Hidden Markov Models in Bioinformatics

1
Hidden Markov Models in Bioinformatics

• A General Survey
• Project Proposal
• Colin Cherry (colinc_at_cs)

2
What are Hidden Markov Models?

• In a nutshell
• A statistical model based on the assumption that
  an observed sequence can be generated by a random
  walk through some series of hidden Markovian
  states

3
Huh?

• Assume that there is some hidden process that
  generates the observed sequence
• (Simple) Example
• Sequence ACCATC
• Assume model was trained on something of interest
• Fit to model 0.8 x 0.8 x 0.2 x 0.8 x 0.8 x 0.8
  0.065

4
So what?

• Things get exciting once we start allowing
  transitions to multiple states
• We can follow different paths through the model
• Established algorithms exist to
• Train these models on a set of observed sequences
• Use trained models to analyze and classify new
  sequences

5
Wheres the bioinformatics?

• These models are used throughout bioinformatics,
  but especially in
• Protein Family Characterization
• Gene Finding
• Many more details to come in the final project
  presentation

6
What Im Going to Do (Part 1)

• A tutorial on HMMs in bioinformatics, including
• A primer on HMM theory
• How to apply HMMs to solve biological problems
• Limitations of HMMs and how they can be overcome
• The general HMM techniques used for the two
  domains mentioned earlier will be discussed in
  depth as examples

7
What Im Going to Do (Part 2)

• A survey of the current state of the art
  techniques that make use of HMMs
• I will concentrate on those techniques and
  programs in the fields of Protein Classification
  and Gene Finding
• I will also briefly outline other applications
  where HMMs have been useful
• One state of the art program will be demoed in
  front of the class in the final presentation

8
Thats all for now!

• Any questions?