Impact of data resolution on three-dimensional structure inference methods.

Area of Science:

• Genomics
• Computational Biology
• Biophysics

Background:

• Genome-wide chromatin interaction assays generate vast datasets, advancing understanding of 3D genome structure.
• Increasing data resolution (down to 1 Kb) presents analytical challenges, particularly an excess of zero contact counts.
• Existing methods, developed for lower resolution, struggle with high-resolution data characteristics.

Purpose of the Study:

• To develop a robust analytical method for 3D chromatin structure reconstruction from high-resolution interaction data.
• To address the challenge of excess zeros in genome-wide contact count data.
• To evaluate the performance of the proposed method against existing approaches.

Main Methods:

• Proposed the truncated Random effect EXpression (tREX) method to handle data with excess zeros across various resolutions.
• Assessed tREX and leading existing methods using in-silico generated data at multiple resolutions.
• Validated methods on a human lymphoblastoid cell line Hi-C dataset with available FISH measurements.

Main Results:

• The tREX method demonstrated consistently good performance across 30 simulated settings, proving robust to resolution and parameter variations.
• tREX showed insensitivity to model misspecification, maintaining accuracy in 3D structure estimation and topologically associated domain preservation.
• Application to human chromosomes 14 and 22 showed tREX-constructed 3D structures consistent with FISH measurements, with higher distance correlation.

Conclusions:

• The tREX method offers superior performance for 3D chromatin structure reconstruction from high-resolution interaction data.
• tREX provides accurate and robust estimation, outperforming existing methods in both simulated and real-world datasets.
• An open-source R-package for tREX is available, facilitating its application in genomic research.