Repository navigation
Releases
Releases · helmutcarter/molentropy
molentropy 0.1.1
- Python 3.13 and 3.14 support: the bindings now use PyO3 0.30 and
rust-numpy 0.30 (from 0.21),requires-pythonno longer stops at 3.12,
and releases build wheels for Python 3.10 to 3.14. - Constrained-coordinate detection (
--drop-constrained,
drop_constrained=Trueinload_system,
CoordinateOptions::drop_constrained): drops every bond and angle that the
trajectory shows to be constrained, keeps free bonds to hydrogen, and keeps
the dropped coordinates' terms in the BAT Jacobian, so one run gives both
the entropy and ln J.System.constrained_atomsand
InternalCoordinates::constrained_coordinateslist what was dropped. - A Bayesian network soft estimator (
estimate_network_entropy,
--soft-estimator network,soft_estimator="network"): MIST with each
coordinate conditioned on up to several earlier ones (5 by default) instead
of one. On peptides it halves MIST's error; see the validation page. - Truncated Gaussian hard estimators for the split estimator
(--hard-estimator qh-mie1|qh-mie2,hard_estimator="qh-mie1"|"qh-mie2",
HardEstimator::QuasiHarmonicMie1|QuasiHarmonicMie2): QH with its
correlation term dropped (first order) or truncated at second order in the
mutual information expansion.
molentropy 0.1.0
First release, under the name molentropy (previously nn-entropy).
- Reads Amber topologies (
.parm7) and NetCDF trajectories (.nc) and converts
every frame to bond-angle-torsion coordinates, with phase angles and an
optional Jacobian term. - Estimators:
- nearest-neighbor mutual information expansion (NN MIE) to fourth order, and
pairwise mutual information; - mutual information spanning tree (MIST);
- joint nearest-neighbor entropy up to eight dimensions;
- quasi-harmonic (QH) and Boltzmann quasi-harmonic (BQH);
- a split estimator: soft torsions by MIE, MIST or the joint estimate, stiff
coordinates by QH or BQH, and a Gaussian (Fourier-harmonic) or per-state
coupling term.
- nearest-neighbor mutual information expansion (NN MIE) to fourth order, and
- Per-coordinate entropies for every estimator except the joint estimate.
- Theiler windows for time-correlated frames, and exact finite-sample constants.
- Results are bit-for-bit reproducible across runs and thread counts.
- A command-line program, a Rust library and Python bindings.
- Validation against HREX/MBAR reference entropies of small molecules and
peptides at 300 K.