FOLDIFF · UMBRELLA SAMPLING
From individual windows to a free-energy profile
Combine time series with known bias potentials. Compare WHAM and MBAR, inspect window overlap, sample size and sensitivity of the result.
One coordinate, one temperature and one physical system. Calculate WT and mutants separately. This analyses existing time series; it does not run molecular dynamics. The troponin profile awaits author data.
Upload a coordinate time series for each window
Provide config.json and separate CSVs with time,cv columns. Time must increase with uniform spacing. Verify and join restarts or continuations before uploading.
Online: up to 2 MB, 3–40 runs, 80000 raw records, 15000 decorrelated samples and 100 bins; matrix-size limits also apply. Up to 3 minutes per calculation, one computation at a time. Data are never automatically thinned to fit these limits.
Review parameters before calculation
| CSV | Centre | k | Replica | Records |
|---|
Files stay in your browser until you submit. They are sent to the Oligoplan server in Russia for computation and deleted after the job finishes. Results remain available through a private key for up to one hour; download a copy. Data are not used for training or published.
Profile and uncertainty
Energy in kcal/mol; the zero is arbitrary. Bars show ±1 analytical MBAR standard error, excluding physical-model uncertainty. Insufficiently sampled regions are left disconnected.
Window overlap
Rows and columns are runs in config.json order. Darker means greater overlap. A disconnected graph prevents interpreting a single global profile.
Quality and interpretation limits
| Sensitivity check | RMSE · kcal/mol | Shared bins |
|---|
Profiles are aligned with a single additive shift over shared supported bins. Independently normalised WT and mutant profiles cannot be subtracted to obtain mutation stability ΔΔG.
Windows: correlation, discarded start and drift
| Run | Centre | Raw | Discarded | g | Effective | Drift |
|---|
g estimates correlation; effective observations are fewer than raw frames. Drift compares the two halves of a series. Single-coordinate checks can miss slow motion along other coordinates.
Open a saved result
JSON viewing happens in your browser without an upload. Benchmark validation does not validate your experiment.
Where the data come from
The first two datasets are generated from known potentials to test numerical calculations. The third is a published simulation example: lysozyme L99A with benzene, 26 side-chain torsion windows with 501 records each. It comes from the official PyMBAR repository at a recorded commit, with source checksums.
PyMBAR · source data ↗ · Mobley et al., 2007 ↗ · Provenance and validation ↗