Goal

• figure out the right easy peak calling

TODO

• [x] get the peak calls for sox2
• [ ] what fraction of the genome do they cover?
• [ ] write a random genome sampler (discard the read in case it overlaps some of the labels)
• [ ] make 2 histograms: number of counts per windows
  • randomly sampled regions
  • peak regions
• [ ] plot the same things for oct4

• compute the count statistics for contiguous bins (100bp) and plot the distribution
• compare the results (overlap) to the currently called peaks with smoothed signal

Next

• [ ] Setup the pipeline for learning from a single bigwig file:
  • pre-compute a set of high-count regions
  • find 'interesting' regions. Sample regions w.r.t. how interesting they are and update your model