ChIP-seq analysis

In comparison to RNA-seq analysis, ChIP-seq analysis can be a bit more complicated. The experimental protocol itself is more complex, and has additional variability imposed by antibody effectiveness.

Fundamentally, ChIP-seq is about finding where on chromatin a protein of interest binds. To do this, we generally convert aligned reads into signal over the genome, and wherever there is a large pileup of reads, we call that as a “peak” and infer that the protein was found there.

This aspect of peak-calling can also be used with other chromatin-related assays like ATAC-seq and CUT&RUN albeit with some minor modifications. In general, peak-calling is more of an art than a science. Multiple peak-calling algorithms will often give different results, so a big part of ChIP-seq analysis is understanding which algorithms are most appropriate for your particular experiment and making well-justified decisions about which peaks to use as the “final” set.

The best one-stop-shop for learning about ChIP-seq analysis is the Harvard Bioinformatics Core’s Introduction to ChIP-seq using high-performance computing starts with an overview of ChIP-seq, and continues through working with Linux, project organization, alignment, QC, peak-calling, and comparing peaks between conditions.

Nakato 2021 has a high-level overview of ChIP-seq, and may be more succinct than the HBC workshop linked above.

Technical details

The current ENCODE guidelines for ChIP-seq can be found here. Here are some highlights:

  • two or more biological replicates

  • for mammalian genomes, 20 million usable fragments for narrow peaks (a fragment is either one read for single-end sequencing, or one pair for paired-end sequencing)

  • for mammalian genomes, 45 million usable fragments for broad peaks

Input or H3 as a control? Flensburg 2014 compares the use of whole-cell extract (input) vs H3 as a control for histone modification ChIP-seq. They found that the called peaks are mostly similar no matter which control is used. As you might expect, when using H3 there is a depletion of signal around the TSS due to expected nucleosome depletion. While there may be corner cases where the particular biological questions may cause one to be more biologically meaningful, in general using input for histone modifications seems well-justified.