Theoretical characterisation of strand cross-correlation in ChIP-seq

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Strand cross-correlation profiles are crucial for pre-analysis and quality control (QC) in ChIP-seq.
• Current understanding of what quality aspects these profiles measure remains limited, despite their peak-calling-independent potential for signal-to-noise ratio (S/N) assessment.

Purpose of the Study:

• To theoretically model and elucidate the factors influencing strand cross-correlation coefficients in ChIP-seq.
• To develop a novel, peak-call-free metric for assessing S/N and improving QC in ChIP-seq data.
• To provide a framework for evaluating peak-calling performance and estimating detectable peaks.

Main Methods:

• Developed a simulation model for ChIP-seq read density to derive theoretical cross-correlation coefficients.
• Evaluated theoretical models using 790 public ChIP-seq datasets.
• Introduced mappability-bias correction to enhance sensitivity.
• Proposed and implemented the virtual S/N (VSN) metric and the PyMaSC tool.

Main Results:

• The maximum cross-correlation coefficient is proportional to total mapped reads and the square of signal read ratio, inversely proportional to peak number and region length.
• Simulation and real data analyses showed high consistency with theoretical predictions.
• Virtual S/N (VSN) demonstrated consistent S/N estimation across various ChIP targets and sequencing depths.
• Mappability-bias correction improved sensitivity and differentiation from noise.

Conclusions:

• This study provides the first theoretical insights into strand cross-correlation in ChIP-seq, defining its potential and limitations.
• The proposed VSN metric offers a robust, peak-call-independent QC framework for ChIP-seq data.
• The VSN metric aids in evaluating peak-calling results and estimating the number of detectable peaks for future experiments.