LiftOver

From Genome Analysis Wiki
Jump to: navigation, search

LiftOver is a necesary step to bring all genetical analysis to the same reference build. LiftOver can have three use cases:

(1) Convert genome position from one genome assembly to another genome assembly

In most scenarios, we have known genome positions in NCBI build 36 (UCSC hg 18) and hope to lift them over to NCBI build 37 (UCSC hg19).

(2) Convert dbSNP rs number from one build to another

(3) Convert both genome position and dbSNP rs number over different versions

It is likely to see such type of data in Merlin/PLINK format.

We will explain the work flow for the above three cases. In the rest of this article, our example is to lift over from lower/older build to newer/higher build, as it is the common practice.

Using different tools, liftOver can be easy. For example, UCSC liftOver tool is able to lift BED format file between builds. With our customized scripts, we can also lift rsNumber and Merlin/PLINK data files.

Contents

Lift genome positions

Genome positions are best represented in BED format. UCSC provides tools to convert BED file from one genome assembly to another.

Binary liftOver tool

We need liftOver binary from UCSC and hg18 to hg 19 chain file.

Provide BED format file (e.g. input.bed)

NOTE: Use the 'chr' before each chromosome name

chr1    743267  743268  rs3115860
chr1    766408  766409  rs12124819
chr1    773885  773886  rs17160939

Run liftOver:

   liftOver input.bed hg18ToHg19.over.chain.gz output.bed unlifted.bed

unlifted.bed file will contain all genome positions that cannot be lifted. The reason for that varies. See Various reasons that lift over could fail

Web interface

Alternatively, you can lift over BED file in web interface at: Link Web interface can tell you why some genome position cannot be lifted if you click "Explain failure messages"

Lift dbSNP rs numbers

rs number is release by dbSNP. UCSC also make their own copy from each dbSNP version. Be aware that the same version of dbSNP from these two centers are not the same. When we convert rs number from lower version to higher version, there are practically two ways.

Use RsMergeArch and SNPHistory

It is necessary to quickly summarize how dbSNP merge/re-activate rs number:

  1. when different rs number are found to refer to the same SNP, then higher rs number will be merged to lower rs number, and the merging will be recorded in RsMergeArch.bcp.gz.
  2. when rs number have to be retracted, rs number will be recorded in SNPHistory.bcp.gz
  3. a retracted SNP can be re-activated in SNPHistory.bcp.gz by adding comment

With the above in mind, we are able to combine these two tables to obtain the relationship between older rs number and new rs number. We have developed a script (for internal use), named liftRsNumber.py for lift rs numbers between builds. This scripts require RsMergeArch.bcp.gz and SNPHistory.bcp.gz, those can be found in Resources.

Example input:

3000
3001
3002

Command:

python liftRsNumber.py input.rs > output.rs

Example output:

unchanged       3000
lifted  2032
unchanged       3002

Lift Merlin/PLINK format

In Merlin/PLINK .map files, each line contains both genome position and dbSNP rs number. Our goal here is to use both information to liftOver as many position as possible. There are 3 methods to liftOver and we recommend the first 2 method. The first method is common and applicable in most cases, and in our observations it lifts the most genome positions, however, it does not reflect the rs number change between different dbSNP builds. The second method is more robust in the sense that each lifted rs number has valid genome position, as it lift over old rs number as the first step by using dbSNP data. The third method is not straigtforward, and we just briefly mention it.

Lift Merlin format

PLINK format and Merlin format are nearly identical. The difference is that Merlin .map file have 4 columns. We will show the lift over procedure for PLINK format, then you can use:

 awk '{print $1,$2,"\t",$3;}' PLINK.map > Merlin.map

to obtain Merlin .map file.

Lift PLINK format

PLINK format usually referrs to .ped and .map files.


Method 1

We mainly use UCSC LiftOver binary tools to help lift over. We have a script liftMap.py, however, it is recommended to understand the job step by step:

(1) Convert .map to .bed file

By rearrange columns of .map file, we obtain a standard BED format file.

(2) LiftOver .bed file

Use method mentioned above to convert .bed file from one build to another.

(3) Convert lifted .bed file back to .map file

Rearrange column of .map file to obtain .bed file in the new build.

(4) Modify .ped file

.ped file have many column files. By convention, the first six columns are family_id, person_id, father_id, mother_id, sex, and phenotype. From the 7th column, there are two letters/digits representing a genotype at the certain marker. In step (2), as some genome positions cannot be lifted to the new version, we need to drop their corresponding columns from .ped file to keep consistency. You can use PLINK --exclude those snps, see Remove a subset of SNPs.

(5) (optionally) change the rs number in the .map file

Similar to the human reference build, dbSNP also have different versions. You may consider change rs number from the old dbSNP version to new dbSNP version depending on your needs. Such steps are described in Lift dbSNP rs numbers.

Method 2

The idea is to use LiftRsNumber.py to convert old rs number to new rs number, use the data file b132_SNPChrPosOnRef_37_1.bcp.gz (a data file containing each dbSNP and its positions in NCBI build 37), and adjust .map and .ped files accordingly.

(1) Extract and lift rs numbers

Use the tools LiftRsNumber.py to lift the rs number in the map file from old build to new build.

(2) Lookup SNP positions from rs number

dbSNP provides a file b132_SNPChrPosOnRef_37_1.bcp.gz which contains rsNumber, chromosome and its position. Use this file along with the new rsNumber obtained in the first step. In practice, some rs numbers do not exist in build 132, or not suitable to be considered ( e.g. they do not reside on human reference, or they are mapped to multiple locations, these scenarios are noted by the chromosome column with values like "AltOnly", "Multi", "NotOn", "PAR", "Un"), we can drop them in the liftover procedure. We will obtain the rs number and its position in the new build after this step.

(3) Lift .map file and .ped file

To lift over .map files, we can scan its content line by line, and skip those not lifted rs number. Accordingly, it is necessary to drop the un-lifted SNP genotypes from .ped file.

Method 3

NCBI dbSNP team has provided a provisional map for converting the genome position of a larget set dbSNP from NCBI build 36 to NCBI build 37. In the second step, we have obtained unlifted genome positions, so we can try to use the table to convert those unlfted dbSNPs. After this step, there are still some SNPs that cannot be lifted, as they are mostly located on non-reference chromosome. Note: due to the limitation of the provisional map, some SNP can have multiple locations. For example, we cannot convert rs10000199 to chromosome 4, 7, 12.

10000199	A/G	4	166142415	166142415	2	3	G	+	4	165922965	165922965	2	3	G	+
10000199	A/G	7	4589694	4589694	2	3	C	-	7	4623168	4623168	2	3	C	-
10000199	A/G	12	57008620	57008620	2	3	C	-	12	58722353	58722353	2	3	C	-
10000199	A/G	5	156018406	156018406	2	3	C	-	5	156085828	156085828	2	3	C	-

We can dissect this method into steps:

(1) Remove invalid record in dbSNP provisional map.

Provisional map have duplicated rs number or the chromsome in the new build can be "Unable to map"(UN), we need to clean this table.

(2) Use provisional map to update .map file

By joining .map file and this provisional map, we can obtain the new genome position in the new build. Note: provisional map uses 1-based chromosomal index. Things will get tricker if we want to lift non-single site SNP e.g. AA/GG Since provisional map provides a range in this case, it is necessary to know the genome position of that single base provided in the .map file, and then we can look up the table, so it is not straigtforward.

(3) Adjust .map and .ped file

For those lifted dbSNP, we need to keep them in the .map files, otherwise, we need to delete them. Accordingly, we need to deleted SNP genotypes for those cannot be lifted.

Various reasons that lift over can fail

Genome position cannot be lifted

When a SNP resides in a contig that only exists in older reference build, liftOver cannot give it new genome.

You can try the following SNP (in BED format) in UCSC online liftOver site:

20 56737667 56737668 rs1073519

The error message will be: "Sequence intersects no chains"

SNP in higher build are located in non-referernce assembly

Some SNP are not in autosomes or sex chromosomes in NCBI build 37. dbSNP does not include them. You cannot use dbSNP database to lookup its genome position by rs number.

Take rs1006094 as an example: In NCBI dbSNP webpage, this SNP is reported as "Mapped unambiguously on non-reference assembly only" Thus it is probably not very useful to lift this SNP.

rs number changed in newer dbSNP build

It is possible that new dbSNP build does not have certain rs numbers. When dbSNp release new build, higher rs number may be merged to lower rs number because of those rs numbers are actually the same SNP. This merge process can be complicate. For short description, see Use RsMergeArch and SNPHistory . For detail, see:

Finding Specific Data in dbSNP’s FTP Files

Merging RefSNP Numbers and RefSNP Clusters

For example:

rs3001 has merged to rs2032.

Different dbSNP build

NCBI released dbSNP132 (VCF format), and UCSC also have their version of dbSNP132 (plain txt). The two database files differ not only in file format, but in content.

For NCBI release, its release will not contain:

  • SNPs listed as microsatellites or named variations
  • SNPs with multibyte alleles and unknown (N) adjacent base pairs
  • SNPs that are not mapped on the reference genome (GRCh37)

For UCSC release, see UCSC dbSNP track note

Use rs1054140 as an example:

NCBI dbSNP website gives 1 location: Link

NCBI dbSNP VCF file has NO record.

UCSC genome browser website gives 2 locations: Link

UCSC dbSNP file give 2 locations:

721     chr10   17842693        17842694        rs1054140       0       +       T       T       A/T     genomic single  by-cluster,by-submitter ...
723     chr10   18089681        18089682        rs1054140       0       +       T       T       A/T     genomic single  by-cluster,by-submitter ...

Resources


Third party contribution

Acknowledge

  • Hyun: provides sample liftOver tool: [/net/wonderland/home/hmkang/prj/Sardinia/MetaboChip/scripts/j01-liftover-metabochip-positions.pl]
  • Alex: careful examines of 0-based index in UCSC data file
  • Adrian: explaination of SNPs omitted in NCBI dbSNP file
  • Goncalo: all other supports

Questions and Comments

Please contact Xiaowei Zhan.