Predicting cell-type specific combinatorial binding of neuronal TF network by Deep learning and Using Single-cell RNA seq to explore bone marrow immune landscape by IL-2 therapy

Abstract

1: Predicting cell-type specific combinatorial binding of neuronal transcription factor network by Deep Learning Transcription factors (TFs) are DNA binding proteins that recognize and bind to specific DNA nucleotide sequence (motifs), which recruit cofactors and RNA Polymerase II to initiate transcription. TFs play a vital role in regulating the expression of target genes and are involved in cellular processes such as development and differentiation. In neurons, cell-specific TFs play a crucial role in the differentiation process by guiding neuronal specification, migration and circuit assembly through transcriptional cascades. Mapping of chromatin marks and TF binding proteins at the genome and single-cell level can identify different cellular sub-types and their role in cell-type specific gene expression. Although deep learning methods have made progress in predicting TF binding sites, our understanding of cell-type specific binding and gene expression at single-cell resolution remains elusive. To understand the underlying mechanism that mediates combinatorial gene regulation and cell fate specification during development, we utilised neuronal-specific TFs in Drosophila early embryo to predict TF binding profiles and gene expression. In this project, we developed a novel Transformer encoded U-Net model to predict the combinatorial binding sites from the DNA sequence and chromatin accessibility and accurately identified the binding of 14 neuronal-specific TFs at single-cell resolution with an accuracy of 0.99. Our DL model employed two approaches, a classification model that predicts the TF binding sites and a regression model that predicts the gene expression. We applied the model to single-cell ATAC-seq data to obtain cell-type specific binding of these neuronal TFs at the single cell level and gene expression pattern from DNA sequence. To interpret the model predictions, we generated saliency scores of the input genomic regions and signals, and mapped them to identify the locations of the TF motifs at the predicted binding regions. The resulting saliency maps provided valuable insights into the mechanisms underlying the predicted binding patterns. We further validated our regression model predictions with the corresponding Precision nuclear run-on sequencing (PRO-Seq) and single-cell RNA-seq data.

This study presents a novel and robust method to determine cell-type-specific binding patterns and gene expression in Drosophila early embryo. We demonstrate that our model can predict the TF motif of various TFs in different cell types using the DNA sequence, shedding lights on the principles of combinatorial gene regulation and cell lineage by analysing TF binding and chromatin accessibility in conjunction.

2: Using Single-cell RNA seq to explore bone marrow immune landscape by IL-2 therapy : Bone marrow (BM) contains multiple immune cell subsets with critical functions and is considered an immune regulatory organ. In autoimmune diseases, inflammation can impair the BM niche, disturb hematopoietic and immune development, and induce osteoporosis. Interleukin-2 (IL-2) is a crucial cytokine that exhibits pleiotropic effect on the immune system and plays a significant role in the maintenance of CD4+ regulatory T cells, along with the differentiation of CD4+ T cells. The strength of the IL-2 signal affects the differentiation process with strong signals leading to short-lived effector T cells and low-level signals leading to follicular helper or central memory T-cells.

This project was designed to address the fundamental question of how low-dose IL-2 therapy modulates BM immune landscape by comprehensively mapping the therapy-induced changes using single-cell technologies and then dissecting the underlying mechanisms by mouse models. I analyzed the scRNA-seq data obtained from BM of CD45+ cells of mice to identify different immune cell types and compared their expression across four experimental conditions. Show more