Identifying Structural Motifs in Proteins

1
Identifying Structural Motifs in Proteins

• Rohit Singh and Mitul Saha
• Pacific Symposium on Biocomputing (PSB) 2003
• Section Informatics Approaches in Structural
  Genomics

2
Outline

• Introduction
• Definition
• Problem Description
• Previous Work
• Method
• Analysis
• Results
• Conclusion
• Future Readings Related Links

3
Introduction

• Problem formulation
• Identifying a given structural motif in a target
  protein and discussing the notion of complete
  matches vis-à-vis partial matches
• Quantitative measures of structural similarity
  between two proteins
• Root mean squared deviation (RMSD) simplest
  case
• The problem of finding a good match for a pattern
  in an example
• To find the best match
• To evaluate if this match is significant enough
  (quality)

compared to or relating to someone or
something.
4
Definition (1)
5
Definition (2)
6
Data Representation

• A protein structure
• (b) A collection of significant points
• (c) An anchor and a reference frame

7
Problem Description (1)

• Problem 1 MATCH

8
Problem Description (2)

• Problem 2 COMPLETE-MATCH

9
Problem Description (3)

• Problem 3 PARTIAL-MATCH

10
M-estimators

• Used in statistics and computer vision for robust
  estimation
• The choice of a particular?depends on the
  situation and biological motivation

11
Reference for M-estimators

• Robust Statistical Estimators
• http//archer.ee.memphis.edu/www.ee.memphis.edu/pe
  ople/faculty/nasraoui/MY_TUTORIALS/RobustStatistic
  s/RobustStatistics.html
• http//www-sop.inria.fr/robotvis/personnel/zzhang/
  Publis/Tutorial-Estim/node24.html

12
Summary

• COMPLETE-MATCH
• To abstract the problem of looking for
  well-preserved matches of a given active site or
  sub-domain in a protein
• PARTIAL-MATCH
• To use an influence function, ?(x), designed so
  that the larger inter-point distances do not
  overshadow the small ones

13
Previous Work

• Kleywegt et al.
• Brute-force approach
• Idea inter-point distances are invariant w.r.t
  rotation and translation
• Wolfson and Nussinov
• Geometric Hashing
• Idea distances are invariant under rotation and
  translation
• An ordered triplet of points can be used to
  defined a reference frame
• Besl and McKay
• Iterative Closest Point Algorithm
• Idea at any point, the (currently) best estimate
  of the optimal transformation can be used to
  improve the (currently) best estimate of the
  optimal set of correspondences and vice-versa

14
Matching Pairwise Distances Good Idea?
15
ICP Algorithm
See an animation example.
16
Method
Page. 7
17
Analysis
18
Results

• Dataset Protein Data Bank
• About 42 different variants of the protease
  trypsin
• About 37 different types of kinases
• Experiment
• Pattern trypsin active site (3 a.a.)
• Target kinases
• Result
• 32 of 42 trypsin-like molecules matches with
  RMSDs less than 0.5Å
• Some other cases structural information in PDB
  was incomplete
• In the remaining 3 cases algorithm hadnt
  converged on to the optimal match

19
COMPLETE-MATCH Results
Trypsin-like Proteins
Kinases-like Proteins
20
Comparison of partial matching methods
21
Conclusion

• To formulate the problem of detecting motifs in
  proteins and present a general method for it
• To provide for a rigorous measure of the best fit
  between a given pattern and an example
• A certain flexibility in the choice of an
  appropriate distance measure

22
Future Readings

• MotifMiner A Toolkit for Mining Common
  Substructures in Molecular Data, Knowledge and
  Information Systems Journal
• Automated construction of structural motifs for
  predicting functional sites on protein
  structures, PSB 2003, pdf
• MotifMiner A General Toolkit for Efficiently
  Identifying Common Substructures in Molecules,
  IEEE BIBE 2003, pdf
• Efficient Discovery of Common Substructures in
  Macromolecules, IEEE ICDM 2002, pdf

23
Related Links

• BioGeometry Project at Duke University
• http//biogeometry.duke.edu/
• Structure and Motion Slide
• http//biogeometry.cs.duke.edu/meetings/ITR/02_nov
  /presentations/latombe-bio.ppt