A Better Method for Length Distribution Modeling in HMMs and Its Application to Gene Finding

  • Broňa Brejová
  • Tomáš Vinař
Conference paper

DOI: 10.1007/3-540-45452-7_17

Part of the Lecture Notes in Computer Science book series (LNCS, volume 2373)
Cite this paper as:
Brejová B., Vinař T. (2002) A Better Method for Length Distribution Modeling in HMMs and Its Application to Gene Finding. In: Apostolico A., Takeda M. (eds) Combinatorial Pattern Matching. CPM 2002. Lecture Notes in Computer Science, vol 2373. Springer, Berlin, Heidelberg

Abstract

Hidden Markov models (HMMs) have proved to be a useful abstraction in modeling biological sequences. In some situations it is necessary to use generalized HMMs in order to model the length distributions of some sequence elements because basic HMMs force geometric-like distributions. In this paper we suggest the use of an arbitrary length distributions with geometric tails to model lengths of elements in biological sequences. We give an algorithm for annotation of a biological sequence in O(ndm2 Δ) time using such length distributions coupled with a suitable generalization of the HMM; here n is the length of the sequence, m is the number of states in the model, d is a parameter of the length distribution, and Δ is a small constant dependent on model topology (compared to previously proposed algorithms with O(n3m2) time [10]). Our techniques can be incorporated into current software tools based on HMMs.

To validate our approach, we demonstrate that many length distributions in gene finding can be accurately modeled with geometric-tail length distribution, keeping parameter d small.

Keywords

computational biology hidden Markov models gene finding length distribution 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer-Verlag Berlin Heidelberg 2002

Authors and Affiliations

  • Broňa Brejová
    • 1
  • Tomáš Vinař
    • 1
  1. 1.School of Computer ScienceUniversity of WaterlooCanada

Personalised recommendations