Pith. sign in

REVIEW 2 major objections 4 minor 42 references

An approximated machine-learned committor can drive enhanced sampling of the first step of membrane fusion, stalk formation, without a hand-built reaction coordinate, yielding converged free energies and a committor-ordered mechanism.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 10:42 UTC pith:TD3PFP7B

load-bearing objection A credible application of the authors' own approximate-committor method to NP-mediated stalk formation, with honest limitations; the main risks are the inherited validity of the variational approximation and a non-independent chain-coordinate comparison. the 2 major comments →

arxiv 2607.20122 v1 pith:TD3PFP7B submitted 2026-07-22 cond-mat.soft physics.comp-ph

Let's Stalk About Membranes: Committor-Based Enhanced Sampling of Stalk Formation

Giorgia Rossi , Enrico Trizio , Davide Bochicchio , Giulia Rossi , Michele Parrinello This is my paper
classification cond-mat.soft physics.comp-ph
keywords membrane fusionstalk formationcommittorenhanced samplingmachine learningcollective variablesnanoparticlecoarse-grained simulation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that a machine-learned committor function, built from only the two endpoint states and a set of lipid-tail coordination descriptors, can drive enhanced-sampling simulations of nanoparticle-mediated stalk formation. If true, it means the first, rare step of membrane fusion can be simulated and its free-energy barrier and transition mechanism characterized without constructing an intuitive collective variable by hand. The authors show that their self-consistent training/sampling loop reaches converged free-energy estimates that match an established chain coordinate, and that committor-ordered configurations reveal the gradual water expulsion, lipid splaying, and tail reorientation that form the hydrophobic stalk. The approach is semi-automatic, transferable to other fusogens, and requires only a reasonable guess of where the stalk is expected to form.

Core claim

The paper claims that an approximated variational functional for the committor—minimizing the descriptor-gradient norm |∇_d q|^4 rather than the exact atomic-gradient functional—yields a collective variable that samples the stalk-forming transition uniformly. Applied to a coarse-grained DOPC bilayer pair with an embedded amphiphilic gold nanoparticle, it produces multiple reactive events, a well-sampled transition-state region, and free-energy estimates consistent with those from the chain-coordinate method. The learned committor also orders configurations along the pathway, showing that stalk formation requires reorientation of an upper-membrane lipid tail toward the nanoparticle, accompani

What carries the argument

The central machinery is the approximated committor functional K~[q(d(x))] = ⟨|∇_d q|^4⟩, Eq. (1), which replaces expensive atomic-coordinate gradients with cheap descriptor-space gradients. Combined with a neural-network representation of q as a function of 17 concentric lipid-tail coordination shells centered on the nanoparticle (restricted by height and 2-nm lateral cutoffs), it defines the committor-related CV z, which is biased by a Kolmogorov transition-state bias and an OPES-explore metadynamics bias in an iterative train-sample loop.

Load-bearing premise

The claim rests on the assumption that the approximated variational functional of Eq. (1), which is not guaranteed to reproduce the exact committor, still produces a collective variable that drives unbiased, converged sampling of the true transition ensemble for a collective membrane rearrangement.

What would settle it

A concrete test would be to take a set of configurations from the transition-state region, run many short unbiased trajectories from each, and compare the measured (exact) committor values to the model's predictions. If the model's ordering diverges significantly from the exact committor for these membrane configurations, the sampling and free-energy estimates would not be trustworthy. Also, running the same protocol with the vertical/lateral cutoffs removed would show whether the stalk's location truly emerges from the simulation or is imposed by the descriptor priors.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Stalk formation free-energy barriers and transition ensembles can be computed for arbitrary fusogen designs without hand-crafted collective variables, enabling systematic screening of nanoparticle, peptide, and protein fusogens.
  • The method provides an operator-independent way to assign mechanistic stages to configurations via committor values, giving a probabilistic rather than geometric reaction coordinate.
  • Because the descriptors are lipid tail coordination shells, the approach transfers to peptide- and protein-mediated fusion with minimal modification.
  • Reliable estimates of stalk barrier heights are important for predicting whether a given fusogen is kinetically competent at inducing fusion.
  • The self-consistent protocol converges in two iterations for a membrane-scale system, suggesting practical applicability to similarly complex collective self-assembly transitions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the method is as transferable as claimed, it could be extended to fusion pore expansion or vesicle rupture, where the region of interest is not known a priori; the height and lateral cutoffs would then need to be relaxed or adaptively determined, a testable extension.
  • The paper's finding that upper-membrane tail reorientation is the committing step suggests that fusogen design strategies might target the distal leaflet rather than only local curvature, a hypothesis worth testing in simulations with different ligand chemistries.
  • The agreement between the approximated committor and the chain coordinate could be checked more stringently by computing the exact committor on a subset of configurations via short unbiased trajectories, quantifying how much is lost by the approximation.
  • Since the approximation is only shown to be an upper bound functionally, the reported free-energy differences could carry a systematic bias; comparing against alternative estimates (e.g., umbrella sampling with the same CV or higher-resolution models) would clarify the magnitude of that bias.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper applies a recently introduced approximate committor-based enhanced sampling framework to study nanoparticle-mediated stalk formation between two DOPC bilayers in Martini coarse-grained simulations. The committor is represented by a neural network whose input descriptors are counts of lipid tail beads in concentric spherical shells above the nanoparticle, restricted to a 2 nm lateral cylinder. Enhanced sampling is driven by a Kolmogorov bias plus OPES on the committor-related variable z, in an iterative self-consistent training loop. After two iterations, five 1 µs production replicas are used to obtain a free-energy surface, which is compared with a chain-coordinate calculation, and the transition mechanism is characterized by binning configurations along z. The central claim is that the approximate functional K~[q(d(x))] = ⟨|∇_d q|^4⟩ yields a collective variable that provides an effective, semi-automatic tool for studying stalk formation and produces accurate free energies and mechanistic insight.

Significance. If the central claim is valid, the paper presents a practically useful approach to an outstanding problem: constructing collective variables for membrane fusion without a hand-built reaction coordinate. The authors deserve credit for shipping reproducible code and inputs, for reporting five independent 1 µs replicas with statistical error bars, for testing robustness to shell geometry (Fig. S2–S3), and for comparing with a chain-coordinate based free energy. These are good internal practices. However, the load-bearing premise—that the approximate functional in Eq. (1), inherited from Ref. 25, produces a physically meaningful committor for membrane-scale rearrangements—is not independently validated in this manuscript. The checks offered are either self-consistency checks of the training loop or comparisons with a chain coordinate that shares the same localization priors and descriptor construction. The free-energy comparison with the chain coordinate is therefore weaker evidence than it first appears. In addition, a lower-wall restraint used during production is not reweighted, which may make the reported free energies conditional on that restraint.

major comments (2)
  1. [Method, Eq. (1); Results, Fig. S4; SI §C] The approximate variational functional K~[q(d(x))] = ⟨|∇_d q|^4⟩ is explicitly stated not to target the exact committor, and its validity is inherited from Ref. 25. The internal convergence check (Fig. S1) only demonstrates self-consistency of the iterative procedure, not that the learned z orders configurations by true commitment probability. The chain-coordinate comparison (Fig. S4, SI §C) uses a CV constructed from the same lipid-tail beads in the same cylindrical region above the NP (ZSHIFT=1.1, RCYLMEM=2.0) and therefore shares the same localization prior; agreement could reflect that common prior rather than validate the learned committor. This is load-bearing for the central claim that the method provides an effective and accurate tool. I recommend adding a direct validation of the learned committor, e.g., computing empirical commitment probabilities from short unbiased trajectori
  2. [Method: Enhanced sampling scheme; SI §A] The production simulations apply a lower-wall restraint on the number of contacts between MUS terminal beads and lipid headgroups of the upper membrane, to prevent excursions to the detached A′ state. This restraint is a bias potential, but it is not included in the reweighting expression w_n^i = exp(β V_eff)/⟨exp(β V_eff)⟩. If the wall is active in the sampled basin or at the transition-state region, the reported FES and ΔG describe a restrained conditional ensemble rather than the unbiased system. Please either reweight with the full bias including the wall, or demonstrate that the wall is inactive in the A and B basins and does not alter their relative free energy.
minor comments (4)
  1. [SI §A, Listing 1; Main text results] The text states that 17 concentric shells are used, but the COORDINATION_MULTI input lists 18 values of R_0 and N_OUT=18. Clarify the relationship between the number of radii and the number of shell coordination numbers used as descriptors.
  2. [Fig. 3B] The definitions of 'splayed lipid tails' and 'horizontally oriented tail segments' are not given in the main text or SI. Please add precise geometric definitions so the reported trends are reproducible.
  3. [Fig. 2B] The label 'free energy difference ΔG between the two states as a function of simulation time' is ambiguous. Specify whether this is a cumulative reweighted estimate and how the two states are defined for the ΔG calculation.
  4. [Discussion] The claim of 'minimal prior knowledge' is qualified by the choice of height and lateral cutoffs that localize descriptors to the region above the NP. The Discussion appropriately acknowledges this prior, but the abstract might overstate the method's autonomy; consider softening 'minimal prior knowledge' in view of the localization assumptions.

Circularity Check

1 steps flagged

Central claim leans on a self-cited approximate committor functional; the main cross-check shares the same localization priors, so the method is not independently validated here.

specific steps
  1. self citation load bearing [Methods: Machine learning the committor; Eq. (1)]
    "we adopt the reformulation we recently introduced in Ref.25, in which the coordinate gradients present in the original variational functional are restricted to the descriptor space (i.e., ∇u → ∇d), thus leading to the approximated and much easier to compute variational functional ... Although this new approach does not formally target the exact committor function, it has been shown that minimizing the modified functional in Eq. 1 yields a committor model that remains effective in driving efficient enhanced sampling simulations."

    The load-bearing premise of the paper is that Eq. (1), although approximate, yields a CV that drives unbiased sampling and correct free energies. That premise is supported only by a citation to Ref. 25, whose authors overlap with the present paper (Trizio, Rossi, Parrinello). No independent exact-committor test, benchmark, or externally falsifiable prediction is offered; the convergence checks (Fig. S1) only show iterative self-consistency. Thus the central claim of an 'effective and semi-automatic tool' rests on a self-citation chain rather than on evidence derived in this paper.

full rationale

The derivation chain is: choose descriptors (tail-coordination shells in an NP-centered cylinder), learn q by minimizing Eq. (1) with boundary conditions, bias with V_K + OPES on z, reweight to obtain the FES, and read the mechanism from committor-ordered bins. Most of this chain is not internally circular: the FES is not equal to the loss function, q is not fitted to the reported free-energy profile, and the mechanistic statistics (water counts, splayed tails) include observables that were not used as inputs. The main circularity is epistemic rather than algebraic: the assertion that the approximate functional (Eq. 1) produces a valid committor is the load-bearing premise, and it is supported only by Ref. 25 from the same group. The paper's own validation is a self-consistency check (Fig. S1) plus agreement with a chain-coordinate calculation that uses the same lipid-tail beads in the same 2-nm cylinder above the NP (SI §C), so the comparison is not an independent test of the approximation. For these reasons the central claim is not forced by definition, but it does lean on a self-citation chain; score 4.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 0 invented entities

The method adds no new physical entities. Its load-bearing ingredients are a hand-built descriptor set with geometric cutoffs, an approximated committor functional inherited from the authors' earlier work, and standard coarse-grained MD assumptions.

free parameters (7)
  • tail-coordination shell geometry = 17 shells, width 0.1 nm, R0 = 1.6–3.3 nm
    Hand-selected descriptor set; robustness checked only against other shell geometries in Fig. S2.
  • height cutoff Z_SHIFT = 1 nm above NP COM
    Excludes tail beads below the NP; encodes the prior that stalk formation occurs above the NP.
  • lateral cutoff XY_RADIUS = 2 nm
    Restricts descriptors to the intermembrane region directly above the NP; same location prior.
  • Kolmogorov bias strength λ = 30 kJ/mol (iterations 0–1), 50 kJ/mol (iteration 2)
    Tuned so that the bias magnitude is comparable to the expected free-energy barrier.
  • OPES BARRIER = 30 kJ/mol
    User-set target-distribution parameter for OPES-explore.
  • NN architecture and training hyperparameters = [17,14,12,1]; α = 10 then 1e-3; γ = 1 then 1e3; 10k/5k/40k epochs; lr 1e-3 with exp decay
    Chosen by hand; no systematic optimization reported.
  • MUS-head contact lower wall = not specified numerically
    Restraint suppressing the detached A′ substate; conditions the sampled ensemble and is not reweighted.
axioms (5)
  • domain assumption Overdamped Langevin dynamics assumption for the committor variational principle
    Methods: 'under the assumption of overdamped Langevin dynamics' the committor minimizes K[q(x)].
  • domain assumption Approximate functional Eq. (1) is an upper bound to the exact functional and yields an effective sampling CV
    Taken from Ref. 25; the paper states it 'does not formally target the exact committor function'.
  • domain assumption MARTINI 2.2 coarse-grained model faithfully represents stalk-formation thermodynamics
    All simulations use this force field; standard but approximate for membrane fusion.
  • ad hoc to paper Stalk formation is localized above the NP and captured by lipid-tail shell counts with the chosen cutoffs
    Explicit in descriptor construction; Discussion admits this is 'prior information about the location and direction of the stalk'.
  • domain assumption Reweighting with exp(β V_eff) converges in five 1-microsecond replicas
    Relies on overlap between biased and unbiased ensembles; argued from replica consistency, not proven rigorously.

pith-pipeline@v1.3.0-alltime-deepseek · 12083 in / 14029 out tokens · 135008 ms · 2026-08-01T10:42:02.961298+00:00 · methodology

0 comments
read the original abstract

Membrane fusion is essential for cellular communication and function, and understanding how two lipid bilayers merge is key to informing therapeutic strategies. Functionalized nanoparticles have recently emerged as synthetic fusogens, but the molecular mechanisms driving this process remain unclear, partly because fusion involves transitions over high free-energy barriers, difficult to capture in molecular simulations. While enhanced sampling methods can address this problem, they also rely on the definition of collective variables, which are especially hard to define for fusion, as it arises from the collective rearrangement of many molecules and cannot be easily reduced to a simple intuitive coordinate. Here, we study stalk formation, the first step of fusion, mediated by an amphiphilic gold nanoparticle, by employing an enhanced sampling strategy based on the committor function, machine-learned through a self-consistent procedure. This method requires minimal prior knowledge of the system and leverages the learned committor function as an effective collective variable, enabling uniform sampling of the entire pathway. From the resulting reactive trajectories and extensive transition region sampling, we obtain converged free-energy estimates and mechanistic insight into stalk formation.

Figures

Figures reproduced from arXiv: 2607.20122 by Davide Bochicchio, Enrico Trizio, Giorgia Rossi, Giulia Rossi, Michele Parrinello.

Figure 1
Figure 1. Figure 1: FIG. 1 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2 [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 20 canonical work pages

  1. [1]

    2015 , journal =

    Abraham, Mark James and Murtola, Teemu and Schulz, Roland and P. 2015 , journal =. doi:10.1016/j.softx.2015.06.001 , urldate =

    doi:10.1016/j.softx.2015.06.001 2015

  2. [2]

    2018 , journal =

    Structure--Property Relationships of Amphiphilic Nanoparticles That Penetrate or Fuse Lipid Membranes , author =. 2018 , journal =. doi:10.1021/acs.bioconjchem.7b00777 , urldate =

    work page doi:10.1021/acs.bioconjchem.7b00777 2018

  3. [3]

    2016 , journal =

    Simulations of Pore Formation in Lipid Membranes: Reaction Coordinates, Convergence, Hysteresis, and Finite-Size Effects , author =. 2016 , journal =. doi:10.1021/acs.jctc.6b00369 , urldate =

    work page doi:10.1021/acs.jctc.6b00369 2016

  4. [4]

    Synaptotagmin-1

    Bartolo, Ary Lautaro Di and Masone, Diego , year =. Synaptotagmin-1. Chemical Science , volume =. doi:10.1039/D1SC06711G , urldate =

    work page doi:10.1039/d1sc06711g

  5. [5]

    ChemRxiv , volume =

    Alessandro Berselli and Enrico Trizio and Umberto Raucci and Maria J Ramos and Maria Cristina Menziani and Francesco Muniz-Miranda and Michele Parrinello , title =. ChemRxiv , volume =. 2026 , doi =

    2026

  6. [6]

    2024 , journal =

    Nanoparticle Induced Fusion of Lipid Membranes , author =. 2024 , journal =. doi:10.1039/D4NR00591K , urldate =

    work page doi:10.1039/d4nr00591k 2024

  7. [7]

    The Journal of Chemical Physics , volume =

    Bonati, Luigi and Trizio, Enrico and Rizzi, Andrea and Parrinello, Michele , title = ". The Journal of Chemical Physics , volume =. 2023 , month =. doi:10.1063/5.0156343 , url =

    doi:10.1063/5.0156343 2023

  8. [8]

    The Journal of Chemical Physics , volume =

    Bussi, Giovanni and Donadio, Davide and Parrinello, Michele , title = ". The Journal of Chemical Physics , volume =. 2007 , month =. doi:10.1063/1.2408420 , url =

    doi:10.1063/1.2408420 2007

  9. [9]

    Nanoparticle-Induced Biomembrane Fusion: Unraveling the Effect of Core Size on Stalk Formation , shorttitle =

    Brosio, Giorgia and Rossi, Giulia and Bochicchio, Davide , year =. Nanoparticle-Induced Biomembrane Fusion: Unraveling the Effect of Core Size on Stalk Formation , shorttitle =. Nanoscale Advances , volume =. doi:10.1039/D3NA00430A , urldate =

    work page doi:10.1039/d3na00430a

  10. [10]

    2020 , journal =

    Amphiphilic Gold Nanoparticles Perturb Phase Separation in Multidomain Lipid Membranes , author =. 2020 , journal =. doi:10.1039/D0NR05366J , urldate =

    work page doi:10.1039/d0nr05366j 2020

  11. [11]

    Cholesterol-

    Canepa, Ester and Bochicchio, Davide and Brosio, Giorgia and Silva, Paulo Henrique Jacob and Stellacci, Francesco and Dante, Silvia and Rossi, Giulia and Relini, Annalisa , year =. Cholesterol-. Small , volume =. doi:10.1002/smll.202207125 , urldate =

    work page doi:10.1002/smll.202207125

  12. [12]

    A Tethering Complex Drives the Terminal Stage of

    D'Agostino, Massimo and Risselada, Herre Jelger and L. A Tethering Complex Drives the Terminal Stage of. 2017 , journal =. doi:10.1038/nature24469 , urldate =

    work page doi:10.1038/nature24469 2017

  13. [13]

    and Ramos, Maria J

    Das, Sudip and Raucci, Umberto and Trizio, Enrico and Kang, Peilin and Neves, Rui P.P. and Ramos, Maria J. and Parrinello, Michele , date =. A. 2025 , journal =

    2025

  14. [14]

    Proceedings of the National Academy of Sciences , volume =

    Yuanpeng Deng and Peilin Kang and Xiang Xu and Hui Li and Michele Parrinello , title =. Proceedings of the National Academy of Sciences , volume =. 2026 , doi =

    2026

  15. [15]

    2015 , journal =

    Mechanics of Membrane Fusion/Pore Formation , author =. 2015 , journal =. doi:10.1016/j.chemphyslip.2014.07.010 , urldate =

    work page doi:10.1016/j.chemphyslip.2014.07.010 2015

  16. [16]

    Living Journal of Computational Molecular Science , author=

    Enhanced Sampling Methods for Molecular Dynamics Simulations , volume=. Living Journal of Computational Molecular Science , author=. 2022 , month=. doi:10.33011/livecoms.4.1.1583 , abstractNote=

    doi:10.33011/livecoms.4.1.1583 2022

  17. [17]

    and Awasthi, Neha , year = 2017, journal =

    Hub, Jochen S. and Awasthi, Neha , year = 2017, journal =. Probing a

    2017

  18. [18]

    2021 , journal =

    Joint Reaction Coordinate for Computing the Free-Energy Landscape of Pore Nucleation and Pore Expansion in Lipid Membranes , author =. 2021 , journal =. doi:10.1021/acs.jctc.0c01134 , urldate =

    work page doi:10.1021/acs.jctc.0c01134 2021

  19. [19]

    The Journal of Physical Chemistry Letters , volume=

    Rethinking metadynamics: from bias potentials to probability distributions , author=. The Journal of Physical Chemistry Letters , volume=. 2020 , publisher=

    2020

  20. [20]

    Exploration vs

    Invernizzi, Michele and Parrinello, Michele , year =. Exploration vs. Journal of Chemical Theory and Computation , volume =. doi:10.1021/acs.jctc.2c00152 , urldate =

    doi:10.1021/acs.jctc.2c00152

  21. [21]

    2021 , journal =

    Amphiphilic Nanoparticles Generate Curvature in Lipid Membranes and Shape Liposome--Liposome Interfaces , author =. 2021 , journal =. doi:10.1039/D1NR05067B , urldate =

    work page doi:10.1039/d1nr05067b 2021

  22. [22]

    2025 , journal =

    Physical Determinants of Nanoparticle-Mediated Lipid Membrane Fusion , author =. 2025 , journal =. doi:10.1039/D4NR04851B , urldate =

    work page doi:10.1039/d4nr04851b 2025

  23. [23]

    2026 , journal =

    Understanding, Mimicking, and Exploiting Lipid Membrane Fusion in Biomedicine and Synthetic Biology , author =. 2026 , journal =. doi:10.1080/23746149.2026.2636806 , urldate =

    arXiv 2026

  24. [24]

    2024 , journal =

    Computing the Committor with the Committor to Study the Transition State Ensemble , author =. 2024 , journal =

    2024

  25. [25]

    and Risselada, H

    Marrink, Siewert J. and Risselada, H. Jelger and Yefimov, Serge and Tieleman, D. Peter and. The. 2007 , journal =. doi:10.1021/jp071097f , urldate =

    doi:10.1021/jp071097f 2007

  26. [26]

    2011 , journal =

    Model Systems for Membrane Fusion , author =. 2011 , journal =. doi:10.1039/C0CS00115E , urldate =

    work page doi:10.1039/c0cs00115e 2011

  27. [27]

    2022 , journal =

    Extracellular Vesicle Fusion Visualized by Cryo-Electron Microscopy , author =. 2022 , journal =. doi:10.1093/pnasnexus/pgac156 , urldate =

    work page doi:10.1093/pnasnexus/pgac156 2022

  28. [28]

    2012 , journal =

    Transition Path from Two Apposed Membranes to a Stalk Obtained by a Combination of Particle Simulations and String Method , author =. 2012 , journal =. doi:10.1103/PhysRevLett.108.228103 , urldate =

    work page doi:10.1103/physrevlett.108.228103 2012

  29. [29]

    Nature Methods , volume=

    Promoting transparency and reproducibility in enhanced molecular simulations , author=. Nature Methods , volume=. 2019 , publisher=

    2019

  30. [30]

    Nature Communications , volume =

    Free Energies of Membrane Stalk Formation from a Lipidomics Perspective , author =. Nature Communications , volume =

  31. [31]

    2014 , journal =

    Expansion of the Fusion Stalk and Its Implication for Biological Membrane Fusion , author =. 2014 , journal =. doi:10.1073/pnas.1323221111 , urldate =

    work page doi:10.1073/pnas.1323221111 2014

  32. [32]

    2022 , publisher =

    Erich Schubert and Lars Lenssen , title =. 2022 , publisher =. doi:10.21105/joss.04183 , url =

    doi:10.21105/joss.04183 2022

  33. [33]

    Monolayer-Protected Anionic

    Simonelli, Federica and Bochicchio, Davide and Ferrando, Riccardo and Rossi, Giulia , year =. Monolayer-Protected Anionic. The Journal of Physical Chemistry Letters , volume =. doi:10.1021/acs.jpclett.5b01469 , urldate =

    work page doi:10.1021/acs.jpclett.5b01469

  34. [34]

    2019 , journal =

    Thermodynamically Reversible Paths of the First Fusion Intermediate Reveal an Important Role for Membrane Anchors of Fusion Proteins , author =. 2019 , journal =. doi:10.1073/pnas.1818200116 , urldate =

    work page doi:10.1073/pnas.1818200116 2019

  35. [35]

    2020 , journal =

    Calcium-Triggered Fusion of Lipid Membranes Is Enabled by Amphiphilic Nanoparticles , author =. 2020 , journal =. doi:10.1073/pnas.1902597117 , pmcid =

    work page doi:10.1073/pnas.1902597117 2020

  36. [36]

    2014 , publisher=

    Tribello, Gareth A and Bonomi, Massimiliano and Branduardi, Davide and Camilloni, Carlo and Bussi, Giovanni , journal=. 2014 , publisher=

    2014

  37. [37]

    arXiv preprint arXiv:2410.18019 , year=

    Advanced simulations with PLUMED: OPES and Machine Learning Collective Variables , author=. arXiv preprint arXiv:2410.18019 , year=

    Pith/arXiv arXiv

  38. [38]

    Everything Everywhere All at Once: A Probability-Based Enhanced Sampling Approach to Rare Events , shorttitle =

    Trizio, Enrico and Kang, Peilin and Parrinello, Michele , date =. Everything Everywhere All at Once: A Probability-Based Enhanced Sampling Approach to Rare Events , shorttitle =. 2025 , journal =

    2025

  39. [39]

    Ceci n'est Pas Un Committor, yet It Samples like One:

    Trizio, Enrico and Rossi, Giorgia and Parrinello, Michele , year = 2026, journal =. Ceci n'est Pas Un Committor, yet It Samples like One:

    2026

  40. [40]

    2008 , journal =

    Surface-Structure-Regulated Cell-Membrane Penetration by Monolayer-Protected Nanoparticles , author =. 2008 , journal =. doi:10.1038/nmat2202 , urldate =

    work page doi:10.1038/nmat2202 2008

  41. [41]

    Annual review of physical chemistry , volume=

    Transition-path theory and path-finding algorithms for the study of rare events , author=. Annual review of physical chemistry , volume=

  42. [42]

    Chemical Reviews , volume =

    Zhu, Kai and Trizio, Enrico and Zhang, Jintu and Hu, Renling and Jiang, Linlong and Hou, Tingjun and Bonati, Luigi , title =. Chemical Reviews , volume =. 2026 , doi =

    2026