THUNDER

THUNDER infers cell type proportions from bulk Hi-C contact data using a reference-free, two-step deconvolution approach based on Non-negative Matrix Factorization (NMF) to resolve cell-type-specific chromatin contact profiles.

Key Features:

• Two-Step Deconvolution Approach: Performs feature selection specific to intra-chromosomal or inter-chromosomal Hi-C readouts to retain informative bin-pairs and then applies deconvolution.
• Unsupervised / Reference-Free: Estimates cell type proportions without prior knowledge of cell type signatures or proportions.
• Non-negative Matrix Factorization (NMF): Uses NMF as the computational framework for deconvolution of bulk Hi-C contact matrices.
• Simulation-Based Validation: Validated via extensive simulations using two published single-cell Hi-C (scHi-C) datasets and compared against MuSiC, TOAST, and standard NMF approaches.
• Application to Adult Human Cortex: Applied to adult human cortex Hi-C data to estimate cell type proportions and identify cell-type-specific interactions.
• Estimated Contact Profiles: Produces estimated contact profiles that enable exploration of cell-type-specific chromatin interactions and the chromatin interactome.
• Adjustment for Population Samples: Enables adjustment for varying cell type compositions in population Hi-C samples to support downstream analyses such as differential chromatin organization.

Scientific Applications:

• Cell composition adjustment: Adjusts for differential cell type proportions across heterogeneous bulk Hi-C samples to improve accuracy of contact analyses.
• Detection of cell-type-specific interactions: Identifies putative cell-type-specific chromatin contacts from bulk Hi-C data.
• Genomics and epigenetics studies: Supports studies of chromatin organization, gene regulation, and the cellular basis of complex traits and diseases using Hi-C data.

Methodology:

Feature selection of informative bin-pairs tailored to intra- or inter-chromosomal contacts, followed by a two-step deconvolution based on Non-negative Matrix Factorization (NMF), with simulation-based validation using two published single-cell Hi-C (scHi-C) datasets.