Fig. 4

Maximum-likelihood alignment parameters derived using expectation-maximization (EM). The process starts with four guide alignments, each generated with a different mapper using tuned parameters. Squares denote error estimates derived from different mappers when used without tuning; circles denote error estimates post-tuning; and triangles denote error estimates post-EM. a Insertion versus deletion rates, expressed as events per aligned base. b Indel events per aligned base versus rate of mismatch per aligned base. Rates varied strongly between different guide alignments; but EM training and realignment resulted in very similar rates (gray shading in circles), regardless of the initial guide alignment. c The matrix for substitution emissions determined using EM reveals very low rates of A-to-T and T-to-A substitutions. The color scheme is fitted on a log scale, and the substitution values are on an absolute scale. (Figure reproduced from Jain et al. [9])