Review : Protein Secondary Structure Prediction Continues to Rise

1
Review Protein Secondary Structure Prediction
Continues to Rise

• Burkhard Rost
• CUBIC, Columbia University
• rost_at_columbia.edu
• present by yfhuang

2
Outline

• Protein Structure
• Protein Secondary Structure
• Prediction Methods
• Using Secondary Structure Predictions, in
  Practice
• Conclusions
• Reference

3
Protein Structure
4
Protein Secondary Structure
5
The Secondary Structure Prediction Problem

• Given a protein sequence
• NWVLSTAADMQGVVTDGMASGLDKD
• Predict a secondary structure sequence
• LLEEEELLLLHHHHHHHHHHLHHHL
• 3-state problem
• ARNDCQEGHILKMFPSTWYVn --gt L,E,Hn

6
Generation of Prediction Methods

• 1st generation single residue statistics
• Base on single amino acid propensity
• 2nd generation segment statistics
• Propensity for segments of 3-51 adjacent residues
• 3rd generation evolution to better predictions
• The use of evolutionary information (evolutionary
  profile)

7
Prediction Methods

• Advanced recursive neural network system. (SSpro)
• Hidden Markov Models for connecting library of
  structure fragments. (HMMSTR)

8
Prediction Methods (Cont)

• Evolutionary profile
• A common strategy for predicting protein function
  is to find another protein with similar sequences
  and known biological function and compare the
  two.
• One way to model a family of proteins is with a
  profile -- a statistical description of the
  protein family where different sequences
  contribute proportionally to their degree of
  divergence.

9
Prediction Methods (Cont)

• PHD
• Multiple levels of computations
• Sequence -gt structure
• Structure -gt structure
• Balanced predictions by balanced training
• Better segment prediction by structure-to-structur
  e networks
• Automatically aligning protein families based on
  profiles

10
Accuracy of Secondary Structure Prediction Methods
11
Using Secondary Structure Predictions, in Practice

• Regions likely to undergo structural change
  predicted successfully
• Classifying proteins based on secondary structure
  predictions in the context of genome analysis

12
Using Secondary Structure Predictions, in
Practice (Cont)

• Aspects of protein function predicted based on
  expert-analysis of secondary structure
• Exploring secondary structure predictions to
  improve database searches
• From 1D predictions to 2D, and 3D structure

13
Conclusions

• What can you expect from secondary structure
  prediction?
• How accurate are the predictions?
• Confusion between strand and helix?
• Strong signal from secondary structure caps?
• Internal helices predicted poorly?
• What about protein design and synthesised
  peptides?

14
Conclusions (Cont)

• How can you avoid pitfalls?
• 70 correct implies 30 incorrect
• Special classes of proteins
• Better alignments yield better predictions
• 1D structure may or may not be sufficient to
  infer 3D structure

15
Reference

• Paper
• http//dodo.bioc.columbia.edu/rost/Papers/2001_re
  v_sec_jsb/paper.pdf
• Web Link
• http//dodo.bioc.columbia.edu/rost/
• http//www.smi.stanford.edu/