Research article

Predictive modeling of plant messenger RNA polyadenylation sites

Guoli Ji¹ , Jianti Zheng¹ , Yingjia Shen² , Xiaohui Wu¹ , Ronghan Jiang¹ , Yun Lin¹ , Johnny C Loke^2,4 , Kimberly M Davis² , Greg J Reese³ and Qingshun Quinn Li²

¹Department of Automation, Xiamen University, Xiamen, Fujian, 361005, P. R. China

²Department of Botany, Miami University, Oxford, OH 45056, USA

³Research Computing Group, IT Services, Miami University, Oxford, OH 45056, USA

⁴Current address: Department of Medicine, Division of Liver Diseases, Mount Sinai Medical Center, 1425 Madison Avenue, RM 1176, New York, NY 10029, USA

author email corresponding author email

BMC Bioinformatics 2007, 8:43doi:10.1186/1471-2105-8-43

Published:	7 February 2007

Abstract

Background

One of the essential processing events during pre-mRNA maturation is the post-transcriptional addition of a polyadenine [poly(A)] tail. The 3'-end poly(A) track protects mRNA from unregulated degradation, and indicates the integrity of mRNA through recognition by mRNA export and translation machinery. The position of a poly(A) site is predetermined by signals in the pre-mRNA sequence that are recognized by a complex of polyadenylation factors. These signals are generally tri-part sequence patterns around the cleavage site that serves as the future poly(A) site. In plants, there is little sequence conservation among these signal elements, which makes it difficult to develop an accurate algorithm to predict the poly(A) site of a given gene. We attempted to solve this problem.

Results

Based on our current working model and the profile of nucleotide sequence distribution of the poly(A) signals and around poly(A) sites in Arabidopsis, we have devised a Generalized Hidden Markov Model based algorithm to predict potential poly(A) sites. The high specificity and sensitivity of the algorithm were demonstrated by testing several datasets, and at the best combinations, both reach 97%. The accuracy of the program, called poly(A) site sleuth or PASS, has been demonstrated by the prediction of many validated poly(A) sites. PASS also predicted the changes of poly(A) site efficiency in poly(A) signal mutants that were constructed and characterized by traditional genetic experiments. The efficacy of PASS was demonstrated by predicting poly(A) sites within long genomic sequences.

Conclusion

Based on the features of plant poly(A) signals, a computational model was built to effectively predict the poly(A) sites in Arabidopsis genes. The algorithm will be useful in gene annotation because a poly(A) site signifies the end of the transcript. This algorithm can also be used to predict alternative poly(A) sites in known genes, and will be useful in the design of transgenes for crop genetic engineering by predicting and eliminating undesirable poly(A) sites.