Closed dhimmel closed 7 years ago
Do we want to split these into multiple records when creating chromosome-symbol-mapper.tsv
?
The X|Y ones are in the pseudoautosomal regions of the X and Y chromosomes. I would not be worried about those and would not split them. These should be retained.
OMS looks like a susceptibility "gene." It's not really a molecular entity, just a set of association signal regions: https://www.ncbi.nlm.nih.gov/gene/?term=619538 . This could be dropped.
The others appear to be on unplaced scaffolds: https://www.ncbi.nlm.nih.gov/gene/?term=105379561 For now, I would probably drop these too, though it's not as clear that these should be dropped as it is for something like OMS.
By the way - if someone picks a gene on the X or Y chromosomes other than those in the X|Y set, you may want to automatically detect it and build separate male and female classifiers. This is a strong signal in expression data, even for unsupervised learning.
The X|Y ones are in the pseudoautosomal regions of the X and Y chromosomes. I would not be worried about those and would not split them. These should be retained.
@cgreene, we're including this file to map PANCAN_mutation
(as has been done by https://github.com/cognoma/cancer-data/pull/12). Therefore I looked how several of the pseudoautosomal genes were coded in that dataset:
sample | chr | start | end | reference | alt | gene | effect | DNA_VAF | RNA_VAF | Amino_Acid_ChangeTCGA-BH-A18P-01 | chrX | 1508405 | 1508405 | G | A | SLC25A6 | Silent | p.F109 | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
TCGA-BH-A18P-01 | chrX | 1508405 | 1508405 | G | A | SLC25A6 | Silent | p.F109 | |||||||||||
TCGA-06-5416-01 | chrX | 1746629 | 1746629 | C | T | ASMT | Silent | 0.276457883369 | |||||||||||
TCGA-CD-A4MI-01 | chrX | 1413254 | 1413254 | G | A | CSF2RA | Missense_Mutation | p.R227H |
So unless we split chromosomes, these genes will not map. I propose splitting with an optional step to include the unsplit rows. Therefore the top row would yield:
entrez_gene_id | symbol | description | chromosome | gene_type | synonyms | |||||
---|---|---|---|---|---|---|---|---|---|---|
263 | AMD1P2 | adenosylmethionine decarboxylase 1 pseudogene 2 | X | pseudo | AMD | AMD2 | AMDP1 | AMDPX | AMDPY | |
263 | AMD1P2 | adenosylmethionine decarboxylase 1 pseudogene 2 | Y | pseudo | AMD | AMD2 | AMDP1 | AMDPX | AMDPY | |
263 | AMD1P2 | adenosylmethionine decarboxylase 1 pseudogene 2 | X | Y | pseudo | AMD | AMD2 | AMDP1 | AMDPX | AMDPY |
Do you think we should even keep the last row?
By the way - if someone picks a gene on the X or Y chromosomes other than those in the X|Y set, you may want to automatically detect it and build separate male and female classifiers. This is a strong signal in expression data, even for unsupervised learning.
May want to open an issue in machine-learning.
The others appear to be on unplaced scaffolds. For now, I would probably drop these too, though it's not as clear that these should be dropped as it is for something like OMS.
Okay leaving these in will in effect drop them because the resource being mapped won't have that symbol-chromosome combination. No need to explicitly filter.
@dhimmel : for the purposes of having a resource to connect potential symbols with chromosomes, I think that retaining at least the first two lines would make the most sense. Maybe the third - I don't know how many resources use X|Y for these regions. I don't see the harm in it, so I guess my inclination would be to leave it as well.
@cgreene in b64fcb4261005cf717d5d5ef4e03540e4a1f361e I retained all three lines.
However, there is another issue -- some genes have no chromosome. For example:
These genes all have type unknown
, so I'm guessing the inability to map them will not be a big deal. In fact they most likely won't be in our datasets?
These - to my knowledge - come from the expectation that there exists a gene for the disease but nobody has found it. They aren't really meaningful molecular entities and expect that you won't see them in practice.
What does it mean for a gene to have multiple chromosomes? Here are all the genes from
genes.tsv
that exhibited multiple chromosomes: