JUCHMME

JUCHMME fits custom Hidden Markov Models with a discrete alphabet to analyze and model biological sequences.

Key Features:

• Discrete-alphabet HMMs: Supports fitting Hidden Markov Models defined over a discrete alphabet of symbols.
• Decoding algorithms: Implements Viterbi, N-Best, posterior-Viterbi, and Optimal Accuracy Posterior Decoder for sequence decoding.
• Model customization: Enables construction and evaluation of user-specified HMM topologies and parameterizations.
• Evaluation procedures: Provides independent test (self-consistency), jackknife, and k-fold cross-validation for model assessment.
• Reliability measures: Associates prediction algorithms with corresponding reliability measures.
• HMM extensions: Includes extensions such as Hidden Neural Networks, models that condition on previous observations, and semi-supervised learning methods.

Scientific Applications:

• Gene prediction: Application of custom HMMs to predict gene structures from biological sequence data.
• Protein structure analysis: Use of HMM-based models and decoders in analyses related to protein structure.
• Sequence modeling: Modeling and interpretation of labeled biological sequences in other sequence-analysis problems requiring detailed probabilistic models.

Methodology:

Fitting of custom discrete-alphabet Hidden Markov Models with standard decoding algorithms (Viterbi, N-Best, posterior-Viterbi, Optimal Accuracy Posterior Decoder), evaluation via independent test (self-consistency), jackknife, and k-fold cross-validation, inclusion of reliability measures, and extensions including Hidden Neural Networks, conditioning on previous observations, and semi-supervised learning.