We hid every protein's name and annotations, asked a language model to read the raw sequence for anything that recurs, and kept only the copies that exist verbatim. It recovers known repeat families, finds short motifs scattered through disordered regions, and turns up a few things we can't yet name.
Repeats in disordered regions are hard to find by alignment because copies are short, degenerate and scattered. Reading sequences the way a person would, scanning for anchored patterns rather than aligning them, picks them up alongside the classic structured arrays. Every result here survived a strict check: each reported copy has to appear, letter for letter, in the real sequence.
The pipeline keeps the reading step honest. Scripts handle cohort selection, blinding and verification and contain no motif-finding logic; the model sees only an anonymous sequence.
Human proteins with ≥100 aa of disorder, from UniProt / MobiDB-lite regions or a TOP-IDP propensity window.
Strip names and annotations, shuffle to anonymous IDs, and keep the key in a separate place.
One scan for long structural repeats, then an independent scan of disordered stretches for short anchored motifs (≥5 aa, ≥3 copies).
Each reported copy must occur exactly in the sequence. Coordinates are computed here; anything that doesn't match is dropped.
Only now attach identities, then check against UniProt features and AlphaFold models.
Wfixed residue[ST]one of thesexany residue
φhydrophobic (AVILMFWY)ψlarge aliphatic (VILM)Ωaromatic (FWY)
+basic (KR)-acidic (DE)opolar (STNQ)(x){4,13}variable loop
Each line is one protein, sorted by length; colored blocks are verified repeat copies. Hover to identify a protein, click to open it in the explorer.
Hand-picked proteins, one tab per category. Each shows the AlphaFold model with the repeat copies colored, the repeat map against UniProt's own annotation, and the copies themselves. Click any copy to fly to it in 3D.
All proteins with at least one verified repeat family. Search by gene, protein name or consensus, then open any row in the explorer above.
| Gene | Repeat families | ||
|---|---|---|---|
| Loading catalog… | |||
This is a preview of work in progress; the screen is still running and numbers will change. Please get in touch before citing.
@misc{mcwhite_repeats_preview,
title = {Reading repeats blind across the disordered proteome},
author = {{McWhite Lab}},
year = {2026},
note = {Research preview},
url = {https://clairemcwhite.github.io/repeats/highlights.html}
}