This research projec aims to contribute to the development of methods that extract informative and interpretable features from high-dimensional datasets, with a focus on uncovering high-level causally related factors that describe meaningful semantics of the data. This, in turn, can help us gain deeper insights into the representations found within advanced generative models, particularly foundation models and LLMs, with the goal of improving their efficiency and safety.

Amongst the leading causes of retinal dystrophies worldwide is Retinitis pigmentosa, a severe disease often caused by splice variants in the USH2A gene. This project aims to develop a safe CRISPR-based therapeutic strategy for correction of such splicing defects. Using enhanced-deletion nucleases, the disease-causing alterations can be eliminated, hereby restoring correct protein synthesis. The focus lies on safety features as well as the development of an inducible viral delivery system for clinical application.

The research project investigates the influence of sterically demanding NHC-gold(I) complexes on the cyclization of diyne derivatives. The focus is on the synthesis of various sterically demanding NHC-gold(I) complexes and their application in diyne cyclizations, particularly examining the reactivity and selectivity in gold-catalyzed reactions. Further investigations include theoretical calculations and practical applications of the synthesized cyclization products for pharmaceuticals or organic materials.

This project seeks to advance rare disease classification using deep neural networks by addressing key challenges such as limited data and high heterogeneity. We will assess existing models and their representations, investigating how technical variations in medical data affect their characteristics.

We aim to develop artificial neural networks with brain-inspired circuits. Like in the brain, artificial neurons and synapses are built with novel memristors, arranged in a crossbar array. By combining these with ultra-fast photonics, we seek to improve signal processing and matrix-vector multiplications to address limitations from state-of-the-art architectures. This approach aims to advance computing solutions by reducing energy consumption, computational time, and system complexity.