STATISTICAL CURVE MODELS FOR INFERRING 3D CHROMATIN ARCHITECTURE.

Area of Science:

• Computational Biology
• Genomics
• Biophysics

Background:

• Reconstructing 3D chromatin structure is crucial for understanding cellular processes.
• Current methods often represent chromatin as polygonal chains, neglecting its smooth, continuous nature.
• Direct imaging of chromatin architecture is experimentally challenging.

Purpose of the Study:

• To develop novel computational methods for reconstructing 3D chromatin structure from Hi-C data.
• To directly model the 1D curve of chromatin in 3D space using B-spline and smoothing spline techniques.
• To enhance statistical models for analyzing sparse Hi-C contact data, including single-cell assays.

Main Methods:

• Development of B-spline and smoothing spline techniques to capture the 1D curve of chromatin.
• Integration of spline methods with a Poisson model for Hi-C contact counts.
• Extension of the distribution-based metric scaling (DBMS) framework to include zero-inflated and Hurdle Poisson models, and negative binomial models for sparse data.

Main Results:

• The proposed spline-based methods effectively capture the complex 1D curve of chromatin.
• The new statistical models provide improved analysis of sparse Hi-C data, including single-cell data.
• Performance comparison on bulk Hi-C data from IMR90 cells and single-cell Hi-C data from mouse embryonic stem cells.

Conclusions:

• Spline-based approaches offer a more accurate representation of 3D chromatin structure compared to polygonal chain models.
• The developed statistical framework enhances the analysis of sparse and complex Hi-C datasets.
• This work provides advanced computational tools for studying genome organization and its functional implications.