Chapter 5. Homology 3D Structure Prediction Chapter 6. Ab Initio

Document technical information

Format pdf
Size 352.1 kB
First found May 22, 2018

Document content analysis

Category Also themed
Language
English
Type
not defined
Concepts
no text concepts found

Persons

Organizations

Places

Transcript

Bioinformatics III
Structural Bioinformatics and Genome Analysis
Chapter 5 Homology 3D Structure Prediction
5.1 Introduction
5.2 Comparative Modeling
Sequence-Sequence Comparison
5.3 Threading
Sequence-Structure Alignment
Chapter 6 Ab Initio Prediction and Molecular Dynamics
6.1 Introduction
6.2 Ab Initio Methods
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.1 Introduction
•
Homology search
– Prediction of proteins 3D structure based on their primary
sequence
– The new sequence has an homolog with the same solved
structure
– Prediction of new structures
•
Process of folding from amino acid sequence into a protein is poorly
understood : many local effects dependent
– Quantum mechanics: to find a minimum energy state of
the amino acid sequence
– Molecular Dynamics
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.1 Introduction
•
Fold recognition/ Structure prediction
– Sequence comparison: No 3D but databases as NR (sequence-sequence,
sequence-profile, profile-profile alignments)
– Secondary structure prediction
– Sequence-Structure alignments / Structures comparison: Threading or
the use of a solved 3D protein structure to search for compatibilities of
sequences with known 3D folds
•
Proteins have limited variety of shapes: most folds are known
 Comparative Modeling success
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5Homology 3D Structure Prediction
5.2 Comparative Modeling
Sequence-Sequence Comparison
To find homologies
–
For high sequence similarities: Pairwise alignment methods (WatermanSmith, FASTA, BLAST, PSI-BLAST)
–
For remote homologous similarities: Alignment-based Methods and
discriminative Methods (only positive examples)
• PSI-BLAST: More than one iteration through NR, profile generated and
used as template for comparing unknown structures, folds and folds
classes
• FPS: Family Pairwise Search based on BLAST (comparisons of new
sequence)
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.2 Comparative Modeling
Sequence-Sequence Comparison (cont. )
For remote homologous similarities
• SVMs based protein homology: relay on a kernel specially designed for protein
sequences
– Fisher kernel: HMMs and alignments
– Mismatch kernel: sequence identities
– SVM-Mismatch kernel applied to profiles (PSI-BLAST and NR)
– SVM- pairwise method: SW score as the feature vector
– SVM using the SW kernel: SW pairwise score as kernel matrix
– SVM using Local Alignment kernel: gap penalties and BLOSUM matrices
– SVM with LA and SW- kernels applied to profiles
– SVM using oligomer based distances: construction of a feature space of
indicative patterns (PROSITE and BLOCKS)
– SVM-HMMSTR: profile construction from SwissProt data base
• LSTM recurrent Network
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.2 Comparative Modeling
Sequence-Sequence Comparison (cont. )

Overview: example from the remote homology detection benchmark:
http://www.cs.columbia.edu/compbio/svm-pairwise
–
–
–
–
Data set: 54 superfamily tasks from SCOP with one family holdig Positive and
Negative examples (in and out of belonging family)
Goal: Detection of examples from outside the fold
Quality by the area under ROC curves: values from 0.5 (random guessing) to 1.0
(perfect prediction)
Quality by the area under ROC50 curves: up to 50 false positives
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.2 Comparative Modeling
Sequence-Sequence Comparison (cont. )
Result on benchmark data (Sensitivity Vs Specificity)
ML based
classical
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.2 Comparative Modeling
Sequence-Sequence Comparison
(cont. )
Profile-profile alignment performs
better for homology remote
detection than sequence-sequence
or sequence-profile alignments
PSSM: Position Scoring Specific Matrix
GSM: Global Scoring Matrix
AF, BF and BV: All Fixed-width, Best
Fixed-width and Best variable-width ωmer
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Sequence-Structure Alignments
Structure prediction from sequence or fold recognition
“..also known as fold recognition, is a method of computational protein structure
prediction used for protein sequences which have the same fold as proteins of known
structures but do not have homologous proteins with known structure. Protein threading
predicts protein structures by using statistical knowledge of the relationship between the
structure and the sequence” Wikipedia
In PDB Ratio sequence to structure 7/1 and structures submitted in the past three years
have similar structural folds
Number of folds is small: Similar structures or folds do not have similar sequences
Proteins with different sequences but do fold into similar structures
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Sequence-Structure Alignments
Dictionary of solved structures are available DSSP
Number of folds is limited (High chance to detect the structure of new
sequence in the dictionary )
Evaluate the fitness of the query sequence for each of the possible structures
(SSEs matching, residue environment matching)
Post-processing of the results need due to the low accuracy (50%) finding the
correct fold (filtering by other predictions or known experimental data)
Goal
From native fold approximation of the energy or part of it and comparison with the
energy of the new sequence squeezed into this fold to determine if it is a suited fold for
the sequence or not
“The prediction is made by "threading" each amino acid contained in the target sequence
to a position in the template structure, and evaluating how well the target fits the
template” Wikipedia
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Sequence-Structure Alignments
Folds as cores or SSEs BUT not loops or turns (high variation)
Decoys generation and evaluation to fix the range of energy values for a native fold
and for sequences not fitting in the fold
Decoys energy values computation to separate the native fold from similar ones :
“Energy of native fold with original sequence should be less than the
energy of a random sequence”
Conformation of non-native Decoys: Parameter-Independent Decoys in which
conformation pairs of torsional angles from native decoys are perturbed by
-30°≤Φ≤ 30°
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Sequence-Structure Alignments
Computational limitations due to empirical physical energy function (water Vs
molecules simulation energies)
Concepts
Energy as values based on potentials: Cβ-Cβ distances from 3 Å to 13 Å
Unknown structure: Problem sequence Target
Known structure: Template sequence
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Threading Method design
A. Structure template database : Size and quality of the cores in the template
dictionary (as high the number higher probabilities to find an existing one)
Domains by CATH or SCOP
Bias introduced by 3D potential function deductions
NMR and x-ray crystallography
B. Scoring Function: Potential and energy function and how it is optimized to evaluate
target fitness into the folds template
Description of core elements: hydrophobic and hydrophilic residues, neighbor
relation, number and types of contacts, environment
Contact potentials: knowledge-based potentials and potential of mean forces
Potentials and configuration of the query sequence to compute the energy
(normalization to obtain the energy)
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
C. Optimization procedure to find the best fold the sequence has in the known
structure
Goal is the energy function
Difficulties due to gaps (loops and turns length variability)
For pairwise contact potentials, procedure as a
NP-hard:
DDP: iteratively a residue is placed in another
position and all other residues are optimized for
the new position
Frozen approximation: template residues are
kept and new query residue is inserted
Sampling and searching methods: Gibs sampling,
Monte Carlo.,
Mean field approaches and branch and bound
algorithms
For singleton
procedure as sequence- sequence alignment:
alignment of new sequences to the new positions
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
D. Final selection of the template once the optimal energy on each structure/fold is
computed
“..construct a structure model by placing the backbone atoms of the target sequence at
their aligned backbone positions of the selected structural template” W.
By Decoys construction
Deviation of the native fold by perturbation in torsional angles of 30°≤Φ≤
30°
Minimizing the energy of native fold with respect the current potential
function
By Z-score to measure how the energy value obtained deviates σ from the
mean value µ
Mean µ and variance σ 2 should be computed
µ and σ estimated: sequences of other folds are threaded through the fold
A Gaussian distribution CAN NOT be assumed!!!!!
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Energy parameter optimization
ai, aj amino acids
positions
For a single pairwise contact potential
Sij contact matrix
Cij contact potential
Z-score
Decoys generated: only the µ and covariance of the contact maps have to be computed
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Substituting Eo, µ and σ
In vector notation
c vector with components c (ai, aj )
s vector with components Sij (analog for s0)
S covariance matrix of s
P-SVM: z-score as a classification problem with native fold as the only member of the
positive class
Maximize Z by
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
5. Homology 3D Structure Prediction
5.3 Threading
Energy
The goal is
can be learned by Perceptron
learning rule or one-class SVM
When different sequences are used
c (ai, aj ) replaced by cij
Sij ai in contact with aj
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
Chapter 6 Ab Initio Prediction and Molecular Dynamics
6.1 Introduction
Ab initio and molecular dynamics : insights into protein folding and stability
Ab Initio
Use of amino acids sequence as the ONLY input for 3D prediction
Experimental data can be included (Rosetta method)
Novel structure to be determined with no homolog known structure (no threading
methods): Prediction of new structures
Molecular dynamics
Force fields not always modeled correctly
Computation of many sums over all atoms or sets of atoms
Simulation of water and its interaction with many molecules
Downscale macroscopic parameters: dielectric constant.,
No simulation of the context in the cell: chaperones not considered
Simulation in femtoseconds: gaps of 10 12
Computing time of 1012 CPU-years
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
6 Ab Initio Prediction
6.2 Ab Initio Methods
Rosetta Method: the way to a fold protein
Local folds
― Constructed based on small fragments
― Library of 3 and 9 residues from which folds are generated
― Sequence and profile-profile method extracts the appropriate fold by sampling
possible conformation by Monte Carlo approach
Scoring function
– Hydrophobic burial
– Pairwise interaction (electrostatic and disulfide bonds)
– α helix and β strand and spherical packing
–
β strand packing
Improvement by
– filtering out non-plausible folds as poorly formed β strand, low contact order or
packed interior
– Information from homologous sequence
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
6 Ab Initio Prediction
6.2 Ab Initio Methods
•
Rigid body models: Secondary structures are predicted and represented as rigid
models where the torsion angles are only changeable at the junctions of those bodies
•
Lattice representations: Residues are restricted to points on a regular 3D lattice
•
Potential functions: Molecular mechanics and force fields are used but
computationally expensive because water must be also modeled
•
Optimization techniques and search methods: Energy landscape of the current
conformation must be sampled (torsion angles variation, direct movements of the
atoms or fragments insertions). Monte Carlo simulation, evolutionary or genetic
algorithms and simulated annealing can be used. The candidate solutions are filtered
and checked for plausibility. As fewer candidates to be considered more detailed the
model
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
Threading, Ab initio

Performance: Threading methods perform better being comparable methods Rosseta
and Ab initio

Threading programs:
– PROSPECT [Xu and Xu, 2000]
– Tasser
– FAMS
– Zhang (threading + clustering )
SS10 Structural Bioinformatics and Genome Analysis Dipl-Ing Noura Chelbat
Wednesday 2.06.2010
×

Report this document