Overview of the split-read strategy. Each read is split into two pieces, or "anchors," of equal length (red and blue), with a gap between them. The anchors are aligned independently, and only the instances in which both align uniquely to the reference sequence are considered. Then, the alignments are extended as long as they still match the reference sequence. The SplitSeek program identifies all candidate junction reads from the split-read alignments where the boundary is located in the gap between the anchors. Then additional junction reads are detected from the set of reads that partly align to a previously detected candidate junction, and where the remaining, nonaligned, part of the read (grey lines) has a 5-bp identical sequence compared with the corresponding part of the same candidate read. SplitSeek then groups all potential junction reads, applies cut-offs, and reports the results.

Number of predicted splice junctions (y-axis) as a function of the total number of processed reads (x-axis). The number of predicted junctions (black line) increases almost linearly with the number of reads. The green and orange lines represent two subgroups of predicted junctions: those where the two boundaries are separated by ≤100 kb, and those connecting two exon boundaries of a RefSeq gene. Predicted insertions and deletions are combined and represented by the red line.

Two long-range SplitSeek predictions (>100 kb) that extend known gene models. (a) A predicted junction that connects an exon in the Ensembl Gene Prediction database with the second exon of the Phactr3 gene, suggesting the presence of an alternative transcription start site. (b) A putative novel exon in the Sorcs2 gene that is currently only supported by EST data.

Comparison of predictions from RNA-MATE and SplitSeek. (a) Venn diagram showing the number of predicted junctions by the two methods. (b) Predicted number of junction reads for all for all 11,395 exon boundaries reported by both RNA-MATE (x-axis) and SplitSeek (y-axis).

SplitSeek results viewed in the UCSC genome browser. (a) Predicted splice junctions in the gene Fpgs. (b) The two grey boxes give a schematic view of how deletions and insertions are detected. The genome browser image below shows the SplitSeek results in the last exon and 3' UTR of the Nol10 gene on chromosome 12. Three events are predicted, a splice junction (to the left), a deletion (in the middle,) and an insertion (to the right). The predicted insertion and deletion are both supported by the mRNA AK148210, as indicated by the orange arrows at the bottom.