Using the Fisher Kernel Method to Detect Remote Protein Homologies
Proceedings of the Seventh International Conference on Intelligent Systems for Molecular Biology
Graphical Models of Residue Coupling in Protein Families
IEEE/ACM Transactions on Computational Biology and Bioinformatics (TCBB)
Segmentation conditional random fields (SCRFs): a new approach for protein fold recognition
RECOMB'05 Proceedings of the 9th Annual international conference on Research in Computational Molecular Biology
Hi-index | 0.00 |
One of the most widely-used methods to date for recognizing protein sequences that are evolutionarily related, termed homologs, has been profile hidden Markov models. For the cases where positive training data for these methods is sparse, Kumar and Cowen in 2009 introduced the paradigm of simulated evolution, a randomized algorithm to construct additional artificial training sequences, which are generated based on a highly simplistic model of how protein sequences evolve. These artificial sequences are then used together with the true positive training sequences to learn the profile hidden Markov model. Kumar and Cowen showed that augmenting the training set with a simple pointwise model of simulated evolution improved the detection of remote homologs for profile hidden Markov models. In 2010, they then constructed a model of simulated evolution that captures the pairwise statistical preferences of residues that are hydrogen bonded in beta-sheets, and showed that it improved the ability of hidden Markov models to recognize remote homologs for beta-structural motifs. In this work, we explore how best to extend the paradigm of simulated evolution to alpha-helical motifs. We determine that simulated evolution can also improve the performance for profile hidden Markov models on detecting remote homologs of alpha-structural proteins.