Goal: Compare the average volume of the major protrusion for untreated cells and CK666 treated cells. 

Got all the major protrusion volumes
untreated.largest.prot = c(c14_protrusions_rv2$V3, c15_protrusions_rv$V3, c10_protrusions_rv2$V3, c12_protrusions_rv$V3)

Found the mean and the standard dev
mean(untreated.largest.prot)
sd(untreated.largest.prot)

Note: for setting all NAs to 0:
x[is.na(x)] <- 0

Results:
The average volume of the major protrusion is higher, but also has a very large standard deviation.

Goal: To see if there is a correlation between the protrusion parameters matrix and the math of the cells using linear regression.

Training matrices:
DMSO treated cells (in order): cell 14, cell 15, cell 12, cell 10
CK666 treated cells (in order): cell 43 ,cell 48 , cell 55 , cell 53 
variables: ​ Count, largest protusion volume, total protustion volume, protrusion fraction, largest protrusion length, largest protrusion angle, and total protrusion angle.

Dependent value (y): derivative using savitzky-golay
variables:

• untreated: training.set
• ck666: ck666.training.matrix

​
To get rid of rows without Y values
training.set.complete<-training.set[complete.cases(training.set),]
I did this because the smoothing window on the sg filter leaves a lot of blank values for the derivative. 

Results: all the parameters were significant except largest protrusion angle, and total angles. The largest protrusion length is only significant for the first few fits. 

Dependent value (y): curve method
variables:

• untreated: training.set.curve
  
• Haven't completed this yet. 

Dependent value (y): bi-modal with cutoff (ie, is the cell turning or not turning?)

• untreated: trainingset.curve.bimodal

​Results:

• untreated: 
  • ​glm (family=binomial): fit.curve.bimodal 
  • all fits loaded into matrix: fits.matrix.curve.bimodal
  • pvalues for all fits loaded into matrix: fits.matrix.curve.bimodal.pv

To do next:

add more cells to the training set
write function to find average turning point prediction