Unformatted text preview: 7.36/7.91/BE.490
Due April 22 at 1:00 PM
Using Deep View to study protein structure
Computer graphics provides a powerful tool for studying
protein structure. You can learn a lot about a protein
just by looking at the properties of its structure. For
computer visualization of structures in this course, we
will use the powerful program Deep View, developed at
GlaxoSmithKline R&D in Geneva and hosted by the ExPASy
proteomics server. This is available for many different
platforms, including Mac, PC, SGI and Linux. Instructions
for starting Deepview on MIT Server can be found at
program was formerly called Swiss PDB Viewer.) Note that
this program only runs on Linux machines and not Sun
machines. You can also download a version for your own
computer, if you wish, from the site
http://www.expasy.ch/spdbv/ If you have access to a PC or
Mac, this is probably preferable, as the Linux version
appears to contain a few bugs (but we describe work-arounds
for them below.) If you use Deep View on MIT Server, try to
get a fast machine, as it’s very
computationally demanding. The IBM’s are the fastest Linux
machines in the clusters. They’re in clusters W20-575 at
the back, 66-080, and 2-225.
The quickest way to learn to use Deep View is to do the
tutorial written by Prof. Gale Rhodes, at the University of
Southern Maine. Go to
http://www.usm.maine.edu/~rhodes/SPVTut/ and, starting with
the Overview section, work your way through the tutorial
parts 1 - 6. Referring to the remaining parts of the
tutorial may also be helpful for this problem set.
Different versions of Deep View may have slightly different
user interfaces, so you may occasionally need to click
around a bit to find what they’re referring to or ask a
teaching assistant. Take time as you are doing this to
play around with different features of the program. Deep
View can be a great resource for you in your research, so
this is time well spent.
1. Familiarizing yourself with Deep View A study by Gidh-Jain et al in 1993 was aimed at determining how mutations in glucokinase (same as hexokinase IV from
problem set 1) altered its function. You can obtain a PDF
version of this article for free at
http://www.pnas.org/cgi/reprint/90/5/1932. They performed
biochemical assays on the wild-type and mutant proteins and
calculated their activity. In addition the crystal
structure of the related yeast hexokinase B (solved in
1978) was available. The human and yeast sequence are ~30%
identical and the authors ‘predicted’ that the arrangements
of a-helices and b-sheets would be similar between the two
proteins. They mapped the sites of mutations to the crystal
structure and concluded that mutations fell into two
groups. One group of mutations was located in the active
cleft separating the two domains of the protein and the
second group was on the outer side of the protein in a
place predicted to undergo a substrate induced
conformational change that results in the closure of the
active site cleft.
A quick search of the Protein Data Bank for human
glucokinases reveals a structure that was added recently
(March 30, 2004). Download the structure 1V4S to your
computer and open it in SwissPDB viewer. In the toolbar
(you can get to the toolbar form the Wind menu) there is an
icon showing a page with lines. Clicking on this will open
the PDB text file for the structure. Using the text file or
the structure viewer answer the following: a.
g. How was the phase problem solved for this structure?
Discuss the pros and cons of this method vs.
What was the free R value for this structure? What
does this value mean? Does it give you confidence in
What was the RMSD of the bond angles and dihedral
angles from ideal values? Which one is larger? Is
this what you would expect? Why is this an important
List the crystallization conditions for this protein.
Is the structure likely to be physiologically
List the order of secondary structural elements
from N to C terminus in this structure.
List the heteroatom(s) (non-protein atoms), if any,
included in this structure.
Now go to the Control Panel (you can get to it form
the Wind menu). This is a quick way to find residues,
change views of the molecule, etc.
By default all
residues and their side chains are visible (denoted with a “v”). What are the N- and C-terminal residues
that appear on the structure?
Table 1 shows a list of common mutations in glucokinase
that lead to non-insulin-dependent (type 2) diabetes
mellitus (from Gidh-Jain et al).
or Gln *
* * Predicted to be near the active site
^ Predicted to participate in the conformational change
that results in cleft closure.
Label and show the sides chains of all of these amino
acids. On the control panel window, check the “Show”,
“side”(to see the side chain) and “labl”(to show the
residue and position) column. Hide the side chains and
labels for the rest of the sequence. Hint: To check or
uncheck all the residues, place the mouse on the column you
want to modify and shift-click. Do this first, then add the
amino acids you want to see.
h. i. Are any of the amino acids that are not adjacent in
the primary structure close in space in the tertiary
structure? Use the move, center, zoom and rotate
tools in the Toolbar menu to look at the structure.
Save this structure view for later (Save layer
command). Print a copy of this layer and submit it
with your homework assignment. (Note: if you have
trouble printing, you may need to take a screen shot
of your protein and print the image. Black and white
print outs are fine.)
Do you think the mutations fall into two classes like
the authors described? Explain. Now that both the human and yeast structures are available,
you can test if the author’s prediction that the yeast and human glucokinase structures were similar was correct.
Download and open the PDB file 2YHX.
j. On the ‘group’ column of the control panels many
aminoacids are labeled UNK because the authors didn’t
have the sequence of the protein they were
crystallizing. Therefore for some residues, the
authors could not unambiguously assign the identity
of the residue. For unknown residue 112, list which
possible amino acids it could be based on the atoms
present. Superimpose the two structures in order to determine how
similar their folds are. To make it easier to see the
structures, first display the molecules without side chains
(hint: deselect the side chains from the Control Panel and
choose ‘Show CA trace only’ from the Display menu on the
tool bar to see the alpha carbons). With the two
structures loaded, the Control Panel can now be toggled to
display information about each structure. Color the alpha
C trace to distinguish the 2 models. Shift-Click on any
box in the Col column and choose a color from the color
pallet (be sure that backbone + side is selected in the
pull down menu to the right of the header in the Col
column.) Select Best fit from the Fit menu. Use the alpha
carbons for the alignment.
2. How similar are the two protein structures? What can
you say about the corresponding location of the
residues listed in Table 1 in yeast hexokinase? Do
they all lie in regions where both structures are
What is the RMSD (considering only alpha carbons) for
the superposition? How similar would you say these
Based on your results, do you think the authors
‘predictions’ were correct? Explain. Homology Modelling Retrieve a fasta file containing the protein sequence
the protein with accession number JQ2288. This is a
transmembrane protein from soybean. It is in general
difficult to obtain structures of membrane proteins.
you would like to construct a model of this protein.
this file to your home directory. of very
Save Start up Deep View without opening a structure file.
Choose SwissModel -> Load Raw Sequence to Model and open
your fasta file.
In Prefs -> SwissModel, enter your name and email address
and click OK.
Choose SwissModel -> Find Appropriate ExPDB templates to
find structures that are similar to your sequence of
interest. If your Mozilla browser is already started, your
window should be redirected to a page that allows you to
search for appropriate PDB templates for your sequence of
interest. Click on submit to begin finding similar
A list of crystal structures of channels should appear.
Download the PDB files of top two BLAST hits and open them
in Deep View. The layers window will enable you to switch
between the three proteins you now have open.
Superimpose the two structures using Magic Fit. Then
select your sequence of interest and repeat a Magic Fit
again. Finally choose SwissModel -> Update Threading
Display Now. (This may be grayed out if “Update Threading
Display Automatically” is set.) You should see your
sequence folded up upon your reference PDB structure. To
verify that you have done these steps correctly, we suggest
that you color the three structures different colors (Color
-> Layer) and verify that they are all superimposed on one
Choose SwissModel -> Submit Modelling Request. You will be
asked for a directory to save your model in. On Linux, the
name of this file is by default tmp.html. It will be saved
with that filename prefixed by proj_. Mozilla will open at
a page that lets you submit your request. Click Browse
button, and navigate to where your file was saved. Once the
filename has been entered into the corresponding field,
click the send request button.
Some time later, you will receive mail from the modeling
server. One of the mails will contain your model in the
form of a pdb file. Save that file to your home directory.
Open your model in Deep View. The three structures will be
displayed superimposed. Hide all but your homology model,
and turn off the ribbon display. a. b. c. d. e. 3. Based on the sequence similarity between your
sequence of interest and your two template
structures, are these valid templates for homology
modeling of your sequence? Explain.
Go to Prefs -> Ramachandran Plot and choose to ignore
ignore glycines and prolines. View the Ramachandran
plot of your homology model (Wind -> Ramachandran
Plot). Note that only currently selected amino acids
appear in the Ramachandran plot so make sure that you
select all amino acids in your model and not any of
the other two structures. Are there any residues
that fall outside of the “typical” phi/psi space? If
there are any, what type of residues are these? Does
this analysis increase or decrease your confidence in
View the Whatcheck results for your model. Whatcheck
is a program that performs various checks on homology
models looking for different kinds of errors a PDB
file. Give three different problems with the
structure of your homology model (other than what was
answered in part b) and briefly explain whether you
think the error is significant or not.
What is the advantage of using more than one template
structure during homology modeling? Would it have
been worthwhile to include additional template
structures from the list of BLAST hits obtained by
Do you think that the soybean channel under
investigation is actually an aquaporin channel?
Discuss. X-ray vs NMR structures The structure of lac repressor has been solved several
times both by X-ray crystallography and by NMR. Open a
crystal structure (PDB ID1EFA) and an NMR structure (PDB ID
1L1M) of lac repressor in Deep View. Initially only load
one NMR structure. Make sure that you are viewing the
Sequences Alignment window and the Layers window.
a. b. Align the two structures using Magic Fit. What is
the all-atom RMSD between the two structures after
this fit? Is this value higher or lower than you
were expecting? Did the result surprise you?
Provide a possible explanation for why you either do
or do not see differences between the two structures.
Go beyond merely stating the fact that one structure c. d.
4. was obtained via X-ray crystallography and the other
Usually, NMR experiments yield a set of solution
structures that satisfy the constraints obtained from
the data. Now open all 20 structures in 1L1M.Where
do you see the most variability between these
In what parts of the crystal structure are the Bfactors (temperature factors) the highest? Is this
the same region where the NMR solutions differ?
In some cases with NMR experiments, an averaged
energy minimized structure is given rather than a set
of structures. Discuss the advantages and
disadvantages of the two approaches.
Would knowing the NMR structure of a protein aid in
solving the crystal structure? Vice versa? Why? Protein side-chain repacking A particular case of the general protein structure
prediction problem is the side-chain repacking problem. In
this problem, the backbone and the sequence are fixed and
the question is to find the combination of amino acid
rotamers (conformational isomers), which makes the lowest
energy structure. Repacking is useful in a number of ways.
It can be used to model the structural response of a
protein to one or more mutations, and it can be used as the
final step when building a homology model. Another way in
which repacking can be used is to determine the reliability
of a particular energy function. If a protein crystal
structure is available, one can take its backbone and
sequence and repack it given any particular energy
function. Comparing the wild type structure to the repacked
structure gives one an idea of whether the energy function
has the ability to distinguish the native (or near native)
conformation as having minimal energy.
Download the file 1FXD.pdb from the class website (not the
PDB). Make another copy of it and call it
“1FXD_repacked.pdb” (on MIT Server “cp 1FXD.pdb
1FXD_repacked.pdb”). Open this second copy in SwissPDB
Viewer (on MIT Server “spdbv 1FXD_repacked.pdb &” – will open
SwissPDB Viewer in the background). This is the structure
of ferredoxin II from Desulfovibrio gigas.
a. The sidechain assignment problem is often referred to
as a ‘combinatorial’ problem. Why is this? b. Let’s suppose you have come up with an amazingly fast
(and accurate) energy function, which can assign an
energy to one combination of rotamers for our
structure in 1 CPU cycle. Lets also suppose you have
access to an array of 1,000 5 GHz CPUs (which run
nothing else but your energy code). Calculate how
much time it would take for you to exhaustively
enumerate all rotamer combinations for this structure
and find the lowest one (assume that each amino acid
has 15 rotamer states). Express this in “graduate
student work-year equivalents”, or some other measure
that is meaningful for you. Go to the Control Panel window and select all amino acids
(alternatively, from the Select menu, choose All). Then go
Tools Æ Fix Selected Sidechains Æ Quick and Dirty. This
function is intended for relieving possible unfavorable
steric interactions of the sidechains with the rest of the
structure (after a mutation is introduced, for example).
This is equivalent to doing a repacking with a very simple
energy function (actually, the score or energy each rotamer
gets is a combination of the number of clashes it makes
with the structure and the number of hydrogen and disulfide
bonds it is involved in – for details, see
“quick and dirty” part means that it applies a rapid
minimization algorithm, which gives reasonable results but
is not guaranteed to give the absolute minimum. Save this
structure by going to File Æ Save Æ Layer, navigate into
the directory where the original file was, click “Save” and
select OK to overwrite the old file (note: do not try to
edit the name of the file – this feature is a bit buggy.
You’ll need this structure for the python problem. Click
Display Æ Show H-bonds to hide the hydrogen bonds. Then
open the original copy of the file (on MIT Server “spdbv
1FXD.pdb &” – this will attach the new file to your
previous session). You now have both structure – the
original and the repacked one in separate layers. For both
layers go to Color Æ By Accessibility (to switch between
layers use the Layers Info window; it is located under
Window Æ Layers Info).
c. Look at the difference between the wild type and the
repacked structures. What regions, or amino acids,
are least similar between the two structures? Give
some specific residue numbers where the sidechain
conformation is wrong. d. e. 5. Which region of the molecule is best repacked? (i.e.
where is the difference between the two structures
smallest?) Give residue numbers for some sites where
the side-chain structure is predicted most
Give three reasons why the repacked structure might
not agree perfectly with the experimentally
determined structure. Analyzing a repacked structure using python Below is a python program, which reads two PDB files of
identical sequence and calculates the room mean square
difference (RMSD) between the coordinates of the
corresponding sidechains. The names of the two input files
should be specified on the command line. It assumes that
the two files have exactly the same sequence (the same
atoms in the same order).
# Check usage
if (len(sys.argv) < 3):
print "Error in usage!"
# read in PDB files
file1 = sys.argv;
file2 = sys.argv;
f = open(file1);
lines1 = f.readlines()
f = open(file2);
lines2 = f.readlines()
# Calculate and print RMSDs
i1 = -1
i2 = -1
pri = "" # previous residue index
prn = "" # previous residue name
sad = # array of squares of sidechain atom deviations of the current
f1 = 0
for i1 in range(i1+1, len(lines1)):
if not(re.match("^ATOM", lines1[i1])):
rn1 = lines1[i1][17:20]
ri1 = lines1[i1][22:26]
an1 = lines1[i1][12:16]
continue f1 = 1
f2 = 0
for i2 in range(i2+1, len(lines2)):
if not(re.match("^ATOM", lines2[i2])):
rn2 = lines2[i2][17:20]
ri2 = lines2[i2][22:26]
an2 = lines2[i2][12:16]
if (re.match("(^\s*H|^\s*C\s*$|^\s*O\s*$|^\s*N\s*$|^\s*CA\s*$)", an2)):
f2 = 1
if (f1 and f2 and not((rn1 == rn2) and (an1 == an2) and (ri1 == ri2))):
print "Error - mismatched lines:\n%s%s" % (lines1[i1], lines2[i2])
# have we started a new residue?
if (not(ri1 == pri) or (not f1) or (not f2)):
if (len(sad) > 0):
rmsd = 0
for i in range(len(sad)):
rmsd += sad[i]
rmsd = math.sqrt(rmsd/len(sad))
print "%s %s %.2f" % (prn, pri, rmsd)
# reset all arrays
sad = prn = rn1
pri = ri1
if (not(f1) or not(f2)):
# now process atom deviation
x1 = float(lines1[i1][30:38])
y1 = float(lines1[i1][38:46])
z1 = float(lines1[i1][46:54])
x2 = float(lines2[i2][30:38])
y2 = float(lines2[i2][38:46])
z2 = float(lines2[i2][46:54])
sad.append((x1-x2)*(x1-x2) + (y1 - y2)*(y1 - y2) + (z1 - z2)*(z1 - z2)) Note that since glycines do not have a side chain (it’s a
hydrogen, which is usually not explicitly present in PDB
files) they are not listed in the output.
a. b. Apply the provided program to 1FXD.pdb and
1FXD_repacked.pdb. Sort the output values in
descending order (you can use Excel or whatever
program you want) and look at the top 10 entries.
What is common about the corresponding positions?
What about the amino acids?
Modify the provided python program so that instead of
computing the sidechain RMSD between two proteins of
identical sequence, it instead computes the c-alpha RMSD between two user-specified sets of residues in
two proteins (possibly of different sequence) and
prints the value to the screen.
> rmsd_backbone.py test1.pdb 20 34 test2.pdb 55 69
would give the C-alpha rmsd between residues 20-34 in
test1.pdb and residues 55-69 in test2.pdb.
Submit your program on line. Call it rmsd_calpha.py. For
this program, you will not be given sample input or output. ...
View Full Document
- Spring '03
- Deep View