Related Experiment Videos
Stochastic motif extraction using hidden Markov model
Y Fujiwara1, M Asogawa, A Konagaya
1Massively Parallel Systems NEC Laboratory, RWCP, Kanagawa, Japan.
Proceedings. International Conference on Intelligent Systems for Molecular Biology
|January 1, 1994
Summary
Hidden Markov models (HMMs) effectively represent protein sequence motifs, achieving 79.3% prediction accuracy for leucine zippers. This stochastic motif approach enhances protein sequence analysis and database validation.
Area of Science:
- Bioinformatics
- Computational Biology
- Structural Bioinformatics
Background:
- Protein motifs are crucial for function and structure.
- Stochastic motifs capture inherent variability in biological sequences.
- Hidden Markov Models (HMMs) offer a probabilistic framework for sequence modeling.
Purpose of the Study:
- To apply HMMs for representing protein sequences as stochastic motifs.
- To develop an effective method for learning optimal HMM topology.
- To evaluate the performance of HMMs in predicting protein motifs and validating databases.
Main Methods:
- Developed the "iterative duplication method" for HMM topology learning.
- Started with a small network, iteratively refining topology and parameters.
- Trained HMMs on specific protein motifs like leucine zippers and zinc fingers.
Main Results:
- Achieved 79.3% prediction accuracy for leucine zipper motifs using HMMs, significantly outperforming symbolic patterns (14.8%).
- Demonstrated HMM applicability to various zinc finger motifs and potential for separating mixed sequence data.
- Validated HMMs for protein database annotation, identifying an outlier leucine-zipper-like sequence.
Conclusions:
- HMMs provide a powerful and accurate method for representing and predicting protein motifs.
- The iterative duplication method is effective for learning discriminative HMM topologies.
- This approach enhances protein sequence analysis, motif discovery, and database curation.