Fig. 1

Posterior log-probability normalization. The left panel shows posterior log-probabilities for 38 781 sequences. The sequences are random sub-sequences of the contax.trim data set, spanning all lengths from 100 bases to more than 1500. Every sequence has been classified using the multinomial model trained on the full-length data, and each dot marks the maximum posterior log-probability for one sequence. There is clearly a linear trend in the values, with larger variance for longer sequences. In the right panel the same values are plotted after the normalization procedure described in the text