The variation between the splice variants identified by different algorithms implied that many unique peptides concluded by one algorithm were not necessarily recognized by another.
However, most of the removed hits belonged to the identifications reported by only one algorithm (see Additional file 2).
It is important to point out that, in contrast to class 2 terminators, the class 3 terminators were defined by only one algorithm.
Therefore, to increase the number of putative terminators of the 320 class 1 terminators we lowered the specificity of the class 2 decision rule as follows: class tc3 — class 1 intrinsic terminators validated by only one algorithm, maintaining rules i, iii and iv, described above; and class tc4 — class 1 intrinsic terminators validated by only one algorithm, maintaining rules i and iii or iv.
The distribution of somatic probability scores for sites unique to each caller and returned by multiple callers, as shown in Figure 1, indicates that filtering out sites returned by only one algorithm would remove sites regarded to have high somatic probability by one such measure.
To avoid over-prediction, genes were only accepted if they were predicted by at least two algorithms or, if they were predicted by one algorithm and were also similar to known ESTs, cDNAs, or proteins.
Of these, 152 genes were predicted by more than one algorithm [Additional file 3] and only 17 are predicted by all algorithms.
Other genes predicted by one algorithm were also notable; for example, ERBB3 (MuSiC) and GNAS and FBXW7 (OncodriveFM) (Table 2).
Beyond this core group, there were marked differences in the probability scores assigned to the same sites and in the characteristics of sites returned by one algorithm only.
Table 2 shows that the majority of the experimentally found binding sites are correctly predicted by at least one algorithm.
