Research Focus

Neuro-symbolic Artificial Intelligence combines the pattern recognition capabilities of powerful deep learning models with the structured knowledge of symbolic approaches, such as those derived from logic or ontologies. In our research group, we investigate how these methods can be applied to reveal and explain complex biological relationships in experimental data. We explore various innovative strategies to align knowledge-driven, logical structures with data-driven, statistically inferred patterns. We are a young, communicative team that values scientific collaboration and is always open to jointly addressing interesting research questions using machine learning, including those beyond the scope of bioinformatics.

For example, in the analysis of large RNA datasets, as well as single-cell or spatial omics data, the combination of data-driven learning and explicit knowledge offers many opportunities – such as more robust cell type classification, integration of biological ontologies, and the generation of biologically plausible hypotheses. A key advantage of neuro-symbolic approaches lies in their explainability: while conventional deep learning models often operate as “black boxes,” neurosymbolic methods enable transparent reasoning behind predictions, for instance by incorporating biological rules, pathways, or gene regulatory relationships. From a biological and medical perspective, this provides direct added value in interpretation: results can be more effectively linked to existing domain knowledge, new hypotheses can be derived more purposefully, and experimental validation can be more precisely guided. In this way, neurosymbolic models not only enhance precision but also strengthen trust and interoperability in bioinformatics analyses within molecular biology research. For further details, see below.

Modern molecular biology increasingly generates multimodal datasets – for example, gene expression profiles, miRNA signatures, epigenetic information, or histological imaging data. Our research group develops and applies machine learning (ML) methods to jointly analyze these diverse data sources, aiming to obtain a more comprehensive view of biomedical processes.

By integrating multimodal information, it is possible to uncover relationships that remain hidden when examining individual data types alone – such as complex disease patterns, regulatory networks, or cellular states. We combine classical bioinformatics analyses with modern ML approaches, including embedding techniques, attention models, and variational autoencoders, to systematically link heterogeneous data. The result is interpretable models that help to elucidate molecular mechanisms and generate new biological hypotheses.

Our research group is dedicated to fundamental research in molecular biology, with the goal of understanding cellular and regulatory processes through data-driven approaches. We analyze large molecular datasets – often in national and international collaborations – such as those from RNA sequencing (bulk and single-cell), addressing fundamental questions on gene regulation, cell type identity, molecular networks, and disease mechanisms in health, aging, and disease.

Example: High-throughput single-cell and spatial transcriptomics

The group focuses on the analysis of high-dimensional molecular data generated by modern high-throughput technologies, in both bulk and single-cell experiments. In particular, we study gene and miRNA expression profiles to better understand biological processes at the transcriptional level. Bulk RNA sequencing provides robust averages across cell populations and is a well-established tool for analyzing differentially expressed genes and regulatory miRNAs under various biological conditions. Complementary to this, single-cell and single-nucleus RNA sequencing allow the investigation of individual cells, offering insights into cellular heterogeneity, rare cell types, and dynamic gene expression patterns at single-cell resolution. We have also gained experience in the emerging field of spatial omics technologies, which enable the integration of spatial and temporal coordinates with molecular measurements.

Example: Neurodegeneration and infectious disease research

We are also engaged in the bioinformatic analysis of molecular biology data at the interface between neurodegenerative and infectious diseases. Our focus includes conditions such as Alzheimer’s disease, Parkinson’s disease, and viral infections, including those caused by SARS-CoV-2. The aim is to better understand the underlying molecular mechanisms and systematically capture disease-associated changes at the cellular and tissue levels.

For all these data types and modalities, we employ classical bioinformatics tools – such as quality control, normalization, and statistical analysis – together with modern machine learning and deep learning methods. These approaches enable us to detect complex patterns, identify cell types, reconstruct gene regulatory networks, and reveal disease-specific signatures. Such signatures allow us to describe disease processes at their root and systematically assess their translational potential for developing new biomarkers or targets for therapeutic intervention. By combining classical analytical methods with modern data-driven approaches, we contribute to advancing the understanding of age- and disease-related changes in the human brain and to elucidating immunological processes in infections.