Methodology article

Strengths and weaknesses of EST-based prediction of tissue-specific alternative splicing

Shobhit Gupta¹ , Dorothea Zink² , Bernhard Korn² , Martin Vingron¹ and Stefan A Haas¹

¹  Computational Molecular Biology, Max Planck Institute for Molecular Genetics, Ihnestr. 73, D-14195 Berlin – Germany

²  German Resource Center for Genome Research, INF 580, 69120 Heidelberg – Germany

author email corresponding author email

BMC Genomics 2004, 5:72doi:10.1186/1471-2164-5-72

Published:	28 September 2004

Abstract

Background

Alternative splicing contributes significantly to the complexity of the human transcriptome and proteome. Computational prediction of alternative splice isoforms are usually based on EST sequences that also allow to approximate the expression pattern of the related transcripts. However, the limited number of tissues represented in the EST data as well as the different cDNA construction protocols may influence the predictive capacity of ESTs to unravel tissue-specifically expressed transcripts.

Methods

We predict tissue and tumor specific splice isoforms based on the genomic mapping (SpliceNest) of the EST consensus sequences and library annotation provided in the GeneNest database. We further ascertain the potentially rare tissue specific transcripts as the ones represented only by ESTs derived from normalized libraries. A subset of the predicted tissue and tumor specific isoforms are then validated via RT-PCR experiments over a spectrum of 40 tissue types.

Results

Our strategy revealed 427 genes with at least one tissue specific transcript as well as 1120 genes showing tumor specific isoforms. While our experimental evaluation of computationally predicted tissue-specific isoforms revealed a high success rate in confirming the expression of these isoforms in the respective tissue, the strategy frequently failed to detect the expected restricted expression pattern. The analysis of putative lowly expressed transcripts using normalized cDNA libraries suggests that our ability to detect tissue-specific isoforms strongly depends on the expression level of the respective transcript as well as on the sensitivity of the experimental methods. Especially splice isoforms predicted to be disease-specific tend to represent transcripts that are expressed in a set of healthy tissues rather than novel isoforms.

Conclusions

We propose to combine the computational prediction of alternative splice isoforms with experimental validation for efficient delineation of an accurate set of tissue-specific transcripts.