Protein Modeling

for a Better Indonesia

Muhammad Yusuf
14 Desember 2017

PUSAT RISET
BIOTEKNOLOGI MOLEKULAR
DAN BIOINFORMATIKA
Jalan Singaperbangsa No. 2 Bandung
What is
Protein?
A Brief Introduction about Protein
From Sequence to Structure
 • Identifying active and binding
sites

• Characterization of the protein’s

mechanism (catalysis &
interactions)

Motivation • Searching for ligand of a given

binding site
to Acquire a
• Understanding the molecular
Structure basis of diseases

• Designing mutants

• Drug design

• And more...
OUTLINE:
Molecular Dynamics Insight into:
Mutasi pada penyakit:
 Degenerative
 Drug resistance
 Pathogen
Enzyme design: computational perspective
 Chemical modification: methylated lysine
 Solvent modification: organic solvent model
 Introducing S-S bond
 Introducing Surface Binding Site
 Biosensor
 Therapeutic
Determining Structure
para ilmuwan menemukan bagian unik dari
struktur patogen virus Zika menggunakan
mikroskop cryo-elektron, mengidentifikasi target
potensial untuk vaksin
Determining Structure
enzim β-galaktosidase

Cryo-EM
resolusi rendah Koordinat atom
resolusi tinggi

Cryo-EM
resolusi tinggi

National Cancer Institute, National Institutes of Health

 Cryo-EM
harganya
US$17 juta
~
Rp230.578.135.203,00

+ biaya
operasional
US$6.000 per hari
A cryo-electron microscope
Northwestern University
 Bagaimana
Indonesia bisa ikut
Virus Influenza
Strain Indonesia
memanfaatkan
untuk
Pengembangan
Cryo-EM?
Vaksin yang
Cocok untuk kita
 “We are good at
guiding
experimentalists.”

“Every year,
hazardous-waste
disposal gets more
expensive,
whereas
computing power
gets cheaper. So
the progress
curves favour the
theoreticians.”
(Nature, 2013)
 Why predict • Experimental methods are slow
protein structure and expensive
if we can use • Some structures were failed to be
experimental solved
tools to • A representative family structure
determine it? can suffice to deduce structures of
the entire family sequences
More Sequences Than Structures
• Discrepancy between the number of known sequences
and solved structures:
5,047,807 UniRef90 entries vs.
25566 90% Non-redundant structures

Computational
methods are
needed to obtain
more structures
 Protein Modeling
for a Better Indonesia

❑ Desain Vaksin (Hewan/Manusia) untuk Strain Indonesia

❑ Mencari Biomarker Penyakit Khas Indonesia
❑ Mengembangkan Diagnosis Patogen Khas Indonesia
❑ Merekayasa Enzim hasil Mikroorganisme Strain Indonesia
❑ Mengembangkan Obat sesuai Farmakogenomik Indonesia
“smart” VACCINE

Many kind of new vaccines:

VLP, Peptide-based Vaccine
Sub-unit Vaccine, etc
Should be also applicable
to the other diseases:
DENGUE,
Neglected disease,
HPV, etc
Yusuf et al., 2013, J. Chem. Inf. Mod. (53) 9
 biomarker: tb drug resistance

❑ Overall, the most frequently seen drug resistance mutations observed for isoniazid, rifampicin,
streptomycin, ethambutol, aminoglycosides and ofloxacin were katG315, rpoB531, rpsL43,
embB306, rrs1401 and gyrA94 respectively

❑ Currently available commercial assays for detection of drug resistance (MTBDRplus and MTBDRsl)
target the rpoB531, 516 and 526, katG315, gyrA94, 91 and 90 and rrs1401 regions.

❑ Based on our data the frequency of drug resistance detection using such assays for these strains
would have been 100% for rifampicin resistance, 92% for isoniazid resistance, 81% for
fluoroquinolone resistance, 78% for aminoglycoside and 62% for ethambutol resistance.

❑ Our data indicates that SNPs in drug resistance genes vary with different MTB lineages. Therefore, it
highlights the relevance for WGS data from prevalent global strain types in order to develop
improved molecular diagnostic assays for resistance in MTB. Additional knowledge of mutations that
are compensatory to those that confer drug resistance will be important for prediction of levels of
drug resistance and therefore, to guide treatment of drug resistance isolates.
 pengembangan agen diagnostik dan terapeutik,
pengembangan metode assay molekular untuk pengujian
herbal, pengembangan vaksin modern, rekayasa enzim atau
protein untuk keperluan industri, pengembangan database
senyawa bahan alam Indonesia, serta pengembangan database
genomik dan struktur dari agen penyakit lokal Indonesia.
Structure prediction
approaches
Structure Prediction Approaches
1. Homology (Comparative) Modeling
Based on sequence similarity with a protein for
which a structure has been solved.

2. Threading (Fold Recognition)

Requires a structure similar to a known structure

3. Ab-initio fold prediction

Not based on similarity to a sequence\structure
 Ab-initio
Structure prediction from “first principals”

Given only the sequence, try to predict the structure

based on physico-chemical properties
(energy, hydrophobicity etc.)

• When all else fails ➔ works for novel folds

• Shows that we understand the process

 1. Molecular Dynamics
• Simulates the forces that governs
the protein within water.
• Since proteins usually naturally fold,
this would lead to the native protein
structure.
Approaches
to Ab-initio Problems:
Prediction • Thousands of atoms
• Huge number of time steps to reach
folded protein
➔ feasible only for very small
proteins
Approaches
to Ab-initio
Prediction
2. Minimal Energy

Assumption: the folded form

is the minimal energy
conformation of a
protein

Main principals:
• Define an energy
function.
• Search for 3D
conformation that
minimize energy.
Ab initio
example
 FoldIt: Protein-folding game
• https://fold.it/
• Basic idea: allow players to optimize the Rosetta all-atom energy function
– Game score is negative of the energy (plus a constant)

28
 Fold Recognition (Threading):
Sequence to structure matching
Given a sequence and a library of folds, thread the sequence
through each fold. Take the one with the highest score.

• Method will fail if new protein does not belong to any fold
in the library.

• Score of the threading is computed based on known

physical chemistry properties & statistics of amino acids.

• In practice, fold recognition methods are often

mixtures of sequence matching and threading.
 Phyre2:
Online server for threading
COMPARATIVE
MODELING
 Comparative Modeling

Good
% identity

*
# residues
 Comparative modeling in short…
Prediction of structure based upon a highly similar structure

NSDSECPLSHDG

NSDSECPLSHDG
|| || | ||
NSYPGCPSSYDG

Alignment of model
and template
Model!
Unknown structure sequence
Known structure Back bone copied
Copy backbone and
Add sidechains, Molecular
conserved residues
Dynamics simulation on model
The 8 steps of Homology
modeling
1: Template recognition
and initial alignment
1: Template recognition
and initial alignment
• BLAST your sequence against PDB

• Best hit → normally template

The Importance of Resolution
• In X-ray crystallography it is
not always possible to
4Å low flawlessly resolve the crystal
density of the protein of
interest.

3Å • This results in a lower

resolution structure.

• The lower the resolution the

2Å more likely the structure is
wrong.

• The resolution of the

high template structure also
1Å
reflects in the quality of the
homology model.
1: Template recognition 2: Alignment correction
and initial alignment
 2: Alignment correction
E E E -A-V
F-D- I V V
• Core residues → conserved
• Use multiple sequence
C
alignments P A
C
C sequence → shift gaps
• Deletions in your
R R M S
R

M G
L
P
P

Known structure FDICRLPGSAEAV

Model FNVCRMP---EAI
Model FNVCR---MPEAI  Correct alignment
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
3: Backbone generation
• Making the model….
• Copy backbone of template to model
• Make deletions as discussed
• (Keep conserved residues)
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
4: Loop
modeling
4: Loop modeling
Known structure GVCMYIEA---LDKYACNC
Your sequence GECFMVKDLSNPSRYLCKC

Loop library,
try different
options
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
4: Loop
modeling

5: Sidechain modeling
5: Side-chain modeling
• Several options
• Libraries of preferred rotamers based upon backbone conformation
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
4: Loop
modeling

5: Sidechain modeling

6: Model
optimization
6: Model optimization

• Molecular dynamics
simulation

• Remove big errors

• Structure moves to
lowest energy
conformation*
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
4: Loop
modeling

5: Sidechain modeling

7: Model
validation

6: Model
optimization
7: Model Validation
• Second opinion by PDBreport /WHAT IF
• Errors in active site? → new alignment/ template

• No errors? → Model!
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
4: Loop
modeling

5: Sidechain modeling
8: Iteration

7: Model
validation

6: Model
optimization
1: Template recognition 2: Alignment correction
and initial alignment

3: Backbone
generation
4: Loop
modeling

5: Sidechain modeling
8: Iteration Model!

7: Model
validation

6: Model
optimization
8 steps of comparative modeling
1: Template recognition and initial Alignment
alignment
2: Alignment correction
3: Backbone generation
Modeling
4: Loop modeling
5: Side-chain modeling
6: Model optimization
Correction
7: Model validation
8: Iteration
LAST BUT NOT LEAST…
We cannot work alone!
 m.yusuf@unpad.ac.id
umibaroroh2@gmail.com
ade.rizqi.r@gmail.com
puslit.bio.inform@unpad.ac.id