I found our lab stats package (Graphpad Prism), and read bits of its very detailed and user-friendly help files. Then I pasted in my data and did some two-way ANOVAs. Then I read the help files some more and decided I should have done 1-way ANOVAs with 'repeated measures'. (That tells the software to consider all the values in the same row as belonging together.)
I first analyzed each group of tripeptides separately (the blue ones as one dataset, then the pink, then the yellow). The blue set had significant differences between the columns in the ANOVA (p=0.01). It also had significant differences between the bright-blue column and all other columns by Tukey's multiple comparison test. I used this rather then the Bonferroni test but I'm not sure which would have been more appropriate - I think this is less sensitive than the experiment deserves, because I had specific comparisons in mind from the start. The pink set had not-quite significant differences (p=0.058) in the ANOVA, and not-significant differences between any pairs of columns in the Tukey's test. The yellow data had very significant differences between the columns in the ANOVA (p<0.0001), and significant differences between the bright-yellow column and all other columns by the Tukey's test.
I then rearranged the data, putting the bright-colour data all in the same column (the 'cognate-proteome' column), and the pale-colour data in the other columns. This let me analyze all three colours together. The ANOVA found very significant differences between the columns (p<0.0001) and the Tukey's test found significant differences between the cognate-proteome column and all the other columns.
The control comparisons ( using reversed tripeptides) were never significant.
So now I can add a sentence to the manuscript, reporting that the effects shown in Figure 1 are statistically significant.
A new kind of problem
12 hours ago in RRResearch