Title: Predictive cell-specific gene regulatory models
Speaker: Hatice Ulku Osmanbeyoglu
Summary and Thoughts:
Dr. Osmanbeyoglu presented at the ISP forum this week on her research on prediction of cell-specific gene regulation using machine learning (ML) models. Her work focuses on techniques to use ML to gain biological and clinical insights from multiomics data. As per the definition, ‘multiomics’ refers to a biological analysis approach in systems biology where data from multiple omics (genomics, proteomics and so on) are combined to study life. While I do not have any experience in systems biology, the talk was an interesting introduction to multiomics data and analysis. The motivation for the research is quite clear, and so is the novelty - when signaling pathways in the body send information to the nucleus, they modulate activity of transcription factors (TFs), leading to up/downregulation of target gene expressions.
The Single cell Proteomic and RNA based Transcription factor Activity Network (SPARTAN) developed by Dr. Osmanbeyoglu integrates data sources to link upstream signaling to downstream response. This allows mapping of surface proteins and gene expressions to infer TF activity and cell surface protein activity. The ML approach involves using gene-level binary features, vectors for cell-specific surface proteins and gene expression data with regularized bilinear regression to make these predictions. With this framework, one can ask key questions such as the following and gain insights regarding TFs that underlie key developmental or differentiation transitions and activation states of cells (e.g. within the immune system).
- What are critical regulators such as TFs underlying cellular identities?
- What ared ifferent or common regulators givena cell types across tissues (B cells in spleen vs lung)?
- What are commonalities as well as differences of cell-specific regulatory programs across healthy individuals and those manifesting a disease?