Skip to main content
. Author manuscript; available in PMC: 2018 Sep 1.
Published in final edited form as: IEEE/ACM Trans Comput Biol Bioinform. 2016 Jan 26;14(5):1070–1081. doi: 10.1109/TCBB.2016.2520919

Fig. 2. Read coverage features are not linearly separable.

Fig. 2

Points colored green represent training examples labeled active and those colored red indicate training examples labeled inactive. The blue shading provides a contour plot of the active state probability given the feature’s average read coverage (x3, y-axis) and the gap length between adjacent contigs (x1, x-axis in log nucleotides). (A) uses logistic regression with a linear kernel function (i.e. d = 1 in equation 3), whereas (B) uses a second-order polynomial kernel function (i.e. d = 2 in equation 3).