Skip to main content
. 2008 Jul 1;24(13):i147–i155. doi: 10.1093/bioinformatics/btn152

Fig. 4.

Fig. 4.

Classification accuracy of MS patients’ response to IFNβ, based on 5 random permutations and 4-fold cross validation (20 different training-testing splits). For the loop HMM, 2 and 3 time points use 1 state, 4 to 6 time points use 2 states, and 7 time points use 4 states. For the jump HMM, 3, 4, 7, 8, 10, 12 states are used from 2 to 7 time points. For both loop and jump HMM, 10 random initialization were carried out and the initialization resulting in the highest likelihood was selected.