< Back to previous page

Project

Resolving the gene-regulatory dimension of the Fly Cell Atlas using single-cell multi-omics and deep learning

Single-cell transcriptomics and single-cell epigenomics provide a powerful means of cataloging known cell types, discovering novel ones, and deciphering their underlying regulatory principles.
Constructing such a cell atlas requires state-of-the-art single-cell technologies that combine droplet microfluidics, cell hashing approaches, and multi-omics. Even more so, building a whole-animal atlas focused on genome regulation involves many computational steps including reproducible analysis pipelines, controlled vocabularies, and powerful methods to integrate information from transcription factor binding sites, accessible chromatin, and target gene expression at single-cell resolution. Here, we propose to generate a single-cell ATAC-seq and multi-omics atlas across tissues of the fruitfly Drosophila melanogaster, and apply new eXplainable Artificial Intelligence (XAI) methods to unravel the gene regulatory programs across cell types. Within this framework we will explore deep learning on cells and enhancers, and develop new ways to combine deep learning with topic modelling, motif discovery, and co-expression network inference. Finally, we will extend this framework towards Generative Adversarial Networks (GAN) to create synthetic enhancers and test these by in vivo transgenic reporter assays. This will allow further fine-tuning our AI models, and will yield novel insight into enhancer architectures that underlie the spatiotemporal control of gene expression.

Date:1 Jan 2021 →  Today
Keywords:Single-cell transcriptomics, single-cell epigenomics, cell atlas, (XAI) methods, Generative Adversarial Networks (GAN), fruitfly
Disciplines:Single-cell data analysis, Computational transcriptomics and epigenomics, Computational biomodelling and machine learning, Invertebrate biology, Epigenomics