About

The analysis of ChIP-seq samples outputs a number of enriched regions (commonly known as "peaks"), each indicating a protein-DNA interaction or a specific chromatin modification.

When replicate samples are analyzed, overlapping peaks are expected. This repeated evidence can therefore be used to locally lower the minimum significance required to accept a peak. MSPC is a method for joint analysis of peaks.

Given a set of peaks from (biological or technical) replicates, MSPC combines the p-values of overlapping enriched regions, and allows the "rescue" of weak peaks occurring in more than one replicate.

Remarks

MSPC rigorously combines the statistical significance of the overlapping enriched regions (ER), in order to "rescue" weak peaks, which would probably be discarded in a single-sample analysis, when their combined evidence across multiple samples is sufficiently strong. [Ref]

MSPC takes as input, for each replicate, a list of ERs and a measure of their individual significance in terms of a p-value. Starting from a permissive call, it divides the initial ERs in "stringent" (highly significant) and "weak" (moderately significant), and it assesses the presence of overlapping enriched regions across multiple replicates. Non-overlapping regions can be penalized or discarded according to specific needs.

The significance of the overlapping regions is rigorously combined with the Fisher's method to obtain a global score. This score is assessed against an adjustable threshold on the combined evidence, and peaks in each replicate are either confirmed or discarded.

Finally, in order to account for multiple testing correction, MSPC applies the Benjamini-Hochberg procedure, and outputs ERs with false-discovery rate smaller than an adjustable threshold.

This flow is captured in the following flowchart for each ER of each replicate:

(This flowchart is a simplified version of the flowchart available in MSPC's manuscript.)

In other words:

• for each replicate, MSPC classifies input ERs as stringent, weak, and background, based on their p-value;
• performs a comparative analysis, and based on the combined stringency test, it classifies stringent and weak ERs as confirmed or discarded;
• based on false-discovery rate, it classifies confirmed ERs as true-positive or false-positive.

The following figure is a schematic view of this procedure.

(The number of ERs in different sets reported in this figure, are the result of wgEncodeSydhTfbsK562CmycStdAlnRep1 and wgEncodeSydhTfbsK562CmycStdAlnRep2 (available for download from Sample Data page) comparative analysis using -r bio -w 1e-4 -s 1e-8 parameters.)