In Silico Protein Structure Prediction

1
In Silico Protein Structure Prediction
New data creates new opportunity
New drug / biological data
New sequence
New structure
Novel drug development
2
The promise of genome projects
Human Genome Watch , 05 Feb. 2003 Draft
3.0 Finished 95.8 Total 98.8
http//www.ncbi.nlm.nih.gov/genome/seq/
3
The promise of genome projects
The genomes of some 800 organisms have now been
either completely or partially sequenced, and
this number will double within the next year
4
The promise of genome projects

• Diagnosing and Predicting Disease and Disease
• Susceptibility
• Disease intervention
• Solutions at the DNA level
• Gene Therapy. Although promising,
• it is not straightforward

5
Proteins The workhorses of living organisms

• Proteins are the primary components of the
  networks that conduct the flows of mass, energy
  and information.
• Protein function
• Biological (phenotypic, cellular)
• Biochemical (molecular)
• The amino acid sequence and the 3D structure
  confer unique functionality.

6
Disease prevention and management at the protein
level

• Diseases can be linked to the aberrant activity
  of proteins (enzymes / receptors).
• Pharmaceutical research is based on the search of
  molecules that will interact in a desirable
  manner with the therapeutic target.

7
Drug Discovery and Development
8
Protein Sequence-Structure -Function
structure
ALA - PHE - CYS - LYS - GLU - GLN - PRO - MET-
TRP - TYR - GLY - ARG

• Reaction / substrate
• Interactions
• Metabolic pathway

sequence
function
9
Protein Structure Initiative (PSI)

• The Structural Genomics Project aims at
  determination of the 3D structure of all
  proteins. This aim can be achieved in four steps
  
• Organize known protein sequences into families.
• Select family representatives as targets.
• Solve the 3D structure of targets by X-ray
  crystallography or NMR spectroscopy.
• Build models for other proteins by homology to
  solved 3D structures.
• www.structuralgenomics.org

10
Protein Structure Initiative (PSI)

• Unfortunately, only a few hundred protein
  structures can be determined every year
  experimentally.
• There are possibly hundreds of thousands of
  protein molecules just in humans.
• There is a need for computational/bioinformatics
  strategies to accelerate protein structure
  prediction

11
Computer simulations to fold proteins

• At physiological conditions, biomolecules undergo
  several movements and changes.
• The time-scales of the motions are diverse,
  ranging from few femtoseconds to few seconds.

12
Computer simulations to fold proteins

• Newtons second law of motion

13
We can fold small peptides using Molecular
Dynamics
14
We cannot fold proteins using MD
PDB structure of CDK_2
20 ns simulation result
15
Sequence-Structure Theoretical Relationship
Folding time scales. Physicochemical principles.
Simulations
ALA - PHE - CYS - LYS - GLU - GLN - PRO - MET-
TRP - TYR - GLY - ARG
Evolutionary time scales. Knowledge-based.
Structural bioinformatics
16
Protein Structure Prediction Using Bioinformatics
Secondary structure prediction is accurate
ALA - PHE - CYS - LYS - GLU - GLN - PRO - MET-
TRP - TYR - GLY - ARG
17
Protein Structure Prediction Using Bioinformatics
It is not straightforward to fold the secondary
structural elements in 3D space
?
18
Protein Structure Prediction Using Bioinformatics
Can we predict pairs of amino acids, distant in
sequence but proximal in structure?
19
Correlated Mutations

• Goal Manipulate multiple sequence alignments of
  protein families to identify residues that are
  close in 3D space.
• Hypothesis During evolution residues that are in
  proximity in 3D space mutate in a covariant
  fashion, so as to retain structural and
  functional properties of the protein.

20
Covariant Mutations

• Seq. 57 132
• 1 MENFQAVEKI..GEGTYGVWY
• 2 KERNKATGEV..VALKKIRWM
• 3 TETEGAPSTA..IREISAFWR
• 4 MEGEGAYRNE..VVATAIIWA
• 5 MENFIALDPV..PSTAIREWI
• 6 REPSTFIREI..SFALPRFHI
• 7 MENGHFTNKH..FCDIGEGHI
• 8 MEALKFVRLT..ETRCVGPHT

21
Measure of Covariance

• Correlation coefficient between positions i and
  j.
• where qil and qjl are the values for some amino
  acid physicochemical vector (volume,
  hydrophobicity, etc.) for sequence l at positions
  i and j. mi, mj, si and sj are the mean values
  and the standard deviations.

22
Covariant Mutations

• Seq. 57 132
• 1 MENFQAVEKI..GEGTYGVWY
• 2 KERNKATGEV..VALKKIRWM
• 3 TETEGAPSTA..IREISAFWR
• 4 MEGEGAYRNE..VVATAIIWA
• 5 MENFIALDPV..PSTAIREWI
• 6 REPSTFIREI..SFALPRFHI
• 7 MENGHFTNKH..FCDIGEGHI
• 8 MEALKFVRLT..ETRCVGPHT

Seq. 57 132 1 ..
52.60 .. 135.4 .. 2 .. 52.60 .. 135.4 .. 3 ..
52.60 .. 135.4 .. 4 .. 52.60 .. 135.4 .. 5 ..
52.60 .. 135.4 .. 6 .. 113.9 .. 91.90 .. 7 ..
113.9 .. 91.90 .. 8 .. 113.9 .. 91.90 ..
rc57,1321.0
23
Covariant Mutations

• Seq. 57 132
• 1 MENFQAVEKI..GEGTYGVWY
• 2 KERNKATGEV..VALKKIRWM
• 3 TETEGAPSTA..IREISAFWR
• 4 MEGEGAYRNE..VVATAIIWA
• 5 MENFIALDPV..PSTAIREWI
• 6 REPSTFIREI..SFALPRFHI
• 7 MENGHFTNKH..FCDIGEGHI
• 8 MEALKFVRLT..ETRCVGPHT
• 9 TETEGFPSTA..IREISAFTR

Seq. 57 132 1 ..
52.60 .. 135.4 .. 2 .. 52.60 .. 135.4 .. 3 ..
52.60 .. 135.4 .. 4 .. 52.60 .. 135.4 .. 5 ..
52.60 .. 135.4 .. 6 .. 113.9 .. 91.90 .. 7 ..
113.9 .. 91.90 .. 8 .. 113.9 .. 91.90 .. 9 ..
113.9 .. 71.20 ..
rc57,1320.97
24
Physicochemical Properties

• AAindex collection of published amino acid
  properties.
• Kawashima S. et al. Nucleic Acids Research, 27,
  368, 1999
• e.g.

25
Physicochemical Descriptors

• 142 descriptors used. Redundancy eliminated
  calculating principal components.
• 12 principal components explain 97 of the
  variance. PCA_1 is associated with
  hydrophobicity, PCA_2 represents a measure of
  size, PCA_3 is related to AA electronic
  properties, etc.
• Four different measures of distance Ca-Ca,
  Cß-Cß, minimum distance, COM-COM. Proximity if lt
  6 Å.
• Which correlation coefficient best predicts
  proximity?

26
Validation Model-System CDK-2

• Cyclin Dependent Kinase 2
• CDKs are the switches that regulate the cell
  cycle.
• CDKs control gene transcription and coordinate
  proliferation. Important therapeutic targets.

27
CDK-2 Multiple Sequence Alignment
28
Correlation Coefficient as Diagnostic Test
29
Accuracy of Diagnostic Tests
AccuracyTP/(TPFP)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
CMA-based constraints for MD
Start with a stretched protein conformation
35
CMA-based constraints for MD
Predict secondary structural elements Add spring
forces between pairs predicted to be proximal
36
CMA-based constraints for MD
37
Fold CDK-2 with 10 constraints
With 6 TP and 4 FP RMSD 14 Angstrom No secondary
structure
With 10 True Positives RMSD 5 Angstrom No
secondary structure
38
Summary

• We need protein structures to harness the
  promise of genome projects
• Further, we need computational tools to
  accelerate protein structure determination
• Use bioinformatics to guide computer simulations
• Proximal residues do mutate in a covariant fashion

39
Next steps

• Improve predictive ability of CMA.
• Systematically study the influence of constraints
  on folding.
• Develop additional methods for non-local contact
  prediction (free energy based techniques)

40
Acknowledgements
Spyros Vicatos (CEMS) Himanshu Khandelia
(CEMS) Eric Fauman (Pfizer) Sangtae Kim (Eli
Lilly) Biotechnology Institute Digital
Technology Center Minnesota Supercomputing
Institute