README.md

metadata

license: cc-by-4.0
tags:
  - biology
  - proteins
  - intrinsically-disordered-proteins
  - molecular-dynamics
  - CALVADOS
  - trajectories
  - BENDER
  - IDP
pretty_name: BENDER — Biological ENsembles of Disordered proteins across kingdoms
size_categories:
  - 10K<n<100K
task_categories:
  - other
language:
  - en

BENDER — Biological ENsembles of Disordered proteins across kingdoms

Raw CALVADOS coarse-grained molecular dynamics trajectories for 11,533 intrinsically disordered proteins spanning 13 kingdoms of life.

Each protein folder contains:

• <uniprot_id>.dcd — CALVADOS Cα trajectory (200 ns+)
• top.pdb — topology file

---

Folder structure

Kingdom.zip/
└── <uniprot_id>/
    ├── <uniprot_id>.dcd
    └── top.pdb

---

Available zip files

File	Kingdom	Sequences
Bacteria.zip	Bacteria	2,850
Plants.zip	Plants	2,480
Fungi.zip	Fungi	1,507
Mammals.zip	Mammals	1,361
Parasites_Protists.zip	Parasites / Protists	1,049
Viruses.zip	Viruses	1,025
Other_Vertebrates.zip	Other vertebrates	711
Insects.zip	Insects	289
Nematodes.zip	Nematodes	95
Other_Invertebrates.zip	Other invertebrates	77
Archaea.zip	Archaea	76
Algae.zip	Algae	10

---

Loading a trajectory

from huggingface_hub import hf_hub_download
import mdtraj as md
import zipfile

# Download zipped kingdom
zip_path = hf_hub_download(
    repo_id="taseef/BENDER",
    filename="Bacteria.zip",
    repo_type="dataset"
)

# Extract specific protein
with zipfile.ZipFile(zip_path, "r") as z:
    z.extract("N0AZA6/N0AZA6.dcd", path="./trajectories")
    z.extract("N0AZA6/top.pdb",    path="./trajectories")

# Load trajectory
traj = md.load(
    "./trajectories/N0AZA6/N0AZA6.dcd",
    top="./trajectories/N0AZA6/top.pdb"
)
print(traj)

---

Simulation protocol

Parameter	Value
Force field	CALVADOS-2 (Cα coarse-grained, Tesei & Lindorff-Larsen 2023)
Ensemble	NVT, 300 K
Ionic strength	0.15 M NaCl
Minimum length	200 ns
Long sequences (>150 res)	Extended — length scaled to residues^1.5
Equilibration	First 50 % of each trajectory discarded
Clustering	Global CD-HIT at 90 % sequence identity across all kingdoms
Scope	Pure complete IDPs only — no IDR fragments, no domain context
Convergence	87.4 % of sequences exceed ν fit R² ≥ 0.99

---

Prediction targets

10 ensemble-level targets per sequence:

Geometric properties

Target	Description
Rg	Radius of gyration
Re	End-to-end distance
nu	Flory scaling exponent
delta	Asphericity
A0	Flory prefactor

Contact network properties

Target	Description
global_efficiency	Global network efficiency
fragmentation_index	Fragmentation index
avg_clustering	Average clustering coefficient
transitivity	Network transitivity
degree_assortativity	Degree assortativity

---

Citation

@misc{taseefr_2026,
    author       = { taseefr },
    title        = { BENDER (Revision 5a8bda5) },
    year         = 2026,
    url          = { https://huggingface.co/datasets/taseef/BENDER },
    doi          = { 10.57967/hf/8692 },
    publisher    = { Hugging Face }
}

---

License

CC BY 4.0 — free to use for any purpose with attribution.

BENDER: making IDPs go supersonic.