Of the 38 distinct peptides that matched a bound ligand most closely, 25 occur as
complete loops in real protein structures. Seven are superposed here in the
conformation the loop actually adopts in its own crystal.
Companion to the forty closest ligand and peptide pairs.
1,524,535 loops were searched, drawn from the same corpus used to train
PharmCast. Every loop shown comes
from a protein unrelated to the one the ligand was solved against.
1,524,535loops searched
25 of 38peptides that exist as a real loop
608loop instances across them
1.80 Åmedian resolution of those loops
4 of 7solvent exposed at RSA 0.25 or better
1,524,535loops searched
25 of 38peptides that exist as a real loop
608loop instances across them
1.80 Åmedian resolution of those loops
4 of 7ligands peptide derived in their own literature
The real loop against a generated conformer
Each sequence can be posed two ways: as an idealized low energy conformer
built for it, or as the geometry the loop genuinely holds in its own crystal. Both were
fitted onto the ligand and scored the same way, by Open3DAlign on Crippen contributions,
where a higher score is a closer fit. Cathepsin K inhibitor FSP is the case where the
crystallographic loop wins outright: it scores 74.3 against 52.3 for the generated
conformer of the same sequence, and it shares eleven pharmacophore features with the
ligand where the generated conformer shares five. Every case below is drawn with the
crystallographic loop.
Seven cases
Pharmacophore similarity is the PharmCast v10 Tanimoto that selected the peptide. Loop fit is the Open3DAlign score of the crystallographic loop onto the ligand, where a higher score is a closer fit, computed independently of that similarity. Shared features counts places where the ligand and the real loop present the same feature type within 2.5 angstroms. The badge records whether the ligand’s own primary literature describes it as peptide derived.
#1
ZGW
→AMVPNpeptidomimetic
0.913pharmacophore similarity
0.167Morgan
79.6loop fit, Open3DAlign
9shared features
nirmatrelvir; best library match AMVP, the real loop is AMVPN
Nirmatrelvir
Ligand complex
8GFU Contribution of the catalytic dyad of SARS-CoV-2 main protease to binding covalent and noncovalent inhibitors. J.Biol.Chem. 2023 299
Loop source
2JHX chain A residues 65 to 69 CRYSTAL STRUCTURE OF RHOGDI E155H, E157H MUTANT Protein Crystallization by Surface Entropy Reduction: Optimization of the Ser Strategy Acta Crystallogr.,Sect.D 2007 63
Loop exposure
relative solvent accessibility 0.679, resolution 1.60 Å
Peptide origin of the ligand
Nirmatrelvir, the Paxlovid active ingredient. A designed peptidomimetic covalent inhibitor built against the enzyme’s peptide substrate.
acceptor ×2donor ×3hydrophobic ×4
Against a generated low energy conformer of the same sequence the fit is 87.3 with 10 shared features. The real loop gives 79.6 with 9, a change of -7.7.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 2JHX, not a generated conformer.
4R5A Structural Insights into Carbonic Anhydrase IX Isoform Specificity of Carbohydrate-Based Sulfamates. J.Med.Chem. 2014 57
Loop source
4MYE chain A residues 68 to 70 Cymosema roseum seed lectin structure complexed with X-man Cymosema roseum seed lectin structure complexed with X-man To be Published
Loop exposure
relative solvent accessibility 0.647, resolution 1.65 Å
Peptide origin of the ligand
A carbohydrate based sulfamate.
acceptor ×3donor ×2
Against a generated low energy conformer of the same sequence the fit is 63.7 with 8 shared features. The real loop gives 50.8 with 5, a change of -12.9.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 4MYE, not a generated conformer.
4RN0 Variable Active Site Loop Conformations Accommodate the Binding of Macrocyclic Largazole Analogues to HDAC8. Biochemistry 2015 54
Loop source
4ADB chain D residues 3 to 5 Structural and functional study of succinyl-ornithine transaminase from E. coli Determination of the Structure of the Catabolic N-Succinylornithine Transaminase (Astc) from Escherichia Coli. Plos One 2013 8
Loop exposure
relative solvent accessibility 0.593, resolution 2.20 Å
Peptide origin of the ligand
A macrocyclic largazole analogue. Largazole is a natural product depsipeptide, so the scaffold is peptide derived. The paper is titled for variable active site loop conformations.
donor ×2hydrophobic ×2
Against a generated low energy conformer of the same sequence the fit is 59.4 with 3 shared features. The real loop gives 68.6 with 4, a change of +9.2.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 4ADB, not a generated conformer.
5J20 Protein conformational flexibility modulates kinetics and thermodynamics of drug binding. Nat Commun 2017 8
Loop source
8JMZ chain A residues 487 to 489 FGFR1 kinase domain with sulfatinib Structural basis and selectivity of sulfatinib binding to FGFR and CSF-1R. Commun Chem 2024 7
Loop exposure
relative solvent accessibility 0.469, resolution 1.99 Å
Peptide origin of the ligand
A triazolone.
acceptor ×2donor ×2hydrophobic ×1
Against a generated low energy conformer of the same sequence the fit is 56.4 with 8 shared features. The real loop gives 60.8 with 5, a change of +4.4.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 8JMZ, not a generated conformer.
6M0K Structure-based design of antiviral drug candidates targeting the SARS-CoV-2 main protease. Science 2020 368
Loop source
4PFT chain A residues 514 to 516 Crystal structure of mannobiose bound oligopeptide ABC transporter, periplasmic oligopepti Duplication of Genes in an ATP-binding Cassette Transport System Increases Dynamic Range While Maintaining Ligand Specificity. J.Biol.Chem. 2014 289
Loop exposure
relative solvent accessibility 0.037, resolution 1.75 Å
Peptide origin of the ligand
A structure based peptidomimetic aldehyde from the Science 2020 series. Its loop match comes from an oligopeptide ABC transporter periplasmic binding protein, a protein whose function is binding peptides.
acceptor ×2donor ×2hydrophobic ×5aromatic ×2
Against a generated low energy conformer of the same sequence the fit is 126.9 with 12 shared features. The real loop gives 86.7 with 11, a change of -40.2.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 4PFT, not a generated conformer.
5B5P Discovery of Novel, Highly Potent, and Selective Matrix Metalloproteinase (MMP)-13 Inhibitors with a 1,2,4-Triazol-3-yl Moiety as a Zinc Binding Group Using a Structure-Based Design Approach J. Med. Chem. 2017 60
Loop source
8SQ7 chain F residues 189 to 192 X-ray crystal structure of Acinetobacter baumanii beta-lactamase variant OXA-82 K83D in co Structural and Dynamic Features of Acinetobacter baumannii OXA-66 beta-Lactamase Explain Its Stability and Evolution of Novel Variants. J.Mol.Biol. 2024 436
Loop exposure
relative solvent accessibility 0.218, resolution 1.78 Å
Peptide origin of the ligand
A triazolyl zinc binding series.
acceptor ×1donor ×2hydrophobic ×2
Against a generated low energy conformer of the same sequence the fit is 74.8 with 10 shared features. The real loop gives 68.7 with 5, a change of -6.1.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 8SQ7, not a generated conformer.
1TU6 Potent and selective P2-P3 ketoamide inhibitors of cathepsin K with improved pharmacokinetic properties via favorable P1', P1, and/or P3 substitutions BIOORG.MED.CHEM.LETT. 2004 14
Loop source
7V1X chain A residues 92 to 94 Difructose dianhydride I synthase/hydrolase (alphaFFase1) from Bifidobacterium dentium in Identification of difructose dianhydride I synthase/hydrolase from an oral bacterium establishes a novel glycoside hydrolase family. J.Biol.Chem. 2021 297
Loop exposure
relative solvent accessibility 0.068, resolution 1.76 Å
Peptide origin of the ligand
Described in its own paper in Schechter and Berger subsite terms, a P2-P3 ketoamide with P1’, P1 and P3 substitutions. That is the explicit language of substrate mimicry.
acceptor ×2donor ×3hydrophobic ×5aromatic ×1
Against a generated low energy conformer of the same sequence the fit is 52.3 with 5 shared features. The real loop gives 74.3 with 11, a change of +22.0.
Heavy dark sticks are the ligand. Light sticks are the loop in the conformation it adopts in 7V1X, not a generated conformer.
Designing from peptide loops is established practice
Two bodies of work document turning a short peptide taken from a folded protein into a drug-like molecule.
Protein epitope mimetics
Robinson and colleagues transplant a beta-hairpin loop sequence out of a folded protein onto a hairpin stabilizing template, typically the D-Pro-L-Pro dipeptide, producing a synthetic molecule that reproduces the loop’s conformation and its biological activity. The approach has produced ligands for CXCR4 and for the bacterial outer membrane protein LptD. This is the direct precedent for reading a drug back to a loop.
The N-terminal tetrapeptide of Smac and DIABLO, Ala-Val-Pro-Ile, binds a groove on the BIR3 domain of XIAP and carries comparable affinity to the full protein. Structure based design converted that tetrapeptide into conformationally constrained small molecules, and the resulting clinical candidates include birinapant, LCL161, GDC-0152 and AT-406. It is the clearest case of a real four residue protein epitope becoming a drug.
Scored against our ligand set, that same AVPI tetrapeptide reaches a pharmacophore Tanimoto of 0.809 with nirmatrelvir, 4.7 standard deviations above nirmatrelvir’s mean over all 3,368,420 peptides. The main protease substrate register Leu-Gln-Ser reaches 0.751 on the same molecule, 4.2 standard deviations above its mean.
Sun H et al. Structure based design of conformationally constrained peptidomimetics as XIAP antagonists, 2004.
Loop corpus. 1,524,535 loops extracted from Protein Data Bank entries, the same corpus used to train PharmCast. Each carries its sequence, parent entry, chain, residue range, resolution and relative solvent accessibility.
Selection. A peptide counts as a real loop when its sequence is the entire loop, not a fragment of a longer one. For each, the instance shown is the most solvent exposed, with resolution breaking ties, since a loop buried in a protein core presents nothing to anything. Across the 608 instances the median relative solvent accessibility is 0.237 and 49 percent are exposed at 0.25 or better.
The loop geometry is not generated. The loop is taken directly from its entry, the named chain and residue range, alternate location A where present. It is used exactly as deposited. Only the ligand is embedded with ETKDGv3, minimized with MMFF94, and kept at its lowest energy conformer of twenty five.
The fit. The loop is superposed onto the ligand with Open3DAlign on Crippen contributions, which aligns on physicochemical character rather than shared atoms. This is required because the pairs share almost no substructure.
Shared features. RDKit base feature definitions on both partners. A feature counts as shared when the two are the same type and their centers fall within 2.5 angstroms, matched greedily closest first. Clouds are drawn at the midpoint of each agreeing pair.
Four of these seven ligands are peptide derived or explicitly peptidomimetic in their own
primary literature, and the method selected a peptide match for each of them from the fingerprint
alone. We are happy to walk through the method and what it is useful for.
