A multi-level bias correction model for bulk and single-cell CUT&Tag data

点击翻译按钮获取中文摘要

Histone modifications (HM) and transcription factors (TF) are key factors in maintaining the cell identity by regulating the type-specific gene expression program and chromatin structure. Both HMs and TFs can be aberrantly regulated in the pathogenesis and are a major class of cancer cell dependencies. Precise detection of HM and TF binding genome-wide is essential for a better understanding of transcriptional regulation. Cleavage Under Targets & Tagmentation (CUT&Tag) is an easy and low-cost epigenomic profiling method that can be performed on a low number of cells or even on the single-cell level. Thousands of CUT&Tag datasets have been generated for profiling TF binding sites and HMs since the advent of this technique, providing a valuable resource for functional genomics and disease research. CUT&Tag experiments rely on the hyperactive transposase Tn5 for tagmentation. Tn5 is subject to intrinsic sequence insertion biases, and enrichment of Tn5 captured reads toward chromatin accessibility regions also confound the distribution of CUT&Tag reads, especially for factors with weak association with chromatin accessibility. Both features bring great biases in the CUT&Tag data that confound the data analysis. For example, Strong CUT&Tag signal enrichment of repressive histone modification H3K27me3 can be observed at actively transcribed gene promoter regions where chromatin is openly accessible but no H3K27me3 signal from ChIP-seq, indicating that the observed CUT&Tag signal is likely false positive. The high-sparsity characteristics of single-cell data makes the intrinsic biases more substantial compared to bulk data, creating additional challenges in computational modeling and data analysis. For example, the average Tn5 intrinsic cleavage bias level varies across individual cells and confound the cell clustering result from single-cell ATAC-seq data, which carries similar Tn5 intrinsic bias as CUT&Tag. Based on these preliminary observations and our group’s existing work, we propose to develop computational models to accurately quantify both the open chromatin bias and the Tn5 intrinsic cleavage bias from CUT&Tag data on both bulk and single-cell levels. Using the new model to be developed, we will characterize how open chromatin and intrinsic cleavage biases affect the detection of HM and TF binding sites in both bulk and single-cell level CUT&Tag data. The bias correction model can be further incorporated in existing or new bioinformatics methods to detect the HM/TF signals, for both bulk and single- cell CUT&Tag data. this project focuses on developing a computational method for bias correction for improving CUT&Tag data analysis. The proposed computational method complements existing bioinformatics tools and will have broad applications in functional genomics and epigenomics research. The results from the proposed work will fill the knowledge gap in single-cell studies of chromatin dynamics and transcriptional regulation and could provide mechanistic insights for both basic science and translational studies.