How to find values in one vector that don't have corresponding values (in the same row) in another

Hi,

I have a very large dataset of sample measurements and I need to plot sample data but without the 'background' data, i.e. data that appears in the control samples. Here is an example of the data:

row a b c sample1 sample2 control1

1 5 2 1 236489 NaN 456932

2 8 3 0 NaN 278695 569202

3 12 8 2 NaN 332465 NaN

I have loaded the whole dataset as a tab delimited txt file and so each column of data is in a different vector. (NaN is where there was no value in the original tab file).

I need to plot a/b against c/b with the sample value on the z axis. I have managed to do this for all the values in the sample vector, but I need to remove all the sample values that have a value on the same row in the control sample. For example, if I want to plot the data for sample2 that does not occur in control1 I have this:

tdfread('file.txt');

ab = a(:,1)./b(:,1);

cb = c(:,1)./b(:,1);

sample2_without_control1 = setdiff(sample2,control1)

scatter3(ab,cb,sample2_without_control1)

What I would aim to plot is the a/b and c/b values but only from row 3 - as this is the only row in which there is data from sample2 but not control1. However setdiff doesn't work - I still need the value to be kept the same (i.e. as 332465) and not altered at all.

(I have thousands of rows of data like this, so I need a way to do this for entire vectors at once please!)

Thank you :)

[Adam]

Can you not just use something like

idx = ( ~isnan( sample2 ) & isnan( control1 ) );

a(idx)./b(idx)

etc?

[Emily Cooper]

I tried that. It works but just gives a binary output in 1 and 0 for where a value occurs/doesnt occur. I need to keep the values as they are, if thats possible. Thank you though!

[Adam]

Which bit gives a binary output?

You will lose the exact positions for your

a(idx)./b(idx)

result, but these can just be reinserted into a full length vector of NaNs or whatever you want for the non-results again afterwards, using the same idx vector for their positions.

[Jan]

I do not understand, what you are asking for. "setdiff doesn't work" - what does this mean? Of course setdiff is working.

You want to get the indices of some rows. So do not use a = setdiff(b,c) but the 2nd output also.

"I need to remove all the sample values that have a value on the same row in the control sample" - this might mean:

keep = not(ismember(sample1, control1) | ismember(sample2, control1));
ab = a(keep,1) ./ b(keep,1);
cb = c(keep,1) ./ b(keep,1);
scatter3(ab, cb, sample2(keep))

Answer 1

[SaiDileep Kola]

Hi Emily,

Please first find the indices of the matrix with control1 values as NaN using the following command:

indices = find(isnan(your_matrix(:,control1)) == 1)

You can also use the 'nansum' comment to add values of sample1 and sample2 in a row-wise fashion:

y = nansum(X,2) %2 is to add along column dim

X is your reformed matrix after removing rows with NAN values and containing only sample1 and sample2 columns

Now you can plot accordingly