search

kundajelab / gecco-variants

0 stars 0 forks source link

Histone-binding site labels #4

Open annashcherbina opened 6 years ago

annashcherbina commented 6 years ago

Some improvement in H3K27ac_V576 (Task #19) observed with new H3K27ac labeling approach:

image

Browser tracks for New H3K27ac labeling scheme for V576 sample. For the bed tracks: Green -- positive Blue -- ambiguous Red -- negative http://epigenomegateway.wustl.edu/browser/?genome=hg19&datahub=http://mitra.stanford.edu/kundaje/annashch/gecco/histone_focus/gecco.datahub.V576.histone_focus.json&tknamewidth=150

Next steps:

  1. Evaluate accuracy of histone identification by using V576 DNAse data as held-out gold standard
  2. Use ENCODE region summits (extended ~300 bases from summit) for sliding windows along H3K27ac peaks. Take the merge of the summits rather than of the full ENCODE DNAse peaks.
  3. Test new V576_H3K27ac model's ability to predict V576_DNAse labels (these should be learned indirectly if the "zeroing in on histone binding site" is working as expected.