"
Team:Calgary/17 August 2009
From 2009.igem.org
| Line 545: | Line 545: | ||
<br> | <br> | ||
<div class="heading"> | <div class="heading"> | ||
| - | + | Sensitivity analysis and parameter optimisation | |
</div> | </div> | ||
<br> | <br> | ||
<div class="desc"> | <div class="desc"> | ||
</html> | </html> | ||
| - | + | My primary goals for today were: | |
| + | * to learn how to perform a sensitivity analysis for a given system | ||
| + | * to use that information to make the parameter estimation more useful and efficient | ||
| + | |||
| + | SENSITIVITY ANALYSIS | ||
| + | |||
| + | I followed the SimBiology example and applied that to the WithoutAI-2 system. While Chinee collected information on what affects the sensitivity of the outputs with respect to different parameters, I worked out how to automate that at the command line level. I explored GFP output specifically, with respect to 5 different parameters. Of those, my results showed GFP displays high sensitivity to parameters 1, 3 and 4; and very low sensitivity to parameters 2 and 5. | ||
| + | |||
| + | PARAMETER ESTIMATION | ||
| + | |||
| + | To support the sensitivity analysis, I performed the parameter optimisation on each of the 5 different parameters to try to fit the curve to output data that I made up. The idea is that if you can optimise just one of the parameters and fit the curve to the data, it means that the system is highly responsive to tweaking of that one parameter. The results of this were encouraging: by modifying either parameters 1, 3 or 4, I was able to achieve a 92% fit. However, even when I optimised parameters 2 or 5 as much as I possibly could, I could only achieve a 40% correlation. This supports the results of the sensitivity analysis, by showing that the system is most responsive to parameters 1, 3 and 4. | ||
| + | |||
| + | I then conducted the optimisation on parameters 1, 3 and 4 all at once. The results were surprising: the R-squared fit was .92 - slightly higher than by just tweaking parameters 3 or 4, but slightly lower than when I only tweaked parameter 1. I will explore this further to see what is going on. | ||
| + | |||
| + | If anyone would like a demonstration, these results can be seen by using the command "do_work" on my computer. | ||
<html> | <html> | ||
Revision as of 15:42, 18 August 2009
|
CAROL
Descriptive Title of What You’re Doing
WIKI CODING HERE
|
|
CHINMOYEE
Working with MATLAB Day1
Researched Sensitivity Analysis:
Intro into sensitivity Analysis , by Lucia Breierova and Mark Choudhari... Content: Examples / Tutorial on how to do sensitivity analysis on data ( No examples related to our work) Learned: Sensitivity investigates the robustness of a study when the study includes some form of math modelling. Sensitivity allows the modeller to determine what level of accuracy is needed for a parameter to make the model sufficiently useful and valid. When the parameter is insensitive in the model , making an estimation is a good idea. With Systems Dynamic models the k parameter doesn't affect the behaviour of the graph . The shape of the graph stays the same . Matlab Programing: Looked into the effect of changing the kinetic constants on the graph. This resulted in no change in the behavior of the graph as predicted by the paper. The max/min/ steepness of the graphs changed as a result of changing k values. The change of some k values effected the steepness of the lines very much while changing the value of other k values didn't change the steepeness all that much. Also looked into the effect of changing the initial concentration of the species . This completely changed the behavior of the graphs. Looked into sensitivity analysis with simbiology interface . Still trying to work out some kinks.
|
|
EMILY
Transformation of J13002-LuxOD47A-B0015 into TOP10 cells, restreaks and overnight cultures of J13002-LuxOD47E-B0015 in KT1144 cells
|
|
FAHD
Descriptive Title of What You're Doing
WIKI CODING HERE
|
|
IMAN
Working on Paper
I am gathering information needed to write down the paper we have been asked to write. I have written some sections so far.
|
|
JAMIE
Descriptive Title of What You're Doing
WIKI CODING HERE
|
|
JEREMY
Descriptive Title of What You're Doing
WIKI CODING HERE
|
|
KATIE
Completion of DNA Replication and More Writing
I was able to complete a blog update that will be posted sometime on Tuesday and DNA replication was completed except for the user prompts that will be included with the notecard. I had to make single stranded binding proteins that attach close to where the helicase is and the primase will only rez one set of primers, which is the same for DNA polymerase until it has been reset. Other things that I made today were:
I also posted wiki updates for the majority of May and also made a couple for July. I have now begun on my wiki notebook for June, which I would like to complete by Wednesday at the latest.
|
|
KEVIN
Construction of Pqrr4+B0034+K082003
Once again, the Pqrr4+B0034+K082003 circuit was constructed. This time, the vector was Pqrr4+B0034 Colony 1. The vector was cut with SpeI and PstI, and the insert, K082003, was cut with XbaI and PstI. The vector was phosphotase treated, and then they were ligated together. Next, this constructed circuit was transformed into TOP10 cells and are now growing on Kanamycin + Chloramphenicol plates.
Modelling Blog
Another Modelling blog was written today. Most of the modelling work last week was done by Vicki and Chinee, working hard to track down those sensitive parameters. Here's the link to the blog:
Sensitivity analysis: Models... have feelings too!
|
|
MANDY
Descriptive Title of What You're Doing
WIKI CODING HERE
|
|
PATRICK
Descriptive Title of What You're Doing
WIKI CODING HERE
|
|
PRIMA
Colony PCR and Marketing
It turns out that I had used the Low-melt agarose on Friday to make my 1% gel. Unfortunately, the gel didn't turn out well and I had to postpone my work to Monday. Today, I made a 1% gel with the normal agarose powder and remade my stock (3 microL PCR product, 2 microL dye & 15 microL ddH2O) and ran my gel at 100V.
The results showed exactly what I was expecting. Since I had used regular BBK CP primers and length of my construct was 885bp, the total length of the bands should've been 1135bp. My bands were just above the 1kb ladder band! I let the students visiting our lab make my overnight cultures. Tomorrow, I'll do a mini-prep and a restriction digest to verify the size of my construct. If all else is good, I'll send it to sequencing.
I followed up with the T-shirt company but unfortunately I couldn't get a hold of her. I'll follow up tomorrow. I also called numerous new companies and sent our sponsorship package to one of them. I left a voice mail for the rest of the companies. I'll follow up with them tomorrow. I also began write-ups for the August newsletter.
|
|
STEFAN
Descriptive Title of What You're Doing
WIKI CODING HERE
|
|
VICKI
Sensitivity analysis and parameter optimisation
My primary goals for today were:
SENSITIVITY ANALYSIS I followed the SimBiology example and applied that to the WithoutAI-2 system. While Chinee collected information on what affects the sensitivity of the outputs with respect to different parameters, I worked out how to automate that at the command line level. I explored GFP output specifically, with respect to 5 different parameters. Of those, my results showed GFP displays high sensitivity to parameters 1, 3 and 4; and very low sensitivity to parameters 2 and 5. PARAMETER ESTIMATION To support the sensitivity analysis, I performed the parameter optimisation on each of the 5 different parameters to try to fit the curve to output data that I made up. The idea is that if you can optimise just one of the parameters and fit the curve to the data, it means that the system is highly responsive to tweaking of that one parameter. The results of this were encouraging: by modifying either parameters 1, 3 or 4, I was able to achieve a 92% fit. However, even when I optimised parameters 2 or 5 as much as I possibly could, I could only achieve a 40% correlation. This supports the results of the sensitivity analysis, by showing that the system is most responsive to parameters 1, 3 and 4. I then conducted the optimisation on parameters 1, 3 and 4 all at once. The results were surprising: the R-squared fit was .92 - slightly higher than by just tweaking parameters 3 or 4, but slightly lower than when I only tweaked parameter 1. I will explore this further to see what is going on. If anyone would like a demonstration, these results can be seen by using the command "do_work" on my computer.
|
