diff-epr provides differentiable kernels for simulating EPR distance distributions (DEER/PELDOR) from structural ensembles and spin-label rotamers.
- Spin-Label Modeling: Differentiable distance calculations between paramagnetic centers.
- Orientation Selection: Support for the Polyhach 5-angle formula (Polyhach et al., 2007) to model relative domain orientations.
-
Time-Domain Simulation: Simulate DEER modulation traces
$V(t)$ with parameterizable background decay and modulation depth. - Rotamer Library Integration: Support for weighted rotamer averages in distance distribution calculations.
- Hardware Acceleration: GPU-optimized distance kernels via JAX.
- Backend: JAX (XLA-compiled).
-
Physics: Dipolar coupling frequency (
$\omega$ ) based kernels. -
Performance:
$O(N)$ scaling for distance distribution integration.
- Core DEER trace simulation kernels.
- Background decay and modulation depth parameters.
- Integration with MMM (Multi-Scale Modeling of Macromolecules) rotamer libraries.
- Full orientation-dependence support.
pip install diff-epr-
Dipolar Frequency Parity: Coupling frequencies verified against the Pake pattern
$1/r^3$ dependence. - Time-Domain Accuracy: DEER traces validated for parity against standard simulation tools (e.g., DeerAnalysis/MMM).
-
Auto-Diff Gradients: Differentiable distance-to-signal kernels verified with JAX
grad.
diff-epr is part of the differentiable biophysics ecosystem:
- diff-biophys — Core differentiable biophysics engine.
- diff-hdx — Differentiable HDX-MS prediction.
- diff-fret — Differentiable FRET modeling.
- synth-nmr — NMR observable simulation.
@software{diff_epr,
author = {Elkins, George},
title = {diff-epr: Differentiable EPR/DEER simulation in JAX},
year = {2026},
url = {https://github.com/elkins/diff-epr},
version = {0.1.0}
}MIT