NIST Inter Laboratory Study 2013

Recently Barry Scheck spoke at some length at a meeting on the New York State Commission on Forensic Science ^[1].  Amongst other matters he stated that a certain software was the only software that gets the NIST Mix 13 ^[2] exercise correct.  We are a little disappointed at Mr Scheck’s statements at this important meeting.  His statement about NIST Mix 13 is untrue.

We did not participate in MIX13 at the time that the study was undertaken.  We have subsequently analysed the five cases within the study.  The STRmix™ results are summarised in the table below.

The outcome is considered ‘correct’ based on a high likelihood ratio for true contributors and a low one for non-contributors. 

Case 5 was challenging.  The mixture is actually a 1:1:1:1 mix of individuals a and b and two other individuals selected to maximise allelic overlap (the genotypes follow).  0.5ng of each was amplified (total 2ng). 

The epg follows.

The allelic overlap and 1:1:1:1 ratio creates a perfectly viable three person solution.  The comparison of the peak fits for the 3 and the actual 4 person solutions is shown below.  The three person solution is a closer fit by least squares.

Looking at the genotypes, if one considers, say, CSF1P0, the four contributors are 12,12 and 12,13 and 11,11 and 11,12.  There is an equivalent solution as a 2:1:1 mix of 11,12 and 11,12 and 12,13.  This is true for all loci. 

STRmix requires the assignment of the number of contributors.  With no prior information, we would assign case 5 as a three person mixture.  STRmix™ gives mixture ratios 2:1:1 for this solution. 

If we assume the presence of reference a, as some software programmes do, we force the four person solution in STRmix and obtain the mixture proportions 1:1:1:1.  This is a valid assumption under H_p and we do not argue with the software programmes that do this.  However it would be wrong to make this assumption under H_d and the most parsimonious H_d solution is three contributors.   The only obvious way to decide that this is a four person mixture is to assume the presence of one of the persons of interest (POI): reference a, which we were not prepared to do.

However, whether assuming three or four contributors all POI have been correctly assigned valid likelihood ratios and included or excluded as contributors to the mixture correctly.

[1] June 19, 2015 meeting available here 

http://www.criminaljustice.ny.gov/pio/openmeetings.htm (external link)(external link)

[2] http://www.cstl.nist.gov/strbase/interlab/MIX13.htm (external link)(external link)

[3] LRs were calculated using the NIST Caucasian allele frequencies