Is removing samples based on clustering for downstream analysis a right choice?

Question

I'm using TCGA Lung cancer data. I'm interested in doing differential analysis between Lung vs Normal. Before DEA, to check the distance between each pair of samples I plotted an MDS plot:

In this, I see some Tumor samples are clustered with Normals. As this is a two dimensional plot, it is a little hard to say.

So, I made a clustering heatmap with the top 10% genes:

(Where brown color is Normal and black color is Tumor). I also see some Normals clustering with Tumor here, and some Tumor are with the Normals.

My question is: Should I exclude those tumor samples which are clustering with the normals for differential analysis? Is it a good idea to remove samples based on clustering?

Note: I didn't see any technical bias of those samples

Answer 1

I would be very hesitant to blindly exclude those samples based on the clustering. Check to see if the clusters actually denote some sort of batch effect, since it's not like all of the TCGA datasets were processed at the same place or time. Check for clinical covariates too. Also, do a PCA and check the genes with high loading on the relevant PC. If it appears that the genes are indicators for cancerous state, then it's likely the samples are just mislabeled and you can reasonably exclude them. Otherwise keep them in so your results better represent the underlying biological variability.

BTW, your heatmap and MDS plots are what I'd expect from a cancer dataset. You occasionally see a tumour with the normal samples and vice versa, since the problematic genes aren't always in the top 10% and there is likely some heterogeneity in the cancer samples.