Dear all,
The dataset is a matrix of enzymatic activities obtained by fluorescence measurements. 16 enzymes (10 measurements per well, 3 well per enzyme. I averaged the triplicates measurements) were measured for 5 different cell lines/groups which I would like ideally to represent on a low dimensional space. PCA and PCA-DA are possible choices, but since some cell line are related (specifically, 2 cells lines belong to the same isogenic line, etc.). Therefore, the basic PCA-DA might be not the best suitable choice to distinguish these cell lines. Also I should maybe not averaging the data at that point (but just the 16 enzymes are fine). I saw that in Ecology NMDS is often used, would that be more robust for related cell lines. Any idea?
Thank you for your suggestions,
Arno
(EDIT: Question was answered, thank you all for your comments!)
I think that I am able to suggest you only for the last part of your question:
*Please don't take a mean value because you will lose useful information!
For other parts, I don't think I can help you.
Hi,
First of all, do not take the mean value. You will lose important information, because you just have a small number of replicates.
Secondly, you should get familiar with your data. Perform some exploratory data analysis, plot some graphs, check some statistics of your data, because it will help you understand what your data will be able to tell you.
Now, to check similarity you can of course go with clustering. There are several methods for checking similarity, and what to use depends much on your familiarity with the methods. The book "Pattern Recognition" by Sergio Theodoridis has a lot of possibilities for clustering with the theory behind them very well explained.
To understand whether you need to standardize or normalize you should take a look at a book on Exploratory Data Analysis. That said, I would say that if you have one variable with everything in the same range, there may not be any need. Note that some methods (e.g. some estimation techniques or sometimes the software itself) may require you to standardize/normalize your data.
You can always take a look on ANOVA methods (from Wikipedia):
Analysis of variance (ANOVA) is a collection of statistical models used in order to analyze the differences between group means and their associated procedures (such as "variation" among and between groups), developed by R. A. Fisher.
Hope this helps,
Cheers, Daniel
Hi Daniel, thanks for the information. I already processed the data through some graphs, statistics, etc. so I do have a good idea on whats going on. Clustering looked good now I wonder about NMDS which is often used in literature on enzymatic analyses when there are many factors / variable take into account, which is not really my case here. The software (JMP SAS institute) propose a check box to normalize or not the data for clustering.
Well, do your clustering results change if you normalize, i.e. if you check the box? If it changes radically, that could tell you there is something interesting/strange about your data.
Not really very familiar with NMDS (non metric multidimensional scaling) but I would think it is primarily used to reduce dimensionality with purpose of visualization, for example. I don't see how that would help in your case.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size