Computational Prediction of Enzyme Active Sites

1
Computational Prediction of Enzyme Active Sites

• 6.874 Project
• 13-May-2004
• Daniel Kim
• Rasika Kumar
• Leopold Parts
• Earl Solis
• Ragu Vijaykumar

2
Outline

• Motivation
• Literature (Serine Hydrolase Family)
• Aspartyl Protease Family
• Method Modification
• Conclusions

3
Motivation

• Structures of many enzymes are being solved
• Identifying active sites
• Structural and functional mechanisms
• Minimize experimental methods through
  computational predictions
• Search for structural analogs of known active
  sites

4
Literature

• Serine hydrolase family
• Chou Cai, Proteins 5577 (2004)
• Ser--His--Asp catalytic triad
• Hydrogen bond formation potential
• Active site recognition
• Distances among active atoms
• (N choose 2) (intra - a.a.) 8
• Standard distances
• Avg. of active site
• Avg. of non-active site

5
Literature

• Distance Metrics
• Least Hamming Distance
• Least Euclidean Distance

n of comparable distances ? active std,
non-active std
6
Literature

• Covariant Discriminant Algorithm
• Hamming and Euclidean distances independent
• Coupling effect among different pairwise
  distances
• Mahalanobis distance

7
Literature

• Results (Serine hydrolase family)
• 88 enzymes in training set
• 50 enzymes in test set
• 47 active sites correctly predicted (99.91
  success rate)

8
Literature

• Describe any significant portion of our program
  that mimics their results

9
Aspartyl Protease Family

• Pepsin active site (1am5.pdb)
• Gadus Morhua (Atlantic Cod)

10
Aspartyl Protease Family

• _ _ enzymes in training set
• _ _ enzymes in test set
• Covariant Discriminant Algorithm
• Computationally expensive to calculate non-active
  site distance scores
• Distributions of training set distance scores to
  match active sites

11
Method Modification

• Distribution Method
• Calculate overall distance scores for active and
  non-active sites on training set
• Calculate mean and std. deviations
• Compare test set enzyme distances to training set
  distributions
• Minimize z-score

12
Method Modification

• Using normal distrib. approximation

13
Method Modification

• Results