09.15 - 09.45 Faculty club (1870-816)

Nikolai Hecker: Deep learning and single cell omics reveals shared enhancer codes in human, mice, and chicken brains

Vertebrate brains exhibit striking anatomical differences such as the mammal-specific neocortex. Despite their differences, vertebrate brains comprise similar cell types. Cell types are defined by transcription factors that regulate gene expression by binding to genomic enhancers. These binding sites form regulatory enhancer codes. We combined deep learning with single cell omics to identify and compare these enhancer codes  between human, mouse, and chicken brain cell types. As a prerequisite, we created a spatial transcriptomics and single cell omics atlas of the chicken telencephalon. We identified specifically accessible genomic regions as a proxy for enhancers. Next, we trained DNA sequence-based deep learning models on these regions to extract the enhancer codes for the different brain cell types. The enhancer codes are highly conserved between non-neuronal cell-types and GABAergic neurons of mammals and birds whereas homologies between excitatory neurons are less well defined. Surprisingly, we detected similar enhancer codes for subsets of mammalian neocortical and chicken excitatory neurons that question aspects of previously proposed evolutionary models. Furthermore, our study shows that deep learning approaches solely based on genomic sequences and ATAC-seq are capable of inferring cell type homologies that are comparable with  transcriptome-based comparisons.

12.00 - 12.30 Faculty club (1870-816)

Lena Collienne: Combining theoretical, empirical, and deep learning approaches to enable a paradigm shift in phylogenetics

Phylogenetic trees display evolutionary relationships across biological scales, ranging from species level to individual viral genomes. Traditional approaches to phylogenetic inference rely on principles developed decades ago and do not scale to large datasets with tens of thousands of sequences, a size now standard in viral genomics and pandemic surveillance.

In this talk I will explain how I aim to enable a paradigm shift in phylogenetic inference by combining a fundamental understanding of phylogenetic tree space, insights from empirical data, and deep learning models that integrate these insights to scale up phylogenetic inference.

First, I show how phylogenetic trees can be structured into a tree space using tree rearrangements and how fundamental theoretical advances can transform a computationally intractable problem in tree space into an efficiently solvable one. I then highlight the role of empirical data in guiding method development using the example of ``online methods'', where an inferred tree is updated as new data arrives - common during viral outbreaks - without rerunning the entire analysis. Finally, I present a deep learning approach that combines our understanding of tree space and empirical data, moving us a step closer to efficiently reconstructing phylogenetic trees for large datasets, thereby addressing a critical need in modern genomics.

15.00-15.30 Online

Alla Mikheenko: Computational approaches to understanding neurodegeneration

As the global population ages, neurodegenerative diseases are becoming increasingly common. In high-income countries, dementia has already become the second leading cause of death. Despite this, our understanding of how these diseases begin and progress remains incomplete, and we still lack effective diagnostic and treatment options. Progress in this area will require not only experimental discoveries but also new bioinformatics methods and pipelines to extract meaning from rapidly growing data.

In this presentation, I will show how computational approaches are helping us investigate neurodegenerative diseases and identify potential biomarkers and therapeutic targets. My research focuses on amyotrophic lateral sclerosis (ALS), a fatal disease characterized by progressive muscle paralysis. I will highlight our recent research on alternative splicing aberrations in the disease and the implications of this work from improving our understanding of disease progression to discovering novel early biomarkers. I will also share our study of the cell-type specificity of ALS, aimed at better understanding the role of RNA-binding proteins in the effects of TDP-43 pathology. Additionally, I will discuss how we use large-scale datasets and machine learning approaches to validate and extend experimental findings.