Biomedical Genomics

Research Strategy

How is transcriptional control of gene expression regulated in such a way that the right transcriptional programs get activated in the right cell type and the right developmental time? And how is this complex fine tuning altered in diseases, such as cancer, mental disorders and infections? These are the fundamental questions that our group is working on, using a combination of statistical and machine-learning approaches to integrate multiple data types and extract molecular multi-omics signatures representing different regulatory or transcriptional states. We apply this to bulk and single-cell datasets and apply transfer learning strategies to identify shared signatures in different data modalities or diseases, in particular to describe mechanisms of comorbidity. We are increasingly working on interpretable deep learning models (in particular auto-encoders) to provide a fine-grained computational phenotyping of single-cells or patients and relate these molecular characteristics with clinical and macroscopic phenotypic features.