Propedia 26: An expanded and updated database of protein-peptide interactions for machine learning applications

Propedia 26: An expanded and updated database of protein-peptide interactions for machine learning applications | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF data-descriptor Propedia 26: An expanded and updated database of protein-peptide interactions for machine learning applications Diego Mariano, Adenilson Arcanjo, Leonardo Henrique Silva, Milenna Machado Pirovani, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8349237/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Propedia is a curated database of protein-peptide interactions. In this new release, we expanded the dataset to more than four times the size of the previous version and introduced a comprehensive set of new annotations. We systematically analyzed protein-peptide complexes to derive predicted physicochemical properties, solvent-accessible surface areas, interaction energies, interface residues, and interchain contacts. In addition, we applied machine learning models to predict the potential therapeutic functions of the peptides. Propedia 26 comprises 98,779 entries, including 78,148 protein–peptide pair complexes (pep-pro) and 20,631 multi-protein complexes featuring a single peptide interacting with one or more protein chains (multipro). We anticipate that this expanded and enriched resource will support the discovery and development of peptide-based drugs and biotechnological products. Propedia 26 is freely accessible at https://bioinfo.dcc.ufmg.br/propedia26 . Figures Figure 1 Figure 2 Figure 3 Figure 4 Background & Summary Peptides are molecules composed of 2 to 50 amino acid residues connected by peptide bonds [ 1 ]. In cells, they play several essential roles in regulation and cellular communication, acting as neurotransmitters, hormones, antimicrobial agents, and immune signaling molecules. Furthermore, they influence various processes such as growth, wound healing, and defense against pathogens. It is estimated that peptides mediate 15% to 40% of molecular interactions between proteins [ 2 ]. Recently, peptides have gained great pharmaceutical and biotechnological importance due to their low toxicity, high structural flexibility, and high interaction specificity [ 3 ]. Therefore, they have been the target of several studies aimed at developing new drugs and targeted therapies, as well as producing new biotechnological products [ 4 , 5 ]. However, screening of peptides with biotechnological potential still relies heavily on high-cost, low-scalability experimental approaches. Hence, computational methods have emerged as an innovative alternative for virtual peptide screening [ 6 ]. In this context, the creation of peptide databases has become essential for such studies. Propedia is a database of protein-peptide interactions. The first version of Propedia was released in early 2021 [ 7 ]. Propedia consists of data retrieved from the experimental repository Protein Data Bank (PDB), which are then processed and curated [ 8 ]. These data were then filtered, and various analyses were performed to extract new features. Initially, Propedia only handled entries consisting of unique protein-peptide pairs. Propedia v1 contained 19,813 protein-peptide complexes clustered by three methods: sequence-based, interface-based, and binding-site-based. Sequence-based clusters were constructed using the Hammock tool v1.2 [ 9 ]. The Hammock pipeline identified 3,495 unique sequences, categorizing them into 771 sequence clusters and 1,074 unique sequences (singletons), resulting in a total of 1,845 peptide sequence clusters. Interface clusters were constructed using the MUSTANG tool [ 10 ], which performs multiple structure alignments. MUSTANG detected 535 clusters, in addition to 1,356 singletons, resulting in 1,891 protein-peptide interfaces. Lastly, clusters were also constructed based on binding-site similarities using the ProBiS algorithm [ 11 ]. ProBIS employs a local alignment algorithm to identify similar binding sites in proteins with distinct folds. To achieve this, it utilizes geometric and functional groups to detect three-dimensional patterns of physicochemical properties on their surfaces. Ultimately, 521 binding site clusters were detected, in addition to 945 singletons, totaling 1,466 different binding sites. Among the main features presented by Propedia v1, the binding-site search engine stands out. In practice, if a user wants to find peptides that bind to a specific known binding site, they would enter the PDB code, the chain, and the residues that comprise the site into the interface. Propedia would then use the ProBIS algorithm and its database of binding-site-based clusters to detect experimentally resolved peptides that potentially bind to that binding site. The second version of Propedia was released in late 2023 [ 12 ]. It presented 49,300 protein-peptide complexes. Among the new features, it introduced a set of subdatasets classified according to the protein class extracted from the Protein Data Bank. Among the classifications, we can mention: antimicrobial, viral, enzyme, membrane, hormone, and plant. Furthermore, structural signatures were calculated for each protein-peptide complex, receptor, and peptide individually using the aCSM algorithm [ 13 ]. Structural signatures are mathematical representations used to encode three-dimensional structural information in a one-dimensional vector space [ 14 ]. In this paper, we present Propedia 26, the third version of the protein-peptide interaction database. The version nomenclature has been changed to reflect the year of release. Propedia 26 contains 4 times as many entries as the first version and introduces several new features to characterize protein-peptide complexes better. Furthermore, Propedia 26 introduces calculations for each entry of seven types of interchain contacts: hydrogen bonds, attractive bonds, repulsive bonds, hydrophobic bonds, aromatic bonds, salt bridges, and disulfide bonds. We also offer several new features that may be beneficial for classification tasks utilizing machine learning algorithms. Data Records Briefly, Propedia 26 contains 98,779 entries, with 78,148 protein-peptide pair complexes (pep-pro) and 20,631 complexes composed of a single peptide interacting with two or more proteins (multipro). Figure 1 summarizes the steps taken to collect the data that make up Propedia 26. First, the data were collected from the PDB database. We selected structures with peptides close to proteins (distance ≤ 6Å). We separated the protein-peptide structure pairs and calculated a set of features using various approaches. For each entry, Propedia 26 provides a comprehensive set of structural, physicochemical, and functional annotations. Complex-level characteristics include chain identifiers, structural descriptions, sequence length (in residues), molecular weight (Da), isoelectric point (pI), instability and aliphatic index, GRAVY (Grand Average of Hydropathy), percentage of hydrophobic residues, number of positively and negatively charged residues, atomic formula, molar extinction coefficient (M − 1 cm − 1 ), and full sequence information. These characteristics were calculated using the ProtParam library [ 15 ]. The dataset also lists structurally and sequentially similar complexes, facilitating comparison and similarity-based analyses. Previous versions of Propedia only considered protein-peptide interaction pairs. For example, consider the PDB entry 1A1R, which contains three chains (A, B, and C). Chain C belongs to a peptide that interacts simultaneously with the protein in chains A and B. By default, the Propedia pipeline saves two entries: 1A1R-C-A (peptide C interacting with the protein chain A) and 1A1R-C-B (peptide C interacting with the protein chain B). In this new version of Propedia, we decided to include a new type of entry that groups all chains interacting with the same peptide. We refer to this dataset as Propedia Multipro. In this case, the entry 1A1R-C would contain the C chains of the peptide and the A and B chains of proteins that are interacting with it. We did this for all entries in Propedia 26. In total, we collected 20,631 Multipro complexes. Propedia 26 further reports the predicted functional profiles of each peptide, indicating the probability of exhibiting anti-angiogenic (AAP), antibacterial (ABP), anticancer (ACP), anti-inflammatory (AIP), quorum-sensing (QSP), or surface-binding (SBP) activity. The original Propedia v1 annotation schema, including Binding Site, Interface, and Sequence classes, is fully retained. A detailed contact surface analysis is also provided. The Accessible Surface Area (ASA), expressed in Ų, quantifies the portion of the molecule exposed to solvent (typically water). The Buried Surface Area (BSA) represents the region effectively engaged at the binding interface. Additionally, we report the Buried Peptide surface Percentage (BPP%), indicating the fraction of the peptide that is not exposed to the solvent. All ASA- and BSA-related measurements were obtained using NACCESS [ 16 ]. Additionally, Propedia 26 also includes predictions of protein-peptide interaction energy, estimating binding affinity (kcal/mol) and dissociation constant (M) at 25°C using the PRODIGY algorithm [ 17 ]. Finally, the database identifies all interface residues, defined as protein residues within 6 Å of the peptide, and provides an atom-level catalog of interchain contacts. The computed contact types include hydrogen bonds, attractive and repulsive interactions, hydrophobic contacts, π-π stacking, salt bridges, and disulfide bonds. All contacts were calculated using the COCαDA algorithm [ 18 ]. A description of the table of entries in Propedia is provided in Supplementary Table S1 . Technical validation Data quality was assessed through random sampling of the dataset. A subset of entries was randomly selected, and the values presented in the Propedia interface were manually cross-checked against their respective primary data sources. To evaluate the descriptors, we consulted the Protein Data Bank page corresponding to the entry. For the analysis of physicochemical parameters, we collected the peptide sequence and used the ProtParam web tool ( https://web.expasy.org/protparam ). To evaluate the interaction energy, we consulted the Prodigy web tool, available at https://rascar.science.uu.nl/prodigy . The list of intra-chain contacts was verified using the COCαDa web tool, available at https://bioinfo.dcc.ufmg.br/cocada-web . The NACCESS surface area data were evaluated using in-house Python scripts. In addition, the Orange Data Mining tool was used to assess the feature statistics and detect distribution, mean, mode, median, dispersion, minimum and maximum values, and missing values. Missing data in Propedia 26 arise from isolated issues during the execution of specific computational tools, as detailed in the Methods section. For example, the presence of non-canonical amino acids or high cutoff distances between chains (by default, Propedia defines the maximum distance between chains as 6Å, but other tools may require smaller distances) can lead to errors in the tools' calculations. Overall, data completeness is high across the dataset. Protein surface area measurements are missing for 256 entries (< 1% of the dataset), while predicted therapeutic properties are unavailable for 1,875 structures (2%). Interaction energy predictions could not be computed for 5,010 entries (6%), primarily due to limitations in structural quality or contact detection required by the prediction algorithm. Predicted physicochemical properties for protein chains are missing for 41 entries (< 1%). Importantly, all entries include experimentally derived complex-level information obtained from the Protein Data Bank, and 100% of peptide chains contain predicted physicochemical descriptors. These statistics indicate that missing values are limited, well-characterized, and confined to specific derived features, ensuring the overall reliability and reusability of the dataset. The data report is available on the supplementary material's GitHub repository. Usage Notes Propedia 26 web interface Propedia 26 has a user-friendly web interface designed to simplify navigation and search of structures. There are three entry search mechanisms: (i) sequence-based search (uses BLAST to perform local alignments with parameters optimized for peptide search); (ii) binding site-based search (uses ProBis to detect structures with cavities similar to the region indicated in the target protein by users); and (iii) search based on Propedia ID, PDB ID, or entry metadata (Fig. 2 - top). Each entry in Propedia has its own page with information such as a description of the entry, a presentation of physicochemical parameters detected in silico , and an interactive visualization of the 3D structure (Fig. 2 - below). The entry page also presents similar structures, predicted characteristics for therapeutic peptides, and the clusters inherited from Propedia v1 (based on sequence, interface, and binding site; Fig. 3 A). Furthermore, information on protein-peptide interactions is calculated using different software. Solvent-accessible area is predicted using NACCESS [ 16 ]. We display information about the solvent-exposed area for ​​the complex, only the peptide, and only the protein (Fig. 3 B). In addition, we show the buried areas of the peptide (BPepA) and the protein (BProA), corresponding to the regions that mediate the protein-peptide interaction. Note that, mostly, these regions occupy different area values ​​in the protein and in the peptide. Therefore, we use the average of these values ​​to indicate the area of ​​the binding interface ( i.e. , buried surface area - BSA). We also report the percentage of the peptide that is buried (BPP%). The higher the BPP% value, the more deeply embedded the peptide is, and therefore, the stronger the bond tends to be. However, it should be noted that the peptide may form bonds with other protein chains (see the complex's multipro page for more details). Interaction energy, binding affinity, and dissociation constant are predicted using Prodigy (Fig. 3 B). We also list the residues in the protein interface (Fig. 3 C). Finally, interchain contacts were calculated using the COCαDA-CLI (Fig. 3 D). The contact list table is integrated into the 3D structure visualization. Clicking on a listed contact displays the interacting residues as sticks (Fig. 3 E). Case study: Searching for peptides that inhibit the SARS-CoV-2 Mpro protein In this section, we illustrate the application of the binding site search algorithm to identify peptides that interact with the SARS-CoV-2 Mpro protein. This section is not intended to provide an in-depth biological analysis of the case studies or predictions; instead, it aims to offer illustrative examples of the database entries and their potential use cases. SARS-CoV-2 remains a global concern due to its complex pathogenesis and its potential to cause severe disease, particularly when viral replication is not effectively controlled in target tissues [ 19 – 21 ]. Mpro (Main Protease), also called 3CLpro, is one of the most essential enzymes for the SARS-CoV-2 life cycle. It plays a central role in viral replication, which is why it has become one of the most critical targets for antiviral drug development. The Propedia binding site search requires as input a PDB code of the target protein, the chain, and a list of residue codes. The tool then uses the ProBis algorithm to search for structures with binding sites that have conformations similar to those of the input structure. The binding site used in our search was defined according to recent structural studies [ 22 ], focusing on the key residues that form the catalytic region of Mpro. Propedia 26 identified 180 protein-peptide complexes matching the defined binding site, whereas Propedia v1 returned only 34. This increase reflects the expansion of the database, in particular, due to the large number of recent studies on SARS-CoV-2. This highlights the importance of updating the Propedia database. Figure 4 presents the top three peptides obtained in our search, ranked by Alignment Score and RMSD. Methods Data collection We collected all structures with a maximum of 10,000 amino acid residues from the Protein Data Bank (PDB) [ 8 ] in CIF format on September 8, 2025 (n = 236,616). Then, we filtered all complexes performing protein-peptide interactions using Python scripts. We defined a complex as exhibiting a protein–peptide interaction when at least one atom from the protein and one atom from the peptide were separated by a distance of 6 Å or less. Additionally, we extracted all combinations of chain pairs where one chain had between 2 and 50 amino acids, while the other had more than 50 amino acids. In total, we extracted 78,148 protein-peptide complexes. Furthermore, we created a sub-dataset comprising all protein chains within a maximum distance of 6 Å from the peptide. We called this the Multipro dataset. In this case, each entry point of the multipro is composed of one peptide interacting with two or more protein chains. In total, we obtained 20,631 multipro entries. Clustering and Redundancy Detection To detect redundancy in protein-peptide complexes, we concatenated the protein and peptide sequences and grouped identical sequence combinations. In total, we detected 51,082 unique protein-peptide complexes. Then, we collected all peptide sequences containing only canonical amino acids and grouped those that were 100% identical. In total, we extracted 17,509 unique sequences composed only of the 20 canonical amino acids. We refer to this peptide dataset as the canonical non-redundant (CNR) dataset. Clusters described in previous versions of Propedia have been recalculated using Python scripts. Other classification information was obtained from the PDB file using scripts developed with the Biopython library (Bio.PDB package) [ 23 , 24 ]. Feature extraction For each protein-peptide complex, physicochemical properties were obtained using ProtParam [ 15 ] embedded in the Biopython library v1.85 [ 23 ]. The collected information included the chain identifier, description, sequence length (number of residues), molecular weight (Da), isoelectric point (pI), instability index, aliphatic index, GRAVY (grand average of hydropathicity), hydrophobicity (%), total number of positively and negatively charged residues, atomic formula, number of atoms, extinction coefficient (with and without disulfide bonds), and the full amino acid sequence. Protein-peptide interactions Protein interaction interface residues were calculated using Python scripts. The IDs of proteins’ residues at a distance ≤ 6 Å from the peptide were annotated. The solvent-accessible surface area (ASA) was calculated using NACCESS v2.1.1 [ 16 ]. For each protein–peptide complex, the total ASA of the complex, the changes in ASA for the protein and the peptide upon binding (ΔASA), and the buried surface area (BSA) were computed and reported in square angstroms (Ų). The Buried Surface Area (BSA), which represents the region effectively engaged at the binding interface, is computed as: $$\:BSA\:=\:\frac{AS{A}_{protein}\:+\:AS{A}_{peptide}\:-\:AS{A}_{complex}}{2}$$ Furthermore, we calculated the buried area of ​​peptides (BPepA) and proteins (BProA). Finally, we calculated the buried peptide percentage (BPP%) using the formula: $$\:BPP\left(\%\right)\:=\:\frac{100\:\times\:\:BPep{A}_{}}{AS{A}_{peptide}}$$ The interaction energy of each protein–peptide complex was calculated using PRODIGY v.2.4.0 [ 17 ]. For each complex, the total number of interfacial contacts (categorized by amino acid type), the predicted binding affinity (kcal/mol), and the predicted dissociation constant (M) at 25°C were computed. Chain contacts were calculated using the COCαDA command-line tool v1.0 [ 18 , 25 ]. For each complex, COCαDA calculated hydrogen bonds, attractive and repulsive interactions, hydrophobic interactions, aromatic interactions, disulfide bonds, and salt bridges. Structural signature For each three-dimensional structure of unique peptides composed only of canonical amino acids (CNR dataset), we calculated structural signatures using the SIGNA library ( https://github.com/LBS-UFMG/signa ). Structural signatures convert the 3D structure into one-dimensional vectors that can be used in machine learning tasks. For the calculation of structural signatures, we used the aCSM-ALL algorithm [ 13 ] with a maximum distance parameter of 10 Å and a step of 0.2 Å. These parameters were based on previous studies [ 12 , 14 , 26 – 28 ]. Sequence signature Furthermore, we extracted features based on the sequences using the iFeature tool ( https://github.com/Superzchen/iFeature ) [ 29 ]. We named the extracted sequence features as sequence signatures. Sequence-based signatures can be crucial for machine learning tasks that involve predictions where three-dimensional structures are unavailable. We selected the 10 following descriptors: AAC, DPC, DDE, GAAC, GDPC, GTPC, CTDC, CTDT, CTDD, and CTriad. The Amino Acid Composition (AAC) descriptor represents the relative frequency of each of the 20 amino acids in a peptide sequence. The Dipeptide Composition (DPC) descriptor extends this idea by quantifying the frequencies of all 400 possible dipeptide pairs, thereby capturing short-range order information. The Dipeptide Deviation from Expected mean (DDE) descriptor measures how much the observed dipeptide frequencies deviate from values expected based on individual amino acid compositions, reflecting pairwise dependency patterns. The Grouped Amino Acid Composition (GAAC) descriptor simplifies the sequence by clustering amino acids into physicochemical groups and computing their overall proportions. Similarly, the Grouped Dipeptide Composition (GDPC) and Grouped Tripeptide Composition (GTPC) descriptors compute the frequencies of dipeptides or tripeptides based on predefined amino-acid groups rather than individual residues, reducing dimensionality while retaining key biochemical patterns. The Composition Transition Distribution Composition (CTDC) descriptor summarizes the fraction of residues belonging to different physicochemical categories, whereas CTD Transition (CTDT) quantifies how often the sequence switches between these categories. CTD Distribution (CTDD) further describes the positional distribution of residues from each category along the sequence. Finally, the Conjoint Triad (CTriad) descriptor groups amino acids into seven classes and measures the frequency of all possible triads of these groups, capturing local physicochemical environments. Finally, iFeature returns a vector of 1248 features for each sequence used as input. In this work, we refer to this vector as a sequence signature. Therapeutic peptides classification We classified each peptide in Propedia into six classes according to their predicted therapeutic properties: Anti-Angiogenic (AAP), Antibacterial (ABP), Anticancer (ACP), Anti-Inflammatory (AIP), Quorum Sensing (QSP), and Surface Binding (SBP). The classification was performed using machine learning models with Orange Data Mining [ 30 ]. To build the models, we used sequence-based signatures and attempted to reproduce the methodology described in the article by Rodrigues et al. [ 31 ]. Details of the methodology are available in our GitHub’s supplementary material repository (see Data Availability section). User-friendly interface development The web interface was built using the same approaches used in [ 32 – 35 ]. The CodeIgniter 4 framework was used for the back-end. The Bootstrap and DataTables libraries were used in the front-end application. Interactive protein visualizations were built using the 3Dmol.js library [ 36 ]. Contact map graphs were built using the Chat.js library. Figures used to illustrate the sections were constructed using the ChimeraX tool [ 37 ]. The sequence-based search system was built using BLAST [ 38 , 39 ]. To enable searches for peptide sequences, the following parameters were used: "-word_size 2 -task blastp-short -seg no -evalue 100000". For protein sequence searches, only the parameter "-word_size 3" was employed. To search for similar binding sites, the Propedia back-end uses the ProBiS tool [ 11 ]. The ProBiS algorithm detects local structural similarities by comparing the three-dimensional surface of the protein's binding site with the 3D surfaces of proteins stored in the Propedia-26 database. To reduce search costs, we selected only non-redundant structures. Additionally, some structures exhibited errors during the extraction of the peptide interaction surface. In total, 32,695 protein-peptide complex surfaces are used in the search for similar binding sites. Declarations Competing interests The authors declare no competing interests. Funding This work was supported by grants from the Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG, projects APQ-02690-22 and APQ-01834-21) and the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, project 440307/2022-8). This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES) – Finance Code 001. Author Contribution P.M., D.M., and R.C.M.M contributed to the conceptualization; D.M. collected the data and developed the web tool; L.H.S., M.M.P, and L.M. performed the case studies; M.M.P wrote the documentation; L.B. performed the surface analysis; A.A. performed the analysis of interaction energy; R.L. performed the contact analysis; D.M. performed the analysis of physical-chemical parameters; R.C.M.M. acquired funds and supervised the research. D.M. prepared the manuscript with the assistance and feedback from all other co-authors. Acknowledgement The authors would like to express their gratitude to the research funding agencies: CAPES, CNPq, and FAPEMIG. The authors thank Felipe Conceição for his valuable contributions. Data availability The web tool can be found at https://bioinfo.dcc.ufmg.br/propedia26 . The dataset and data descriptors can be downloaded at https://bioinfo.dcc.ufmg.br/propedia26/public/download . Full supplementary material is available at https://github.com/LBS-UFMG/propedia26-sm . Data descriptors are available at Supplementary Table S1 . Code availability The source code of the web tool can be found at: https://github.com/LBS-UFMG/propedia26 . The Orange Data Mining machine learning models are available in our GitHub supplementary material repository: https://github.com/LBS-UFMG/propedia26-sm . References Sauvestre, C., Zagury, J.-F. & Langenfeld, F. Comparative Analysis of Deep Learning-Based Algorithms for Peptide Structure Prediction. Proteins Struct. Funct. Bioinforma. n/a, (2025). Neduva, V. et al. Systematic discovery of new recognition peptides mediating protein interaction networks. PLoS Biol. 3, e405 (2005). Lee, A. C.-L., Harris, J. L., Khanna, K. K. & Hong, J.-H. A comprehensive review on current advances in peptide drug development and design. Int. J. Mol. Sci. 20, 2383 (2019). Wang, L. et al. Therapeutic peptides: current applications and future directions. Signal Transduct. Target. Ther. 7, 48 (2022). Cabral, H. Peptides: Molecular and Biotechnological Aspects. Biomolecules 11, 52 (2021). Tripathi, N. M. & Bandyopadhyay, A. High throughput virtual screening (HTVS) of peptide library: Technological advancement in ligand discovery. Eur. J. Med. Chem. 243, 114766 (2022). Martins, P. M. et al. Propedia: a database for protein–peptide identification based on a hybrid clustering algorithm. BMC Bioinformatics 22, 1 (2021). Berman, H. M. et al. The Protein Data Bank. Nucleic Acids Res. 28, 235–242 (2000). Krejci, A., Hupp, T. R., Lexa, M., Vojtesek, B. & Muller, P. Hammock: a hidden Markov model-based peptide clustering algorithm to identify protein-interaction consensus motifs in large datasets. Bioinforma. Oxf. Engl. 32, 9–16 (2016). Konagurthu, A. S., Whisstock, J. C., Stuckey, P. J. & Lesk, A. M. MUSTANG: a multiple structural alignment algorithm. Proteins 64, 559–574 (2006). Konc, J. & Janežič, D. ProBiS algorithm for detection of structurally similar protein binding sites by local structural alignment. Bioinformatics 26, 1160–1168 (2010). Martins, P. et al. Propedia v2.3: A novel representation approach for the peptide-protein interaction database using graph-based structural signatures. Front. Bioinforma. 3, (2023). Pires, D. E. V., de Melo-Minardi, R. C., da Silveira, C. H., Campos, F. F. & Meira, W. aCSM: noise-free graph-based signatures to large-scale receptor-based ligand prediction. Bioinforma. Oxf. Engl. 29, 855–861 (2013). Mariano, D. et al. A Computational Method to Propose Mutations in Enzymes Based on Structural Signature Variation (SSV). Int. J. Mol. Sci. 20, (2019). Gasteiger, E. et al. Protein identification and analysis tools on the ExPASy server. in The proteomics protocols handbook 571–607 (Springer, 2005). Hubbard, S. J. & Thornton, J. naccess. Dept Biohem Mol Biol UCL (1993). Vangone, A. & Bonvin, A. M. Contacts-based prediction of binding affinity in protein–protein complexes. elife 4, e07454 (2015). Lemos, R., Mariano, D., Silveira, S. & Melo-Minardi, R. COCαDA - Large-Scale Protein Interatomic Contact Cutoff Optimization by Cα Distance Matrices . 70 (2024). doi: 10.5753/bsb.2024.245545 . Ibrahim, M. et al. Why is the Omicron main protease of SARS-CoV-2 less stable than its wild-type counterpart? A crystallographic, biophysical, and theoretical study of the free enzyme and its complex with inhibitor 13b-K. 2024.03.04.583178 Preprint at https://doi.org/10.1101/2024.03.04.583178 (2024). Lamers, M. M. & Haagmans, B. L. SARS-CoV-2 pathogenesis. Nat. Rev. Microbiol. 20, 270–284 (2022). Hashemi, Z. S. et al. In silico Approaches for the Design and Optimization of Interfering Peptides Against Protein–Protein Interactions. Front. Mol. Biosci. 8, (2021). Chaves Carvalho, F. et al. A hybrid evolutionary and structural method for AI-guided peptide inhibitor design using AlphaFold and Rosetta. Sci. Rep. https://doi.org/10.1038/s41598-025-28061-y (2025) doi:10.1038/s41598-025-28061-y. Cock, P. J. A. et al. Biopython: freely available Python tools for computational molecular biology and bioinformatics. Bioinforma. Oxf. Engl. 25, 1422–1423 (2009). Hamelryck, T. & Manderick, B. PDB file parser and structure class implemented in Python. Bioinformatics 19, 2308–2310 (2003). Lemos, R. P., Mariano, D., Silveira, S. de A. & de Melo-Minardi, R. C. COCαDA-A Fast and Scalable Algorithm for Interatomic Contact Detection in Proteins Using Cα Distance Matrices. Front. Bioinforma. 5, 1630078 (2025). Moreira, E. U. et al. Predicting Mutation-Driven Changes in the SARS-CoV-2 Spike Protein Using Structural Signatures and Neural Networks. in 167–178 (SBC, 2024). Mariano, D. et al. Using Machine Learning and Graph-Based Signatures to Evaluate Normal and Decreased Function CYP2D6 Haplotypes. in 90–105 (Springer, 2025). Cioletti, A. G., Lemos, R. P., dos Santos, L. M., Mariano, D. & de Melo-Minardi, R. C. Using graph-based structural signatures and machine learning algorithms for molecular docking assessment of histone deacetylases and small ligands. in 223–234 (SBC, 2024). Chen, Z. et al. iFeature: a Python package and web server for features extraction and selection from protein and peptide sequences. Bioinformatics 34, 2499–2502 (2018). Demšar, J., Zupan, B., Leban, G. & Curk, T. Orange: From experimental machine learning to interactive data mining. in 537–539 (Springer, 2004). Rodrigues, C. H. M., Garg, A., Keizer, D., Pires, D. E. V. & Ascher, D. B. CSM-peptides: A computational approach to rapid identification of therapeutic peptides. Protein Sci. 31, e4442 (2022). Silva, M. F. M. et al. Proteingo: Motivation, user experience, and learning of molecular interactions in biological complexes. Entertain. Comput. 29, 31–42 (2019). Fassio, A. V., Santos, L. H., Silveira, S. A., Ferreira, R. S. & de Melo-Minardi, R. C. nAPOLI: a graph-based strategy to detect and visualize conserved protein-ligand interactions in large-scale. IEEE/ACM Trans. Comput. Biol. Bioinform. https://doi.org/10.1109/TCBB.2019.2892099 (2019) doi:10.1109/TCBB.2019.2892099. Barroso, J. R. M. S. et al. Proteus: An algorithm for proposing stabilizing mutation pairs based on interactions observed in known protein 3D structures. BMC Bioinformatics 21, 275 (2020). Pimentel, V. et al. VTR: a web tool for identifying analogous contacts on protein structures and their complexes. Front. Bioinforma. 1, 28 (2021). Rego, N. & Koes, D. 3Dmol.js: molecular visualization with WebGL. Bioinformatics 31, 1322–1324 (2015). Meng, E. C. et al. UCSF ChimeraX: Tools for structure building and analysis. Protein Sci. 32, e4792 (2023). Altschul, S. F., Gish, W., Miller, W., Myers, E. W. & Lipman, D. J. Basic local alignment search tool. J. Mol. Biol. 215, 403–410 (1990). Camacho, C. et al. BLAST+: architecture and applications. BMC Bioinformatics 10, 421 (2009). Additional Declarations No competing interests reported. Supplementary Files supplementarymaterialv2.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8349237","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"data-descriptor","associatedPublications":[],"authors":[{"id":596813863,"identity":"ab0607ed-bf0c-4791-9126-8d4a893846c8","order_by":0,"name":"Diego Mariano","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzUlEQVRIiWNgGAWjYDCCwyCighnM/gDECURqOQPWwjiDOC0HQGrbSNHCd5z58IuP86zl+Rt4DJtu/GLIM28goEXyMFua5cxt6YYzDvAYNuf2MRTLHCCgxeAwj5kx77bDCQYMPOaPc3sYEmcQcpjBYf5vxrxzwFqAthCnhYf5MW8DVEvODyK0AP1ixjjjGNAvh9kKm3MbJIolCGnhO3/48YcPNcAQa2/e2JzzxyaPoBYgYIMoAkUNYxsxGoBqPyDYf4jSMQpGwSgYBSMMAAD8Oj98W/4F8AAAAABJRU5ErkJggg==","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":true,"prefix":"","firstName":"Diego","middleName":"","lastName":"Mariano","suffix":""},{"id":596813864,"identity":"1d1f6593-60b4-4f86-b670-540ac0d8c9bd","order_by":1,"name":"Adenilson Arcanjo","email":"","orcid":"","institution":"Instituto Federal de Educação, Ciência e Tecnologia do Ceará","correspondingAuthor":false,"prefix":"","firstName":"Adenilson","middleName":"","lastName":"Arcanjo","suffix":""},{"id":596813865,"identity":"95751909-b02a-4372-8aea-0782039d6680","order_by":2,"name":"Leonardo Henrique Silva","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Leonardo","middleName":"Henrique","lastName":"Silva","suffix":""},{"id":596813866,"identity":"d84a70a3-f57c-4c5c-960b-7010cf08b033","order_by":3,"name":"Milenna Machado Pirovani","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Milenna","middleName":"Machado","lastName":"Pirovani","suffix":""},{"id":596813867,"identity":"caea26ab-6934-419d-85ce-1575c3861cfd","order_by":4,"name":"Leandro Morais","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Leandro","middleName":"","lastName":"Morais","suffix":""},{"id":596813868,"identity":"6cd7a2b7-cd71-4235-a4f7-39450b68e637","order_by":5,"name":"Luana Luiza Bastos","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Luana","middleName":"Luiza","lastName":"Bastos","suffix":""},{"id":596813869,"identity":"46654dd1-b713-4e03-be3a-2adb15ec5b67","order_by":6,"name":"Rafael Pereira Lemos","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Rafael","middleName":"Pereira","lastName":"Lemos","suffix":""},{"id":596813870,"identity":"d49883f7-4475-45c8-b67d-874bc132c1ea","order_by":7,"name":"Pedro Martins","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Pedro","middleName":"","lastName":"Martins","suffix":""},{"id":596813871,"identity":"4b8e92eb-c139-4396-8a57-42473ebb1114","order_by":8,"name":"Raquel Cardoso de Melo-Minardi","email":"","orcid":"","institution":"Universidade Federal de Minas Gerais – UFMG","correspondingAuthor":false,"prefix":"","firstName":"Raquel","middleName":"Cardoso","lastName":"de Melo-Minardi","suffix":""}],"badges":[],"createdAt":"2025-12-12 22:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8349237/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8349237/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103593276,"identity":"1f202a3c-7bf1-42ea-b35d-9cd30b9e870f","added_by":"auto","created_at":"2026-02-27 12:33:40","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":255545,"visible":true,"origin":"","legend":"\u003cp\u003ePropedia 26 overview. 3D structures were collected from the PDB database. Complexes containing protein-peptide structures were filtered. Then, we predicted and calculated several features, such as signatures (based on sequence and structures), physical/chemical parameters, surface area, interaction energy, interatomic contacts, and grouped structures based on their function. The search engine implemented in the interface was developed using a sequence approach (using BLAST) and a binding site approach (using ProBis).\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8349237/v1/7bbaab814686b5935331f51e.png"},{"id":104398861,"identity":"85bd895e-c62e-427a-ae96-2aafc83551ef","added_by":"auto","created_at":"2026-03-11 12:04:00","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":832200,"visible":true,"origin":"","legend":"\u003cp\u003ePage from a Propedia entry (ID: 1A1M-C-A). At the top, we can see the three search engines, accessible on any Propedia page. Below, we can see entry details, such as entry description, physicochemical parameters, and interactive 3D structure visualization.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8349237/v1/62678a0c56f48515b1f8c92d.png"},{"id":104398850,"identity":"d6dd1a24-3f9f-4c22-8d9b-26ba6f36f6b3","added_by":"auto","created_at":"2026-03-11 12:03:57","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":1241462,"visible":true,"origin":"","legend":"\u003cp\u003ePredicted functions and protein-peptide interaction data are shown in the Entry page (ID: 1A1M-C-A). (A) Similar structures and classifications; (B) Surface and protein-peptide interactions; (C) Interface protein residues; (D) Protein-peptide contacts; and (E) Interactive 3D structure view of the complex. In this example, we are highlighting the contact between D114 and Y3 (hydrogen bond).\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8349237/v1/3c5567b03f3222c51c08fa98.png"},{"id":103593281,"identity":"6d403351-3d0c-4c67-9eec-072dec16f989","added_by":"auto","created_at":"2026-02-27 12:33:41","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":1315985,"visible":true,"origin":"","legend":"\u003cp\u003eIdentification of inhibitory peptides for SARS-CoV-2 Mpro. The target structure (left) highlights the binding site surface in green (PDB ID: 6LU7). Residues that compound the binding site are: 24, 25, 26, 27, 41, 49, 140, 141, 142, 143, 144, 145, 164, 165, 166, 167, 168, 172, 187, 188, 189, 190, 191, 192. The search returned complexes with high structural compatibility, represented by the three peptides with more compatibility to the binding site: ATVRLQAGNA, KVATVQSKMS, and AVKLQNNE. The orange ribbons illustrate how these ligands fit into the active site, illustrating how the binding-site similarity search retrieves peptides with structural compatibility to a target site based on low RMSD values ​​and high Alignment Score values.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8349237/v1/361ef2a9bc33da7ddfce8492.png"},{"id":104410291,"identity":"fd5a6638-2106-4eba-8b69-4dbcb0ce2df8","added_by":"auto","created_at":"2026-03-11 12:51:08","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4154567,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8349237/v1/202825ec-5966-4a7d-a6ef-18159e74d2a1.pdf"},{"id":103593278,"identity":"c3e7538d-33b3-4b83-b2ed-214d76d6421a","added_by":"auto","created_at":"2026-02-27 12:33:40","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":146359,"visible":true,"origin":"","legend":"","description":"","filename":"supplementarymaterialv2.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8349237/v1/36cb8b3d47cc4e9672f0d8e9.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Propedia 26: An expanded and updated database of protein-peptide interactions for machine learning applications","fulltext":[{"header":"Background \u0026 Summary","content":"\u003cp\u003ePeptides are molecules composed of 2 to 50 amino acid residues connected by peptide bonds [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. In cells, they play several essential roles in regulation and cellular communication, acting as neurotransmitters, hormones, antimicrobial agents, and immune signaling molecules. Furthermore, they influence various processes such as growth, wound healing, and defense against pathogens. It is estimated that peptides mediate 15% to 40% of molecular interactions between proteins [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Recently, peptides have gained great pharmaceutical and biotechnological importance due to their low toxicity, high structural flexibility, and high interaction specificity [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Therefore, they have been the target of several studies aimed at developing new drugs and targeted therapies, as well as producing new biotechnological products [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, screening of peptides with biotechnological potential still relies heavily on high-cost, low-scalability experimental approaches. Hence, computational methods have emerged as an innovative alternative for virtual peptide screening [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. In this context, the creation of peptide databases has become essential for such studies.\u003c/p\u003e \u003cp\u003ePropedia is a database of protein-peptide interactions. The first version of Propedia was released in early 2021 [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Propedia consists of data retrieved from the experimental repository Protein Data Bank (PDB), which are then processed and curated [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. These data were then filtered, and various analyses were performed to extract new features. Initially, Propedia only handled entries consisting of unique protein-peptide pairs. Propedia v1 contained 19,813 protein-peptide complexes clustered by three methods: sequence-based, interface-based, and binding-site-based.\u003c/p\u003e \u003cp\u003eSequence-based clusters were constructed using the Hammock tool v1.2 [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The Hammock pipeline identified 3,495 unique sequences, categorizing them into 771 sequence clusters and 1,074 unique sequences (singletons), resulting in a total of 1,845 peptide sequence clusters. Interface clusters were constructed using the MUSTANG tool [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e], which performs multiple structure alignments. MUSTANG detected 535 clusters, in addition to 1,356 singletons, resulting in 1,891 protein-peptide interfaces. Lastly, clusters were also constructed based on binding-site similarities using the ProBiS algorithm [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. ProBIS employs a local alignment algorithm to identify similar binding sites in proteins with distinct folds. To achieve this, it utilizes geometric and functional groups to detect three-dimensional patterns of physicochemical properties on their surfaces. Ultimately, 521 binding site clusters were detected, in addition to 945 singletons, totaling 1,466 different binding sites. Among the main features presented by Propedia v1, the binding-site search engine stands out. In practice, if a user wants to find peptides that bind to a specific known binding site, they would enter the PDB code, the chain, and the residues that comprise the site into the interface. Propedia would then use the ProBIS algorithm and its database of binding-site-based clusters to detect experimentally resolved peptides that potentially bind to that binding site.\u003c/p\u003e \u003cp\u003eThe second version of Propedia was released in late 2023 [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. It presented 49,300 protein-peptide complexes. Among the new features, it introduced a set of subdatasets classified according to the protein class extracted from the Protein Data Bank. Among the classifications, we can mention: antimicrobial, viral, enzyme, membrane, hormone, and plant. Furthermore, structural signatures were calculated for each protein-peptide complex, receptor, and peptide individually using the aCSM algorithm [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. Structural signatures are mathematical representations used to encode three-dimensional structural information in a one-dimensional vector space [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eIn this paper, we present Propedia 26, the third version of the protein-peptide interaction database. The version nomenclature has been changed to reflect the year of release. Propedia 26 contains 4 times as many entries as the first version and introduces several new features to characterize protein-peptide complexes better. Furthermore, Propedia 26 introduces calculations for each entry of seven types of interchain contacts: hydrogen bonds, attractive bonds, repulsive bonds, hydrophobic bonds, aromatic bonds, salt bridges, and disulfide bonds. We also offer several new features that may be beneficial for classification tasks utilizing machine learning algorithms.\u003c/p\u003e"},{"header":"Data Records","content":"\u003cp\u003eBriefly, Propedia 26 contains 98,779 entries, with 78,148 protein-peptide pair complexes (pep-pro) and 20,631 complexes composed of a single peptide interacting with two or more proteins (multipro). Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e summarizes the steps taken to collect the data that make up Propedia 26. First, the data were collected from the PDB database. We selected structures with peptides close to proteins (distance ≤ 6Å). We separated the protein-peptide structure pairs and calculated a set of features using various approaches.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFor each entry, Propedia 26 provides a comprehensive set of structural, physicochemical, and functional annotations. Complex-level characteristics include chain identifiers, structural descriptions, sequence length (in residues), molecular weight (Da), isoelectric point (pI), instability and aliphatic index, GRAVY (Grand Average of Hydropathy), percentage of hydrophobic residues, number of positively and negatively charged residues, atomic formula, molar extinction coefficient (M\u003csup\u003e− 1\u003c/sup\u003ecm\u003csup\u003e− 1\u003c/sup\u003e), and full sequence information. These characteristics were calculated using the ProtParam library [\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]. The dataset also lists structurally and sequentially similar complexes, facilitating comparison and similarity-based analyses.\u003c/p\u003e \u003cp\u003ePrevious versions of Propedia only considered protein-peptide interaction pairs. For example, consider the PDB entry 1A1R, which contains three chains (A, B, and C). Chain C belongs to a peptide that interacts simultaneously with the protein in chains A and B. By default, the Propedia pipeline saves two entries: 1A1R-C-A (peptide C interacting with the protein chain A) and 1A1R-C-B (peptide C interacting with the protein chain B). In this new version of Propedia, we decided to include a new type of entry that groups all chains interacting with the same peptide. We refer to this dataset as Propedia Multipro. In this case, the entry 1A1R-C would contain the C chains of the peptide and the A and B chains of proteins that are interacting with it. We did this for all entries in Propedia 26. In total, we collected 20,631 Multipro complexes.\u003c/p\u003e \u003cp\u003ePropedia 26 further reports the predicted functional profiles of each peptide, indicating the probability of exhibiting anti-angiogenic (AAP), antibacterial (ABP), anticancer (ACP), anti-inflammatory (AIP), quorum-sensing (QSP), or surface-binding (SBP) activity. The original Propedia v1 annotation schema, including Binding Site, Interface, and Sequence classes, is fully retained.\u003c/p\u003e \u003cp\u003eA detailed contact surface analysis is also provided. The Accessible Surface Area (ASA), expressed in Ų, quantifies the portion of the molecule exposed to solvent (typically water). The Buried Surface Area (BSA) represents the region effectively engaged at the binding interface. Additionally, we report the Buried Peptide surface Percentage (BPP%), indicating the fraction of the peptide that is not exposed to the solvent. All ASA- and BSA-related measurements were obtained using NACCESS [\u003cspan class=\"CitationRef\"\u003e16\u003c/span\u003e]. Additionally, Propedia 26 also includes predictions of protein-peptide interaction energy, estimating binding affinity (kcal/mol) and dissociation constant (M) at 25°C using the PRODIGY algorithm [\u003cspan class=\"CitationRef\"\u003e17\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFinally, the database identifies all interface residues, defined as protein residues within 6 Å of the peptide, and provides an atom-level catalog of interchain contacts. The computed contact types include hydrogen bonds, attractive and repulsive interactions, hydrophobic contacts, π-π stacking, salt bridges, and disulfide bonds. All contacts were calculated using the COCαDA algorithm [\u003cspan class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eA description of the table of entries in Propedia is provided in Supplementary Table \u003cspan class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e "},{"header":"Technical validation","content":"\u003cp\u003eData quality was assessed through random sampling of the dataset. A subset of entries was randomly selected, and the values presented in the Propedia interface were manually cross-checked against their respective primary data sources. To evaluate the descriptors, we consulted the Protein Data Bank page corresponding to the entry. For the analysis of physicochemical parameters, we collected the peptide sequence and used the ProtParam web tool (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://web.expasy.org/protparam\u003c/span\u003e\u003cspan class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). To evaluate the interaction energy, we consulted the Prodigy web tool, available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://rascar.science.uu.nl/prodigy\u003c/span\u003e\u003cspan class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. The list of intra-chain contacts was verified using the COCαDa web tool, available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://bioinfo.dcc.ufmg.br/cocada-web\u003c/span\u003e\u003cspan class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. The NACCESS surface area data were evaluated using in-house Python scripts. In addition, the Orange Data Mining tool was used to assess the feature statistics and detect distribution, mean, mode, median, dispersion, minimum and maximum values, and missing values.\u003c/p\u003e\u003cp\u003eMissing data in Propedia 26 arise from isolated issues during the execution of specific computational tools, as detailed in the Methods section. For example, the presence of non-canonical amino acids or high cutoff distances between chains (by default, Propedia defines the maximum distance between chains as 6Å, but other tools may require smaller distances) can lead to errors in the tools' calculations. Overall, data completeness is high across the dataset. Protein surface area measurements are missing for 256 entries (\u0026lt; 1% of the dataset), while predicted therapeutic properties are unavailable for 1,875 structures (2%). Interaction energy predictions could not be computed for 5,010 entries (6%), primarily due to limitations in structural quality or contact detection required by the prediction algorithm. Predicted physicochemical properties for protein chains are missing for 41 entries (\u0026lt; 1%). Importantly, all entries include experimentally derived complex-level information obtained from the Protein Data Bank, and 100% of peptide chains contain predicted physicochemical descriptors. These statistics indicate that missing values are limited, well-characterized, and confined to specific derived features, ensuring the overall reliability and reusability of the dataset. The data report is available on the supplementary material's GitHub repository.\u003c/p\u003e"},{"header":"Usage Notes","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003ePropedia 26 web interface\u003c/h2\u003e \u003cp\u003ePropedia 26 has a user-friendly web interface designed to simplify navigation and search of structures. There are three entry search mechanisms: (i) sequence-based search (uses BLAST to perform local alignments with parameters optimized for peptide search); (ii) binding site-based search (uses ProBis to detect structures with cavities similar to the region indicated in the target protein by users); and (iii) search based on Propedia ID, PDB ID, or entry metadata (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e - top). Each entry in Propedia has its own page with information such as a description of the entry, a presentation of physicochemical parameters detected \u003cem\u003ein silico\u003c/em\u003e, and an interactive visualization of the 3D structure (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e - below).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe entry page also presents similar structures, predicted characteristics for therapeutic peptides, and the clusters inherited from Propedia v1 (based on sequence, interface, and binding site; Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Furthermore, information on protein-peptide interactions is calculated using different software. Solvent-accessible area is predicted using NACCESS [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. We display information about the solvent-exposed area for ​​the complex, only the peptide, and only the protein (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). In addition, we show the buried areas of the peptide (BPepA) and the protein (BProA), corresponding to the regions that mediate the protein-peptide interaction. Note that, mostly, these regions occupy different area values ​​in the protein and in the peptide. Therefore, we use the average of these values ​​to indicate the area of ​​the binding interface (\u003cem\u003ei.e.\u003c/em\u003e, buried surface area - BSA). We also report the percentage of the peptide that is buried (BPP%). The higher the BPP% value, the more deeply embedded the peptide is, and therefore, the stronger the bond tends to be. However, it should be noted that the peptide may form bonds with other protein chains (see the complex's multipro page for more details).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eInteraction energy, binding affinity, and dissociation constant are predicted using Prodigy (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB). We also list the residues in the protein interface (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eC). Finally, interchain contacts were calculated using the COCαDA-CLI (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eD). The contact list table is integrated into the 3D structure visualization. Clicking on a listed contact displays the interacting residues as sticks (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eE).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eCase study: Searching for peptides that inhibit the SARS-CoV-2 Mpro protein\u003c/h3\u003e\n\u003cp\u003eIn this section, we illustrate the application of the binding site search algorithm to identify peptides that interact with the SARS-CoV-2 Mpro protein. This section is not intended to provide an in-depth biological analysis of the case studies or predictions; instead, it aims to offer illustrative examples of the database entries and their potential use cases.\u003c/p\u003e \u003cp\u003eSARS-CoV-2 remains a global concern due to its complex pathogenesis and its potential to cause severe disease, particularly when viral replication is not effectively controlled in target tissues [\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Mpro (Main Protease), also called 3CLpro, is one of the most essential enzymes for the SARS-CoV-2 life cycle. It plays a central role in viral replication, which is why it has become one of the most critical targets for antiviral drug development.\u003c/p\u003e \u003cp\u003eThe Propedia binding site search requires as input a PDB code of the target protein, the chain, and a list of residue codes. The tool then uses the ProBis algorithm to search for structures with binding sites that have conformations similar to those of the input structure.\u003c/p\u003e \u003cp\u003eThe binding site used in our search was defined according to recent structural studies [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e], focusing on the key residues that form the catalytic region of Mpro. Propedia 26 identified 180 protein-peptide complexes matching the defined binding site, whereas Propedia v1 returned only 34. This increase reflects the expansion of the database, in particular, due to the large number of recent studies on SARS-CoV-2. This highlights the importance of updating the Propedia database. Figure\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e presents the top three peptides obtained in our search, ranked by Alignment Score and RMSD.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eData collection\u003c/h2\u003e \u003cp\u003eWe collected all structures with a maximum of 10,000 amino acid residues from the Protein Data Bank (PDB) [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] in CIF format on September 8, 2025 (n\u0026thinsp;=\u0026thinsp;236,616). Then, we filtered all complexes performing protein-peptide interactions using Python scripts. We defined a complex as exhibiting a protein\u0026ndash;peptide interaction when at least one atom from the protein and one atom from the peptide were separated by a distance of 6 \u0026Aring; or less. Additionally, we extracted all combinations of chain pairs where one chain had between 2 and 50 amino acids, while the other had more than 50 amino acids. In total, we extracted 78,148 protein-peptide complexes.\u003c/p\u003e \u003cp\u003eFurthermore, we created a sub-dataset comprising all protein chains within a maximum distance of 6 \u0026Aring; from the peptide. We called this the Multipro dataset. In this case, each entry point of the multipro is composed of one peptide interacting with two or more protein chains. In total, we obtained 20,631 multipro entries.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eClustering and Redundancy Detection\u003c/h3\u003e\n\u003cp\u003eTo detect redundancy in protein-peptide complexes, we concatenated the protein and peptide sequences and grouped identical sequence combinations. In total, we detected 51,082 unique protein-peptide complexes.\u003c/p\u003e \u003cp\u003eThen, we collected all peptide sequences containing only canonical amino acids and grouped those that were 100% identical. In total, we extracted 17,509 unique sequences composed only of the 20 canonical amino acids. We refer to this peptide dataset as the canonical non-redundant (CNR) dataset.\u003c/p\u003e \u003cp\u003eClusters described in previous versions of Propedia have been recalculated using Python scripts. Other classification information was obtained from the PDB file using scripts developed with the Biopython library (Bio.PDB package) [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e\n\u003ch3\u003eFeature extraction\u003c/h3\u003e\n\u003cp\u003eFor each protein-peptide complex, physicochemical properties were obtained using ProtParam [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] embedded in the Biopython library v1.85 [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. The collected information included the chain identifier, description, sequence length (number of residues), molecular weight (Da), isoelectric point (pI), instability index, aliphatic index, GRAVY (grand average of hydropathicity), hydrophobicity (%), total number of positively and negatively charged residues, atomic formula, number of atoms, extinction coefficient (with and without disulfide bonds), and the full amino acid sequence.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eProtein-peptide interactions\u003c/h2\u003e \u003cp\u003eProtein interaction interface residues were calculated using Python scripts. The IDs of proteins\u0026rsquo; residues at a distance\u0026thinsp;\u0026le;\u0026thinsp;6 \u0026Aring; from the peptide were annotated.\u003c/p\u003e \u003cp\u003eThe solvent-accessible surface area (ASA) was calculated using NACCESS v2.1.1 [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. For each protein\u0026ndash;peptide complex, the total ASA of the complex, the changes in ASA for the protein and the peptide upon binding (ΔASA), and the buried surface area (BSA) were computed and reported in square angstroms (\u0026Aring;\u0026sup2;). The Buried Surface Area (BSA), which represents the region effectively engaged at the binding interface, is computed as:\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:BSA\\:=\\:\\frac{AS{A}_{protein}\\:+\\:AS{A}_{peptide}\\:-\\:AS{A}_{complex}}{2}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eFurthermore, we calculated the buried area of ​​peptides (BPepA) and proteins (BProA). Finally, we calculated the buried peptide percentage (BPP%) using the formula:\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equb\" name=\"EquationSource\"\u003e\n$$\\:BPP\\left(\\%\\right)\\:=\\:\\frac{100\\:\\times\\:\\:BPep{A}_{}}{AS{A}_{peptide}}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e \u003cp\u003eThe interaction energy of each protein\u0026ndash;peptide complex was calculated using PRODIGY v.2.4.0 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. For each complex, the total number of interfacial contacts (categorized by amino acid type), the predicted binding affinity (kcal/mol), and the predicted dissociation constant (M) at 25\u0026deg;C were computed.\u003c/p\u003e \u003cp\u003eChain contacts were calculated using the COCαDA command-line tool v1.0 [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]. For each complex, COCαDA calculated hydrogen bonds, attractive and repulsive interactions, hydrophobic interactions, aromatic interactions, disulfide bonds, and salt bridges.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eStructural signature\u003c/h2\u003e \u003cp\u003eFor each three-dimensional structure of unique peptides composed only of canonical amino acids (CNR dataset), we calculated structural signatures using the SIGNA library (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/LBS-UFMG/signa\u003c/span\u003e\u003cspan address=\"https://github.com/LBS-UFMG/signa\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Structural signatures convert the 3D structure into one-dimensional vectors that can be used in machine learning tasks. For the calculation of structural signatures, we used the aCSM-ALL algorithm [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e] with a maximum distance parameter of 10 \u0026Aring; and a step of 0.2 \u0026Aring;. These parameters were based on previous studies [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e, \u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eSequence signature\u003c/h2\u003e \u003cp\u003eFurthermore, we extracted features based on the sequences using the iFeature tool (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/Superzchen/iFeature\u003c/span\u003e\u003cspan address=\"https://github.com/Superzchen/iFeature\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003cspan type=\"Underline\" class=\"Underline\" name=\"Emphasis\"\u003e)\u003c/span\u003e [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. We named the extracted sequence features as sequence signatures. Sequence-based signatures can be crucial for machine learning tasks that involve predictions where three-dimensional structures are unavailable. We selected the 10 following descriptors: AAC, DPC, DDE, GAAC, GDPC, GTPC, CTDC, CTDT, CTDD, and CTriad.\u003c/p\u003e \u003cp\u003eThe Amino Acid Composition (AAC) descriptor represents the relative frequency of each of the 20 amino acids in a peptide sequence. The Dipeptide Composition (DPC) descriptor extends this idea by quantifying the frequencies of all 400 possible dipeptide pairs, thereby capturing short-range order information. The Dipeptide Deviation from Expected mean (DDE) descriptor measures how much the observed dipeptide frequencies deviate from values expected based on individual amino acid compositions, reflecting pairwise dependency patterns. The Grouped Amino Acid Composition (GAAC) descriptor simplifies the sequence by clustering amino acids into physicochemical groups and computing their overall proportions. Similarly, the Grouped Dipeptide Composition (GDPC) and Grouped Tripeptide Composition (GTPC) descriptors compute the frequencies of dipeptides or tripeptides based on predefined amino-acid groups rather than individual residues, reducing dimensionality while retaining key biochemical patterns. The Composition Transition Distribution Composition (CTDC) descriptor summarizes the fraction of residues belonging to different physicochemical categories, whereas CTD Transition (CTDT) quantifies how often the sequence switches between these categories. CTD Distribution (CTDD) further describes the positional distribution of residues from each category along the sequence. Finally, the Conjoint Triad (CTriad) descriptor groups amino acids into seven classes and measures the frequency of all possible triads of these groups, capturing local physicochemical environments.\u003c/p\u003e \u003cp\u003eFinally, iFeature returns a vector of 1248 features for each sequence used as input. In this work, we refer to this vector as a sequence signature.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eTherapeutic peptides classification\u003c/h2\u003e \u003cp\u003eWe classified each peptide in Propedia into six classes according to their predicted therapeutic properties: Anti-Angiogenic (AAP), Antibacterial (ABP), Anticancer (ACP), Anti-Inflammatory (AIP), Quorum Sensing (QSP), and Surface Binding (SBP). The classification was performed using machine learning models with Orange Data Mining [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. To build the models, we used sequence-based signatures and attempted to reproduce the methodology described in the article by Rodrigues et al. [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. Details of the methodology are available in our GitHub\u0026rsquo;s supplementary material repository (see Data Availability section).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eUser-friendly interface development\u003c/h2\u003e \u003cp\u003eThe web interface was built using the same approaches used in [\u003cspan additionalcitationids=\"CR33 CR34\" citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]. The CodeIgniter 4 framework was used for the back-end. The Bootstrap and DataTables libraries were used in the front-end application. Interactive protein visualizations were built using the 3Dmol.js library [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Contact map graphs were built using the Chat.js library. Figures used to illustrate the sections were constructed using the ChimeraX tool [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe sequence-based search system was built using BLAST [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]. To enable searches for peptide sequences, the following parameters were used: \"-word_size 2 -task blastp-short -seg no -evalue 100000\". For protein sequence searches, only the parameter \"-word_size 3\" was employed.\u003c/p\u003e \u003cp\u003eTo search for similar binding sites, the Propedia back-end uses the ProBiS tool [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. The ProBiS algorithm detects local structural similarities by comparing the three-dimensional surface of the protein's binding site with the 3D surfaces of proteins stored in the Propedia-26 database. To reduce search costs, we selected only non-redundant structures. Additionally, some structures exhibited errors during the extraction of the peptide interaction surface. In total, 32,695 protein-peptide complex surfaces are used in the search for similar binding sites.\u003c/p\u003e \u003c/div\u003e "},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eCompeting interests\u003c/h2\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis work was supported by grants from the Funda\u0026ccedil;\u0026atilde;o de Amparo \u0026agrave; Pesquisa do Estado de Minas Gerais (FAPEMIG, projects APQ-02690-22 and APQ-01834-21) and the Conselho Nacional de Desenvolvimento Cient\u0026iacute;fico e Tecnol\u0026oacute;gico (CNPq, project 440307/2022-8). This study was financed in part by the Coordena\u0026ccedil;\u0026atilde;o de Aperfei\u0026ccedil;oamento de Pessoal de N\u0026iacute;vel Superior \u0026ndash; Brasil (CAPES) \u0026ndash; Finance Code 001.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eP.M., D.M., and R.C.M.M contributed to the conceptualization; D.M. collected the data and developed the web tool; L.H.S., M.M.P, and L.M. performed the case studies; M.M.P wrote the documentation; L.B. performed the surface analysis; A.A. performed the analysis of interaction energy; R.L. performed the contact analysis; D.M. performed the analysis of physical-chemical parameters; R.C.M.M. acquired funds and supervised the research. D.M. prepared the manuscript with the assistance and feedback from all other co-authors.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors would like to express their gratitude to the research funding agencies: CAPES, CNPq, and FAPEMIG. The authors thank Felipe Concei\u0026ccedil;\u0026atilde;o for his valuable contributions.\u003c/p\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eData availability\u003c/h2\u003e \u003cp\u003eThe web tool can be found at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://bioinfo.dcc.ufmg.br/propedia26\u003c/span\u003e\u003cspan address=\"https://bioinfo.dcc.ufmg.br/propedia26\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. The dataset and data descriptors can be downloaded at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://bioinfo.dcc.ufmg.br/propedia26/public/download\u003c/span\u003e\u003cspan address=\"https://bioinfo.dcc.ufmg.br/propedia26/public/download\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Full supplementary material is available at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/LBS-UFMG/propedia26-sm\u003c/span\u003e\u003cspan address=\"https://github.com/LBS-UFMG/propedia26-sm\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Data descriptors are available at Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec17\" class=\"Section2\"\u003e \u003ch2\u003eCode availability\u003c/h2\u003e \u003cp\u003eThe source code of the web tool can be found at: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/LBS-UFMG/propedia26\u003c/span\u003e\u003cspan address=\"https://github.com/LBS-UFMG/propedia26\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. The Orange Data Mining machine learning models are available in our GitHub supplementary material repository: \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://github.com/LBS-UFMG/propedia26-sm\u003c/span\u003e\u003cspan address=\"https://github.com/LBS-UFMG/propedia26-sm\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e \u003c/div\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eSauvestre, C., Zagury, J.-F. \u0026amp; Langenfeld, F. Comparative Analysis of Deep Learning-Based Algorithms for Peptide Structure Prediction. \u003cem\u003eProteins Struct. Funct. Bioinforma.\u003c/em\u003e n/a, (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNeduva, V. \u003cem\u003eet al.\u003c/em\u003e Systematic discovery of new recognition peptides mediating protein interaction networks. \u003cem\u003ePLoS Biol.\u003c/em\u003e 3, e405 (2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee, A. C.-L., Harris, J. L., Khanna, K. K. \u0026amp; Hong, J.-H. A comprehensive review on current advances in peptide drug development and design. \u003cem\u003eInt. J. Mol. Sci.\u003c/em\u003e 20, 2383 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, L. \u003cem\u003eet al.\u003c/em\u003e Therapeutic peptides: current applications and future directions. \u003cem\u003eSignal Transduct. Target. Ther.\u003c/em\u003e 7, 48 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCabral, H. Peptides: Molecular and Biotechnological Aspects. \u003cem\u003eBiomolecules\u003c/em\u003e 11, 52 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTripathi, N. M. \u0026amp; Bandyopadhyay, A. High throughput virtual screening (HTVS) of peptide library: Technological advancement in ligand discovery. \u003cem\u003eEur. J. Med. Chem.\u003c/em\u003e 243, 114766 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartins, P. M. \u003cem\u003eet al.\u003c/em\u003e Propedia: a database for protein\u0026ndash;peptide identification based on a hybrid clustering algorithm. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e 22, 1 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerman, H. M. \u003cem\u003eet al.\u003c/em\u003e The Protein Data Bank. \u003cem\u003eNucleic Acids Res.\u003c/em\u003e 28, 235\u0026ndash;242 (2000).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrejci, A., Hupp, T. R., Lexa, M., Vojtesek, B. \u0026amp; Muller, P. Hammock: a hidden Markov model-based peptide clustering algorithm to identify protein-interaction consensus motifs in large datasets. \u003cem\u003eBioinforma. Oxf. Engl.\u003c/em\u003e 32, 9\u0026ndash;16 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKonagurthu, A. S., Whisstock, J. C., Stuckey, P. J. \u0026amp; Lesk, A. M. MUSTANG: a multiple structural alignment algorithm. \u003cem\u003eProteins\u003c/em\u003e 64, 559\u0026ndash;574 (2006).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKonc, J. \u0026amp; Janežič, D. ProBiS algorithm for detection of structurally similar protein binding sites by local structural alignment. \u003cem\u003eBioinformatics\u003c/em\u003e 26, 1160\u0026ndash;1168 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMartins, P. \u003cem\u003eet al.\u003c/em\u003e Propedia v2.3: A novel representation approach for the peptide-protein interaction database using graph-based structural signatures. \u003cem\u003eFront. Bioinforma.\u003c/em\u003e 3, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePires, D. E. V., de Melo-Minardi, R. C., da Silveira, C. H., Campos, F. F. \u0026amp; Meira, W. aCSM: noise-free graph-based signatures to large-scale receptor-based ligand prediction. \u003cem\u003eBioinforma. Oxf. Engl.\u003c/em\u003e 29, 855\u0026ndash;861 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMariano, D. \u003cem\u003eet al.\u003c/em\u003e A Computational Method to Propose Mutations in Enzymes Based on Structural Signature Variation (SSV). \u003cem\u003eInt. J. Mol. Sci.\u003c/em\u003e 20, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGasteiger, E. \u003cem\u003eet al.\u003c/em\u003e Protein identification and analysis tools on the ExPASy server. in \u003cem\u003eThe proteomics protocols handbook\u003c/em\u003e 571\u0026ndash;607 (Springer, 2005).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHubbard, S. J. \u0026amp; Thornton, J. naccess. \u003cem\u003eDept Biohem Mol Biol UCL\u003c/em\u003e (1993).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVangone, A. \u0026amp; Bonvin, A. M. Contacts-based prediction of binding affinity in protein\u0026ndash;protein complexes. \u003cem\u003eelife\u003c/em\u003e 4, e07454 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLemos, R., Mariano, D., Silveira, S. \u0026amp; Melo-Minardi, R. \u003cem\u003eCOCαDA - Large-Scale Protein Interatomic Contact Cutoff Optimization by Cα Distance Matrices\u003c/em\u003e. 70 (2024). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5753/bsb.2024.245545\u003c/span\u003e\u003cspan address=\"10.5753/bsb.2024.245545\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIbrahim, M. \u003cem\u003eet al.\u003c/em\u003e Why is the Omicron main protease of SARS-CoV-2 less stable than its wild-type counterpart? A crystallographic, biophysical, and theoretical study of the free enzyme and its complex with inhibitor 13b-K. 2024.03.04.583178 Preprint at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1101/2024.03.04.583178\u003c/span\u003e\u003cspan address=\"10.1101/2024.03.04.583178\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLamers, M. M. \u0026amp; Haagmans, B. L. SARS-CoV-2 pathogenesis. \u003cem\u003eNat. Rev. Microbiol.\u003c/em\u003e 20, 270\u0026ndash;284 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHashemi, Z. S. \u003cem\u003eet al.\u003c/em\u003e In silico Approaches for the Design and Optimization of Interfering Peptides Against Protein\u0026ndash;Protein Interactions. \u003cem\u003eFront. Mol. Biosci.\u003c/em\u003e 8, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChaves Carvalho, F. \u003cem\u003eet al.\u003c/em\u003e A hybrid evolutionary and structural method for AI-guided peptide inhibitor design using AlphaFold and Rosetta. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s41598-025-28061-y\u003c/span\u003e\u003cspan address=\"10.1038/s41598-025-28061-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2025) doi:10.1038/s41598-025-28061-y.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCock, P. J. A. \u003cem\u003eet al.\u003c/em\u003e Biopython: freely available Python tools for computational molecular biology and bioinformatics. \u003cem\u003eBioinforma. Oxf. Engl.\u003c/em\u003e 25, 1422\u0026ndash;1423 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHamelryck, T. \u0026amp; Manderick, B. PDB file parser and structure class implemented in Python. \u003cem\u003eBioinformatics\u003c/em\u003e 19, 2308\u0026ndash;2310 (2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLemos, R. P., Mariano, D., Silveira, S. de A. \u0026amp; de Melo-Minardi, R. C. COCαDA-A Fast and Scalable Algorithm for Interatomic Contact Detection in Proteins Using Cα Distance Matrices. \u003cem\u003eFront. Bioinforma.\u003c/em\u003e 5, 1630078 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMoreira, E. U. \u003cem\u003eet al.\u003c/em\u003e Predicting Mutation-Driven Changes in the SARS-CoV-2 Spike Protein Using Structural Signatures and Neural Networks. in 167\u0026ndash;178 (SBC, 2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMariano, D. \u003cem\u003eet al.\u003c/em\u003e Using Machine Learning and Graph-Based Signatures to Evaluate Normal and Decreased Function CYP2D6 Haplotypes. in 90\u0026ndash;105 (Springer, 2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCioletti, A. G., Lemos, R. P., dos Santos, L. M., Mariano, D. \u0026amp; de Melo-Minardi, R. C. Using graph-based structural signatures and machine learning algorithms for molecular docking assessment of histone deacetylases and small ligands. in 223\u0026ndash;234 (SBC, 2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, Z. \u003cem\u003eet al.\u003c/em\u003e iFeature: a Python package and web server for features extraction and selection from protein and peptide sequences. \u003cem\u003eBioinformatics\u003c/em\u003e 34, 2499\u0026ndash;2502 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDemšar, J., Zupan, B., Leban, G. \u0026amp; Curk, T. Orange: From experimental machine learning to interactive data mining. in 537\u0026ndash;539 (Springer, 2004).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRodrigues, C. H. M., Garg, A., Keizer, D., Pires, D. E. V. \u0026amp; Ascher, D. B. CSM-peptides: A computational approach to rapid identification of therapeutic peptides. \u003cem\u003eProtein Sci.\u003c/em\u003e 31, e4442 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilva, M. F. M. \u003cem\u003eet al.\u003c/em\u003e Proteingo: Motivation, user experience, and learning of molecular interactions in biological complexes. \u003cem\u003eEntertain. Comput.\u003c/em\u003e 29, 31\u0026ndash;42 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFassio, A. V., Santos, L. H., Silveira, S. A., Ferreira, R. S. \u0026amp; de Melo-Minardi, R. C. nAPOLI: a graph-based strategy to detect and visualize conserved protein-ligand interactions in large-scale. \u003cem\u003eIEEE/ACM Trans. Comput. Biol. Bioinform.\u003c/em\u003e \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TCBB.2019.2892099\u003c/span\u003e\u003cspan address=\"10.1109/TCBB.2019.2892099\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2019) doi:10.1109/TCBB.2019.2892099.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarroso, J. R. M. S. \u003cem\u003eet al.\u003c/em\u003e Proteus: An algorithm for proposing stabilizing mutation pairs based on interactions observed in known protein 3D structures. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e 21, 275 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePimentel, V. \u003cem\u003eet al.\u003c/em\u003e VTR: a web tool for identifying analogous contacts on protein structures and their complexes. \u003cem\u003eFront. Bioinforma.\u003c/em\u003e 1, 28 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRego, N. \u0026amp; Koes, D. 3Dmol.js: molecular visualization with WebGL. \u003cem\u003eBioinformatics\u003c/em\u003e 31, 1322\u0026ndash;1324 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMeng, E. C. \u003cem\u003eet al.\u003c/em\u003e UCSF ChimeraX: Tools for structure building and analysis. \u003cem\u003eProtein Sci.\u003c/em\u003e 32, e4792 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAltschul, S. F., Gish, W., Miller, W., Myers, E. W. \u0026amp; Lipman, D. J. Basic local alignment search tool. \u003cem\u003eJ. Mol. Biol.\u003c/em\u003e 215, 403\u0026ndash;410 (1990).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCamacho, C. \u003cem\u003eet al.\u003c/em\u003e BLAST+: architecture and applications. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e 10, 421 (2009).\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8349237/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8349237/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePropedia is a curated database of protein-peptide interactions. In this new release, we expanded the dataset to more than four times the size of the previous version and introduced a comprehensive set of new annotations. We systematically analyzed protein-peptide complexes to derive predicted physicochemical properties, solvent-accessible surface areas, interaction energies, interface residues, and interchain contacts. In addition, we applied machine learning models to predict the potential therapeutic functions of the peptides. Propedia 26 comprises 98,779 entries, including 78,148 protein\u0026ndash;peptide pair complexes (pep-pro) and 20,631 multi-protein complexes featuring a single peptide interacting with one or more protein chains (multipro). We anticipate that this expanded and enriched resource will support the discovery and development of peptide-based drugs and biotechnological products. Propedia 26 is freely accessible at \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://bioinfo.dcc.ufmg.br/propedia26\u003c/span\u003e\u003cspan address=\"https://bioinfo.dcc.ufmg.br/propedia26\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e","manuscriptTitle":"Propedia 26: An expanded and updated database of protein-peptide interactions for machine learning applications","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-27 12:33:35","doi":"10.21203/rs.3.rs-8349237/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1ae1cc77-7051-4f7a-a98b-5dead662fbea","owner":[],"postedDate":"February 27th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-05T14:09:13+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-27 12:33:35","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8349237","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8349237","identity":"rs-8349237","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}