v0.4.0 — 3D backbone reconstruction (NeRF)
What's new
3D backbone reconstruction from torsion trajectories — convert AlphaDynamics rollout .npz to multi-model PDB you can open in PyMOL / VMD / ChimeraX.
pip install -U alphadynamics
alphadynamics predict --sequence KLVFFAE --output klvffae.npz
alphadynamics rebuild klvffae.npz -s KLVFFAE -o klvffae.pdb --diagnostics
pymol klvffae.pdb # animate the trajectoryOr from Python:
from alphadynamics import predict_torsion_ensemble, trajectory_to_pdb
traj = predict_torsion_ensemble("KLVFFAE", n_ensemble=4, rollout_steps=200)
trajectory_to_pdb(traj[0], "KLVFFAE", "klvffae.pdb")Implementation
- New module
alphadynamics.geometry— deterministic NeRF (Parsons 2005) backbone reconstruction with Engh-Huber 1991 standard bond geometry. - Functions:
torsions_to_backbone(),trajectory_to_pdb(),radius_of_gyration(),end_to_end_distance(),trajectory_diagnostics(). - New CLI subcommand
alphadynamics rebuild. - 13 new unit tests, all passing.
Output: backbone heavy atoms (N, Cα, C, O) only. No side chains, no hydrogens. Suitable for visualizing dynamics; not for docking.
Diagnostic tool, not structure prediction
Torsion errors accumulate along the chain — for long peptides (N > ~50) end-to-end displacement may be substantial even for small per-residue errors. This is intrinsic to torsion-space reconstruction, not a flaw of the algorithm. Use as diagnostic visualization ("does the model produce a sensible chain or spaghetti?"), not as high-resolution structure.
Defaults
- ω = 180° (all-trans peptide bond). For cis-proline, supply per-residue
omega_degarray with appropriate 0°. - Terminal angles
phi[0]andpsi[-1]are undefined geometrically — replaced with conventional defaults (-60° / +120°).
References
- Parsons et al. 2005, J. Comput. Chem. 26: 1063-1068 (NeRF algorithm)
- Engh & Huber 1991, Acta Crystallogr. A 47: 392-400 (bond constants)
Links
- 📦 PyPI: https://pypi.org/project/alphadynamics/0.4.0/
- 🤗 HuggingFace: https://huggingface.co/krissss0/alphadynamics
- 📄 Paper v2: https://doi.org/10.5281/zenodo.19877815