Hidden Markov Models: Bio Prediction

Machine learning and Hidden Markov Models application to the transmembrane protein secondary structure prediction. Awarded as the best final-year Informatics project in Serbia.

The publication can be found here.

What are Hidden Markov Models?

A Markov Chain is a system that expresses transitions from one state to another, with certain probabilities.

A Hidden Markov Model (HMM) is a Markov Chain in which the state sequence is unobservable. What could be observed is a sequence of outputs. Every produced output at a given moment depends only on the current state.

Example of a Markov Chain	Example of a Hidden Markov Model

How can HMMs be applied to the protein structure prediction?

A protein structure could be modeled as a HMM, where every amino acid is an output, and its position a state.

Example of a protein and its position, relative to a cell membrane

What are the algorithms implemented in this project?

Among others, the Baum-Welch algorithm is implemented for the unknown parameters estimation, while the Viterbi algorithm is implemented for finding the most likely sequence of hidden states.