Missing value for SVCLAIM in DEL/DUP in VCFv4.4: clarification and addition

[khetherin]

We are seeking clarification on the validity of a missing value ('.') in SVCLAIM for DEL/DUP.
The reason we ask this: Our use case is that we have legacy data sets in GVF format that we are converting to VCF and we would like to use VCFv4.4 given its improved support for structural variants. With legacy data we may not necessarily have the information to determine the SVCLAIM value. In such a case, the lack of an option for a missing value in SVCLAIM is not very practical. A missing value ('.') for SVCLAIM for DEL and DUP would allow us to represent unknown evidence for SVCLAIM but still make use of the improved support for structural variants in VCFv4.4.

The rule in the VCF specification v4.4 (Section 3, page 16) states "DEL/DUP: SVCLAIM must be specified and can [be] D, J, or DJ. J and DJ claims indicate a breakpoint between the start and end of the DEL/DUP".
However, the specification does not state explicitly if a missing value is accepted for SVCLAIM in DEL and DUP.

Question 1: Is a missing value ('.') valid for SVCLAIM in DEL and DUP? Please could we kindly request that the specification is amended to explicitly state whether or not this is accepted.

Question 2: If it is not currently accepted, please could we add a missing value of ('.') to the accepted values for SVCLAIM for DEL and DUP?

[d-cameron]

The purpose of the SVCLAIM field is the remove the ambiguity as to whether a <DEL> or <DUP> comes from an SV caller or a CNV caller. Usually you know if your data providence so can add SVCLAIM=J or SVCLAIM=D to everything from a given source but there are indeed cases where you don't know whether a record is a SV or CNV record.

VCF 4.4 does require SVLCLAIM for <DEL> and <DUP> and the omission of the missing value from the allowed values in intentional.

Your options are:

• Violate the specs and just write . in the SVCLAIM field.
• Write a non-standard symoblic allele for the records of unknown providence (e.g. <AMBIGUOUS_DEL> instead of <DEL>)
• Assume everything is either an SV or CNV and write the appropriate SVCLAIM record.
• Further investigation of your data sources to detemine which records are SVs and which are CNVs.