What is the criteria to select a threshold for any somatic mutation in tumor to observe a functional consequence?

Question

[abhiramitbalaji]

We have the RNAseq data of oral cancer(OSCC) tumor samples without matched normal. The filtered reads were of high base quality and more than 95% of base matching was considered for mutation calling. We have found several mutations in these reads (with reference to hg38). For a particular position, we have 2% of reads having mutations. For others, we have less than 2% of the reads showing mutations. On what basis do we validate the functionality of the mutation? Is there a specific range (number of reads having the mutation) which implies that the mutation has a consequence on the function.

[abhiramitbalaji]

It is true that we cannot ascertain the protein function by mutations in RNAseq. However, we should know which mutation to consider for experimental validation. This selection of the mutation must be based on some criteria, one of them that we want to consider, is its frequency. So is there a range which can tell us valid mutations and help in ruling out transcription/sequencing errors?

Answer 1

[ATpoint]

There is (by best knowledge) no connection between read numbers (both total and relative) as well as base quality etc. towards the functional impact of the mutation. Sequencing is simply a diagnostic or analytic tool and its quality and output is not related to what is qualitatively going on in the cell in this regard. First of all, even if the mutation is high-quality and real, this does not imply alterations of the protein or RNA that is produced from the respective gene. Second, even if e.g. an open reading frame is affected, you would need to show by experiment that the alteration indeed has a relevant phenotype. There are approaches to predict functional impact of mutations, but still, this is prediction, not validation. Be also aware that the mutations you find without matched normals can be germline. The hg38 is only a reference, though an excellent one, but still only a reference that is not representative on a per-base scale for any individual. I suggest you at least filter against common polymorphisms based on dbSNP and TOPMED VCFs.