diff --git a/README.md b/README.md index d9e8254..363bc4d 100644 --- a/README.md +++ b/README.md @@ -26,7 +26,7 @@ This is the decomposition used in transition interface sampling and forward-flux sampling [2, 3]: $$ -k_{UF} \;=\; \Phi_{U,Q^\ddagger}\;\cdot\;P(Q_f \mid Q^\ddagger), +k_{UF} = \Phi_{U,Q^\ddagger} \cdot P(Q_f \mid Q^\ddagger), $$ the flux of attempts through Q‡ times the probability that an attempt reaches F @@ -34,7 +34,7 @@ before returning to U. Per trajectory we record the number of failed attempts k_ and the folding time T_i. The model success probability per attempt is estimated as $$ -\hat p \;=\; \frac{N_\text{done}}{N_\text{done} + \sum_i k_i}. +\hat p = \frac{N_\text{done}}{N_\text{done} + \sum_i k_i}. $$ --- @@ -47,7 +47,7 @@ $$ variables. The reference path measure factorises as $$ -P(\omega) \;=\; P(\omega \mid Y)\,P(Y),\qquad P(Y)=\prod_j p^{s_j}(1-p)^{1-s_j}, +P(\omega) = P(\omega \mid Y) P(Y),\qquad P(Y)=\prod_j p^{s_j}(1-p)^{1-s_j}, $$ where Y = (s_1, s_2, …) is the sequence of attempt outcomes (s_j = 1 for success). @@ -63,8 +63,8 @@ constraints, the one closest to a reference measure P. "Closest" means minimum relative path entropy (Kullback–Leibler divergence): $$ -\min_{P'}\; \sum_\omega P'(\omega)\ln\frac{P'(\omega)}{P(\omega)} -\quad\text{s.t.}\quad \langle g(Y)\rangle_{P'} = G,\;\; \sum_\omega P'(\omega)=1 . +\min_{P'} \sum_\omega P'(\omega)\ln\frac{P'(\omega)}{P(\omega)} +\quad\text{s.t.}\quad \langle g(Y)\rangle_{P'} = G, \sum_\omega P'(\omega)=1 . $$ Take as constraint the mean number of successful crossings in a fixed number of @@ -72,27 +72,27 @@ attempts, g(Y) = Σ_j s_j. This fixes the crossing flux, i.e. the barrier. The Lagrangian solution is the exponential tilt $$ -P'(\omega) \;=\; \frac{1}{Z}\,P(\omega)\,e^{-\lambda\, g(Y)} . +P'(\omega) = \frac{1}{Z} P(\omega) e^{-\lambda g(Y)} . $$ Two consequences follow directly. 1. **Conditionals are untouched.** Because the tilt depends on ω only through Y, - P'(ω | Y) = P(ω | Y). Given the outcome of each attempt, the dwell times, excursion - shapes and structures are exactly those of the model. This is assumption (A2) - derived as the least-biased choice rather than postulated. + P'(ω | Y) = P(ω | Y). Given the outcome of each attempt, the dwell times, excursion + shapes and structures are exactly those of the model. This is assumption (A2) + derived as the least-biased choice rather than postulated. 2. **The outcome process stays Bernoulli, with shifted log-odds.** For an i.i.d. - Bernoulli reference, the tilt e^{-λ Σ s_j} factorises over attempts. Normalising - each factor gives an i.i.d. Bernoulli(p′) process with + Bernoulli reference, the tilt e^{-λ Σ s_j} factorises over attempts. Normalising + each factor gives an i.i.d. Bernoulli(p′) process with $$ -\operatorname{logit} p' \;=\; \operatorname{logit} p \;-\; \lambda, +\mathrm{logit} p' = \mathrm{logit} p - \lambda, \qquad -p' \;=\; \frac{p\,e^{-\lambda}}{1-p\,(1-e^{-\lambda})}. +p' = \frac{p e^{-\lambda}}{1-p (1-e^{-\lambda})}. $$ - The Lagrange multiplier λ is therefore the shift in log-odds per attempt. + The Lagrange multiplier λ is therefore the shift in log-odds per attempt. **Detailed balance and ΔG.** Counting crossings in both directions tilts U→F and F→U attempts by the same factor. In the rare-event limit both rates scale by the same @@ -105,16 +105,16 @@ probability (1−p)^k p. Under the tilt it has (1−p′)^k p′. The likelihood therefore $$ -w_\text{done}(k) \;\propto\; \Big(\tfrac{1-p'}{1-p}\Big)^{k}\,\frac{p'}{p} -\;=\; c^{\,k+1}\,e^{-\lambda}, +w_\text{done}(k) \propto \Big(\tfrac{1-p'}{1-p}\Big)^{k} \frac{p'}{p} += c^{k+1} e^{-\lambda}, $$ and for a censored trajectory with k failures and no success $$ -w_\text{cens}(k) \;\propto\; c^{\,k}, +w_\text{cens}(k) \propto c^{k}, \qquad -c \;=\; \frac{1}{1-p\,(1-e^{-\lambda})}. +c = \frac{1}{1-p (1-e^{-\lambda})}. $$ These are the weights computed by `log_weights`. The unit test @@ -148,8 +148,8 @@ Under (A1) and the conditional invariance of §2.2, a tilted first-passage time sampled directly. Draw $$ -K \sim \text{Geom}(p') \;\;(\text{failures}),\qquad -T' \;=\; \sum_{j=1}^{K} C_j \;+\; S, +K \sim \text{Geom}(p') (\text{failures}),\qquad +T' = \sum_{j=1}^{K} C_j + S, $$ where each C_j is resampled from the pool of observed cycles ("return to U → end of @@ -160,9 +160,9 @@ it contains the relaxation from the starting structures. The moments of this compound geometric sum are $$ -\langle T'\rangle = \frac{1-p'}{p'}\,\mu_C + \mu_S, +\langle T'\rangle = \frac{1-p'}{p'} \mu_C + \mu_S, \qquad -\operatorname{Var}T' = \frac{1-p'}{p'}\,\sigma_C^2 + \frac{1-p'}{p'^2}\,\mu_C^2 + \sigma_S^2 . +\mathrm{Var} T' = \frac{1-p'}{p'} \sigma_C^2 + \frac{1-p'}{p'^2} \mu_C^2 + \sigma_S^2 . $$ As p′ → 0 the second term dominates, CV → 1, and p′T′/μ_C converges in distribution @@ -185,11 +185,11 @@ Evidence for two-state behaviour must come from quantities that thinning cannot manufacture: - **Attempt independence.** Test whether the observed k_i follow a geometric - distribution (χ² test) and whether successive cycle durations are uncorrelated - (lag-1 Spearman correlation). + distribution (χ² test) and whether successive cycle durations are uncorrelated + (lag-1 Spearman correlation). - **Timescale separation.** Compare the relaxation time within U with the cycle time. - **Committor.** Check that committor values on the transition-state surface are - peaked around 1/2 [10, 11]. + peaked around 1/2 [10, 11]. ### 2.7 Interpretation of λ @@ -226,18 +226,18 @@ of exponential statistics follows Salvalaglio *et al.* [16]. ## 3. Statistics used - **Survival function.** Weighted Kaplan–Meier estimator [17], handling censored - trajectories. Weights act as frequency weights: a weight of 2 is equivalent to - duplicating a trajectory, which is unit-tested. + trajectories. Weights act as frequency weights: a weight of 2 is equivalent to + duplicating a trajectory, which is unit-tested. - **Exponential fit.** Maximum-likelihood time constant with censoring, - τ̂ = Σ w_i T_i / Σ w_i δ_i, where δ_i = 1 for completed trajectories. + τ̂ = Σ w_i T_i / Σ w_i δ_i, where δ_i = 1 for completed trajectories. - **Exponentiality test.** Kolmogorov–Smirnov statistic against an exponential with - a fitted scale. Because the scale is estimated, standard KS p-values are - anti-conservative. The null distribution is simulated once per sample size - (Lilliefors [18]); it is scale-free. + a fitted scale. Because the scale is estimated, standard KS p-values are + anti-conservative. The null distribution is simulated once per sample size + (Lilliefors [18]); it is scale-free. - **Uncertainty.** Nonparametric bootstrap over trajectories [19] gives confidence - intervals on λ*. + intervals on λ*. - **Stitching test power.** The stitched sample is subsampled to the original N - before testing, so the test has the same power as on the original data. + before testing, so the test has the same power as on the original data. --- @@ -246,8 +246,8 @@ of exponential statistics follows Salvalaglio *et al.* [16]. ```bash git clone git@github.com:multi-ego/MaxCal.git cd MaxCal -pip install -e ".[test]" # installs the maxcal-poisson and maxcal-joint commands -pytest -q # fast tests; add --runslow for the validation tests +pip install -e ".[test]" # installs the maxcal-poisson and maxcal-joint commands +pytest -q # fast tests; add --runslow for the validation tests ``` The scripts can also be run directly with `python maxcal_poisson.py ...`. @@ -258,7 +258,7 @@ using the test datasets: - the reweighting weights, their saturation, and the resulting CV and N_eff; - stitching and the emergence of Poisson statistics; - validation against a simulated higher barrier, including what goes wrong with Q‡ - on the barrier top; + on the barrier top; - the joint-mode checks (kinetic ΔG, transition-path symmetry across temperatures). ```bash @@ -273,7 +273,7 @@ committed outputs: ```bash python maxcal_poisson.py "runs/q_*.xvg" --qu 0.3 --qf 0.8 \ - --qts 0.40 --qtse 0.55 --out results + --qts 0.40 --qtse 0.55 --out results ``` Input files contain either one column (Q; give `--dt`) or two columns (time, Q). @@ -293,8 +293,8 @@ Lines starting with `#` or `@` are ignored, so GROMACS `.xvg` files work directl **Outputs** - `summary.csv`: one row per Q‡, with p̂, geometric-test p, lag-1 correlation, CV, - KS p-values, λ* with bootstrap CI, N_eff, stitching threshold, and - `cv_stitch0`/`ksp_stitch0` (stitching at λ = 0). + KS p-values, λ* with bootstrap CI, N_eff, stitching threshold, and + `cv_stitch0`/`ksp_stitch0` (stitching at λ = 0). - `tse_frames_*.csv`: TSE frame per completed trajectory, with its weight. **Status columns.** Every NaN in `summary.csv` comes with a reason code, so it can be @@ -311,11 +311,11 @@ told apart from a failure of the code. **Reading the results** 1. If `geom_p` < α, attempts have memory. Poisson statistics after tilting are then - imposed, not recovered. + imposed, not recovered. 2. If stitching at λ = 0 already passes, the original deviation from Poisson was the - initial relaxation, not the barrier. + initial relaxation, not the barrier. 3. If CV ≥ 1 at λ = 0, a missing barrier does not explain the deviation; look for - intermediates or parallel pathways. + intermediates or parallel pathways. 4. Only Q‡ values on the U side of the barrier region are meaningful (§2.8). --- @@ -323,38 +323,38 @@ told apart from a failure of the code. ## 5. Tests ```bash -pytest -q tests/ # 46 fast tests (~8 s) -pytest -q tests/ --runslow # + 13 validation tests and the demo notebook (~25 s) -MAXCAL_UPDATE_REF=1 pytest tests/test_regtest.py # regenerate the reference +pytest -q tests/ # 46 fast tests (~8 s) +pytest -q tests/ --runslow # + 13 validation tests and the demo notebook (~25 s) +MAXCAL_UPDATE_REF=1 pytest tests/test_regtest.py # regenerate the reference ``` - **`test_unit.py`**: parsing, attempt counting, tilt identities (§2.2–2.4), - Kaplan–Meier, KS/Lilliefors size and power, geometric test, stitching moments - (§2.5), and I/O. + Kaplan–Meier, KS/Lilliefors size and power, geometric test, stitching moments + (§2.5), and I/O. - **`test_regtest.py`**: full command-line run on 100 seeded overdamped Langevin - trajectories, compared with `tests/regtest/reference_summary.csv`. Deterministic - columns use rtol 1e-5; RNG-dependent stitching columns use loose absolute - tolerances. + trajectories, compared with `tests/regtest/reference_summary.csv`. Deterministic + columns use rtol 1e-5; RNG-dependent stitching columns use loose absolute + tolerances. - **`test_reweight_path.py`**: end-to-end test of the path where reweighting - succeeds, which the Langevin data never reach. It uses hand-built trajectories - (`tests/synthetic.py`) in which the weighted CV crosses 1 at λ* ≈ 1.1 with - N_eff ≈ 115. It checks that: - - λ* is finite and bracketed by its bootstrap confidence interval; - - N_eff and the KS p-value at λ* are finite; - - the weighted CV equals 1 at λ*; - - the TSE weight ratio between trajectories with 3 and 0 failures is exactly c³; - - every NaN has the right reason code. + succeeds, which the Langevin data never reach. It uses hand-built trajectories + (`tests/synthetic.py`) in which the weighted CV crosses 1 at λ* ≈ 1.1 with + N_eff ≈ 115. It checks that: + - λ* is finite and bracketed by its bootstrap confidence interval; + - N_eff and the KS p-value at λ* are finite; + - the weighted CV equals 1 at λ*; + - the TSE weight ratio between trajectories with 3 and 0 failures is exactly c³; + - every NaN has the right reason code. - The regression test also checks that every NaN in the summary has a non-`ok` - reason code, and it compares the reason codes with the reference. + reason code, and it compares the reason codes with the reference. - **`test_validation.py`**: low-barrier and high-barrier Langevin runs; the high - barrier is the low one plus a Gaussian bump of 2 or 3 kT on the barrier top. It - checks four things. - - The high-barrier ground truth is Poissonian. - - Stitching the low-barrier cycles with the true p′ reproduces the high-barrier - folding-time distribution: mean within 15% (1–7% observed across seeds), and a - two-sample KS test passes. - - The blind Poisson threshold is a lower bound on the true λ. - - The construction fails, as it should, with Q‡ on the barrier top (§2.8). + barrier is the low one plus a Gaussian bump of 2 or 3 kT on the barrier top. It + checks four things. + - The high-barrier ground truth is Poissonian. + - Stitching the low-barrier cycles with the true p′ reproduces the high-barrier + folding-time distribution: mean within 15% (1–7% observed across seeds), and a + two-sample KS test passes. + - The blind Poisson threshold is a lower bound on the true λ. + - The construction fails, as it should, with Q‡ on the barrier top (§2.8). --- @@ -369,9 +369,9 @@ relations between the two directions hold. The script uses one collective variable (CV) with basins A and B. - **If `--qa < --qb`:** A is Q < qa and B is Q ≥ qb, as for native or intermolecular - contacts. + contacts. - **If `--qa > --qb`:** the orientation is reversed automatically, as for a - distance: A (unbound) is d > qa, B (bound) is d ≤ qb. + distance: A (unbound) is d > qa, B (bound) is d ≤ qb. All thresholds are given in the original units. Backward runs are analysed in the mirrored coordinate −Q, so all of §2 applies unchanged. Each direction has its own @@ -384,14 +384,14 @@ sits in between, at the barrier top. **Detailed balance.** At equilibrium, the rates between core sets satisfy [23, 24] $$ -\frac{k_{AB}}{k_{BA}} \;=\; \frac{\pi_B}{\pi_A} \;=\; e^{-\beta\Delta G_{AB}}, +\frac{k_{AB}}{k_{BA}} = \frac{\pi_B}{\pi_A} = e^{-\beta\Delta G_{AB}}, $$ where π_A and π_B are the committor-split populations. The script estimates each rate as the inverse of the renewal mean first-passage time of §2.5, k = 1/M, with $$ -M(p) \;=\; \mu_C\,\frac{1-p}{p} + \mu_S . +M(p) = \mu_C \frac{1-p}{p} + \mu_S . $$ This excludes the initial relaxation. The rates are ratios on the same clock, so the @@ -400,7 +400,7 @@ scaled time of the model cancels. **Kinetic ΔG.** The quantity $$ -\Delta G_\text{kin} = -kT\ln\!\left(\frac{M_{BA}}{M_{AB}}\right) +\Delta G_\text{kin} = -kT\ln\left(\frac{M_{BA}}{M_{AB}}\right) $$ must equal the equilibrium ΔG for two-state behaviour [25]. It is reported @@ -413,8 +413,8 @@ removes one of the two multipliers. The script scans the forward tilt λ_f (p′_f = tilted p_f) and fixes the backward success probability through $$ -M'_{BA} = R\,M'_{AB},\qquad -p'_b = \frac{\mu_{C,B}}{R\,M'_{AB} - \mu_{S,B} + \mu_{C,B}}, +M'_{BA} = R M'_{AB},\qquad +p'_b = \frac{\mu_{C,B}}{R M'_{AB} - \mu_{S,B} + \mu_{C,B}}, $$ where R = exp(−βΔG_target). By default ΔG_target is the model's own kinetic value, @@ -432,7 +432,7 @@ ensemble is the time reverse of the A→B one [11, 26]. Transition paths run fro last frame in the starting basin to the first frame in the target. Two consequences can be tested using the CV alone: - the distributions of transition-path durations must be identical (two-sample KS - test); + test); - the numbers of Q_tse recrossings per path must be identical too. Under the MaxCal tilt, transition paths are unchanged (§2.2, conditional @@ -474,7 +474,7 @@ box equilibrium constant is K_box = P_bound/P_unbound = k′_on/k_off, where k the pseudo-first-order rate at [L] = 1/V. The standard binding free energy is [30] $$ -\Delta G^\circ = \Delta G_\text{box} - kT\ln\!\left(V\,C^\circ\right), +\Delta G^\circ = \Delta G_\text{box} - kT\ln\left(V C^\circ\right), \qquad C^\circ = 1\ \text{M} = 0.6022\ \text{nm}^{-3}. $$ @@ -489,129 +489,129 @@ missing barrier may be desolvation or induced fit. ```bash # folding / unfolding at the same temperature, experimental DeltaG imposed python maxcal_joint.py --fwd "fold/*.xvg" --bwd "unfold/*.xvg" \ - --qa 0.3 --qb 0.8 --qts-fwd 0.4 --qts-bwd 0.7 --qtse 0.55 \ - --temperature-relation same --dG -2.5 --out joint + --qa 0.3 --qb 0.8 --qts-fwd 0.4 --qts-bwd 0.7 --qtse 0.55 \ + --temperature-relation same --dG -2.5 --out joint # unfolding simulated at a higher temperature python maxcal_joint.py --fwd "fold/*.xvg" --bwd "unfold_hot/*.xvg" \ - --qa 0.3 --qb 0.8 --qts-fwd 0.4 --qts-bwd 0.7 --qtse 0.55 \ - --temperature-relation different --out joint_hot + --qa 0.3 --qb 0.8 --qts-fwd 0.4 --qts-bwd 0.7 --qtse 0.55 \ + --temperature-relation different --out joint_hot # binding / unbinding on a distance CV (A = unbound at d > 2.0 nm) python maxcal_joint.py --fwd "on/*.xvg" --bwd "off/*.xvg" --system binding \ - --qa 2.0 --qb 0.6 --qts-fwd 1.4 --qts-bwd 0.8 --qtse 1.0 \ - --temperature-relation same --dG -8.0 --box-volume 343 --out joint_bind + --qa 2.0 --qb 0.6 --qts-fwd 1.4 --qts-bwd 0.8 --qtse 1.0 \ + --temperature-relation same --dG -8.0 --box-volume 343 --out joint_bind ``` **Outputs** - `joint_summary.json`: per-direction statistics (p, geometric test, CV, KS, - renewal mean, mean TP duration and crossings); TP symmetry p-values; kinetic ΔG - (box and, for binding, standard); joint or independent λ with the tilted p′ in - both directions; the scan; TSE feature comparison; notes. + renewal mean, mean TP duration and crossings); TP symmetry p-values; kinetic ΔG + (box and, for binding, standard); joint or independent λ with the tilted p′ in + both directions; the scan; TSE feature comparison; notes. - `tse_forward.csv`, `tse_backward.csv` (frames with weights), and `tse_features.csv` - if features are present. + if features are present. - `survival_joint.png`, `tp_durations.png`, `tse_features.png`. ### 6.6 Joint-mode tests - **`test_joint.py`** (fast, 13 tests): - - orientation and mirroring; - - transition-path extraction and exact time-reversal symmetry of the TSE frame - rule; - - renewal mean and its inverse; - - standard-state conversion; - - the detailed-balance constraint holding along the whole scan; - - TSE path weighting; - - required `--temperature-relation`; - - end-to-end runs at the same and at different temperatures; - - invariance under reversing the CV (Q vs d = 1.5 − Q); - - binding requiring a volume. + - orientation and mirroring; + - transition-path extraction and exact time-reversal symmetry of the TSE frame + rule; + - renewal mean and its inverse; + - standard-state conversion; + - the detailed-balance constraint holding along the whole scan; + - TSE path weighting; + - required `--temperature-relation`; + - end-to-end runs at the same and at different temperatures; + - invariance under reversing the CV (Q vs d = 1.5 − Q); + - binding requiring a volume. - **`test_joint_validation.py`** (slow, 5 tests), on an asymmetric double well - (ΔG ≈ −0.6 kT) with a low-barrier model, a 2 kT-bump truth, and a hot backward set: - - kinetic ΔG equals the equilibrium committor-split ΔG within 0.3 kT (observed - ≤ 0.1 kT); - - the detailed-balance coupling predicts the true backward success probability - within 25% (observed ≤ 11%) and the backward mean folding time within 20% - (observed ≤ 8%); - - transition paths are time-reversal symmetric at the same temperature; - - a different temperature is detected, and misuse of "same" is flagged. + (ΔG ≈ −0.6 kT) with a low-barrier model, a 2 kT-bump truth, and a hot backward set: + - kinetic ΔG equals the equilibrium committor-split ΔG within 0.3 kT (observed + ≤ 0.1 kT); + - the detailed-balance coupling predicts the true backward success probability + within 25% (observed ≤ 11%) and the backward mean folding time within 20% + (observed ≤ 8%); + - transition paths are time-reversal symmetric at the same temperature; + - a different temperature is detected, and misuse of "same" is flagged. --- ## References 1. Scalone E., Broggini L., Visentin C., Erba D., Bačić Toplek F., Peqini K., - Pellegrino S., Ricagno S., Paissoni C., Camilloni C. "Multi-eGO: An in silico lens - to look into protein aggregation kinetics at atomic resolution." *PNAS* **119**, - e2203181119 (2022). + Pellegrino S., Ricagno S., Paissoni C., Camilloni C. "Multi-eGO: An in silico lens + to look into protein aggregation kinetics at atomic resolution." *PNAS* **119**, + e2203181119 (2022). 2. van Erp T. S., Moroni D., Bolhuis P. G. "A novel path sampling method for the - calculation of rate constants." *J. Chem. Phys.* **118**, 7762 (2003). + calculation of rate constants." *J. Chem. Phys.* **118**, 7762 (2003). 3. Allen R. J., Warren P. B., ten Wolde P. R. "Sampling rare switching events in - biochemical networks." *Phys. Rev. Lett.* **94**, 018104 (2005). + biochemical networks." *Phys. Rev. Lett.* **94**, 018104 (2005). 4. Jaynes E. T. "The minimum entropy production principle." *Annu. Rev. Phys. Chem.* - **31**, 579–601 (1980). + **31**, 579–601 (1980). 5. Pressé S., Ghosh K., Lee J., Dill K. A. "Principles of maximum entropy and maximum - caliber in statistical physics." *Rev. Mod. Phys.* **85**, 1115–1141 (2013). + caliber in statistical physics." *Rev. Mod. Phys.* **85**, 1115–1141 (2013). 6. Ghosh K., Dixit P. D., Agozzino L., Dill K. A. "The maximum caliber variational - principle for nonequilibria." *Annu. Rev. Phys. Chem.* **71**, 213–238 (2020). + principle for nonequilibria." *Annu. Rev. Phys. Chem.* **71**, 213–238 (2020). 7. Kish L. *Survey Sampling*. Wiley, New York (1965). 8. Rényi A. "A characterization of Poisson processes." *Magyar Tud. Akad. Mat. Kutató - Int. Közl.* **1**, 519–527 (1956). + Int. Közl.* **1**, 519–527 (1956). 9. Kalashnikov V. *Geometric Sums: Bounds for Rare Events with Applications*. Kluwer, - Dordrecht (1997). + Dordrecht (1997). 10. Du R., Pande V. S., Grosberg A. Yu., Tanaka T., Shakhnovich E. I. "On the - transition coordinate for protein folding." *J. Chem. Phys.* **108**, 334 (1998). + transition coordinate for protein folding." *J. Chem. Phys.* **108**, 334 (1998). 11. Bolhuis P. G., Chandler D., Dellago C., Geissler P. L. "Transition path sampling: - throwing ropes over rough mountain passes, in the dark." *Annu. Rev. Phys. Chem.* - **53**, 291–318 (2002). + throwing ropes over rough mountain passes, in the dark." *Annu. Rev. Phys. Chem.* + **53**, 291–318 (2002). 12. Kramers H. A. "Brownian motion in a field of force and the diffusion model of - chemical reactions." *Physica* **7**, 284–304 (1940). + chemical reactions." *Physica* **7**, 284–304 (1940). 13. Hänggi P., Talkner P., Borkovec M. "Reaction-rate theory: fifty years after - Kramers." *Rev. Mod. Phys.* **62**, 251–341 (1990). + Kramers." *Rev. Mod. Phys.* **62**, 251–341 (1990). 14. Voter A. F. "Hyperdynamics: accelerated molecular dynamics of infrequent events." - *Phys. Rev. Lett.* **78**, 3908 (1997). + *Phys. Rev. Lett.* **78**, 3908 (1997). 15. Tiwary P., Parrinello M. "From metadynamics to dynamics." *Phys. Rev. Lett.* - **111**, 230602 (2013). + **111**, 230602 (2013). 16. Salvalaglio M., Tiwary P., Parrinello M. "Assessing the reliability of the - dynamics reconstructed from metadynamics." *J. Chem. Theory Comput.* **10**, - 1420–1425 (2014). + dynamics reconstructed from metadynamics." *J. Chem. Theory Comput.* **10**, + 1420–1425 (2014). 17. Kaplan E. L., Meier P. "Nonparametric estimation from incomplete observations." - *J. Am. Stat. Assoc.* **53**, 457–481 (1958). + *J. Am. Stat. Assoc.* **53**, 457–481 (1958). 18. Lilliefors H. W. "On the Kolmogorov–Smirnov test for the exponential - distribution with mean unknown." *J. Am. Stat. Assoc.* **64**, 387–389 (1969). + distribution with mean unknown." *J. Am. Stat. Assoc.* **64**, 387–389 (1969). 19. Efron B. "Bootstrap methods: another look at the jackknife." *Ann. Stat.* **7**, - 1–26 (1979). + 1–26 (1979). **Related MaxCal approaches to kinetics** (alternatives to the attempt-level tilt used here): 20. Dixit P. D., Dill K. A. "Caliber corrected Markov modeling (C2M2): correcting - equilibrium Markov models." *J. Chem. Theory Comput.* **14**, 1111–1119 (2018). + equilibrium Markov models." *J. Chem. Theory Comput.* **14**, 1111–1119 (2018). 21. Brotzakis Z. F., Vendruscolo M., Bolhuis P. G. "A method of incorporating rate - constants as kinetic constraints in molecular dynamics simulations." *PNAS* - **118**, e2012423118 (2021). + constants as kinetic constraints in molecular dynamics simulations." *PNAS* + **118**, e2012423118 (2021). 22. Fersht A. R., Matouschek A., Serrano L. "The folding of an enzyme. I. Theory of - protein engineering analysis of stability and pathway of protein folding." - *J. Mol. Biol.* **224**, 771–782 (1992). (Φ-value analysis, for comparing the TSE - with experiment.) + protein engineering analysis of stability and pathway of protein folding." + *J. Mol. Biol.* **224**, 771–782 (1992). (Φ-value analysis, for comparing the TSE + with experiment.) **Joint mode** 23. Hummer G. "From transition paths to transition states and rate coefficients." - *J. Chem. Phys.* **120**, 516–523 (2004). + *J. Chem. Phys.* **120**, 516–523 (2004). 24. E W., Vanden-Eijnden E. "Towards a theory of transition paths." - *J. Stat. Phys.* **123**, 503–523 (2006). + *J. Stat. Phys.* **123**, 503–523 (2006). 25. Jackson S. E., Fersht A. R. "Folding of chymotrypsin inhibitor 2. 1. Evidence for - a two-state transition." *Biochemistry* **30**, 10428–10435 (1991). + a two-state transition." *Biochemistry* **30**, 10428–10435 (1991). 26. Best R. B., Hummer G. "Reaction coordinates and rates from transition paths." - *PNAS* **102**, 6732–6737 (2005). + *PNAS* **102**, 6732–6737 (2005). 27. Hammond G. S. "A correlation of reaction rates." *J. Am. Chem. Soc.* **77**, - 334–338 (1955). + 334–338 (1955). 28. Leffler J. E. "Parameters for the description of transition states." - *Science* **117**, 340–341 (1953). + *Science* **117**, 340–341 (1953). 29. Donati L., Hartmann C., Keller B. G. "Girsanov reweighting for path ensembles and - Markov state models." *J. Chem. Phys.* **146**, 244112 (2017). + Markov state models." *J. Chem. Phys.* **146**, 244112 (2017). 30. Gilson M. K., Given J. A., Bush B. L., McCammon J. A. "The statistical-thermodynamic - basis for computation of binding affinities: a critical review." - *Biophys. J.* **72**, 1047–1069 (1997). + basis for computation of binding affinities: a critical review." + *Biophys. J.* **72**, 1047–1069 (1997).