NLM DIR Seminar Schedule

Leveraging a deep learning model to assess the impact of regulatory variants on traits and diseases

Deep learning has revolutionized genomics by enabling the extraction of complex patterns from large-scale biological datasets. We leveraged a CNN-based model to characterize the impact of genetic variants on enhancer activity in the human genome, focusing on two key contexts that I will discuss in my talk.
First, I will discuss how we used our model to quantify the genome wide impact of regulatory variants on enhancer turnover which has driven phenotypic divergence during past speciation events and continues to facilitate environmental adaptation. Specifically, we explored how the fate of the cis-regulatory landscape changes in the face of accumulating variants by addressing key questions, such as, what fraction of the genome undergoes regulatory turnover due to accumulating variants, and, are certain genes or transcription factor binding sites more susceptible to regulatory changes than others? We investigated regulatory turnover across three evolutionary contexts: 1) recent evolutionary history – variants arising from human-chimpanzee substitutions, 2) modern population variation – genetic diversity observed in present-day human populations, and 3) mutational susceptibility – the effect of random mutations on enhancer activity. I will highlight key findings of our study which broadly support a model of regulatory innovation within cell type–specific enhancers with potential to accommodate adaptive changes while preserving the core regulatory network that is driven by cell ubiquitous evolutionarily constrained enhancers.
In the second part, I will discuss how the model can enhance the identification of genetic variants associated with Polycystic Ovary Syndrome- a complex neuroendocrine disorder affecting millions of women worldwide. By predicting the regulatory impact of PCOS-associated variants, our model improves the prioritization of functional variants and provides new insights into the genetic basis of this disorder.

TBD

Decoding tumor heterogeneity: computational methods for scRNA-seq and spatial omics

Abstract. Characterizing and understanding drivers of tumor transcriptional and epigenetic heterogeneity is key to advancing personalized medicine and developing effective therapies. In this presentation, I will discuss our recent work on designing a computational methodology to extract gene signatures for distinct transcriptional states of cancer cells from single-cell RNA sequencing data (scRNA-seq) and show examples of linking intratumor transcriptional heterogeneity to tumor microenvironment and clinical variables. In this context, I will talk about the best-performing existing approaches for the integration of scRNA-seq data from malignant cells across cancer patients and also present our recently developed scRNA-seq-based CancerFoundation model, which, in addition to being capable of data integration across patients, can be used for predicting drug responses. I will conclude with our latest efforts in spatial transcriptomics and demonstrate how supervised machine-learning approaches that use spatial information can further resolve the complexity of cancer and provide explainable clinical biomarkers.

Bio. Prof. Dr. Valentina Boeva is a Tenure Track Assistant Professor at the Department of Computer Science, ETH Zurich, where she leads the Computational Cancer Genomics Group. Her research focuses on developing computational methods for multi-omics data integration to understand the epigenetic and transcriptional plasticity of cancer cells. Before joining ETH Zurich in 2019, Prof. Boeva led the Computational Epigenetics of Cancer laboratory at Inserm's Cochin Institute in Paris. She holds a Ph.D. in Bioengineering and Bioinformatics from Lomonosov Moscow State University. Throughout her career, Prof. Boeva has made contributions to the field of computational cancer genomics with developing methods for the analysis of DNA sequencing data, bulk and single-cell transcriptomics and epigenomics data, and, recently, spatial transcriptomics and proteomics.