Figure 6
From: CONTRAST: a discriminative, phylogeny-free approach to multiple informant de novogene prediction
Features that score coding region boundaries. CONTRAST contains two types of feature for scoring coding region boundaries. The first, shown in red, maps the output of a classifier to a score using a piecewise linear function learned during CRF training. In this example, the score from the GT splice donor classifier falls between the fourth and fifth control points for the function, with interpolation coefficients of 0.312 and 0.688. The second type of feature, shown in green, scores a coding region boundary based on the EST sequence characters that flank it.