Skip to main content

Table 4 Re-scaling models improves prediction - Gamma linker length mode l

From: Predicting nucleosome positioning using a duration Hidden Markov Model

model update re-scaled total sensitivity (%) FDR(%) update total sensitivity (%) FDR(%)
1st 0 8746 (28) 60 (0.7) 31 (0.6) 0 10640 (19) 70 (0.3) 35 (0.3)
  1 9471 (42) 67 (0.7) 30 (0.6) 1 11513 (14) 60 (0.6) 48 (0.6)
  2 9787 (38) 68 (0.5) 30 (0.4) 2 11812 (18) 55 (0.4) 53 (0.3)
4th 0 8886 (18) 63 (0.3) 29 (0.4) 0 10461 (25) 73 (0.7) 30 (0.7)
  1 9533 (26) 70 (0.8) 27 (0.8) 1 11190 (32) 66 (0.7) 41 (0.7)
  2 9775 (33) 72 (0.8) 27 (0.9) 2 11443 (26) 63 (0.5) 45 (0.5)
  1. Total predictions, sensitivity, and false discovery rate (FDR) are the averages (standard deviations in parentheses) based on 10 repeated simulations. For each simulation a maize-like genomic sequence consisting of 10000 nucleosomes and 10001 linkers were simulated using the re-scaled 1st and 4th order yeast models. Each sequence was scanned using the true models (re-scaled, 1st or 4th order) and the yeast models with an initial uniform linker length distribution on 1,..., 200. The results after 0, 1, 2 updates of linker length distribution are compared.