Features that define the best ChIP-seq peak calling algorithms.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Chromatin immunoprecipitation followed by sequencing (ChIP-seq) is crucial for studying gene regulation.
• Peak calling is a critical initial step in analyzing ChIP-seq data, involving candidate identification and statistical significance testing.

Purpose of the Study:

• To survey and evaluate 30 peak calling methods for ChIP-seq data.
• To identify key features that differentiate method performance.
• To provide guidance on selecting optimal peak callers for specific applications.

Main Methods:

• Surveyed 30 peak calling methods, identifying 12 distinguishing features.
• Selected six representative methods (GEM, MACS2, MUSIC, BCP, TM, ZINBA) spanning the feature space.
• Utilized 300 simulated and 3 real ChIP-seq datasets, alongside mathematical analyses.

Main Results:

• Methods not explicitly combining ChIP and input signals showed higher power.
• Variable window size methods outperformed fixed window size methods.
• Poisson tests were more powerful than Binomial tests for ranking candidate peaks.
• BCP and MACS2 demonstrated superior performance on simulated transcription factor binding data.
• GEM identified motifs in 50% of top peaks within 10 bp.
• BCP and MUSIC performed best on histone modification data.

Conclusions:

• Methodological choices significantly impact ChIP-seq peak calling accuracy.
• Specific methods like BCP, MACS2, and MUSIC are recommended for transcription factor and histone data, respectively.
• Findings offer a rationale for selecting appropriate peak callers based on data type and desired outcomes.