This is a test version of Biostars. For the public version, visit https://www.biostars.org.
single cell processed data

I'm asking regarding the following: i want to do meta analysis for single cell data from different studies, some studies used human genome reference hg19 in alignment step of raw data, other studies used human genome 38. so, will this be a problem when i merged studies together ? if yes how can i overcome this ?

single-cell
• 315 views • link
updated by marco.barr • marco.barr • written by Aya • Aya

1 answer

Hi Aya,

Yes, it can be a problem because hg19 and hg38 have different coordinates for the same genes, so raw data isn't directly comparable and the count matrices from different studies may have non-identical gene sets. Also watch the annotation version (e.g., GENCODE v19 vs v38), not just the build.

You can get around this, but it depends on the type of data you have available. If you have raw FASTQs re-align everything with the same reference or If you only have count matrices, harmonize gene IDs (Ensembl, not just symbols) first, then proceed with standard integration to correct the batch.

• 0 views • link
written by marco.barr • marco.barr

Thanks Marco, you mean that also annotation version make a change. how to overcome this ? is there a tool in r to harmonize gene IDs? before harmonizing how to which gene IDs version in the study i work in?

• 0 views • link
written by Aya • Aya

Basically, it depends on where the data came from and how much info is available. If you give me some details I can be of more help to you.

To figure out which annotation version a study used, you could start with the methods section of the paper. If not, check the GEO/ArrayExpress metadata or supplementary files, since the CellRanger version used often hints at the default reference. As a last resort, the gene count in the matrix itself is a clue, in fact older annotations have fewer genes annotated than newer ones.

For harmonizing gene IDs in R, biomaRt is very useful, it lets you convert between gene symbols, Ensembl IDs, and Entrez IDs, and you can even point it to older Ensembl archives (like grch37.ensembl.org) if a study used an older build. On top of that, HGNChelper is handy for updating outdated/renamed gene symbols, and if you ever need to convert actual genomic coordinates rather than just gene IDs, liftOver (via rtracklayer) is a standard tool.

• 0 views • link
written by marco.barr • marco.barr

thanks again, i rechecked the studies i work on them the human genome version is GRCh38 except in one study mentioned GRCh38.93. what does 93 refer to? and now i think i don't i need harmonization genes IDs as i excluded studies with older genome reference version

• 0 views • link
written by Aya • Aya

"93" is the Ensembl release/annotation version, not the genome itself. The GRCh38 sequence stays the same across releases; what changes is the annotation (gene models, added/removed/renamed genes).

So you've avoided the build mismatch (hg19 vs hg38), but if your other studies used a different Ensembl release, there could still be minor annotation differences. Take the gene intersection across studies, if you only lose a handful out of ~20k+ genes, you're fine to just use that intersection. If it's a lot, better to map everything to Ensembl IDs from one release with biomaRt first.

• 0 views • link
written by marco.barr • marco.barr

Log in to answer this question.

More posts like this