Embryonic development is a remarkably robust process that displays high degrees of adaptation between species. Therefore, it is important to quantify and compare the developmental progression of related species. Light-sheet fluorescent microscopy (LSFM) is routinely used at the AK Stelzer, Buchmann Institute for Molecular Life Sciences, to perform in-vivo 3D imaging of developing insect embryos. Multiple insect species have been recorded at the AK Stelzer, with a focus on Tribolium Castaneum, commonly known as the red flour beetle. Recently we have observed a phenomenon previously undescribed in T. Castaneum - contraction waves of the serosa cells previous to dorsal closure. In order to detect the serosa contraction waves, we performed 3D particle image velocimetry (PIV) analyses with quickPIV, an open source 3D PIV software developed at the AK Matthäus, Frankfurt Institute for Advanced Studies. From the PIV vector fields we found a clear correlation between migration speed and the onset of the contraction waves. In addition, by quantifying the change in the direction of cellular migration during contraction waves, we could locate the position of the waves in time and space. In order to further investigate the contraction waves, we plan to integrate the contraction wave detection algorithms into the recording setup, allowing us to increase the temporal resolution of the microscope to obtain detailed recordings of the contraction waves.

The periodic arrangement of hair follicles on the skin is determined during embryogenesis. It relies on a reaction-diffusion driven pre-patterning in the epidermal cell layer, as well as on a mesenchymal self-aggregation in the dermal cell layer.  

Both cellular layers exhibit distinct behaviours, that are driven by their bio mechanical properties. We implemented a mathematical model revolving around a set of parameters controlling cellular parameters to investigate their influence on the periodic patterning of hair follicles. Later we use these insights to formulate a hybrid model combining reaction-diffusion driven patterning in the epithelial layer with an agent-based model of directed cellular motion.  

Our implemented model allows us to asses informations and self-organisation features that are hard to observe in the laboratory. Additionally, based on the model we can quantify different processes and their impact on the hair follicle formation as well as on follow-up processes.

Protein folding, misfolding, and conformational changes are ubiquitous in our cells. They play key roles in metabolic pathways and are involved in numerous diseases. In-depth knowledge of the underneath mechanisms is crucial for understanding those processes and can lead to medical applications. However, conformational transitions are rare and fast. In vivo and in vitro observation cannot capture the folding/unfolding dynamics with adequate time and size resolution, while atomistic molecular dynamics simulations are limited by the available computational power.  

Transition path sampling (TPS) is a technique for simulating unbiased transition  trajectories; the resulting transition path ensemble provides information about the dynamics and mechanism. Yet, TPS leaves other challenges unaddressed. It is especially important to encode the progress of a transition in a reaction coordinate, as it provides quantitative insight into its mechanism. Furthermore, generating transition paths with high efficiency is still a challenge. In our method, AIMMD, we let a neural network  control TPS simulations. This significantly speeds up the sampling while learning an optimal reaction coordinate: the committor probability.  

We applied AIMMD to the folding of the mini protein chignolin. We collected hundreds of transition paths while learning the protein folding mechanism through the committor. We extracted the most significant features from the committor and turned it into an interpretable and quantitative expression with symbolic regression. We combined TPS, umbrella sampling, and equilibrium simulations to estimate the system’s transition rates and the free-energy profile along the committor. Our approach can be directly brought to larger and more complex systems and customized to many conformational changes and structural reorganization processes.

Dendritic spines are small morphological protrusions from dendrites. They receive excitatory input from neighboring axons and their size is used as a direct readout of the synaptic strength. Changes in spine size - and thus synaptic strength - are associated with learning processes, but also occur under basal conditions, i.e. without any learning. To investigate these spine dynamics, large-scale analyses are required. However, the detection and tracking of individual spines is mostly done manually and thus a major bottleneck in their analysis. Here, we propose a machine-learning-based framework that enables the automated detection and tracking of spines from 3D image stacks acquired through two-photon microscopy. This tool provides scientists with the power to identify thousands of spines and track their evolution over time under different conditions. This will help develop models and investigate cognitive processes, like learning or forgetting, on the level of cortical neuronal networks.

Researches are pursuing new discoveries of superheavy elements (SHEs) and much experimental progress has been made regarding their synthesis and related stability aspects. At present the heaviest element observed is Z = 118, known as Ognesson (Og). Both theoreticians and experimentalists are working in this field to find the feasible incoming channels for the synthesis of new superheavy elements. In nuclear physics, several theoretical models exist which move in this direction to give predicted cross sections. For the first time within the frame work of the dynamical cluster-decay (DCM) by Gupta and collaborators. We provide reasonable estimations for cross sections of nuclear systems which will give rise to new superheavy elements. The important point and the new approach presented in our work are the theoretical calculations done without tracking any reference to experimental data; i.e., these are independent calculations. We will provide a method for reasonable estimates for the cross sections of unobserved possible decay channels in a nuclear reaction. We will demonstrate the capability of the DCM to give a probable/sensible range of the cross section. This exercise of theoretical calculations within the DCM may benefit experiments for the discovery of new superheavy elements. Our results can provide hints to the experimentalist to choose proper incoming channels in order to perform experiments.

The human tyrosine-protein kinase Met is a trans-membrane receptor placed on the plasma membrane. Also called c-Met receptor, it is crucial in the signalling process that regulates cell migration, replication and growth, both in embryos and adults. Its deregulation leads to a plethora of diseases among which feature cancer and bacterial invasion. Indeed, while the overexpression of the c-Met receptor strongly correlates with poor prognosis in cancer, the binding with a invasion protein of L. Monocytogenes, the Internalin-B (InlB), triggers the c-Met-mediated internalisation of the bacterium. With the final aim of understanding how the L. Monocytogenes induces the c-Met receptor activation, we employed Molecular Dynamics Simulations. The simulated trajectories enabled us to quantify the conformational changes on the c-Met monomer driven by the InlB binding in terms of a collective variable extracted through unsupervised learning. In parallel, we were able to build a structural hypothesis for the c-Met/InlB homodimer to validate via new FRET experiments by our experimental collaborators.

To facilitate vital reactions, the cell is compartmentalized into membrane-confined organelles. However, reaction or signalling pathways are shared between organelles and a variety of metabolites need to be exchanged. Therefore, cells developed a complex interaction and transport system which is largely based on the cytoskeleton and its motor proteins. In summary, movement in plants can either be directional along specific filaments or undirectional by cytoplasmic streaming.

The peroxisome is a cell organelle involved in multiple pathways and serves different functions depending on the tissue type. Previous work suggests that with ranging functions ranging movement types occur. In this project, light sheet microscopy is used to gain insights into the behaviour of A. thaliana root-cell peroxisomes. The timelapse data was used to track individual peroxisomes over time, calculate their speed, abundance and size and describe their movement patterns as well as interactions with one another. The data was analysed with particular regard to individual, rare movement events, and four events were actually identified.

Monte Carlo simulations play a crucial role in all stages of a particle collider experiment. Depending on the accuracy requirements of required simulations or the amount of simulated collisions, their computational complexity can greatly vary. In order to be able to handle this computational workload, a range of different distributed computing approaches can be employed by HEP experiments. In addition to giving an overview to traditional approaches, this talk presents a novel distributed computing approach for HEP which is based on blockchain technology. It features the description of a Proof-of-Useful-Work consensus algorithm which aims to both support real-world HEP experiments with the production of required MC simulation data and to secure the underlying blockchain infrastructure at the same time. Instead of being an alternative to CERNs WLCG or BOINC projects that rely on volunteer computing, this approach aims to be a complementary source of additional computing power that can further support real-world HEP experiments.

Representational drift is a widely observed phenomenon occurring even under basal conditions (i.e. without any explicit learning paradigm), however, the underlying mechanism is currently unclear. Here, by combining network modeling with imaging in mouse auditory cortex, we ask whether representational drift can be fully accounted for by spontaneous (random) synaptic changes or involves a Hebbian-type plasticity. Moreover, we ask how fear learning alters these dynamics.  

Using chronic two-photon recordings in mouse auditory cortex, we demonstrate that under basal conditions both signal and noise correlations change significantly across imaging days. Furthermore, high signal correlations on one day are predictive of increased noise correlations on the next day, but not vice versa. To explain these observations, we study a linear recurrent network, which receives stimulus-driven input. We found that this model can fully account for these observations, but only if i) the feedforward connections change (either through spontaneous drift, or stimulus-driven plasticity occurring despite basal conditions) and ii) recurrent connections are subject to Hebbian-like plasticity. Random changes in the recurrent connections appear neither necessary nor sufficient for reproducing the empirical results.  

During fear learning both signal and noise correlations continue to be highly volatile, but their mutual dependency switches, such that noise correlation is now predictive of signal correlation on the next day. In the network model, this effect can neither be reproduced by changes in the input nor random changes in recurrent connections, but requires reducing the synaptic learning rate.  

Thus, based on our data and a generic network model, we conclude that representational drift during basal conditions involves both changes in the input and Hebbian-like plasticity of recurrent connections. Fear learning, instead, leads to a decrease of this plasticity, thus protecting recurrent connections from ‘overwriting’, which may serve the transfer to long-term memory.