Building high-resolution structural models of parasites to understand their biophysics
Self-supervised denoising that measures the noise first and routes automatically between structured and plain blind-spot training
Deep learning workflows for pixel-level annotation of electron tomography volumes using U-Net architecture
Complete models of T. brucei cytoskeleton amenable for finite element analysis
Open code releases and a validated open benchmark (PhantEM on Zenodo) — workflows, data, and models available to collaborators
The parasitic life cycle involves a multitude of physical interactions with the host microenvironment during stages of motility and adhesion. This requires optimal adaptation of the mechanical properties of the parasite to its environment.
The shape and elasticity of unicellular parasites such as Trypanosoma brucei are largely defined by their cytoskeleton, including a subpellicular array of microtubule filaments that forms a corset around the entire cell. How exactly the interaction between the beat of the flagellum and the mechanical response of the cell body gives rise to the intricate rotational motility patterns is not fully understood.
In this project, we build "virtual parasites" from high-resolution image data as the basis for a precise data-driven mechanical understanding of parasite biophysics.
A web-based platform bringing advanced machine learning and image analysis to researchers without requiring programming expertise. Features include:
A Python package for automated structural noise discovery and routed self-supervised denoising of volumetric electron microscopy. Key features:
Realism-validated synthetic benchmark pairs for denoising volumetric electron microscopy. Key features:
This project is funded as part of the DFG Priority Programme SPP2332 "Physics of Parasitism".