Concern on tumor samples with mixed cell population

[waihungh]

I'm working on cancer exome and transcriptome sequencing project. Sequencing data on tumor samples are derived from a mixed population of cells. Is there any good way to handle this issue?

[krobison]

MuTect (and many others) published to deal with this

http://www.broadinstitute.org/cancer/cga/mutect

[msalm]

For SNV calling, I would also highly recommend VarScan2 (http://www.ncbi.nlm.nih.gov/pubmed/22300766). Others include SomaticSniper (http://www.ncbi.nlm.nih.gov/pubmed/22155872) and Strelka (http://www.ncbi.nlm.nih.gov/pubmed/22581179). In all cases, a matched reference tissue is required.
It is also important to be aware of/accomodate varying sources of contamination. For example, contamination levels (normal sample in the tumour sample and vice-versa) vary significantly between liquid and solid tumours, and the sampling schema. Unfortunately though there is no magic bullet, yet.

[Richard Finney]

Magic bullet might be single cell sequencing. The results of edge tumor vs. core samples would be interesting as would, of course, differences between adjacent malignant cells. After talking with leukemia researchers, I'm under the impression that they do , somehow, sort the cells into "evil" vs. (presumably) "not so evil" and harvest the evil ones.

[krobison]

Yes, with leukemias that might actually be feasible. For solid tumors, really challenging -- the "good" and "evil" are so intertwined it is hard to pick them apart. Probably will need single-cell sequencing that is cheap enough that you can just sequence many & then decide downstream which are "good" and "evil".

If you look closely at the original Mardis lab publication on leukemia WGS (if not first, one of the early ones), they actually encountered the converse problem -- because the patient had a very severe blast crisis, the normal samples were significantly contaminated with leukemic cells!