Translating between various gene IDs can be a bit of a nightmare. I've spent many days trying to disambiguate datasets with different gene IDs, but I have not yet come across a completely generalizable and reliable method of doing this. Perhaps someone that is intimately familiar with the evolution of various gene annotations could chime in and enlighten us both.

However, I can say definitively that you should NOT merge the expression values of genes that have the same symbol. Gene symbols seem to shuffle around more haphazardly than Ensembl IDs or Entrez IDs, and so it is pretty common for completely different genes to be labeled with the same gene symbol depending on which gene annotation builds are being used.

Which brings me to my next point. You will save yourself some headaches if you know what gene annotation build was used in originally mapping your RNA-seq reads to Ensembl IDs. Use the same build to translate from Ensembl IDs to gene symbols. This won't necessarily get rid of all issues, but it will be a big help.

Next, be aware that Ensembl IDs that start with "ENSGR" indicate genes on the Y chromosome. Therefore, you can have some genes with matching gene symbols that come from the Y vs. the X chromosomes.

Another trick I use is to first filter my dataset to get rid of unexpressed or very lowly expressed genes. There are a lot of poorly validated non-coding genes that seem to be a lot messier and more transient in gene annotation builds, which can make settling on their correct gene symbol challenging. Thankfully, many of these are not well expressed in a given dataset, so by first filtering by expression you can save yourself that headache.

Speaking of ncRNAs, many are not typically polyadenylated, which means that if your RNA-seq library was enriched for polyA+ transcripts (i.e. mRNA-seq), then you cannot really trust the accuracy of the expression measurements for these polyA- ncRNAs. In such cases, I will also typically throw out snRNA, snoRNAs, rRNAs, miRNAs, etc. unless I have a particular interest in a ncRNA that I believe is, at least initially, polyadenylated.

In my experience these various tricks can help reduce the number of ambiguities, but I'm pretty much always left with some genes that resist a definitive translation between gene IDs. At that point I tend to just resort to manual disambiguation by looking up the problematic genes on NCBI gene (https://www.ncbi.nlm.nih.gov/gene/) and Ensembl browser (https://www.ensembl.org). Usually a little digging around will reveal why there is ambiguity and how you should resolve the issue (e.g. use the most up-to-date gene symbol, or throw it away if it is a discontinued entry, etc.).

Sorry I don't have a simple answer for you, but hopefully this helps you on your way.

The Ensembl team are very responsive to their helpdesk, I recommend you ask there: [email protected]

However, a brief look suggests that Ensembl has provided different IDs because they map to different chromosomes: one to the main chromosome and the other to Haplotypic regions

For example I have selected a case in case of human:

If you search NCR1 gene in ensemble you will get the following hit:

ENSG00000189430 NCR1

ENSG00000273535 NCR1

ENSG00000275156 NCR1

ENSG00000275521 NCR1

ENSG00000275637 NCR1

ENSG00000276450 NCR1

ENSG00000277334 NCR1

ENSG00000277442 NCR1

ENSG00000278025 NCR1

ENSG00000278362 NCR1

ENSG00000284113 NCR1

ENSG00000284208 NCR1

But if you look for their chromosomal locations you will get something like this:

Gene stable ID Gene name Chromosome/scaffold name

ENSG00000275156 NCR1 CHR_HSCHR19_4_CTG3_1

ENSG00000277334 NCR1 CHR_HSCHR19KIR_FH06_BA1_HAP_CTG3_1

ENSG00000278362 NCR1 CHR_HSCHR19LRC_COX2_CTG3_1

ENSG00000277442 NCR1 CHR_HSCHR19KIR_ABC08_A1_HAP_CTG3_1

ENSG00000276450 NCR1 CHR_HSCHR19KIR_FH05_A_HAP_CTG3_1

ENSG00000278025 NCR1 CHR_HSCHR19KIR_FH05_B_HAP_CTG3_1

ENSG00000275521 NCR1 CHR_HSCHR19KIR_FH06_A_HAP_CTG3_1

ENSG00000189430 NCR1 19

ENSG00000275637 NCR1 CHR_HSCHR19KIR_FH08_A_HAP_CTG3_1

ENSG00000273535 NCR1 CHR_HSCHR19LRC_PGF1_CTG3_1

ENSG00000284113 NCR1 CHR_HSCHR19KIR_CA01-TB01_CTG3_1

ENSG00000284208 NCR1 CHR_HSCHR19KIR_HG2393_CTG3_1

As you can see only one highlighted NCR1 gene is coming from main chromosome others are coming from haplotypic regions.

Now what is a haplotypic region, so as per the ensemble's FB post:

When you’re building a genome assembly for a single species with very diverse populations or strains, it is difficult to capture all of that lovely variability in a single reference sequence. That’s where haplotypic regions come in.

Haplotyptic regions are defined by the Genome Reference Consortium (GRC) and are visible on all their genome assemblies for human, mouse, and zebrafish. They can also be called ‘alternate locus’, ‘partial chromosomes’, and ‘alternate alleles’ or ‘assembly exceptions’ (although this definition also includes patches). Most often these are added because through sequencing a variety of individuals different versions of regions are identified. These regions can have small sequence differences, contain different gene structures or different genes entirely, or contain genes in a different order compared to the reference genome assembly.

We display haplotypic regions in Ensembl as red stripes overlapping the chromosome, and you can easily compare these with the reference assembly. The haplotypes themselves have a unique identifier, which links them to the region on the primary assembly. In human these start CHR_HSCHR followed by the chromosome number. (https://www.facebook.com/151892474871670/posts/tipoftheweek-seeing-red-in-the-genome-haplotypes-and-haplotypic-regions-when-you/2177783548949209/)

Now to your question

How to deal with the multiple mappings in the dataset?

I will suggest that you take only that id into consideration which is coming from main chromosome, untill unless you dont want to look variability of gene in haplotypes. However another solution is that you can consider only that ensemble id which is having maximum expression as there are very chances that your sample is close to that haplotype intead of main reference.

I agree to Nitesh Kumar Sharma on accepting the ENSG ID which has the highest gene expression. I have recently used the same to reduce ambiguity in the data. Also, this ID multi-mapping issue is more seen in older reference annotations. My suggestion is to use only the Primary annotation file (I prefer Gencode Primary Genome assembly and annotations) for any analysis, rather than the one with Haplotype regions, to avoid this.

I'm not an expert in the field, but when you have scientific data these inevitably flow into the famous databases called Big Data, from which we obtain new knowledge through new scientific models. A fundamental condition is that the individual data are accurate, reliable and non-redundant. When we load different functions on a single reference (directly or through annotations) we do nothing but make the context inaccurate, that is, we are polluting Big Data. Therefore, we expect that the new models that imply this specific knowledge may have a certain percentage of inaccuracy, depending on the quantity of pollution. I don't know if I understood your problem well, but the problem of logical pollution already exists.

I would suggest to translate ENSEMBLE IDs into GENE IDs (official gene names) using the AnnotationDbi package in R (https://bioconductor.org/packages/release/bioc/html/AnnotationDbi.html)

I have not faced such problem yet but I expect that using some functions of this package (such as mapIds), all the ENSEMBLE IDs referring to the same gene will be simply converted into a single GENE SYMBOL.

Rather than doing it through R, I have no clues how else this problem can be fixed.

Best of luck!

It would be very helpful if you can also provide your views on how genes having multiple ensembl gene ids should be treated while differential analysis by various programs. ENSEMBL gene ID from primary assembly or haplotype showing highest read counts/expression or sum of read counts/expression from all ensembl gene ids for the same gene (symbol), what should I choose to deal with such a situation in best possible ways.?

Read about Haplotypes

You can convert Ensembl gen IDs to gene symbol by creating a query with the Python mygene library (https://pypi.org/project/mygene/).

Sanket Desai Hi Sanket, thank you for you answer! I've noticed that even using primary assembly, there are situations when multiple gene ids map to the same gene name (for exampe, one can come from chromosome and one can come from unlocalized sequence). Would you give some recommendations on how to deal with such situations, for example whether I should add up their read counts?

Changfeng Chen, I would not expect two genes (one from chromosome and other from unlocalized) to have same official gene name (HGNC). Although if you come across any such name, please post. In a typical analysis, I usually exclude all the genes which map to unlocalized contigs. Hence, I am not aware of any such gene names.

Sanket Desai Hi Sanket, I am using gencode mouse genome mm10 (v25) and there's ~100 such gene names that have multiple ids pointing to it. A few such examples: Ccl21a, Ccl21b, Ccl19, Ccl27a, Il11ra2.