Figure 2

34-feature Bayesian network. Note that BN in fact has 35 nodes. The class node, which has an edge to all other nodes and makes the actual number of edges 67, is omitted for ease of visualization. Thus, this is just the augmenting tree in the TAN classifier. The features associated with the nodes are as follows: 1: 1 if exon length is divisible by 3, otherwise 0. 2: Length of the best alignment in the 3' 100 nt intronic region. 3: Length of the best alignment in the 5' 100 nt intronic region.4: Percent identity of the best alignment in the 5' 100 nt intronic region. 5: Length of the 5' intron. 6: Ratio of the lengths of the 3' intron and the exon. 7: Product of the identities of the exon and both 100-nt intronic flanks with their mouse orthologs. 8: 1 if G at +4 of the 5'ss, otherwise 0. 9: T at +4, 10: A at +6; 11: MAXENTSCAN score of the 5'ss. 12: Sum of the MAXENTSCAN scores of the 3' and 5'ss. 13: Average of the NI scores of all the hexamers with a negative NI score. 14: Variance of the NI scores of all the hexamers with a "strong" (≥ 0.8 or ≤ -0.8) score. 15: Average of the NI scores of all the hexamers with a "strong" (≤ -0.8) negative score. 16: Density of single-stranded (PU value ≥ 0.6), "trusted" ESEs (NI score = 1). 17: Ratio of the number of "trusted" ESEs (NI score = 1) to the number of ESSs (NI score = -1). 18: Density of ISREs enriched in the flanks of AS exons, in the 5'intron flank. 19: Density of single-stranded (PU value ≥ 0.6), intronic splice regulatory elements (ISREs) enriched in the flanks of AS exons, in the 5'intron flank. 20: PTB-binding site TCTT density in the exon. Dimer density in the exon:21:CC, 22: GA; 23: Dimer GA density in the 3' intron flank; Trimer density in the exon: 24: AAG, 25: AGG, 26: ATG, 27: CAA, 28:CCA, 29: CGG, 30: CTC, 31: GCA, 32: GGT, 33: TAG, 34: TCC.