Statistically it is valid to average the three triplicates since they are three different extractions. It might be better for t-test or anova test purposes, not so sure that the PCA will benefit greatly from it unless the replicate data is very noisy. Unfortunately I don't know the MarkerLynx software so can't help you there.

Thank you Sigurður Smárason!

Hello Chris

From my point of view, I would have the following considerations:

1. If you use PCA for comparison, it would be nice to build one model also withouth averaging. Since PCA shows you similarities, you would expect to have the 3 points for the same plant as close as possible from one another.

2. The problem, in this case, is not only about averaging the 3 replicates, but also that you have too many groups (40!). Due to that, seeing patterns in PCA can be really difficult and not clear. As Sigurður suggested, I also believe that univariate statistics would be more interesting in your case. You can perform Heat Map Analysis on samples, and also calculate correlation coefficients in order to have a better view of the similarities and differences among the 40 types of plants.

As Sigurður, I do not have experience with MarkerLynx, but these univariate analysis I suggested can be performed online for free in Metaboanalyst

Hope this helps you.