Alternative splice detection and classification. (a) Splice inference. Coding regions of RefSeq mRNAs were aligned to the genomic sequence to determine canonical splicing patterns. EST alignments to the genomic sequence confirmed the canonical splices and indicated alternative splices. Canonical (RefSeq) splices are indicated above the exons, whereas alternative splices are indicated below the exons. When an alternative splice introduced a stop codon >50 nucleotides upstream of the final exon–exon splice junction of an inferred mRNA isoform, the stop codon was classified as a premature termination codon and the corresponding mRNA isoform was labeled a NMD candidate. In the NMD-candidate example shown, an exon skip caused a frameshift, resulting in the introduction of a premature termination codon. Restricting the analysis to coding regions assured high alignment quality, but this excluded alternative splicing in noncoding regions, such as that which occurs with splicing factor SC35. Intron retentions were also excluded because ESTs indicating intron retention are indistinguishable from incompletely processed transcripts, a common dbEST contaminant. (b) Splice mode classification. Alternative splices were categorized according to splice site usage and effects on the coding sequence. “Splice sites introduced” shows the number of splice donor/acceptor sites that were observed in the alternative splice but were not included in the canonical splice. “Splice sites lost” shows the number of splice donor/acceptor sites that were included in the canonical splice but were absent in the alternative splice. “Coding region change” indicates whether an alternative splice added (red) or subtracted (green) coding sequence to the alternative isoform relative to the canonical isoform. By our method, mutually exclusive exon usage appears as exon inclusion. Our analysis excluded intron retentions, which would be classified as zero splice sites introduced, two sites lost, and addition of coding sequence. (c) Alternative isoform inference from splice pairs. Splice pairs are splice donor/acceptor sites (▴) inferred from the alignments. Alternative splice pairs are those indicated by ESTs, but not by a RefSeq mRNA. The exon composition of an isoform was determined from EST-demonstrated splice pairs, which may be covered by multiple ESTs. Coverage of splice pairs is indicated in each ▴. Coverage for a complete isoform is not meaningful because of the variability in coverage of its splice pairs. (d) Alternative splice pairs by mode and coverage. The total number of alternative splice pairs associated with each splicing mode is shown at various levels of EST coverage. The distance from the y axis to the right edge of each box corresponds to the total number of splice pairs with coverage greater than or equal to the number indicated. Note that each exon inclusion event involves two splice pairs. (e) Alternative splice pairs generating NMD candidates by mode and coverage. The panel shows the subset of alternative splice pairs that produce premature termination codons. These splice pairs are involved in generating NMD-candidate mRNA isoforms. The numbers of splice pairs are displayed as in d. Also shown are the NMD-candidate splice pairs at coverage ≥1 and ≥2 as a percentage of all alternative splice pairs for each splicing mode. (f) Isoforms of alternatively spliced RefSeq-coding genes. Shown are the total numbers of isoforms of the RefSeq-coding genes for which alternative isoforms were found. These are subdivided into the following categories: all isoforms including canonical, alternative isoforms (i.e., all isoforms excluding canonical), and NMD candidates.