TREAD — Protein Repeat Annotation

Predict tandem-repeat regions directly from an amino-acid sequence. Paste a single protein sequence below, select a model, and click Predict repeats (GPU). ProtT5 embeddings are generated automatically.

  • General repeat model (RepeatsDB): identifies repeat regions and provides probabilities for six broad repeat-fold classes (Alpha solenoid, TIM-barrel, Beta-propeller, Beta-barrel, Beta-solenoid, Alpha/beta solenoid).
  • Beta-propeller blade model: specializes in individual beta-propeller blades.

Online demo limits: up to 1200 residues on GPU or 600 residues on CPU. These are shared-compute limits of this Space, not intrinsic limits of TREAD. For longer proteins, multi-FASTA input, and custom model training, use the local TREAD command-line tool (tread predict input.fasta).
Prediction model
Prediction profile

General repeat model only: display the overall repeat probability or one of six repeat-fold class probabilities. Predicted regions correspond to the selected profile. Ignored for the beta-propeller blade model.
0 1
1 100

GPU access in this shared demo may be subject to quota or availability limits. CPU inference can be considerably slower.

Prediction results

Predicted repeat regions

Notes

  • This online demo analyses one protein per run; use the local TREAD CLI for multiple sequences.
  • The 1200-residue GPU and 600-residue CPU limits apply only to this shared-compute demo. The local tool supports longer proteins using overlapping ProtT5 chunks.
  • For the general repeat model, the dropdown selects which probability profile to display and use when calling regions; it does not show all seven profiles at once.
  • Region start and end positions use 1-based, inclusive coordinates.
  • Non-standard residues U, Z, O, and B are converted to X for ProtT5 embedding.
  • The web demo uses a pinned TREAD implementation for reproducibility.

Code, installation, long-sequence analysis, and custom training: KYQiu21/TREAD