Structure Prediction

1
Chapter 9

• Structure Prediction

2
Motivation

• Given a protein, can you predict molecular
  structure
• Want to avoid repeated x-ray crystallography, but
  want accuracy
• You could use nucleotide alignment, but what do
  you do with the gapped regions?
• More complex methods are only justified if they
  can be shown to perform better than simpler
  methods
• Simpler methods are only justified if they can
  perform better than basic sequence alignment

3
First Step

• Some structure comparison methods use secondary
  structures of the new sequence
• Predict location of secondary structure elements
  along the proteins backbone and the degree of
  residue burial
• Supervised learning has been shown to perform
  well in this task

4
Artificial Neural Network

• Predicts
• Structure
• at this
• point

5
Danger

• You may train the network on your training set,
  but it may not generalize to other data
• Perhaps we should train several ANNs and then let
  them vote on the structure

6
Profile network from HeiDelberg

• family (alignment is used as input) instead of
  just the new sequence
• On the first level, a window of length 13 around
  the residue is used
• The window slides down the sequence, making a
  prediction for each residue
• The input includes the frequency of amino acids
  occurring in each position in the multiple
  alignment (In the example, there are 5 sequences
  in the multiple alignment)
• The second level takes these predictions from
  neural networks that are centered on neighboring
  proteins
• The third level does a jury selection

7
PHD

• Predicts 4

Predicts 5
Predicts 6
8
Threading

• Threading matches structure to sequence
• True threading considers 3D spatial interactions

9
3D-1D Matching (Bowie et al.)

• Convert 3D structure into a string
• Include ?-helix, ?-sheet or neither
• Include buried or solvent accessible (6 levels)
• Total of 3X618 distinct states
• With Paj probability of finding amino acid (a)
  in environment (j) and Paprobability of finding
  (a) anywhere

10
3D-1D

• Calculate the information values score on a
  training set of multiple alignments and the score
  was used as a profile for each column
• When applied to the globin family an clearly
  identified myoglobins from nonglobins but not
  from other globins

11
Methods using 3D interactions

• Residues that have large separation in the
  sequence may end up next to each other when the
  protein is folded.
• Define a measure of contact between residues (two
  atoms within 5Å) and count frequency of contact
  between all pairs in PDB
• Use measure in alignment to evaluate cost, or to
  select the best alignment

12
3D interactions
13
Potentials of mean force (POMF)

• Since the notion of contact is somewhat
  arbitrary, a more general formulation can be
  tried
• Derive an empirical function for the propensity
  of each of the 400 pairs of residues to be any
  given distance apart.

14
Multiple Sequence Threading

• Multiple Sequence Alignment
• Align the most similar to create a consensus
  sequence
• Align consensus sequences to create overall
  alignment
• Use the same strategy with structures
• Assume that conserved hydrophobic positions
  should pack in the core
• This appears to be work in progress (1997)

15
Example

• Two small hydrophobic residues alanine (A) and
  valine (V), both of which favor packing in the
  core of the protein.
• The POMF would have a peak around 5A
• Aspartate (D) and valine since do not often pack
  together
• The POMF will have a dip around 5A

POMF(A,V)
Probability
Distance
5A
POMF(D,V)
Probability
Distance
5A
16
Sequence-Structure Alignment

• For all know structures
• Align the unknown sequence to that structure
• Find the best alignment
• Return the structure with the best global
  alignment
• Unfortunately, we cant use dynamic programming
  (NP Complete)
• Heuristics must be used to explore the space.

17
Evaluating Methods

• Is the complexity worth it?
• This is difficult without a benchmark
• Few comparative studies have been performed
• When they have been performed, authors of
  competing methods have complained that wrong
  parameters were used
• Critical Assessment of Structure Prediction (CASP
  1994) releases protein structures prior to
  publication.
• All methods submit their predictions
• Predictions are analyzed based on fold
  recognition, modeling accuracy and alignment
  accuracy.
• No one method or approach is obviously superior