Not a Big Deal, GRCh38: A Semi-Casual Comparison of the New Human Reference Genome

[FractalExpression]

Over christmas GRCh38, the newest human reference genome assembly, was released.Internally we have been using chromosome 20 of human reference builds to benchmark tools and pipelines with datasets. A BWA sequence alignment of the same dataset, generated on a HiSeq 2500, across the last major release GRCh37.69 and the new GRCh38 was performed.

Quantifiably, GRCh38 is very similar to the later GRCh37 releases, showing a change rate of 1 change every 159,558 bases on 37.69 and 1 change every 156,779 bases on 38 for our chromosome 20 dataset.Ts/Tv ratios between the two alignments of the same data across the two references to be quite similar at 0.3527 and 0.3445, respectively. Back of the envelope math seems to give a Δ of +19,359 between GRCh37.69 and 38.
Annotations have a large deviation, to be expected for now.


Read the rest here: http://wp.me/pJHIj-nm

[Michael.James.Clark]

A BWA sequence alignment

Do you think this an appropriate aligner to use to fully take advantage of the new reference genome given its structure?

It seems incorrect to take a tool optimized for the linear GRCh37, apply it to the graph GRCh38, and then state that the difference between 38 and 37 is "not a big deal".

[FractalExpression]

Discussing the primary assemblies themselves, that is not to include alternate regions available for 38, BWA is an appropriate choice- the aligner isn't necessarily trained for GRCh37.

But you're right, how best to utilize the alternate regions, and updating current aligners to take advantage of that is a different question than a quick comparison for anyone considering a primary assembly remap of their data.