Feature extraction for DNA base-calling using NNLS

We present new algorithms that can be used to extract features from a DNA chromatogram prior to base-calling. The algorithms assume that the inter-base distance has already been equalized using methods such as those presented in L. Andrade-Cetto and E. Manolakos (2005). We show first how a good estimate of the peak diffusion (spread) can be calculated from the raw trace and without having to known the underlying base sequence. Using the estimated inter-peak distance and peak spread parameters a non-negative least squares problem can be formulated in order to find the weight factors of the multiple shapes immersed in broad peaks, typically found towards the end of the chromatogram. The two algorithms combined provide peak hypotheses that are tested by the subsequent base decisions and scoring stage of the base-caller using probabilistic methods