I have data for 380 individuals and 29 proteins. I'd like to visualise a couple of things and I have some related questions.
- First, there appears to be some correlation between the individuals based on their protein profile (of those 29 proteins). I just used the Pearson-correlation option in "cor" and than plotted a heatmap using "pheatmap" based on the correlation matrix. For this I used the euclidean distance and the ward.D2 linkage method. Now I have to questions related to this.
- Are these two metrics (distance and linkage) valid for this matrix/datatype? I've been reading a lot online and in this forum. And there isn't much of a consensus it appears. Moreover if I play around with various methods it doesn't really change the overall picture. Intuitively I think taking the average distance and a clustering method based on variances makes more sense - but to be honest I couldn't really put this intuition into words.
- If I use the "cutree_row/col" option, on what is this cutting up the heatmap based?
- I've made the correlation plot, and so I see (when I use the cutree-option) basically three patients groups. Now, for the second part. I'd like to plot the heatmap of the protein levels along side the correlation plot - I know this may be difficult in R. So let's just say, I want to plot a heatmap of the proteinlevel-individuals matrix with in the rows the patients in the order of the correlation-heatmap en than cluster the columns of this matrix (i.e. the proteins). That way I'd be able to see which groups of proteins in which patients are causing the clustering of the patients. Similar to Figure 1 from this paper: http://www.nejm.org/doi/pdf/10.1056/NEJMoa040465. I just don't need the data on the diagonal - we can do this with the annotation option pheatmap.
- Do you have a suggestion as to how to accomplish this?
I've come a long way: I've managed to get a correlation heatmap-plot, and a protein level heatmap-plot. I just need to get a little help to steer me in the right direction for this combinatory effort.