Jonas Wagner
Selected projects

03 Master’s thesis · Heidelberg University · 2026

Predation at the microscale

How do small openings change a predator’s access to its prey? I built a multimodal microscopy pipeline to quantify how soil protists move and how bacterial abundance changes in microfluidic assays.

See the thesis figures

Motivation

Can prey shelter in smaller pores?

In soil, pore openings can limit which predators reach their prey. To study that effect, EMBL Heidelberg’s Van Gestel group ran assays with 1, 2, and 10 µm openings in microfluidic chips. I developed the analysis pipeline for the microscopy data from those experiments.

Quantitative approach

Measure predators, prey, and movement

I built a Nextflow workflow that aligns time series and applies channel-specific models. Custom YOLO models segmented objects in label-free microscopy (DIC) and red exclusion-dye images. An ilastik classifier measured the area covered by fluorescently labeled bacteria. The workflow yielded more than 300,000 protist detections across the assays.

Method

Link the planes, keep the clearest view

One protist can appear in several optical sections. Within each time point, I linked masks across the 11 DIC focal planes and chose one plane per object for measuring its size, shape, and texture. The final focus score combined edge sharpness and contrast with the local background, using model confidence to break ties.

Three-step schematic of matching object masks across Z planes, scoring focus per plane, and extracting the highest-scoring plane
FIG. 06 The thesis schematic shows how the workflow compares mask overlap among nearby detections, scores the candidate planes, and selects the best-focused plane.

Result

Smaller openings limited predator access

At 10 µm, activity extended across the barrier and fluorescent bacterial coverage declined fastest. At 2 µm, activity beyond the barrier was reduced but still detectable. The 1 µm openings appeared largely impassable to protists in these assays. Object counts from DIC and the red exclusion-dye channel also fell as openings narrowed.

Three plots of fluorescent bacterial area over time. The mean area falls most for 10 µm openings, falls less for 2 µm openings, and stays closer to the reference for 1 µm openings.
FIG. 15 Segmented green fluorescence area served as a proxy for bacterial abundance. Bold lines show the geometric mean for each barrier size. The gray band shows variation in controls without soil protists.
Three space-time heatmaps of mean DIC activity for 10, 2, and 1 µm openings. Activity spans much of the large-barrier image and is increasingly confined for smaller barriers.
FIG. 18 Kymographs of movement-related signal in label-free DIC images. Color represents mean activity across replicates at each horizontal image position and time point. L, M, and S correspond to 10, 2, and 1 µm openings.

Phenotypic diversity

Size explained the most variation

I compared measurements of size, shape, texture, and intensity with features learned by a Cell-DINO–inspired model. Size was the main axis of variation in the measured features. Both approaches gave exploratory maps of protist phenotypes, with taxonomic identity still unresolved.

Thesis title

Development of a multi-modal image analysis pipeline to phenotype predator-prey dynamics in microfluidic predation assays. Jonas Wagner, Heidelberg University, 2026.