Finding Alternative Splicing Patterns with Strong Support From Expressed Sequences

摘要

In this paper, we consider the problem of finding the alternative splicing patterns from a set of expressed sequences (cDNAs and ESTs). An effective approach is to align the ESTs to the genomic sequence and concatenate compatible ESTs to form possible splicing patterns. ClusterMerge [4] is a popular software along this direction. Based on our large-scale experiment, ClusterMerge can reveal 81.0% of known splicing patterns, but includes many false positives (with specificity of 9.7%). We propose a simple model to measure the reliability of each EST and develop an algorithm to report those patterns in which each exon/intron is covered by enough reliable ESTs. Our approach is effective in eliminating false positives (with specificity of 36.7%) while maintaining a slightly higher sensitivity as Cluster-Merge (with sensitivity of 84.9%). We also find our approach can achieve a higher specificity with slightly better sensitivity than ECgene database on the same dataset.