---

title: Cyclic Peptide–Protein Binding Affinity Dataset Curation Summary tags: [data-scraping, dataset, summary, cyclic-peptide, binding-affinity] created: 2026-05-11 updated: 2026-05-27 status: active related:

• “data-scraping-curation-plan”
• “data-scraping-pipeline-plan”

---

Cyclic Peptide–Protein Binding Affinity Dataset: Curation Summary

Overview

Search completed across Paperclip (23 queries), web searches, and database checks. Total papers collected: 80 (62 with experimental binding data)

---

Key Databases (Downloadable)

1. PPIKB — PRIMARY RESOURCE

• URL: https://ppikb.duanlab.ac
• Downloads: https://ppikb.duanlab.ac/downloads/
• Content: 40,347 entries total (19,509 research + ~2,815 patent entries)
• Data includes: Kd, Ki, IC50 values, peptide sequences, protein targets, PDB IDs
• Files available:
  • Main.xlsx — Full affinity dataset (sequences, affinity data, DOIs, PMIDs)
  • Branch.xlsx — Refined dataset with matched PDB crystal structures
  • research.xlsx — Research paper–derived affinity data
  • structure.xlsx — PDB structural data
• Filtering needed: Filter for cyclic peptides (look for cyclic/cyclized/disulfide/macrocycle keywords in sequence or annotation)
• License: CC-BY 4.0

2. CPSea / CPBind — COMPUTED AFFINITY + STRUCTURES

• URL: https://github.com/YZY010418/CPSea
• Kaggle: https://www.kaggle.com/datasets/ziyiyang180104/cpsea
• Zenodo: https://zenodo.org/records/17324994
• Content: 2.7M cyclic peptide-receptor complexes (AFDB-derived)
• Key subsets:
  • CPBind_affinity.csv — 476K entries with Rosetta ddG < -25 and Vina score < -6
  • CPBind_basic.tsv — Basic info and filter metrics
  • CPBind_validity.csv — Ramachandran and interaction quality
  • CPBind_hydrophobic.csv — GRAVY, logP, rTPSA
  • CPBind_cluster.tsv — Foldseek cluster assignments
• Note: Affinity is COMPUTED (Rosetta ddG + Vina), not experimental. Useful for training but needs experimental validation benchmark.
• License: Research/academic use

3. CyclicPepedia — NATURAL + SYNTHETIC CYCLIC PEPTIDES

• URL: https://www.biosino.org/iMAC/cyclicpepedia/
• Downloads: https://www.biosino.org/iMAC/cyclicpepedia/download
• Content: 8,751 cyclic peptides, 59 targets, 821 sources
• Key files:
  • Peptide_basic_info.xlsx (2.4 MB)
  • Peptide_to_target.xlsx — Maps peptides to protein targets
  • Target.xlsx — Target protein info
  • Bioassay.xlsx — Bioassay data (280 KB)
  • Peptide_sequence_info.xlsx
  • structure_complex.zip — 3D PDB files of peptide-target complexes
  • structure_pdb.zip — 3D PDB files of cyclic peptides
• Note: Bioassay data likely includes some binding affinity measurements but may not be comprehensive.

4. CycPeptMPDB — PERMEABILITY (NOT BINDING)

• URL: http://cycpeptmpdb.com
• Content: 7,991 cyclic peptides with permeability data
• Note: Permeability only, NOT binding affinity. Useful as supplementary ADME data but does not directly serve our purpose.

5. PEPBI — THERMODYNAMIC DATA (LINEAR PEPTIDES)

• URL: Nature Scientific Data 2025
• Content: 329 peptide-protein complexes with ΔG, ΔH, ΔS (ITC)
• Note: Mostly linear peptides (5-20 aa). Could filter for cyclic entries but likely very few.

6. PepBenchmark — ML BENCHMARK

• URL: https://github.com/ZGCI (PepBenchmark)
• Content: 29 canonical + 6 non-canonical peptide datasets
• Note: Includes some binding affinity prediction tasks. Check for cyclic peptide subsets.

---

Paper Collection Summary

80 papers catalogued in paper_collection.json

By cyclization type:

Type	Count	Notes
Thioether macrocyclic	~12	Most common in mRNA/RaPID display papers
Hydrocarbon stapled	~8	MDM2/p53, helix mimetics
Disulfide constrained	~6	Phage display, DRPs
Lactam stapled	~2	Helical peptides
Head-to-tail cyclic	~5	Native cyclization
Bicyclic	~6	Including bismuth, CPP-conjugated
Multiple/various	~15	Comparison studies
Other (thioether-bipyridyl, CPPC, etc.)	~6	Specialized chemistries

By assay type:

Assay	Count
SPR (Kd, kon, koff)	~25
FP (Kd, IC50)	~15
ITC (Kd, ΔG, ΔH, ΔS)	~8
AS-MS (bind/no-bind)	~3
Enzymatic IC50	~10
Competition binding	~8
Yeast/phage display enrichment	~5
Computed (ddG, Vina score)	~5

By target protein (top):

Target	Papers
MDM2/MDMX	~5
BET bromodomains	~3
SARS-CoV-2 Spike	~3
FGF receptors	~3
PTP1B	~2
Keap1	~2
TNFα	~2
Various kinases	~3
GPCRs	~2
Other diverse targets	~40+

---

Recommended Dataset Assembly Strategy

Phase 1: Download & Filter Existing Databases (HIGHEST PRIORITY — immediate data)

1. PPIKB: Download Main.xlsx and Branch.xlsx. Filter for cyclic peptides (search for cyclic disulfide, macrocyclic, stapled, thioether, head-to-tail cyclization keywords in peptide name/description)
2. CyclicPepedia: Download Peptide_to_target.xlsx, Target.xlsx, Bioassay.xlsx, Peptide_basic_info.xlsx. Join tables to extract binding affinity entries
3. CPSea/CPBind: Download CPBind_affinity.csv for computed affinity scores (Rosetta ddG + Vina). Use as supplementary training data

Phase 2: Extract Data from Key Papers

• Read the 62 papers with experimental binding data using Paperclip’s full-text access
• Extract individual (peptide, protein, affinity, assay, conditions) tuples
• Prioritize papers with large-scale datasets (e.g., RaPID libraries, DEL screening, phage display hit validation)

Phase 3: Normalize & Combine

• Standardize affinity values to Kd (nM) where possible
• Convert IC50/Ki to approximate Kd using Cheng-Prusoff where Kd type not directly reported
• Add binary binder/non-binder labels at 10 µM threshold
• Deduplicate across sources (PPIKB, CyclicPepedia, literature extraction)

Phase 4: Quality Control

• Flag computed vs. experimental affinity
• Flag assay type (SPR > ITC > FP > competition > enzymatic > computed)
• Remove entries with contradictory affinity values
• Annotate cyclization type, peptide length, modification type

---

Data Schema (Planned)

Field	Type	Description
entry_id	str	Unique identifier
peptide_sequence	str	One-letter code or HELM notation
cyclization_type	str	disulfide, head-to-tail, thioether, stapled, lactam, etc.
peptide_length	int	Number of residues
has_noncanonical	bool	Contains N-methyl, D-amino acids, etc.
target_protein	str	Protein name
target_uniprot	str	UniProt accession
target_pdb	str	PDB ID of complex (if available)
affinity_value	float	Numeric affinity value
affinity_unit	str	nM, µM, mM
affinity_type	str	Kd, Ki, IC50, EC50
assay_method	str	SPR, ITC, FP, MST, BLI, AS-MS, etc.
assay_conditions	str	pH, temp, buffer (if available)
is_binder	bool	Binary label (affinity < 10 µM)
is_strong_binder	bool	Kd < 100 nM
is_moderate_binder	bool	100 nM < Kd < 1 µM
is_weak_binder	bool	1 µM < Kd < 10 µM
is_non_binder	bool	Explicitly non-binding or Kd > 100 µM
source	str	PPIKB, CyclicPepedia, literature DOI, CPBind
source_doi	str	DOI of source paper
source_year	int	Year of publication
notes	str	Additional context

---

Files Created

File	Description
plan.md	Original curation plan
paper_collection.json	80 papers with metadata
summary.md	This file — curation summary and strategy