Correction of sequence-based artifacts in serial analysis of gene expression

Motivation: Serial Analysis of Gene Expression (SAGE) is a powerful technology for measuring global gene expression, through rapid generation of large numbers of transcript tags. Beyond their intrinsic value in differential gene expression analysis, SAGE tag collections afford abundant information on the size and shape of the sample transcriptome and can accelerate novel gene discovery. These latter SAGE applications are facilitated by the enhanced method of Long SAGE. A characteristic of sequencing-based methods, such as SAGE and Long SAGE is the unavoidable occurrence of artifact sequences resulting from sequencing errors. By virtue of their low-random incidence, such tag errors have minimal impact on differential expression analysis. However, to fully exploit the value of large SAGE tag datasets, it is desirable to account for and correct tag artifacts. Results: We present estimates for occurrences of tag errors, and an efficient error correction algorithm. Error rate estimates are based on a stochastic model that includes the Polymerase chain reaction and sequencing error contributions. The correction algorithm, SAGEScreen, is a multi-step procedure that addresses ditag processing, estimation of empirical error rates from highly abundant tags, grouping of similar-sequence tags and statistical testing of observed counts. We apply SAGEScreen to Long SAGE libraries and compare error rates for several processing scenarios. Results with simulated tag collections indicate that SAGEScreen corrects 78% of recoverable tag errors and reduces the occurrences of singleton tags.