BENDER — Biological ENsembles of Disordered proteins across taxa

Raw CALVADOS 2 coarse-grained molecular dynamics trajectories for 10,282 intrinsically disordered proteins from 2,156 organisms in six taxa (mammals, plants, fungi, protists, bacteria and viruses).

Each protein folder contains:

• <uniprot_id>.dcd — CALVADOS 2 Cα trajectory (100 ns or longer; see Simulation protocol)
• top.pdb — topology file

Taxa

Taxa are assigned from each protein's NCBI taxonomy lineage (checked against NCBI Taxonomy). Organisms are unique NCBI taxonomy identifiers (the total counts each organism once).

Taxon	Sequences	Organisms
Bacteria	2,858	654
Plants	2,438	224
Fungi	1,537	413
Mammals	1,361	155
Viruses	1,025	464
Protists	1,063	246
Total	10,282	2,156

Taxonomic composition

Inner ring, the six taxa. Middle ring, clades from the NCBI Taxonomy (phylum for bacteria, class for fungi, order for plants and mammals, eukaryotic supergroup for protists and family for viruses). Outer ring, organisms, with every organism of at least 45 sequences named with its sequence count.

Folder structure

One folder per taxon, one sub-folder per protein; every protein appears once.

<Taxon>/
└── <uniprot_id>/
    ├── <uniprot_id>.dcd
    └── top.pdb

Loading a trajectory

from huggingface_hub import hf_hub_download
import mdtraj as md

dcd = hf_hub_download(repo_id="taseef/BENDER", filename="Bacteria/N0AZA6/N0AZA6.dcd", repo_type="dataset")
pdb = hf_hub_download(repo_id="taseef/BENDER", filename="Bacteria/N0AZA6/top.pdb", repo_type="dataset")
traj = md.load(dcd, top=pdb)
print(traj)

Simulation protocol

• Force field: CALVADOS 2 (Cα coarse-grained, Tesei & Lindorff-Larsen 2023)
• Temperature: 300 K, NVT (Langevin); ionic strength 0.15 M
• Time step 10 fs
• Simulation length: 100 ns for ≤150 residues; 100·(N/150)^1.5 ns for longer chains (up to 2.2 µs)
• Frames: 10,000 (≤196 residues), 15,000 (197–312) or 20,000 (>312)
• The first half of each trajectory is equilibration; analyses use the second half (≥50 ns)
• Global CD-HIT clustering at 40% sequence identity — non-redundant across all taxa
• Pure complete IDPs only — no IDR fragments, no domain context

Companion resources

• MPIPI-Recharged trajectories of the same 10,282 proteins, in the same layout: 👉 taseef/MPIPI-Recharged
• MPIPI-GG trajectories of a 957-sequence subset, and the per-sequence properties of both MPIPI models: 👉 taseef/BENDER-MPIPI
• Simulation inputs, analysis code, per-sequence tables and figure scripts: 👉 github.com/Sabian98/BENDER
• Feature dataset (80+ computed features per protein): 👉 taseef/BENDER-features
• KESTREL model (sequence → ensemble property predictor): 👉 taseef/KESTREL

License

CC BY 4.0 — free to use for any purpose including commercial, provided you cite BENDER.

Contributing

Open a GitHub Issue with your FASTA file and simulation parameters. Same CALVADOS protocol required for consistency.

BENDER: making IDPs go supersonic.