Authors:

Yuval Mazor

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Single-cell ATAC-seq maps chromatin accessibility at a single-cell resolution, identifying regions of open chromatin. In the context of bone marrow hematopoiesis, multiome data, which pairs chromatin accessibility with gene expression levels, holds great promise for elucidating the epigenetic regulation of cell fate decisions and transcriptional dynamics. Conventional ATAC-seq pipelines primarily focus on analyzing chromatin accessibility by detecting accessible genomic regions, often overlooking the rich information encoded in the actual shapes of accessibility peaks. These peak shapes can offer a more nuanced understanding of chromatin dynamics, revealing insights into transcription factor (TF) binding loci, RNA polymerase recruitment, and broader transcriptional machinery interactions. Here, we analyzed subtle patterns in single-cell bone marrow data by modeling the distribution of ATAC-seq reads, which result from the unique pattern of enzymatic (Tn5 transposase) cutting of accessible DNA. Specifically, we generated histograms of Tn5 cutsites centered at transcription start sites (TSS) to investigate whether shape-based information regarding chromatin accessibility patterns can predict gene expression levels. To capture shape-based features in a lower-dimensional space, we trained an autoencoder that preserves differences between cutsite distributions by encouraging Euclidean distances in the latent space to approximate the histograms' pairwise Wasserstein distances, a measure from optimal transport theory well-suited for comparing distributions. We then performed linear regression on both the latent embeddings and the histogram features to predict gene expression. These analyses revealed that promoter shape may encode information regarding transcriptional levels, supporting the idea that accessibility pattern shapes carry regulatory signals lost in conventional ATAC-seq analysis pipelines.

Source:

Columbia / Psychology / 2027

Topics:

No topics listed

Co-authors:

Yuval Mazor