Normal number of variants to lose during liftover: GRCh38 to hg19

Question

[JourneyToAbyss]

I am using the 1000 genome files, provided by the plink 2 author, and using liftover to convert the positions to hg19. To do this, I am first converting to a VCF file, sorting with bcftools, and then using CrossMap to perform the liftover. Of 70,692,015 (only chr1-22 and XY included), 16,559,055 failed to map.

Is this to be expected? Or is something suspect with my pipeline?

[Pierre Lindenbaum]

The liftover tool should provide a log about the variant that failed the process

[JourneyToAbyss]

It did and nothing stood out with a quick look. I am curious about other people's experience with liftover and the expected loss in variants from GRCh38 to hg19.

[Pierre Lindenbaum]

how do you liftover ?

[JourneyToAbyss]

I used CrossMap to convert the VCF file, which I generated from plink2's 1000-genome files.

https://crossmap.sourceforge.net/#convert-vcf-format-files

As this program uses the reference genome (not the overchain), maybe this isn't appropriate. I am going to attempt using the overchain and merely update the positions.

[DBScan]

You should definitely provide the chain file AND the reference genome to CrossMap, how else is CrossMap supposed the know how to lift the variants?

[Giulio Genovese]

Using a version of the high coverage 1000 Genomes project callset with 63,993,411 non-singleton bi-allelic SNVs and 9,459,059 non-singleton bi-allelic together indels with the hg38ToHg19.over.chain.gz UCSC chain file, I get 916,020 SNVs and 63,685 indels dropped using CrossMap/VCF while I get 872,258 SNVs and 55,590 indels dropped using BCFtools/liftover so what you are seeing is not expected