PeakSeg: Peak detection via constrained optimal Segmentation

This repos contains the code for the long version of https://github.com/tdhock/PeakSegDP-NIPS – the main differences are

• more discussion of constrained DP and penalty learning.
• comparison with unsupervised AIC/BIC/mBIC and oracle penalty.
• L1-regularized 41-parameter model.

To make sure to use the same package versions, please install the works_with_R function by copying the code in works_with.R to your ~/.Rprofile which will load it at the beginning of every R session.

Then type “make” to run the code. There are two big steps:

1. Download the ChIP-seq benchmark data set from https://rcdata.nau.edu/genomic-ml/chip-seq-chunk-db
2. Run the https://github.com/tdhock/PeakSegDP algorithm on each profile in the benchmark.

On my computer it took about 1 week (using one 1.6GHz CPU) to run the PeakSegDP algorithm on all the profiles in the benchmark.