How drastic would the result be if I use fasta or reference assembly from ucsc and gtf from gencode?

Question

There are difference annotation file for UCSC and gencode.

But if I use the reference assembly from UCSC and the GTF from Genocode or vice versa would my downstream results would be wrong?

Answer 1

Never use genomes or annotations from UCSC, they're poorly versioned and only recently with mouse and human have they even included all of the contigs. For fasta/GTF files from early in the GRCh38 release, you can tell whether you're using UCSC or Gencode by the presence/absence of _random contigs, which will only exist for UCSC. These were mostly later split into the actual contigs, so recent download from UCSC should more closely match what you find at Gencode/Ensembl. Further, that time predated UCSC beginning to adopt Gencode's vastly superior annotations, so if your GTF file has instances where the same gene ID is on either multiple strands or multiple chromosomes (this is obviously biologically impossible) then you have a UCSC GTF file.

In general, with early GRCh38 releases your only real issues will be with _random contigs, which are a minority of the genome and don't have all that many genes. But really, you should be keeping track of the sources of your files and ensuring that they're compatible.

Update: I should expand a bit on my "Never use genomes or annotations from UCSC" comment. In point of fact the genomes themselves aren't so terrible. Early on UCSC had the bad habit of concatenating contigs together into _random "chromosomes", but they seem to have mostly kicked that habit as of late. Note, however, that there are no versions for their genomes. Since reference genomes continue to get updates over time (mostly through the addition of patches) the lack of actual versions means you have to manually check if a recently downloaded file matches what may have been downloaded either previously or by someone else. This has obvious consequences when it comes to reproducibility. The same issue occurs for annotations from UCSC, but they have the additional problem of historically having biologically incoherent concepts of genes. That is, they will contain the same gene in multiple places with multiple orientations, which will break many tools in both obvious and completely unclear ways. For example, DEXSeq will simply break with an error message if given a UCSC annotation, since they break biological plausibility. If you were to use these annotations files with deepTools, you wouldn't get an error message, but the resulting output would be only partial, due to the biologically impossible annotation effectively corrupting most obvious ways of storing annotation data in a data structure (i.e., you can no longer treat IDs as unique). This could have downstream ramifications on biological interpretation of results.

Answer 2

The main difference is in the way the chromosomes are named - UCSC uses the "chr" prefix (so chromosome 1 is "chr1") while in gencode the "chr" isn't used (so chromosome 1 is just "1"). Depending on your use case, this can obviously cause problems - if you're trying to match a locus (e.g. from gencode 1:1000002) between them, whatever tool you use is going to be looking in your aligned data for "1:1000002", but in your aligned data, it'll be named "chr1:1000002", so it won't match the two up.