I joined the Organic Electronics group in February 2022 as a postdoctoral researcher. My current position is funded by the collaborative research center TRR 146: Multiscale Simulation Methods for Soft Matter Systems. I focus on machine learning methods to assist molecular dynamics simulations.
I received my Ph.D. from the Johannes Gutenberg University Mainz in July 2022 within the theory group of Prof. Kremer. My research as a Ph.D. candidate was settled at the intersection of multiscale modeling an deep learning. In particular, I developed a python tool based on deep learning for the reverse-mapping of condensed-phase molecular structures.
Published in the group
2022
Benchmarking coarse-grained models of organic semiconductors via deep backmapping
M. Stieffenhofer, C. Scherer, F. May, T. Bereau, D. Andrienko
Frontiers in Chemistry,
10,
982757,
2022,
[doi]
[abstract]
The potential of mean force is an effective coarse-grained potential which is often approximated by pairwise potentials. While the approximated potential reproduces certain distributions of the reference all-atom model with remarkable accuracy important cross-correlations are typically not captured. In general the quality of coarse-grained models is evaluated at the coarse-grained resolution hindering the detection of important discrepancies between the all-atom and coarse-grained ensembles. In this work the quality of different coarse-grained models is assessed at the atomistic resolution deploying reverse-mapping strategies. In particular coarse-grained structures for Tris-Meta-Biphenyl-Triazine are reverse-mapped from two different sources: (1) All-atom configurations projected onto the coarse-grained resolution and (2) snapshots obtained by molecular dynamics simulations based on the coarse-grained force fields. To assess the quality of the coarse-grained models reverse-mapped structures of both sources are compared revealing significant discrepancies between the all-atom and the coarse-grained ensembles. Specifically the reintroduced details enable force computations based on the all-atom force field that yield a clear ranking for the quality of the different coarse-grained models.