As we see more examples of ‘probabilistic genotyping’ in casework, we grow more sceptical and alarmed about both the likelihood ratio and these programmes.

This case was a murder.  A hair had been found in a vehicle.  The prosecution asserted that the hair was from the victim.  Nuclear and mitochondrial DNA analysis was performed.

The sample from the hair underwent three ‘enhancement’ processes intended to improve the detection of DNA.  At each stage, three analyses or ‘runs’ of the DNA were performed.  The results within each stage were similar, but different for each stage.  The three final stage runs were used as the input for the LikeLTD probabilistic genotyping programme.  At four loci there were no results in any run.   Despite this, the LikeLTD programme calculated a Likelihood Ratio (LR) greater than 1 at three loci.  In other words, despite there being no data whatever at these loci, the software calculated a LR favouring the prosecution. 

At one locus, the software calculated a LR less than 1; favouring the defence.

We consider these results illustrate that probabilistic genotyping is not yet sufficiently tested in casework, at least insofar as LikeLTD is concerned, regardless of the number of publications and cases claimed to support the reliability of the results.

There were further issues in this case that we discuss elsewhere.