adaptyvbio/protenix
> Structure prediction with Protenix, an open AlphaFold3 reproduction. Use this (2) Wanting an open alternative to AF3 alongside Boltz and Chai, (3) Validating designed binder-target complexes. For QC thresholds, use protein-qc. For ipSAE ranking, use ipsae.
npx skills add https://github.com/adaptyvbio/protein-design-skills --skill protenixProtenix is ByteDance's open PyTorch
reproduction of AlphaFold3 (Apache 2.0). It is an AF3-class complex predictor, useful
next to boltz and chai for cross-checking designed complexes. Runnable through
biomodals.
Use Protenix-v2 for antibody-antigen complexes. The v2 model (464M params, April
2026) adds 9 to 13 percentage points of antibody-antigen accuracy over v1 at the
DockQ > 0.23 threshold and is more sample-efficient (v2 at 5 seeds exceeds v1 at 1000).
Select it with --model-name protenix-v2. For general complexes, the v1 base model is
fine.
| Requirement | Value |
|-------------|-------|
| Runner | Modal (biomodals) |
| GPU | L40S (default; GPU env var) |
| Setup | See Getting started |
git clone https://github.com/hgbrian/biomodals && cd biomodals
printf '>protein|A\nMAWTPLLLLLLSHCTGSLSQ...\n' > target.faa
uv run --with modal modal run modal_protenix.py \
--input-faa target.faa \
--seeds 42 \
--no-use-msa
| Parameter | Default | Description |
|-----------|---------|-------------|
| --input-faa | one required | FASTA input (or --input-json) |
| --seeds | 42 | Comma-separated seeds |
| --use-msa / --no-use-msa | MSA on | Pass --no-use-msa for single-sequence |
| --model-name | v1 base | Set protenix-v2 for antibody-antigen complexes |
| --use-mini | off | Switch to the smaller protenix_mini model |
| --out-dir | ./out/protenix | Output directory |
| Need | Tool |
|------|------|
| Affinity head (small molecules) | boltz (Boltz-2) |
| Fastest, ligand support | chai |
| Open AF3 reproduction | protenix (v1 base) |
| Antibody-antigen complexes | protenix-v2 |
Ranking a shortlist across more than one predictor is more reliable than trusting a
single model.
| Issue | Cause | Fix |
|-------|-------|-----|
| Missing input error | No --input-faa/--input-json | Provide one |
| Slow run | MSA enabled | Add --no-use-msa |
| OOM | Large complex | Use --use-mini or a larger GPU |
Next: Rank with ipsae, filter with protein-qc.
Take adaptyvbio/protenix from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.