (i.e., favorable) energy. This fragment is then extended until a residue is found that does not interact favorably with the VSP, denoted by a positive interaction energy value. At this point, a configurable lookahead window is scanned for favorably interacting residues. If such a residue is found and its inclusion outweighs the penalty of incorporating the residues within the gap, the fragment is extended. If no such residue is found, the fragment is terminated. This procedure is illustrated in Figure 2.

FIG. 2.

A schematic view of the FragmentJoiner logic. Fragments are identified via a gap-tolerant forward scan of the residues and then combined to maximize the total energy. Only termini of different type (N/C, shown as gray and white bars) are eligible to be combined.

Once all fragments are identified, FragmentJoiner attempts to find the best combination of fragments, that is, the combination with the lowest energy, while staying within the user-defined peptide length limit. Of note, the algorithm can also join connected subsequences of fragments, such that the full-length budget may be used instead of an “all-or-nothing” approach. FragmentJoiner selects the next best-interacting fragment and only adds residues from this fragment up to the residue count limit, starting from the terminus closest to the current set of candidate residues. Therefore, since FragmentJoiner tries to add all residues in a fragment, the only fragments that may be joined only in part are the ones that end up at either end of the final combination of fragments.

Furthermore, FragmentJoiner only considers combinations of fragments where the N-terminus of one fragment is no further away from the C-terminus of the other fragment than a user-defined distance. This distance is equal to the approximate length of the configurable amino acid sequence used as a linker between fragments, whose length can be specified in the configuration options. This constraint ensures that only fragments which could potentially maintain their original relative positions in the receptor protein are joined in the final peptide. For example, two fragments that are 80 Å apart in the receptor would not be joined if the linker is set to be only 10 Å long.

2.3. Iterative improvement

The steps outlined above generate a single candidate peptide derived from segments of the native human receptor sequence. However, introducing single amino acid substitutions at select spots in the sequence might result in a peptide with a stronger binding affinity for the VSP. Furthermore, conformational peptide stability is another important metric that should be factored in when ranking the different peptides generated by VIPER.

Therefore, to further improve the constructs provided by VIPER’s FragmentJoiner, we implemented a genetic algorithm (GA) optimization approach. The fitness function is a combination of the Rosetta energy function (Alford et al., 2017) and a modified spatial side chain interaction index (sSCII), which provides an estimate of the conformational stability of a peptide. The sSCII is used to derive a configurable percentage-based modifier for the Rosetta energy function, ranging from $- 20 %$ to $+ 30 %$ by default. The default settings, which we used for this work, were five generations of 10 individuals with a weighted parent selection, random crossover, and a $5 %$ mutation rate at each position.

The individuals are generated through the GA operators: (1) crossover of the amino acid sequence of two parents and (2) mutation (i.e., an amino acid substitution). The first generation starts with the amino acid sequence generated through the previously described residue selection and randomly mutated derivatives of this “seeding” sequence. The individual peptides all have their tertiary structure predicted and subsequently relaxed.

All specific settings are explained in detail among the other VIPER settings in the manual in the GitHub repository.2 The default run configuration provided in the repository was used for all case studies, only updating the chain identifiers (“vsp_chain,” “partner_chain”) according to the corresponding PDB files.

2.3.1. Side chain interaction index

The sSCII is an extension of the side chain interaction index developed by Gehenn et al. (Gehenn et al., 2004, 2006) that we modified to take conformational information into account. The original SCII works by averaging an index value computed for every amino acid in the peptide. This index value is computed by summing up all contact propensities that are larger than one and where the associated partner amino acid is present in the peptide and dividing this sum by the sum of all contact propensities larger than one (Gehenn et al., 2004, 2006). The contact propensity is a measure introduced by Singh and Thornton (1992) to quantify how often and how well amino acid side chains pack against each other compared with the expected frequencies derived from a representative experimental dataset. A contact propensity larger than one indicates a more frequently observed interaction, meaning a favorable interaction. The original formulation is given in Equation 1. $SCII = \frac{1}{n} [{(\frac{\sum i_{f}}{\sum i_{t}})}_{a_{1}} + {(\frac{\sum i_{f}}{\sum i_{t}})}_{a_{2}} + \dots + {(\frac{\sum i_{f}}{\sum i_{t}})}_{a_{n}}]$ (1)

Here $\sum i_{f}$ refers to the total sum of unique favorable contact propensities available to the n-th amino acid of the peptide; $\sum i_{t}$ corresponds to the total sum of advantageous contact propensities available for the n-th amino acid of the peptide among all other 19 canonical amino acids; $a_{i}$ represents the i-th amino acid of the peptide, and n the total number of amino acids in the peptide (Gehenn et al., 2006).

In the original SCII (see Equation 1), an index value is calculated for each amino acid in the sequence, which is then averaged. This per-amino acid value incorporates all inter-residue interactions, regardless of the distance between them or even the order of amino acids in the sequence. Therefore, the SCII for a sequence and a permutation of the same sequence are identical. Since protein sequences have been optimized by evolution, we reasoned that the structural information as well as the order of the amino acids in the sequence is important to the tertiary structure and the function of the protein. Hence, we modified the original sSCII to incorporate structural information by only considering residues within a certain distance of the residue in question during the per-residue index calculation step of the SCII. The rationale behind this choice is that only residues that are close to each other can reasonably have their side chains interact. We discuss how we determined this distance and how we validated our modified sSCII in section 2.4.

2.3.2. Secondary structure analysis

VIPER can perform secondary structure analysis of the unbound and bound forms of candidate peptides using Dictionary of Secondary Structure in Proteins (DSSP). VIPER saves the DSSP data, logs differences in secondary structure between bound and unbound states, and generates a json file detailing residues with mismatched secondary structure assignments. It also calculates and logs the percentage of residues affected. This feature is particularly valuable for identifying candidates that may incur an entropic penalty when transitioning from a disordered to an ordered state upon binding. Please consult the manual for additional details.

2.4. Computational validation

2.4.1. Spatial side chain interaction index

The sSCII was validated by gathering peptides between 7 and 25 residues in length from UniProt (The UniProt Consortium, 2023) and StraPep (Wang et al., 2018) datasets, randomly scrambling their sequence, predicting the tertiary structure for the newly generated peptide, and comparing the sSCII values derived from the scrambled sequences against those of the original sequences.

Only peptides consisting of a single chain and containing canonical amino acids were used, resulting in 1309 peptides. The original structures of these peptides were collected from the PDB (Burley et al., 2023) based on the PDB accession code listed in the databases, whereas the structure of the scrambled peptides was predicted using OmegaFold (Wu et al., 2022), which can run locally and return a prediction within seconds. The sSCII was then calculated for the original and the scrambled peptide, at 7, 10, 12, 15, and 25 Å radii. To identify the optimal threshold to discern “unstable” from “stable” peptides, the $F_{2}$ score (Chinchor, 1992) was calculated for each threshold for the radius maximizing the area under the receiver operating characteristic (ROC) curve. The score is given by the formula shown in Equation 2. $F_{β} = \frac{(β^{2} + 1.0) \cdot precision \cdot recall}{β^{2} \cdot precision + recall}$ (2)with $precision = \frac{true positives}{true positives + false positives}$ (3) $recall = \frac{true positives}{true positives + false negatives}$ (4)

As it would be more detrimental if a potentially stable candidate would be excluded than if a potentially unstable candidate included, a β of 2 was chosen to assign more importance to the recall of the classifier.

2.4.2. Case studies

VIPER was run on SARS-CoV-2 (PDB code: 6m0j), HIV (PDB code: 6meo), and HSV (PDB code: 1jma). The run configuration can be viewed in the GitHub repository. In short, the maximum length was set to 18 residues and the resultant peptide was improved through five generations of 10 individuals using the built-in GA optimization.

To evaluate the generated peptides and whether the GA improved on the original solution, 100-nanosecond long molecular dynamics (MD) simulations of the peptide-VSP complex were performed using GROMACS 2023.4 (Abraham et al., 2015, 2024; Páll et al., 2015) with the OPLS/AA forcefield (Jorgensen et al., 1996) (as implemented in GROMACS) and the binding free energy calculated using the gmx_mmpbsa package v.1.6.3 (Valdés-Tresanco et al., 2021) and the Poisson-Boltzmann model, with default parameters. This procedure was applied to the original peptide before any GA modifications, to the best peptide as reported by the GA, and to an intermediary solution produced by the GA optimization, if available.

2.5. Implementation

VIPER was developed with Python 3.8 and incorporates several other tools as external dependencies, most notably the RosettaCommon software suite (Leaver-Fay et al., 2011) and PEPstrMOD (Kaur et al., 2007; Singh et al., 2015) for the prediction of the tertiary structure of peptides. VIPER also makes use of the Biopython package (Cock et al., 2009), among others which are listed in the code repository.

Furthermore, VIPER includes extensive logging to ensure reproducibility and to make the automated decisions transparent and traceable. It provides a supported mechanism for customizing its behavior aside from configuration options, where the user can write their own logic in a special source file in which function hooks are already provided, sparing the user from manually finding and modifying the specific part in the code base where they would like their code injected. Additionally, the GA implementation can evolve multiple populations in parallel and uses the Python built-in multiprocessing module to speed up execution.

3. RESULTS

3.1. Spatial side chain interaction index

To computationally validate the sSCII peptide stability metric and to identify the optimal radius for the sSCII calculations, we obtained a total of 1309 unique peptide sequences, identified through the procedure outlined in Section 2. We then scrambled the sequences and compared the ability of the sSCII metric to discriminate between the native sequences and the scrambled ones, building different ROC curves for increasing values of the radius used in the sSCII calculation (see Section 2).

A plot of the ROC curves for the sSCII with a radius of 7, 10, 12, 15, or 25 Å is shown in Figure 3. It can be seen that the 7 Å radius has an area under the curve of 0.87, the highest of all tested configurations. Therefore, this configuration was selected for further use. As the sSCII returns a value between 0 and 1, it is necessary to define a threshold above which peptides are considered stable. The threshold that maximizes the $F_{2}$ score is circled in this plot and corresponds to 0.4063, with an $F_{2}$ score of approximately 0.7413. Incidentally, this threshold also maximizes Youden’s J statistic (Youden, 1950).

FIG. 3.

A plot of the receiver operator characteristic curves for different radii and the corresponding area under the curve (AUC). An at-random classifier is shown as an unmarked line. As a guide, dashed lines on which every point has the same $F_{2}$ score are included. The threshold maximizing this score is circled on the 7 Å curve.

3.2. Case studies

To showcase the applicability of VIPER to clinically relevant viral infections, we ran three case studies for SARS-CoV-2, Human Immunodeficiency Virus (HIV), and Herpes Simplex Virus (HSV), respectively, using VIPER with standard parameters. The final, optimized putative peptide inhibitors for SARS-CoV-2 and HSV are shown in Figure 4, while results for HIV are shown in Supplementary Data (Supplementary Figure S3). As shown in Figure 4, the peptides approximate the conformation of the receptor at the main binding interface of the proteins.

FIG. 4.

Visualizations for SARS-CoV-2 and HSV. The left-hand side shows the peptide bound to the viral surface protein, whereas the right-hand side shows the per-residue analysis of the receptor bound to the viral surface protein. The residues selected by VIPER to be part of the initial peptide are shown with a magenta outline.

3.2.1. SARS-CoV-2

The “SARS-Residue Selection” panel in Figure 4 shows that VIPER with its default configuration mainly selects an $α$ -helix situated deeply in the binding site, which is associated with strong interaction energy scores between the receptor and the SARS-CoV-2 VSP. Notably, this section of the receptor was also selected as the major part of the inhibitory peptide designed and experimentally validated by Renzi et al. (2023). The initial peptide generated by VIPER had the sequence of EQAKTFLDKFNHGGED after insertion of the polyglycine (GG) linker, which becomes EQAKTWV DKM NHGN ED after optimization.

3.2.2. HSV

The surface protein of HSV, a protein called gD, can utilize several different cellular receptors, namely herpesvirus entry mediator (HVEM), nectin-1, or 3-O-sulfated heparan sulfate (Hilterbrand et al., 2021). As a proof of concept, we chose HVEM as the target for VIPER. The “HSV” panels in Figure 4 show that the binding between gD and HVEM is mainly mediated by a long, linear interface between a β-strand in HVEM and a less structurally defined coil region in HSV gD. VIPER selects most of this section, as it is the largest one where multiple close residues interact strongly with the VSP. It is the only selected section though, as the sections are too far from each other to be joined by the standard linker. The initial amino acid sequence generated by VIPER was CGELTGTVCEP, which became CGELTGTVCV P after optimization.

3.3. Genetic algorithm

To computationally test whether the GA optimization resulted in increased peptide binding affinities for the viral proteins, we performed binding energy calculations with gmx_mmpbsa on MD simulation data (see Section 2 for more details), in addition to the GA score (Rosetta scoring function) proxy. In all tested scenarios, the GA resulted in peptides with higher binding affinity (i.e., lower Rosetta energy units) than the original peptide. Overall, the improvements in the score were sizable, consistent, and well aligned with the results obtained from MD simulation data (Fig. 5). While all peptides initially identified by VIPER had a total negative (i.e., favorable) energy, the GA optimization with default parameters identified at least one peptide for each case study with lower energy, demonstrating the benefit of using the optimization step. Remarkably, values obtained from the GA score (which uses the Rosetta score function) were in very good agreement with those obtained from the MD simulations (Fig. 5).

FIG. 5.

Binding free energy calculation results compared with the GA score for the initial, intermediate (if available), and final peptide. The curves in black refer to the gmx_mmpbsa calculations (kcal/mol), whereas the gray curves refer to the GA score (Rosetta energy units). GA, genetic algorithm.

The actual values for the simulations and projected values (GA scores) are listed in Supplementary Data (Supplementary Table S1).

4. DISCUSSION

In this work, we introduce our tool VIPER to design biomimetic viral entry inhibitors using complexes of viral proteins bound to human receptors. VIPER is fully automated and can optionally improve the initial construct by introducing mutations via GA optimization. As the biological activity of the peptide is closely linked to its structure, we sought to include and improve upon a metric to determine the conformational stability of a peptide. We reasoned that improving not just the interaction energy terms but also the conformational stability of a peptide might lead to more desirable candidates. Therefore, we also extended and computationally validated an existing conformational stability measure for peptides (Gehenn et al., 2004, 2006), for use in the GA fitness function. We then showcased VIPER’s use on three model viruses.

While this tool provides users with a structured framework for deriving decoy peptides, there are several limitations that we wish to highlight. Most notably, the computational validation was performed entirely in silico. However, in a previous work (Renzi et al., 2023), we used a similar strategy to the automated workflow in VIPER to design an inhibitory peptide for SARS-CoV-2 that was experimentally validated and shown to actually inhibit the viral entry into human cells, demonstrating the feasibility of the approach.

Another caveat is that the structure of the peptide is predicted independently of the structure of the VSP, which likely neglects any possible interactions that might influence the distribution of conformations adopted by the peptide. Additionally, as an external web server is used for the prediction of the tertiary structure, there is some degree of non-determinism in this tool, although VIPER sets all random seeds for all local tools. This could be an area for future work, especially as computational structure prediction methods have recently made large advances, as can be seen with the AlphaFold family of models (Abramson et al., 2024; Jumper et al., 2021). Another limitation is the fact that the input structure only represents a single point in time, whereas protein-protein interactions can be highly dynamic. Moving from singular peptides to peptide ensembles may alleviate this issue and is therefore a promising avenue for future research.

As the AlphaFold family of models (Abramson et al., 2024; Jumper et al., 2021) supports the structure prediction of multimer proteins, a potential application of these and other forthcoming models is to predict the structure of VSP-receptor complexes if no experimentally solved structure is available and to use these models to generate multiple conformations of the complex to more closely model the dynamic nature of protein-protein interactions.

Deep learning may also help improve the accuracy of VIPER in other ways. Currently, the interaction between proteins is scored using the Rosetta score function (Alford et al., 2017), which relies on physics- and knowledge-based score terms. These, however, do not explicitly model some non-canonical interactions like $π$ - $π$ or cation- $π$ (Alford et al., 2017), which have been recently shown to be abundant in and significantly contribute to many protein interactions (Tsuboyama et al., 2023). Indeed, Biel et al. showed that these interactions can be crucial for designing proteins that bind to other proteins (Biel et al., 2017). Deep learning-based score functions may not suffer from this issue and should be straightforward to implement, given VIPER’s modular structure. Furthermore, a user could choose to forego the biomimetic aspect of VIPER and develop a module that incorporates a deep learning-based de novo design mechanism like RFdiffusion (Watson et al., 2023) to broaden the design space explored by VIPER.

Another area for future work is incorporating advanced structural and post-translation modifications like cyclization or acetylation of the peptide, which may improve the pharmaceutical properties of the peptide (Ebrahimi and Samanta, 2023; Wang, 2012; Wang et al., 2022).

Finally, VIPER is agnostic to the host and virus and works purely on the basis of the receptor protein and VSP. This makes VIPER applicable in a wide variety of scenarios, including the development of new antivirals for agriculture, where viruses can cause substantial damage to crops (Tatineni and Hein, 2023), or in animal husbandry (Cui et al., 2023).

Footnotes

ACKNOWLEDGMENTS

The authors thank Austin Seamann and all members of the Ghersi lab for helpful discussions and suggestions.

AUTHORS’ CONTRIBUTIONS

D.G.: Conceptualization, methodology, writing. A.S.K.: Methodology, software, writing. All authors have read and approved the article.

CODE AND DATA AVAILABILITY

All code is available at https://github.com/A-Klingenberg/VIPER/tree/main with an extensive handbook documenting all configuration options and internals of the software tool. The tool is released under the MIT license and also indexed and archived on Zenodo at for archival stability.

AUTHOR DISCLOSURE STATEMENT

The authors have no conflicts of interest to declare.

FUNDING INFORMATION

The authors received no financial support for this research.

SUPPLEMENTARY MATERIAL

1

2

References

Abraham

, Murtola

, Schulz

, et al. Gromacs: High performance molecular simulations through multi-level parallelism from laptops to supercomputers. SoftwareX, 2015; 1–2:19–25; ISSN 2352-7110; doi: 10.1016/j.softx.2015.06.001

Abraham

, Alekseenko

, Basov

, et al. Gromacs 2024. 2 Manual, 2024; doi: 10.5281/ZENODO.11148638

Abramson

, Adler

, Dunger

, et al. Accurate structure prediction of biomolecular interactions with alphafold 3. Nature, 2024; 630(8016):493–500; ISSN 1476-4687; doi: 10.1038/s41586-024-07487-w

Alford

, Leaver-Fay

, Jeliazkov

, et al. The rosetta all-atom energy function for macromolecular modeling and design. J Chem Theory Comput, 2017; 13(6):3031–3048; ISSN 1549-9626; doi: 10.1021/acs.jctc.7b00125

Basit

, Karim

, Asif

, et al. Designing short peptides to block the interaction of sars-cov-2 and human ace2 for covid-19 therapeutics. Front Pharmacol, 2021; 12:731828.

Biel

, Thompson

, Cunningham

, et al. Flexibility and design: Conformational heterogeneity along the evolutionary trajectory of a redesigned ubiquitin. Structure, 2017; 25(5):739–749.e3; ISSN 0969-2126; doi: 10.1016/j.str.2017.03.009

Burley

, Bhikadiya

, Bi

, et al. Rcsb protein data bank (rcsb.org): delivery of experimentally-determined pdb structures alongside one million computed structure models of proteins from artificial intelligence/machine learning. Nucleic Acids Res, 2023; 51(D1):D488–D508; ISSN 1362-4962; doi: 10.1093/nar/gkac1077

Chatterjee

, Ponnapati

, Kramme

, et al. Targeted intracellular degradation of sars-cov-2 via computationally optimized peptide fusions. Commun Biol, 2020; 3(1):715.

Chinchor

. Muc-4 evaluation metrics. In Proceedings of the 4th conference on Message understanding—MUC4 ‘92, MUC4 ‘92. Association for Computational Linguistics, 1992; doi: 10.3115/1072064.1072067

10.

Chitsike

, Krstenansky

, Duerksen-Hughes

. Ace2: S 1 rbd interaction-targeted peptides and small molecules as potential covid-19 therapeutics. Adv Pharmacol Pharm Sci, 2021; 2021:1828792.

11.

Cock

PJA

, Antao

, Chang

, et al. Biopython: Freely available Python tools for computational molecular biology and bioinformatics. Bioinformatics, 2009; 25(11):1422–1423; ISSN 1367-4803; doi: 10.1093/bioinformatics/btp163

12.

Conway

, Tyka

, DiMaio

, et al. Relaxation of backbone bond geometry improves protein energy landscape modeling. Protein Sci, 2014; 23(1):47–55; doi: 10.1002/pro.2389 ISSN 1469-896X

13.

Cui

, Fan

, Liang

, et al. Virus diversity, wildlife-domestic animal circulation and potential zoonotic viruses of small mammals, pangolins and zoo animals. Nat Commun, 2023; 14(1):2488.

14.

Curreli

, Victor

SMB

, Ahmed

, et al. Stapled peptides based on human angiotensin-converting enzyme 2 (ace2) potently inhibit sars-cov-2 infection in vitro. mBio, 2020; 11(6):e02451.

15.

Ebrahimi

, Samanta

. Engineering protein-based therapeutics through structural and chemical design. Nat Commun, 2023; 14(1):2411; ISSN 2041-1723; doi: 10.1038/s41467-023-38039-x

16.

Fosgerau

, Hoffmann

. Peptide therapeutics: Current status and future directions. Drug Discov Today, 2015; 20(1):122–128; doi: 10.1016/j.drudis.2014.10.003 ISSN 1359-6446.

17.

Gehenn

, Pipkorn

, Reed

. Successful design and synthesis of a polarity-triggered beta alpha conformational switch using the side chain interaction index (scii) as a measure of local stuctural stability. Biochemistry, 2004; 43(3):607–612; ISSN 1520-4995; doi: 10.1021/bi0301744

18.

Gehenn

, Stege

, Reed

. The side chain interaction index as a tool for predicting fast-folding elements and the structure and stability of engineered peptides. Anal Biochem, 2006; 356(1):12–17; ISSN 0003-2697; doi: 10.1016/j.ab.2006.06.021

19.

Grisoni

, Neuhaus

, Gabernet

, et al. Designing anticancer peptides by constructive machine learning. ChemMedChem, 2018; 13(13):1300–1302; doi: 10.1002/cmdc.201800204

20.

Hilterbrand

, Daly

, Heldwein

. Contributions of the four essential entry glycoproteins to hsv-1 tropism and the selection of entry routes. mBio, 2021; 12(2); ISSN 2150-7511; doi: 10.1128/mbio.00143-21

21.

Huang

, Pearce

, Zhang

. De novo design of protein peptides to block association of the sars-cov-2 spike protein with human ace2. Aging (Albany NY), 2020; 12(12):11263–11276.

22.

Jackman

. Antiviral peptide engineering for targeting membrane-enveloped viruses: Recent progress and future directions. Biochim Biophys Acta Biomembr, 2022; 1864(2):183821; ISSN 0005-2736; doi: 10.1016/j.bbamem.2021.183821

23.

Jaiswal

, Yaduvanshi

, Kumar

. A potential peptide inhibitor of sars-cov-2s and human ace2 complex. J Biomol Struct Dyn, 2022; 40(14):6671–6681.

24.

Jorgensen

, Maxwell

, Tirado-Rives

. Development and testing of the opls all-atom force field on conformational energetics and properties of organic liquids. J Am Chem Soc, 1996; 118(45):11225–11236; ISSN 1520-5126; doi: 10.1021/ja9621760

25.

Jumper

, Evans

, Pritzel

, et al. Highly accurate protein structure prediction with alphafold. Nature, 2021; 596(7873):583–589; ISSN 1476-4687; doi: 10.1038/s41586-021-03819-2

26.

Karoyan

, Vieillard

, Gómez-Morales

, et al. Human ace2 peptide-mimics block sars-cov-2 pulmonary cells infection. Commun Biol, 2021; 4(1):197.

27.

Kaur

, Garg

, Raghava

. Pepstr: A de novo method for tertiary structure prediction of small bioactive peptides. Protein Pept Lett, 2007; 14(7):626–631; doi: 10.2174/092986607781483859

28.

Khatib

, Cooper

, Tyka

, et al. Algorithm discovery by protein folding game players. Proc Natl Acad Sci U S A, 2011; 108(47):18949–18953; ISSN 1091-6490; doi: 10.1073/pnas.1115898108

29.

Larue

, Xing

, Kenney

, et al. Rationally designed ace2-derived peptides inhibit sars-cov-2. Bioconjug Chem, 2021; 32(1):215–223.

30.

Leaver-Fay

, Tyka

, Lewis

, et al. Rosetta3. Elsevier; 2011. 545–574; doi: 10.1016/b978-0-12-381270-4.00019-6

31.

Matthews

, Salgo

, Greenberg

, et al. Enfuvirtide: The first therapy to inhibit the entry of hiv-1 into host cd4 lymphocytes. Nat Rev Drug Discov, 2004; 3(3):215–225; ISSN 1474-1784; doi: 10.1038/nrd1331

32.

Müller

, Hiss

, Schneider

. Recurrent neural network model for constructive peptide design. J Chem Inf Model, 2018; 58(2):472–479; doi: 10.1021/acs.jcim.7b00414 ISSN 1549-960X

33.

Mustafa

, Balkhy

, Gabere

. Current treatment options and the role of peptides as potential therapeutic components for middle east respiratory syndrome (MERS): A review. Journal of Infection and Public Health, 2018; 11(1):9–17; ISSN 1876-0341; doi: 10.1016/j.jiph.2017.08.009

34.

Naeimi

, Bahmani

, Afshar

. Investigating the role of peptides in effective therapies against cancer. Cancer Cell Int, 2022; 22(1):139; ISSN 1475-2867; doi: 10.1186/s12935-022-02553-7

35.

Nivón

, Moretti

, Baker

. A pareto-optimal refinement method for protein design scaffolds. PLoS One, 2013; 8(4):e59004; ISSN 1932-6203; doi: 10.1371/journal.pone.0059004

36.

Odolczyk

, Marzec

, Winiewska-Szajewska

, et al. Native structure-based peptides as potential protein-protein interaction inhibitors of sars-cov-2 spike protein and human ace2 receptor. Molecules, 2021; 26(8):2157.

37.

Páll

, Abraham

, Kutzner

, et al. Tackling Exascale Software Challenges in Molecular Dynamics Simulations with GROMACS. Springer International Publishing; 2015. PP. 3–27; doi: 10.1007/978-3-319-15976-8_1

38.

Panda

, Sen Gupta

, Biswal

, et al. Ace-2-derived biomimetic peptides for the inhibition of spike protein of sars-cov-2. J Proteome Res, 2021; 20(2):1296–1303.

39.

Plisson

, Ramírez-Sánchez

, Martínez-Hernández

. Machine learning-guided discovery and design of non-hemolytic peptides. Sci Rep, 2020; 10(1):16581; ISSN 2045-2322; doi: 10.1038/s41598-020-73644-6

40.

Pountos

, Panteli

, Lampropoulos

, et al. The role of peptides in bone healing and regeneration: A systematic review. BMC Med, 2016; 14(1):103; ISSN 1741-7015; doi: 10.1186/s12916-016-0646-y

41.

Rajpoot

, Ohishi

, Kumar

, et al. A novel therapeutic peptide blocks sars-cov-2 spike protein binding with host cell ace2 receptor. Drugs R D, 2021; 21(3):273–283.

42.

Renzi

, Seamann

, Ganguly

, et al. Engineering an ace2-derived fragment as a decoy for novel sars-cov-2 virus. ACS Pharmacol Transl Sci, 2023; 6(6):857–867; doi: 10.1021/acsptsci.2c00180

43.

Singh

, Thornton

. Atlas of Protein Side-Chain Interactions. Oxford University Press: London, England; 1992.

44.

Singh

, Singh

, Tuknait

, et al. Pepstrmod: Structure prediction of peptides containing natural, non-natural and modified residues. Biol Direct, 2015; 10(1):73; ISSN 1745-6150; doi: 10.1186/s13062-015-0103-4

45.

Tatineni

, Hein

. Plant viruses of agricultural importance: Current and future perspectives of virus disease management strategies. Phytopathology®, 2023; 113(2):117–141; ISSN 1943-7684; doi: 10.1094/phyto-05-22-0167-rvw

46.

The UniProt Consortium. UniProt: The Universal Protein Knowledgebase in 2023. Nucleic Acids Res, 2023; 51(D1):D523–D531; ISSN 0305-1048; doi: 10.1093/nar/gkac1052

47.

Tsuboyama

, Dauparas

, Chen

, et al. Mega-scale experimental analysis of protein folding stability in biology and design. Nature, 2023; 620(7973):434–444; ISSN 1476-4687; doi: 10.1038/s41586-023-06328-6

48.

Tyka

, Keedy

, André

, et al. Alternate states of proteins revealed by detailed energy landscape mapping. J Mol Biol, 2011; 405(2):607–618; ISSN 0022-2836; doi: 10.1016/j.jmb.2010.11.008

49.

Valdés-Tresanco

, Valdés-Tresanco

, Valiente

, et al. gmx_mmpbsa: A new tool to perform end-state free energy calculations with gromacs. J Chem Theory Comput, 2021; 17(10):6281–6291; doi: 10.1021/acs.jctc.1c00645

50.

Wang

. Post-translational modifications of natural antimicrobial peptides and strategies for peptide engineering. Curr Biotechnol, 2012; 1(1):72–79; ISSN 2211-5501; doi: 10.2174/2211550111201010072

51.

Wang

, Yin

, Xiao

, et al. StraPep: A structure database of bioactive peptides. Database (Oxford), 2018; 2018:bay038; ISSN 1758-0463; doi: 10.1093/database/bay038

52.

Wang

, Wang

, Zhang

, et al. Therapeutic peptides: Current applications and future directions. Signal Transduct Target Ther, 2022; 7(1):48; ISSN 2059-3635; doi: 10.1038/s41392-022-00904-4

53.

Watson

, Juergens

, Bennett

, et al. De novo design of protein structure and function with rfdiffusion. Nature, 2023; 620(7976):1089–1100; ISSN 1476-4687; doi: 10.1038/s41586-023-06415-8

54.

, Ding

, Wang

, et al. High-resolution de novo structure prediction from primary sequence. bioRxiv, 2022; doi: 10.1101/2022.07.21.500999

55.

Yoshida

, Hinkley

, Tsuda

, et al. Using evolutionary algorithms and machine learning to explore sequence space for the discovery of antimicrobial peptides. Chem, 2018; 4(3):533–543; ISSN 2451-9294; doi: 10.1016/j.chempr.2018.01.005

56.

Youden

. Index for rating diagnostic tests. Cancer, 1950; 3(1):32–35; ISSN 1097-0142; doi: 10.1002/1097-0142(1950)3:1<32::aid-cncr2820030106>3.0.co;2-3

57.

Zakharova

, Orsi

, Capecchi

, et al. Machine learning guided discovery of non‐hemolytic membrane disruptive anticancer peptides. ChemMedChem, 2022; 17(17):e202200291; ISSN 1860-7187; doi: 10.1002/cmdc.202200291

58.

Zambaldi

, La

, Chu

, et al. De novo design of high-affinity protein binders with alphaproteo, 2024. Available from: https://arxiv.org/abs/2409.08022

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.81 MB

0.04 MB

0.40 MB

0.04 MB

VIPER: Virus Inhibition Via Peptide Engineering and Receptor Mimicry

Abstract

1. INTRODUCTION

2. METHODS

2.2. Residue selection step

2.3.1. Side chain interaction index

2.4. Computational validation

2.4.1. Spatial side chain interaction index

2.5. Implementation

3. RESULTS

3.1. Spatial side chain interaction index

3.2.2. HSV

3.3. Genetic algorithm

Footnotes

ACKNOWLEDGMENTS

AUTHORS’ CONTRIBUTIONS

CODE AND DATA AVAILABILITY

AUTHOR DISCLOSURE STATEMENT

FUNDING INFORMATION

SUPPLEMENTARY MATERIAL

1

2

References

Supplementary Material