Calculation of s(Cseg)

The per-nucleotide read coverage data is used to evaluate this score. To evaluate the probability that the contig originates from a single transcript (i.e., it is not chimeric), a Bayesian segmentation analysis of the per-nucleotide coverage depth is performed. For a correctly assembled contig, it is assumed that the distribution of per-nucleotide coverage values in that contig is best described by a single Dirichlet distribution, i.e., all nucleotides in the same transcript should have the same expression level, and thus should be best modeled as a stochastic sample from a single distribution. In contrast, a contig that is a chimera derived from concatenation of two or more transcripts will have per-nucleotide coverage values that are best described by two or more different Dirichlet distributions. The probability that the distribution of per-nucleotide read coverage values comes from a single Dirichlet distribution is evaluated using a Bayesian segmentation algorithm previously developed for analysis of changes in nucleotide composition (Liu and Lawrence 1999). To facilitate the use of this method, the per-nucleotide coverage along the contig is encoded as a sequence of symbols in an unordered alphabet by taking log₂ of the read depth rounded to the nearest integer. As the probability will be a value between 0 and 1, this probability is used directly as s(C_seg).