Open calls in doctoral studies, the first period

Supervisors: Angela Ivask, Elin Org

The increasing interest in antimicrobials, particularly non-antibiotic ones, for hygiene is reflected in the rise of scientific publications and market share. One preventative measure to reduce the spread of unwanted microbes via touch transfer is application of antimicrobial surfaces. Such surfaces are valued for their effectiveness against microbes, but their efficacy testing often uses limited bacterial strains and conditions that don't match real-world use. This PhD thesis aims to identify key microbial groups on high-touch surfaces and examine how these populations change on surfaces with antimicrobial agents (e.g., silver, copper, quaternary ammonium compounds, reactive oxygen radicals). The study will optimize RNA extraction and 16S rRNA sequencing methods for low biomass samples, comparing results with 16S rDNA sequencing to distinguish active from inactive microbes. An artificial microbial community will be created to observe structural changes during antimicrobial treatment. Similar to antibiotic treatments, antimicrobial agents may select for tolerance or resistance genes. The study will use metagenomic analysis to detect and analyze the frequency of these genes in touch surface samples after antimicrobial exposure.

Supervisors: Asad Munir, Sulev Reisberg, Bernadette Stolz (Max-Planck Gesellschaft zur Förderung der Wissenschaften e.V), Tobias Heimann (Siemens Healthineers AG)

This project develops methods to turn messy, irregular clinical measurements from
electronic health records into machine-readable representations that improve
patient-level predictions. First, it creates several types of numerical and temporal
embeddings — from simple scalar tokens to learned continuous vectors and
window-based summaries — and benchmarks how much information they preserve.
These embeddings are tested on baseline tasks such as short-term event prediction to
show that they capture more clinical signals than naïve approaches.
Next, the project focuses on modelling how measurements change over time. It trains
temporal models that handle irregular sampling, missingness, and variable data
quality, producing time-sensitive embeddings that summarize patient trajectories.
Finally, these embeddings are integrated into a pretrained clinical Foundation Model
and fine-tuned for outcome prediction. By comparing predictions across raw data,
non-temporal embeddings, and the full integrated model, the project demonstrates
that learned representations of continuous measurements can meaningfully enhance
clinical forecasting.