Considering that your cell array only contains one matrix I don't see why you're bothering with a cell and all these cellfun.
Anyway, the problem is simple: columns 6987 to 7984 of that matrix are just NaNs, so of course when you sum across the columns you get NaN. Possibly you could fix that by adding the 'omitnan' option to your sums but most likely the proper fix is for you to find out why there are NaNs in the first place.
Note that even if there were more than one matrix in your cell array, your comp_gm_fv could be simplified. The first cellfun doesn't need 'UniformOutput', false and there's no point in transposing a matrix before summing across the columns, simply sum across the rows instead:
function [gm, gv] = comp_gm_gv(data)
% computes the global mean and variance of data
nframes = sum(cellfun(@(x) size(x, 2), data));
gm = cellfun(@(x) sum(x, 2), data, 'UniformOutput', false);
%gm = cellfun(@(x) sum(x, 2, 'omitnan'), data, 'UniformOutput', false);
gm = sum(cell2mat(gm), 1)/nframes;
gv = cellfun(@(x) sum(bsxfun(@minus, x, gm).^2, 2), data, 'UniformOutput', false);
%gv = cellfun(@(x) sum(bsxfun(@minus, x, gm).^2, 2, 'omitnan'), data, 'UniformOutput', false);
gv = sum(cell2mat(gv), 1)/( nframes - 1 );
end
Commented lines are with the 'omitnan' option.
