CROCS

CROCS implements unconstrained multiple changepoint segmentation and alternative noise modeling to improve peak detection in ChIP-seq data with over-dispersed count distributions for epigenetic analysis of histone modifications such as H3K36me3 and H3K4me3.

Key Features:

• Unconstrained multiple changepoint detection: Employs a changepoint detection model that does not assume a Poisson noise distribution to flexibly identify peaks in ChIP-seq count data.
• Alternative noise assumptions: Incorporates noise models that account for over-dispersion commonly observed in histone modification count data.
• Supervised penalty optimization: Uses supervised learning to optimize the segmentation penalty parameter for improved peak prediction accuracy.
• Implementation: Implemented as an R package.

Scientific Applications:

• Histone modification analysis: Improves peak prediction tracks for epigenetic studies of histone marks such as H3K36me3 and H3K4me3.

Methodology:

Applies unconstrained multiple changepoint detection with alternative noise models and supervised learning to optimize penalty parameters, and compares performance against traditional algorithms on seven reference histone-modification datasets.