Figure 2
From: Adaptive bandwidth kernel density estimation for next-generation sequencing data
Effects of different signal-reconstruction approaches. (A) From top to bottom: fixed-bandwidth Gaussian estimators for a portion of the Rad21 data using h = 16 (close to the h = 15 that is default in the QuEST software) and h = 362 (which appears optimal based on evaluation of held-out tuning data), adaptive-bandwidth Gaussian estimators using k = 7 (optimal based on held-out tuning data) and k = 14 (double the optimal choice, and intended to obtain more aggressive smoothing). In each plot, the curve shows the reconstructed density. The short black vertical marks indicate the read positions. (B) Histograms of heights of local maxima for fixed-bandwidth Gaussian (h = 362) and adaptive-bandwidth Gaussian (k = 7). Note that the vertical axes are in log scale.