Table 4.
Gain in accuracy from different levels of modeling extra residual effects
Click on table to view larger version.
-
In BayesCall, extra residual effects are captured by the parameter αt. We used cycle-dependent parameters dt, σt2, Xt, and Σt in all three cases of BayesCall. Even without modeling extra residual effects (i.e., with αt = 0 for all t), BayesCall achieved an improvement of 38% over Bustard in the by-base error rate. When the full model with cycle-dependent αt was used, the improvement over Bustard increased to 55%.
