MyAbs / README.md
GPUburnout's picture
Deploy MyAbs v0.2 (Docker Space)
9a17899 verified
|
Raw
History Blame Contribute Delete
2.21 kB
metadata
title: MyAbs
emoji: 🧬
colorFrom: indigo
colorTo: purple
sdk: docker
app_port: 7860
pinned: false
short_description: Fold an antibody, see its CDR loops, flag its liabilities.

MyAbs — "My Antibodies"

Design antibodies. Watch them fold. Flag the risks. In your browser. Free.

Paste an antibody variable domain (VH and VL amino-acid sequence) and MyAbs will:

  1. Fold it with ABodyBuilder2 (4-model ensemble, IMGT-numbered output).
  2. Show the 3D structure with the six CDR loops highlighted (rotate/zoom/pan in the browser).
  3. Flag developability liabilities — PTM hotspots (deamidation, isomerization, oxidation), N-glycosylation sequons, unpaired cysteines, and long CDR-H3 — each mapped to its exact IMGT position via ANARCI and painted onto the structure.

Honest positioning

MyAbs predicts in silico. It does not test. It hands you a spatially-reasoned shortlist and a liability map; wet-lab validation (BLI for binding, cell assays for function) is still required. The liability flags are heuristic screens, not disqualifiers. Outputs are in-silico candidates, not patent-ready antibodies.

Privacy: pasted sequences are folded in-process and never stored. No account required.

Included therapeutic library

A small, hand-curated set of famous therapeutics (Trastuzumab, Adalimumab, Pembrolizumab, Rituximab, Bevacizumab), variable domains only, each cross-checked against two independent public sources with provenance shown in the app.

Stack

Docker Space on a miniforge base: ABodyBuilder2 (ImmuneBuilder) · ANARCI + HMMER (IMGT numbering) · OpenMM + pdbfixer (refinement) · py3Dmol (client-side viewer) · Gradio (UI). Runs on free CPU — first fold is slow (cold-load of the ensemble), subsequent folds take ~10-20 s.

Licensing note

ImmuneBuilder / ABodyBuilder2 weights are installed at build time from PyPI. Verify the ImmuneBuilder license terms before any commercial use. The curated therapeutic sequences are self-assembled from public sources (PDB SEQRES, KEGG/DrugBank/GSRS), not redistributed from any restrictively-licensed database.