PeakSegJoint: fast supervised peak detection via joint segmentation of multiple count data samples

TL;DR

PeakSegJoint is a supervised, fast, and interpretable peak detection method for multiple genomic samples, capable of handling any number of sample types with overlapping peaks, outperforming existing algorithms in speed and interpretability.

Contribution

It introduces a novel constrained maximum likelihood segmentation model for multiple samples and a supervised penalty learning approach for peak number selection.

Findings

Achieves similar accuracy to state-of-the-art methods

Operates faster than existing algorithms

Provides more interpretable overlapping peak models

Abstract

Joint peak detection is a central problem when comparing samples in genomic data analysis, but current algorithms for this task are unsupervised and limited to at most 2 sample types. We propose PeakSegJoint, a new constrained maximum likelihood segmentation model for any number of sample types. To select the number of peaks in the segmentation, we propose a supervised penalty learning model. To infer the parameters of these two models, we propose to use a discrete optimization heuristic for the segmentation, and convex optimization for the penalty learning. In comparisons with state-of-the-art peak detection algorithms, PeakSegJoint achieves similar accuracy, faster speeds, and a more interpretable model with overlapping peaks that occur in exactly the same positions across all samples.