Spring 2014 Computational Journal
March 20, 2014
- Ran the Michaelis-Menten simulation using solely the dZAP1 data in order to produce the MAT file necessary to output graphs that compare the sigmoid and Michaelis-Menten models for dZAP1.
- It might be beneficial to consider putting a line in estimation_driver.m that specifies if one strain will be run or multiple strains. Otherwise, as the code is written now, multiple models and multiple sets of graphs (one for each strain) are produced even when a one strain simulation is run (by setting qq = 1 in general_least_squares_error.m).
March 28, 2014
Began merging the different scripts for the sigmoid and Michaelis-Menten deterministic models.
- In the "optimization_parameters" spreadhseet in the input workbook, added a row with the header Sigmoid. There is a 1 if a sigmoid model is to be computed or a 0 if the Michaelis-Menten model is to be determined.
- At the point in Parameters.m where the network thresholds b are imported as a vector bvec, changed the command to the following for loop to fix problems with exporting the network b's to the output in Excel:
if Sigmoid == 1 [bvec,TX6] = xlsread(input_file,'network_b'); b = bvec; end
- In LSE, altered the for loop to do the forward simulation by putting in a switch between the scripts for the dynamics between the two models.
- In general_least_squares_error.m, altered the for loop to solve the differential equation to include a switch between the scripts for the dynamics between the two models.
- In output, created an if statement to output the network thresholds b only if the sigmoid model is being computed.
Katrina Sherbina 14:32, 28 March 2014 (EDT)
April 29, 2014
Normalizing dSWI4 Microarray Data
- Created a target file to normalize microarray data collected for the SWI4 deletion strain by Biology 478 using R 3.1.0 and limma 3.20.1.
- Ran the within array and between array normalization.
- In the final normalized output, the only values in the columns corresponding to data collected using the GCAT chips were "NA".
- This issue has nothing to do with the actual normalization procedures. Rather, the problem lies in the gene names within the GPR files for the GCAT chips. Those gene names that would match those on the Ontario chips do not because the name is followed by "-01".
- As a result of the aforementioned issue, only outputted the within and between array normalized data for the Ontario chips before the start of class (BIOL 478) today.
- During class, I found that the following line of code removed the "_01":
- Modified the dry lab protocol on Dahlquist's page for normalizing only the Ontario chip data to be able to order the columns of the data by strain, then time point, then flask.
Rerunning Model Simulations
In compiling tables for my thesis of the parameters that are estimated and outputted for the sigmoid model, I noticed that the MATLAB output files (.mat files) that I have in my backups have an empty array for the threshold values. As this seemed odd to me, I checked the value of fix_b in these .mat files and it turned out to be 1. So, I decided to rerun the sigmoid model simulations with the correct optimization parameters, namely fix_b is 0.
In order to repeat both the one strain and all strain simulations, I created two sets of the MATLAB scripts that merge that can toggle between the sigmoid model and the Michaelis-Menten model depending on the value of the parameter "Sigmoid" in the input sheet. If Sigmoid = 1, then the simulation is run for the sigmoid model. If Sigmoid = 0, then the simulation is run for the Michaelis-Menten model. One set was designated for the all strains simulation and no changes were made to any of the scripts or functions. The other set was designated for the one strain simulation. Some of the functions and scripts in the set were changed as follows:
- estimation_driver.m : User is asked to specify for which strain the model should be found as the variable "strain_of_interest".
- Parameters.m, LSE.m, general_least_squares_error.m, Output.m : The index for the data structure log2FC that delineated different data sets for different strains was removed as the model is only being found for one strain. Thereby, some for loops that either create the data structure log2FC use any component of it were removed.
The all strain simulation for the sigmoid model not including generating the graphs and output ran in 1386.875550 on one of the computers in the UHall research room (3.30GHz processor + 32 GB RAM).
The one strain simulation for the sigmoid model not including generating the graphs and output ran at the following times for the following strains on my personal computer (2.20 GHz processor + 8 GB RAM):
- wt: 229.375374 sec
- dCIN5: 275.781335 sec
- dGLN3: 294.691639 sec
- dHMO1: 250.482563 sec
- dZAP1: 316.326074 sec
The one strain simulation for the Michaelis-Menten model not including generating the graphs and output ran using the dZAP1 data for 202.944254 sec on my personal computer (2.20 GHz processor + 8 GB RAM).
May 1, 2014
- Made modifications to the code previously used to graph the sigmoid model versus Michaelis-Menten model for each data set.
- Produced 5 sets of these graphs, one for each data set. Noticed that the sigmoid model was very different between this set of graphs and a previous set produced in the Summer 2012. I suspect that this is because whatever .mat files were used were outputted after running the sigmoid model simulation with the b parameter fixed.