If a crime stain isn't matched in the offender database, there's still a good chance that its relative is there. The relative can likely be found by "kinship" searching if you know how to search. If the search results in a list of even dozens of potential relatives, sort them out with Y-haplotyping. This approach could significantly increase the effective reach of offender database. It could also raise some hackles and some eyebrows.
Comments We discuss a hybrid searching approach consisting of two steps:
There has been some misunderstanding about the nature of the suggested searching step, partly from colleagues who assume what we propose is related to the "low stringency" method presently implemented in the CODIS software used by law enforcement in the US. It isn't. It's radically different both in implementation and, more importantly, in performance.
Counting alleles to locate siblings is even worse. The number of common alleles between two samples varies across a very wide range both for sibling and for non-siblings, the distributions overlap a great deal, and therefore depending on where one sets the threshold of number of common alleles, either one will overlook the majority of true sibs or will be overwhelmed by wrong answers. (See WTC prospective DNA analysis presentation).
The CODIS method is to count alleles. For example, a possible parent of a crime sample is an offender who shares an allele at each locus with the crime sample. This approach is hopeless. The rate of false positives is about 40/50000, which means that if you search for crime stain in a 50,000 offender list you will find 40 spurious matches in addition to the possible 1 true match, and the true lead will not stand out.
In particular, our simulations show that about half the time true father or brother has the highest of all the likelihood ratios for relationship.
The searching is done using the set of markers on which the database was built. The only hurdles to begin searching tomorrow are
A simulation not included in the paper showed that 200 SNPs would be almost infallible in distinguishing relatives from non-relatives, so follow-up analysis by these instead of Y-haplotypes would be excellent. However, despite some worthwhile exceptions this is not a technology that is being actively pursued in many forensic labs.
Indeed, 200 SNPs are enough to do the whole job if they were implemented for every sample in the database. But for that to happen is a long way off at the soonest.
Y-haplotyping is all the rage. Many forensic laboratories have Y-haplotyping capability or are interested in developing it. Therefore this method of reduction is immediately feasible for many states.