Maximum likelihood and conditional maximum likelihood learning algorithms for hidden Markov models with labeled data: application to transmembrane protein topology prediction

  • Authors:

  • P. G. Bagos;TH. D. Liakopoulos;S. J. Hamodrakas

  • Affiliations:

  • Department of Cell Biology and Biophysics, Faculty of Biology, University of Athens, Panepistimiopolis, Athens 15701, Greece;Department of Cell Biology and Biophysics, Faculty of Biology, University of Athens, Panepistimiopolis, Athens 15701, Greece;Department of Cell Biology and Biophysics, Faculty of Biology, University of Athens, Panepistimiopolis, Athens 15701, Greece

  • Venue:
  • ICCMSE '03 Proceedings of the international conference on Computational methods in sciences and engineering
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

Hidden Markov Models (HMMs) have been widely used in applications in computational biology, during the last few years. In this paper we are reviewing the main algorithms proposed in the literature for training and decoding a HMM with labeled sequences, in the context of the topology prediction of bacterial integral membrane proteins. We evaluate the Maximum Likelihood algorithms traditionally used for the training of a Hidden Markov Model, against the less commonly used Conditional Maximum Likelihood-based algorithms and, after combining results previously obtained in the literature, we propose a new variant for Maximum Likelihood training. We compare the convergence rates of each algorithm showing the advantages and disadvantages of each method in the context of the problem at hand. Finally, we evaluate the predictive performance of each approach, using state of the art algorithms proposed for Hidden Markov Model decoding and mention the appropriateness of each one.