Misrepresentation of STRmix by Mr Dominic Saraceno ESQ in People of New York v. David Smith

Misrepresentation of STRmix in an affidavit by Mr Dominic Saraceno ESQ. in: Superior Court of the State of New York

County of Niagara: Criminal Term

The People of the State of New York versus David SMITH

Indictment 2015-041

The affirmation can be found on the Cybergenetics website at https://www.cybgen.com/information/newsroom/2015/dec/New-York-motion-filed-on-why-STRmix-is-not-TrueAllele.shtml (external link)(external link)

Mr Saraceno has confirmed by email that he consulted Dr Perlin.

There have been multiple instances of misrepresentation in this affirmation written by Mr Saraceno of data appearing in Taylor, Buckleton, and Evett [1] Table 2 (this table is repeated below).

Table 2: Results of H_d true tests for a four person 0.25:0.25:0.25:0.25 mix at 50pg total input

162 template. Average peak height for the profile was 89rfu.

In the affidavit of Dominic Saraceno a considerable comparison is made of STRmix vs TrueAllele.  Mr Saraceno is an attorney and has formed the affidavit from advice given to him. 

Mr Saraceno states:  “Most worrisome is STRmix’s high false inclusion rate.  By STRmix’s own estimate, false positives that erroneously include the wrong person occur in .01%-.05% of STRmix analyses.  Falsely including an innocent man is contrary to a “reliable” or “generally accepted” forensic method.”

The person advising Mr Saraceno is significantly misrepresenting the data.  The data above show the results of examination of one very difficult mixture.  It is a 4 person mixture 1:1:1:1 at low template.  The results to look at are in the column labelled STRmix^TM  V2.3.  The results under STRmix^TM lite are for a product we made to mimic semicontinuous models. Even for this very difficult mixture STRmix^TM V2.3 assigned an LR = 0 (excluded) 99.958% of false donors.  The LR for the true donor considered was 374,104.  The affidavit of Mr Saraceno (an attorney) concentrates on the number 0.0173% and describes this, inaccurately, as STRmix’s false exclusion rate.  This is the number of LRs greater than 1 for this very difficult mixture.  Most of these LRs are very small, and although strictly are evidence for inclusion the inference is so weak that they could reasonably be termed inconclusive.  However more significantly it is not STRmix’s false inclusion rate. 

An LR > 1 could, in theory, occur because the false donor, by chance has the correct alleles to fit the mixture or because the software has contributed in some way.  We could term these “DNA inclusions” and “software errors”.  For a difficult mixture such as this LRs > 1 are expected at a small positive rate based on the DNA false inclusion rate alone.  There is no evidence that STRmix has added anything to this rate at all.  In fact the average LR of 1.005 is evidence that it has not.  This number is expected to be 1.000 and 1.005 is very close to this.  This was the main message of the paper:  That is LRs > 0 are coming at about the rate expected from the DNA and there is evidence that the software is adding very little or nothing to this rate.

Later Mr Saraceno states:  “STRmix, however, requires the defendant’s genotype as part of its operation”

The person advising Mr Saraceno is simply misinformed.  STRmix does not use the defendant’s genotype during the deconvolution at all.  It does use it to calculate the LR after the deconvolution but so must every other method or software.

There are a great many other errors in Mr Saraceno’s document which we do not discuss here.

References

[1] Taylor D, Buckleton J, Evett I. Testing likelihood ratios produced from complex DNA profiles. Forensic Science International: Genetics. 2015;16:165-71.