Protein Classification using Hidden Markov Models and Randomised Decision Trees

Since the introduction of next generation sequencing there is a demand for sophisticated methods to classify proteins based on sequence data. Two main approaches for this task are to use the raw sequence data and align them against other sequences, or to extract discrete high level features from the protein sequences and compare the features. Two machine learning methods are demonstrated to show each approach. Profile Hidden Markov Models are built from multiple alignment of raw sequence data and learn amino acid emission and transition parameters for a given alignment and effectively harness the power of aligning a test protein to a model built form many proteins. Random Forests on the other hand are used to discriminate between two sets of proteins based on features such as functional amino acid groups and physiochemical properties extracted from the raw sequences. The strengths and limitations of each method are presented and discussed, focussing on the individual merits and how they could work possibly compliment each other rather than just being compared by their classification accuracy.