Segment boundary scale sensitivity, 4-track segmentation

To test the sensitivity of segmentation boundaries to wavelet scale, we use the notion that generally there are fewer, bigger segments as the wavelet scale increases. We therefore compared the published 64kb scale segmentation with a 128kb segmentation and a 32kb segmentation. In each case, we looked at the segment boundaries of the coarser segmentation and computed the distance to its nearest neighbor in the finer segmentation.

64kb vs. 32kb

There are 115 segment boundaries in the 64kb segmentation.

• The sorted distances to nearest neighbors in the 32kb segmentation are:

      0      0      0      0      0      0      0      0      0      0
      0      0      0      0      0      0      0      0      0      0
      0      0      0      0      0      0      0      0      0      0
      0      0      0      0      0      0      0      0      0      0
      0      0      0      0      0   1000   2000   2000   3000   3000
   3000   4000   4000   4000   4000   4000   5000   5000   5000   5000
   5000   6000   6000   6000   6000   7000   7000   7000   7000   8000
   8000   8000   9000   9000   9000   9000  10000  10000  10000  10000
  11000  11000  11000  11000  11000  11000  12000  12000  12000  12000
  12000  12000  12000  12000  13000  14000  14000  14000  15000  15000
  16000  16000  17000  18000  18000  19000  22000  22000  23000  27000
  27000  33000  40000  84000 110000
  

• Median distance = 5000bp
• Mean distance = 8261bp
• The two largest distances, 110kb and 84kb, correspond to a single segment in ENm010, a region in which there are actually more segments in the coarser 64kb segmentation than in the finer segmentation, breaking the general rule.

  bdry.64kb[which(dist == 110000),]
  ## 171 chr7 26912869
  
  bdry.64kb[which(dist == 84000),]
  ## 21 chr7 26886869
  

64kb vs. 128kb

There are 86 segment boundaries in the 128kb segmentation.

• The sorted distances to nearest neighbors in the 64kb segmentation are:

      0     0     0     0     0     0     0     0     0     0     0     0
      0     0     0     0     0     0     0     0     0     0     0     0
      0     0     0     0     0     0     0     0     0     0     0     0
      0     0     0     0     0     0     0  1000  1000  1000  2000  2000
   4000  5000  5000  6000  6000  6000  6000  7000  8000  8000  9000  9000
  11000 11000 12000 12000 12000 12000 13000 13000 16000 17000 17000 18000
  18000 25000 25000 26000 27000 27000 27000 27000 28000 29000 31000 34000
  42000 43000
  

• Median distance = 500bp (tie between 0 and 1000).
• Mean distance = 7663bp