normGAM: an R package to remove systematic biases in genome architecture mapping data

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Genome architecture mapping (GAM) is a technique for genome-wide chromatin interaction analysis.
• Raw GAM data contains systematic biases, including a newly identified fragment length bias.
• Existing normalization methods may not effectively remove all identified biases.

Purpose of the Study:

• To identify and characterize new systematic biases in GAM data.
• To develop and evaluate effective normalization methods for GAM data.
• To provide a computational tool for bias correction in GAM analysis.

Main Methods:

• Detection and analysis of fragment length bias in GAM data.
• Development of the normGAM R package implementing five normalization methods: Knight-Ruiz 2-norm (KR2), normalized linkage disequilibrium (NLD), vanilla coverage (VC), sequential component normalization (SCN), and iterative correction and eigenvector decomposition (ICE).
• Evaluation of normalization methods' performance in removing biases related to fragment length, window detection frequency, mappability, and GC content.

Main Results:

• A novel fragment length bias was detected in GAM data, distinct from previously identified biases.
• The existing normalization method in GAM was insufficient for removing fragment length bias.
• The normGAM package effectively eliminates four key biases in GAM data.
• Vanilla coverage (VC) and Knight-Ruiz 2-norm (KR2) demonstrated superior performance in bias removal.

Conclusions:

• The developed normGAM package and its normalization methods successfully correct systematic biases in GAM data.
• KR2-normalized GAM data show higher correlation with KR-normalized Hi-C data, indicating improved cross-technique consistency.
• Normalized GAM data exhibit better agreement with fluorescence in situ hybridization (FISH) experimental results compared to raw GAM data.