Julija Zavadlav

Julija Zavadlav

Auto-generated profile, based on publicly available information

Julija Zavadlav

Technical University of Munich

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
25 November 2021 in Nature Communications
Verified
Journal Article
Open access
PDF file
STStephan Thaler
Stephan Thaler
JZJulija Zavadlav
Julija Zavadlav
In molecular dynamics (MD), neural network (NN) potentials trained bottom-up on quantum mechanical data have seen tremendous success recently. Top-down approaches that learn NN potentials directly from experimental data have received less attention, typically facing numerical and computational challenges when backpropagating through MD simulations. We present the Differentiable Trajectory Reweighting (DiffTRe) method, which bypasses differentiation through the MD simulation for time-independent observables. Leveraging thermodynamic perturbation theory, we avoid exploding gradients and achieve around 2 orders of magnitude speed-up in gradient computation for top-down learning. We show effectiveness of DiffTRe in learning NN potentials for an atomistic model of diamond and a coarse-grained model of water based on diverse experimental observables including thermodynamic, structural and mechanical properties. Importantly, DiffTRe also generalizes bottom-up structural coarse-graining methods such as iterative Boltzmann inversion to arbitrary potentials. The presented method constitutes an important milestone towards enriching NN potentials with experimental data, particularly when accurate bottom-up data is unavailable.

6 February 2014 in The Journal of Chemical Physics
Verified
Journal Article
Open access
PDF file
JZJulija Zavadlav
Julija Zavadlav
MMManuel Nuno Melo
Manuel Nuno Melo
SMSiewert-Jan Marrink
Siewert-Jan Marrink
MPMatej Praprotnik
Matej Praprotnik
We present an adaptive resolution simulation of protein G in multiscale water. We couple atomistic water around the protein with mesoscopic water, where four water molecules are represented with one coarse-grained bead, farther away. We circumvent the difficulties that arise from coupling to the coarse-grained model via a 4-to-1 molecule coarse-grain mapping by using bundled water models, i.e., we restrict the relative movement of water molecules that are mapped to the same coarse-grained bead employing harmonic springs. The water molecules change their resolution from four molecules to one coarse-grained particle and vice versa adaptively on-the-fly. Having performed 15 ns long molecular dynamics simulations, we observe within our error bars no differences between structural (e.g., root-mean-squared deviation and fluctuations of backbone atoms, radius of gyration, the stability of native contacts and secondary structure, and the solvent accessible surface area) and dynamical properties of the protein in the adaptive resolution approach compared to the fully atomistically solvated model. Our multiscale model is compatible with the widely used MARTINI force field and will therefore significantly enhance the scope of biomolecular simulations.

28 December 2022 in The Journal of Chemical Physics
Verified
Journal Article
Open access
PDF file
STStephan Thaler
Stephan Thaler
MSMaximilian Stupp
Maximilian Stupp
JZJulija Zavadlav
Julija Zavadlav
Neural network (NN) potentials are a natural choice for coarse-grained (CG) models. Their many-body capacity allows highly accurate approximations of the potential of mean force, promising CG simulations of unprecedented accuracy. CG NN potentials trained bottom-up via force matching (FM), however, suffer from finite data effects: They rely on prior potentials for physically sound predictions outside the training data domain, and the corresponding free energy surface is sensitive to errors in the transition regions. The standard alternative to FM for classical potentials is relative entropy (RE) minimization, which has not yet been applied to NN potentials. In this work, we demonstrate, for benchmark problems of liquid water and alanine dipeptide, that RE training is more data efficient, due to accessing the CG distribution during training, resulting in improved free energy surfaces and reduced sensitivity to prior potentials. In addition, RE learns to correct time integration errors, allowing larger time steps in CG molecular dynamics simulation, while maintaining accuracy. Thus, our findings support the use of training objectives beyond FM, as a promising direction for improving CG NN potential’s accuracy and reliability.

5 April 2024 in npj Computational Materials
Verified
Journal Article
Open access
PDF file
SRSebastien Röcken
Sebastien Röcken
JZJulija Zavadlav
Julija Zavadlav
Machine Learning (ML)-based force fields are attracting ever-increasing interest due to their capacity to span spatiotemporal scales of classical interatomic potentials at quantum-level accuracy. They can be trained based on high-fidelity simulations or experiments, the former being the common case. However, both approaches are impaired by scarce and erroneous data resulting in models that either do not agree with well-known experimental observations or are under-constrained and only reproduce some properties. Here we leverage both Density Functional Theory (DFT) calculations and experimentally measured mechanical properties and lattice parameters to train an ML potential of titanium. We demonstrate that the fused data learning strategy can concurrently satisfy all target objectives, thus resulting in a molecular model of higher accuracy compared to the models trained with a single data source. The inaccuracies of DFT functionals at target experimental properties were corrected, while the investigated off-target properties were affected only mildly and mostly positively. Our approach is applicable to any material and can serve as a general strategy to obtain highly accurate ML potentials.

10 December 2021 in Journal of Chemical Theory and Computation
Verified
Journal Article
Closed access
PVPantelis R. Vlachas
Pantelis R. Vlachas
JZJulija Zavadlav
Julija Zavadlav
Matej PraprotnikMatej Praprotnik
Matej Praprotnik
PKPetros Koumoutsakos
Petros Koumoutsakos

6 May 2014 in Journal of Chemical Theory and Computation
Verified
Journal Article
Open access
JZJulija Zavadlav
Julija Zavadlav
MMManuel N. Melo
Manuel N. Melo
ACAna V. Cunha
Ana V. Cunha
AVAlex H. de Vries
Alex H. de Vries
SMSiewert-Jan Marrink
Siewert-Jan Marrink
Matej PraprotnikMatej Praprotnik
Matej Praprotnik
We present adaptive resolution molecular dynamics simulations of aqueous and apolar solvents using coarse-grained molecular models that are compatible with the MARTINI force field. As representatives of both classes of solvents we have chosen liquid water and butane, respectively, at ambient temperature. The solvent molecules change their resolution back and forth between the atomistic and coarse-grained representations according to their positions in the system. The difficulties that arise from coupling to a coarse-grained model with a multimolecule mapping, for example, 4-to-1 mapping in the case of the Simple Point Charge (SPC) and MARTINI water models, could be successfully circumvented by using bundled water models. We demonstrate that the presented multiscale approach faithfully reproduces the structural and dynamical properties computed by reference fully atomistic molecular dynamics simulations. Our approach is general and can be used with any atomistic force field to be linked with the MARTINI force field.