TREAD — Protein Repeat Annotation

Paste a protein sequence and run one of the two pretrained TREAD models.

  • RepeatsDB repeat annotation: residue-wise repeat segmentation plus six repeat-fold heads.
  • Beta-propeller blade annotation: residue-wise blade prediction.

Recommended: use Run with ZeroGPU.
Fallback: use Run on CPU if GPU quota/availability is a problem; CPU inference is substantially slower.

ProtT5 embeddings are generated on the fly with Rostlab/prot_t5_xl_half_uniref50-enc.

Model
RepeatsDB profile to display

Used only for the RepeatsDB model. The default matches the public Colab example.
0 1
1 100

Results

Predicted motif ranges

Notes

  • ZeroGPU accepts sequences up to 1200 aa in this web demo.
  • CPU fallback accepts sequences up to 600 aa by default because ProtT5-XL is very slow on the free CPU backend.
  • The range-calling logic uses the same thresholding rule as the public Colab/TREAD utility.
  • The web table reports 1-based inclusive residue coordinates for readability.
  • Rare/ambiguous residues U, Z, O, and B are mapped to X for ProtT5 embedding, matching the Colab demo.

Source code and pretrained TREAD checkpoints: KYQiu21/TREAD