Transcript Project Update

Project Update
Gary Johnston
Project Update
30th March 2012
Overview
• Aims – Very brief.
• Implementation- Primary focus.
• Results – touch on this only.
Project Update
30th March 2012
Project Aims
Investigate the effects of ionising Radiation on cancer cells.
• PC3(Highly metastatic), DU145(moderately metastatic) and
PNT2(normal).
• AFM response to investigate the mechanical effects.
• Confocal images to look at overall actin cytoskeleton morphology.
• Find if differences exist, and if so, model them.
Project Update
30th March 2012
Implementation
Completed Automated Tasks.
•
•
•
•
•
•
Unwanted files - Removing bad AFM data.
Read Ibw files - Converting from Igor binary wave to force-indentation curves.
Elastic modulus – Hertz Eb and Extended moduli Eb,k1,k2 for the force indentation curves.
Contact point - Calculate reliably for curves fitting normal characteristics.
Database Creation - Based on the directory structure and results.
Database Script function – Performs a series of statistical tests in a methodical manner.
Non-Automated.
• The Frontend – A user tool for the database. (Contains lots of statistical outputs and
visualisation methods for the data contained within the database).
• Contact point Editor – Murphy’s law - For when automation goes wrong.
Project Update
30th March 2012
Unwanted files - Removing AFM data
Approximately 40 subdirectories
Each containing around 100 files to
be selected. So around 20000 files
are removed.
Only the last files from the number are wanted, the unwanted readings occur when
performing the AFM reading and the tip is too far from the cell. These are removed
automatically from all the subdirectories.
Project Update
30th March 2012
Force curves.
Important points.
F
o
r
c
e
• They vary a lot. Same cell line.
• No visible adhesion for all tested cells on
Fibronectin. PNT2, PC3 and DU145. This isn’t
always the case with other cell lines.
Indentation.
Human hair 40-50 microns wide.
The graph to the right is a tenth of
this. (FN)
• Scales for force are sub 20nN. Although this is measured to 20nN here typical scaled
used for the data is normally less than half that at a maximum of 1.75mm. This is less
than 1/20th of the width of a human hair, and in AFM circles, normal use is just 0.50mm.
• Insanity is doing the same thing over and over again but expecting different results, Yet
probe the same cell twice and you will get two different results, and if you wait for it to
promote a strong stress response it could be quite different or just to be awkward not!
Project Update
th March 2012
possibly true fact : Your fingernails grow one nanometer every second!
30Interesting
Elastic Moduli
Hertz Model
𝐹=A2δ2
𝐹=
2 𝐸𝑏
𝑡𝑎𝑛𝛼 𝛿 2
2
𝜋 1−𝜈
0.5
Eb = Constant for particular cantilever tip.
Extended Hertzian model
𝐹=A0δ4+A1δ3+A2δ2
• Linked data
• Fitted in matlab by A/b
Rationale : Standard Polyfit requires
constant. In more recent matlab versions
can be fitted using the curvefit toolbox but
my code also avoids scaling issues with the
matrices due to the scale (10-9)4.
Contact point.
• Initially this was done based on the angle of a short length. (Redundant)
• Changed to be based on smoothing and symbolic differentiation.
• Partial Lin et al. Method. (Partial because the curves don’t adhere)
• An ad-hoc method based on deviation from a localised straight line.
• A Bayesian change point method. (Rudoy et al. 2010)
No method works perfectly all the time, each has weaknesses and strengths.
A review of various methods is contained in Lin et al (2007) and for the Bayesian change point.
(Rudoy et al. 2010)
Care must still be taken and a quick look is still a good idea. Next up an example of why visually
checking is advisable.
Project Update
30th March 2012
The switching regression model : A metropolised Gibbs Rao-Blackwellized sampler.
(Rudoy et al. 2010).
• Normally works well in identifying the
contact point.
• Care though needs to be taken with
short start lines.
• Takes a very long time. 5 Days preoptimisation.
• Still takes a complete day to do.
• Needs some aftercare hence a contact
point editor to visually check the points.
The importance of the contact point.
1
x 10
-14
-
Control Energy diff at hour based on cp movement of +//- 1.0*10.0
7
0.9
0.8
Energy (J)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2
4
6
8
Hours
12
24
48
• At 100nm the difference is over 300%. A 1nm difference is at least1.5%.
Project Update
30th March 2012
Database Creation
Select the top level directory to create the database from.
(From the user point of view that is it)
The function itself uses most of the previous things I’ve just mentioned.
• Remove the extra AFM files stored. (automated)
• Read the IBW files (automated)
• Work out the contact point by using one of the 4 methods. (automated, but
selectable)
• Work out the Elastic moduli based on the contact point. (automated)
• Convert them to force indentation curves. (automated)
• Store the result into an SQL database. (automated)
Project Update
30th March 2012
Database Frontend
Project Update
30th March 2012
Standard Graphs available from the frontend.
normal curve fit Eb1
18
16
14
12
10
8
6
4
2
0
-1
0
1
2
3
4
5
x 10
-9
Standard Graphs available from the frontend.
Andrews plot.
Parallel plot.
Project Update
30th March 2012
Non- Standard Frontend Graphs
3d rotating plots.
Project Update
30th March 2012
Force contribution plot.
Non- Standard Frontend Graphs
Method for looking at time.
The flock plot looks at the data in a different way by using the different times as a
basis and moving between them using the shortest distance.
Non- Standard Frontend Graphs
Plot Title Text
4
3.5
3
• Eb,k1,k2 as relative distances to look for
patterns.
• We can’t do actual distance (As scales are
very different)
• We can do ratio distances, to look for
patterns. Left at rough stages as no benefit
was seen for the current data.
BI
2.5
2
1.5
1
30
20
10
delays
0
0.2
0.3
Project Update
30th March 2012
0.4
0.5
ratio
0.6
0.7
0.8
0.9
1
Script functions – some of the output.
Shown below is the total energy.
Blue is Eb, which is x2
Green is k2, which is x3
Red is k1, which is x4
Contact Point Editor – For when things go wrong!
File selected
Point Of Contact
File selection
Fit red
Green point in fit
for 50nm.
File adjuster and saver
Problem Results : Weekly variation Issues.
-15
Control median - Exposed Median energy at each hour
1.5
1.5
1
1
0.5
0.5
Energy (J)
Energy (J)
2
x 10
0
-0.5
-1
-1
2
4
6
8
Hours
12
24
48
-15
Control median - Exposed Median energy at each hour
0
-0.5
-1.5
x 10
-1.5
2
4
6
8
Hours
12
24
48
Future Work.
• Force Eb to be positive. (Doesn’t make physical sense to really be negative)
• Fix Eb to be fitted over 50nm and then adjust k1,k2 accordingly.
• Finite Element modelling of the median AFM force response and confocal images.
Recap.
Record the AFM data an Place into a database.
Select the top level directory.
• Remove the extra AFM files stored. (automated)
• Read the IBW files and convert them to force indentation. (automated)
• Work out the contact point by using one of the 4 methods. (automated, but selectable)
• Work out the Elastic moduli. (automated)
• Store the result into a database. (automated)
Frontend – To enable a user to select whatever plot and parameters they want to test.
Contact point editor – To enable a user to determine if the point is out and to change it.
Statistical script function – To automatically generate a host of stats.
Project Update
30th March 2012
Matlab Bits’n’Bobs.
I figured I’d take a little tangent at the end to talk briefly about matlab.
1)
Deployment of a matlab project makes it slower!
2) http://www.mathworks.co.uk/matlabcentral/. Chances are it’s here already.
3) Matlab does not contain the Anderson-Darling test for normality.
4) X or Y axis naming the numbers. Often you don’t want 1,2,3 so
use…set(gca,'XTickLabel',xaxis_labels); instead to replace the numbers with cell strings.
xaxis_labels = {‘foo1’, ’foo2’, …. ‘foon’};
5)
Speed-ups.
… There’s lots more but I’ll leave those for another time.