# Multidimensional similarity analysis - graph similarity?

I study several chemicals that their effects on different reference surfaces is measured and plotted individually (on a semi-log plot of effect vs concentration of chemical).

Is there a way to find out that chemical A and C are quite similar while A and B are 25% similar using all datapoints on the graphs, and not one representative 50%-effect point (${E}_{50}$)?

Currently the way usually done is to calculate a representative ${E}_{50}$ effect for each plot, and calculate a Pearson coefficient for each two chemicals. However, some may not have ${E}_{50}$ representative points and hence the analysis fails.

Schematics:

To clarify a bit, each chemical has profile I, II, III and IV, and these profiles should be compared 1 to 1, as in the response of Chem A on surface I should be similar to Chem C on surface I, and so on; comparing Chem A on II with Chem C on III is irrelevant.

-