HMM

Question 1

Expectation Maximisation

Answer 1

Make initial guess at transition and emission probabilities
Use these to estimate posterior probabilities (Expectation)
Calculate new transition and emission probabilities given state posterior probabilities calculated (Maximisation)

2 and 3 repeated until satisfied with estimation

Question 2

Pseudocounts

Answer 2

HMM may want to allow transitions which we don’t see much in the training data

Question 3

HMM structure choice (too few states)

Answer 3

Fail to model significant differences
Miss details
Underfit

Question 4

HMM structure choice (Too many states)

Answer 4

Model noise
Poor generalisation
Overfit

Question 5

Primary protein complex

Answer 5

Amino acid sequence

Question 6

Secondary protein structure

Answer 6

alpha helices

beta sheets/coil

Question 7

Tertiary protein structure

Answer 7

3d structure of polypeptide

Question 8

Quaternary protein structure

Answer 8

Protein complex

Question 9

Difficulties of using amino acid to generate the secondary structure

Answer 9

Non local
Different lengths of input
Representation

Question 10

Benefits of using amino acid sequence to generate the secondary structure

Answer 10

Now good training data

Can use windows to get fixed input