Research

We combine machine learning with epigenome profiling to decode enhancer logic. To test enhancers we developed a massively parallel enhancer-reporter assay, called CHEQ-seq. Our enhancer modeling focuses on mammalian TFs, such as TP53, SOX10/SOX9, GRHL1/2/3, AP-1, and TEADs; as well as on Drosophila TFs involved in eye development (e.g. Glass, Optix, sine oculis), epithelial development (Grainyhead), and tumour development (AP-1, STAT92E, and Scalloped).

We study neuronal and glial cell types in the ageing Drosophila brain using single-cell RNA-seq, and compare normal cell states with disease mutations involved in Parkinson’s and Alzheimer’s disease.

Single-cell transcriptomics (scRNA-seq) and single-cell epigenomics (scATAC-seq) data revolutionize the field of regulatory genomics. We combine new computational strategies (e.g., SCENIC, cisTopic) with state-of-the-art single-cell measurements (Drop-seq, 10X, InDrops, SeqWell) to decipher cis-regulatory “programs”, to reverse engineer gene regulatory networks, and to better define cell types and cell state transitions.

• Cis-regulatory variation is a major driver of phenotypic diversity and is associated to many diseases. By comparing chromatin accessibility across Drosophila inbred lines we aim to further our understanding of CRM divergence and plasticity, and the consequential divergence of gene expression and regulatory networks
• Similar techniques are applied to cancer genomes, where we sift through non-coding mutations to identify cis-regulatory driver mutations that have an impact on enhancer function and/or perturb the normal gene regulatory network in a cell.

The eye-antennal disc is a classical model system to study cellular differentiation. We use this system to unravel new genomic regulatory “recipes” that control cell fate decisions, such as photoreceptor specification and differentiation. We also perturb this system using irradiation, transcription factor perturbations, and RasV12-driven malignant transformation, to study cancer-related transcriptional changes, controlled by JNK, EGFR, and Hippo signaling pathways.

We develop new computational approaches that exploit single-cell technologies to link genome variation with changes in epigenome, transcriptome, proteome, and phenome. We apply this to human melanoma (e.g., phenotype switching), to the mouse liver, to the developing Drosophila eye and to ageing/neurodegeneration in the Drosophila brain. See also our collaborations.

By comparing transcriptomes, chromatin state and cis-regulatory modules across species, we learn about enhancer logic and the evolution of gene regulatory networks. We use RNA-seq, FAIRE-seq, and ATAC-seq across Drosophila species, alongside Ornstein-Uhlenbeck models to connect CRM evolution with variation in chromatin accessibility. We have also studied the evolution of epidermal and metabolic GRNs between Drosophila and Daphnia.

Data-driven research in our lab is powered by machine learning and artificial intelligence (AI) to help us guide and understand more about biological systems and processes. Here is a non-exhaustive list what the lab has been and is currently working on:

• Deep learning for genomics; hybrid convolutional and recursive neural networks trained on epigenomes and whole genomes to predict non-coding variation in disease (including cancer).
• Machine-learning applied to single-cell genomics to predict therapy choice and patient outcome.
• Support vector machines, random forest, and deep learning to classify enhancers.
• Gradient boosting machines and random forestregression to predict gene networks from single-cell transcriptomics data.
• Latent Dirchlet Allocation, collapsed Gibbs Sampling, and topic modeling for the analysis of single-cell epigenomics data.

We develop new bioinformatics tools for motif and CRM detection, and for gene regulatory network inference, such as i-cisTarget, iRegulon, and TOUCAN. We also maintain a large collection of curated position weight matrices (currently > 20.000). We exploit single-cell RNA-seq and single-cell ATAC-seq data to improve the identification of GRNs and enhancers, with our tools SCENIC and cisTopic.

We are interested in deciphering regulatory programs of transcriptional state switches in mammalian systems, including human and mouse. To study the cis-regulatory code in mammalian genomes we mainly use cancer cells as model system. During cancer progression, gene expression profiles can change, causing regulatory heterogeneity in tumors. This heterogeneity has an important impact on therapy response, since some cell states may be more or less vulnerable to a particular drug therapy.