ProteinScout
Every per-residue biophysical track in one run — and an ML layer that beats the raw scales.
Paste a protein sequence or a UniProt accession to map transmembrane helices, aggregation-prone regions,
disorder, charge / pI, signal peptide and coiled-coils on a single axis. Published scales run instantly in your browser;
the full server pipeline adds WALTZ-DB / DisProt / UniProt-trained ML models with honest held-out benchmarks.
checking backend…
What it does
- Transmembrane helices + in/out topology (positive-inside rule)
- Aggregation-prone regions (AGGRESCAN scale + WALTZ-DB amyloid ML)
- Disorder, charge-hydropathy (Uversky), NCPR / FCR, MoRF hints
- Signal peptide, coiled-coil, low-complexity, pI & net charge
- Self-contained HTML report + an honest ML-vs-scale benchmark table
Who it's for
- Protein engineers triaging aggregation
- Structural / membrane biologists
- Neurodegeneration & amyloid labs
- Intrinsic-disorder (IDP) researchers
Hydropathy — Kyte–Doolittle, windowed; >1.6 ⇒ transmembrane
Aggregation propensity — AGGRESCAN a3v; peaks ⇒ aggregation-prone regions
Disorder propensity — TOP-IDP; >0 ⇒ disorder-promoting
Net charge per residue — +K/R, −D/E, windowed
Transmembrane segments
Aggregation-prone regions
Disordered regions
⚠ The live ML backend is not active right now. The
instant in-browser
analysis above still works. You can also see the
validated example report,
grab the
code on GitHub, and run it locally:
python proteinscout.py analyze "DAEFRHDSGYEVHHQKLVFFAEDVGSNKGAIIGLMVGGVVIA"
Instant = published scales computed in your browser for fast feedback.
Full ML analysis runs the real CPU-only pipeline on the server — the ML layer that
beats the single-scale
baselines (WALTZ-DB / DisProt / UniProt-trained, with held-out ROC-AUC), UniProt-annotation overlay, signal-peptide &
coiled-coil prediction. No GPU, no model download, no alignment database.
github.com/gianyrox/proteinscout ·
example report →