PAIR: Reconstructing Single-Cell Open-Chromatin Landscapes for Transcription Factor Regulome Mapping

Area of Science:

• Genomics and Bioinformatics
• Epigenetics
• Computational Biology

Background:

• Single-cell ATAC-seq (scATAC-seq) provides cellular resolution of chromatin accessibility but suffers from low sequencing depth, sparsity, and missing data.
• These limitations hinder accurate cell-state classification and the inference of transcription factor (TF) regulatory networks.
• Existing methods struggle to effectively address the inherent noise and sparsity in scATAC-seq data.

Purpose of the Study:

• To introduce PAIR, a novel probabilistic framework designed to restore and impute scATAC-seq accessibility profiles.
• To improve the robustness of cell-state delineation and the inference of TF regulatory programs from sparse scATAC-seq data.
• To provide a computational tool that enhances the utility of scATAC-seq for biological discovery.

Main Methods:

• PAIR employs a bipartite graph encoder to learn joint representations of cells and peaks.
• It incorporates a variational latent layer to model uncertainty from sparse and noisy measurements.
• Two decoders (qualitative and quantitative) reconstruct cell-peak incidences and accessibility counts, respectively, using Negative Binomial likelihood.

Main Results:

• PAIR significantly improves clustering performance and increases sensitivity for differential accessibility analysis on simulated and real scATAC-seq datasets.
• The framework effectively restores regulatory signals from both promoter-proximal and distal elements.
• PAIR-derived peak embeddings facilitate locus-centric regulatory interrogation, revealing structured regulatory neighborhoods and identifying clinically relevant gene sets.

Conclusions:

• PAIR offers a robust probabilistic framework for imputing and enhancing scATAC-seq data, overcoming key technical limitations.
• The method demonstrably improves downstream analyses, including cell-state classification and regulatory network inference.
• PAIR enables deeper biological insights into gene regulation and cellular specialization from scATAC-seq experiments.