CART Classification of Human 5′ UTR Sequences

  1. Ramana V. Davuluri1,
  2. Yutaka Suzuki2,
  3. Sumio Sugano2, and
  4. Michael Q. Zhang1,3
  1. 1Cold Spring Harbor Laboratory, Cold Spring Harbor, New York 11724, USA; 2Department of Virology, Institute of Medical Sciences, University of Tokyo, Tokyo 108-8639, Japan

Abstract

A nonredundant database of 2312 full-length human 5′-untranslated regions (UTRs) was carefully prepared using state-of-the-art experimental and computational technologies. A comprehensive computational analysis of this data was conducted for characterizing the 5′ UTR features. Classification and regression tree (CART) analysis was used to classify the data into three distinct classes. Class I consists of mRNAs that are believed to be poorly translated with long 5′ UTRs filled with potential inhibitory features. Class II consists of terminal oligopyrimidine tract (TOP) mRNAs that are regulated in a growth-dependent manner, and class III consists of mRNAs with favorable 5′ UTR features that may help efficient translation. The most accurate tree we found has 92.5% classification accuracy as estimated by cross validation. The classification model included the presence of TOP, a secondary structure, 5′ UTR length, and the presence of upstream AUGs (uAUGs) as the most relevant variables. The present classification and characterization of the 5′ UTRs provide precious information for better understanding the translational regulation of human mRNAs. Furthermore, this database and classification can help people build better computational models for predicting the 5′-terminal exon and separating the 5′ UTR from the coding region.

Footnotes

  • 3 Corresponding author.

  • E-MAIL mzhang{at}cshl.org; FAX (516) 367-8461.

  • Article published online before print: Genome Res.,10.1101/gr.146000.

  • Article and publication are at www.genome.org/cgi/doi/10.1101/gr.146000.

    • Received May 2, 2000.
    • Accepted August 9, 2000.
| Table of Contents

Preprint Server