The multinationals, surprise
surprise, do not release this information and no-one outside
the industry, unbelievably, would seem to have done such validation.
For all I know this industry could be a house of cards build on sand.
The nearest I have is from FSI Vol 86,1997,p25-33
Source reference:
http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6T6W-3W0G0CK-3&_user=10&_handle=W-WA-A-A-AY-MsSAYVW-UUW-AUDDEAAUEW-WZZZEWCAW-AY-U&_fmt=summary&_coverDate=04%2F18%2F1997&_rdoc=3&_orig=browse&_srch=%23toc%235041%231997%23999139998%2375689!&_cdi=5041&view=c&_acct=C000050221&_version=1&_urlVersion=0&_userid=10&md5=4d7a6643c27f043e66bfe63e3729a47c
Abstract of the FSI article
7 samples divided ,sent to 16 labs and checked blind on the 2 loci D21 and FGA
then as 2 alleles per locus 448 data-points in total.
Bear in mind all the labs knew these were validation samples so
could easily have been on their "best behaviour". E.g. not (say)
using an unbalanced centrifuge in the lab at the same time, that
could mechanically vibrate the DNA processor, or (say) not using a
known electrically noisy motorised bit of kit that pumps
noise spikes down the mains etc etc.
There were 18 errors in that returned data the worst being (23,23)
returned as (21,21) i.e. 2 bins away from the "correct".
'Invalidation ' of DNA profiling - FGA
'Correct' results (majority in agreement) are in the first two columns.
Erroneous results highlighted next to red blobs. Lab '13' would not have been aware of any problems
as for all they knew some false de-natured samples could have been included. HUMFIBRA is also known as FGA. Note [8] is Urquart to Moeller conversion formula.
'Invalidation ' of DNA profiling - D21
So with an error rate of 18 in 448 or 1 in 25, one of the figures in my (un-fudged)
profile could easily be incorrect - who would know ? Until it is repeated
on different machines, with different personel, at different times, at different labs, with different reagent
batches, and a concencus "correct" result emerges, then no-one knows.
Again there is the same uncertainty to any SoCo sample, one figure in 20
can easily be wrong by one or more.
The "correct result" [i.e. most labs in agreement (there is a whole treatise just on this aside, as everything in DNA "science" is inferred ) ] from 13 of
the labs was for locus HUMFIBRA ( FGA ) 2 alleles per sample
and 7 samples
21,25 ; 23,24 ; 22,22.2 ; 23,23 ; 18,23 ; 18,22 ; 23.2,24
but for lab "1" ; 23,24 returned as 22,26
for lab "11" ; 22,22.2 returned as 22,22
and for lab " 13" ; 23,23 returned 21,21 ; 18,22 returned 17,19.2 and
23,24 returned as no result obtained and 23.2,24 returned as no result obtained
i.e. just 3 of the 7 samples matched.
Well at least they were honourable enough to report it all.
On the second locus D21S11 better agreement
7 consensus 'correct' results 59,63;63,67;63,65;61,65;61,70;61,63;59,65
- only one lab
at variance the 61,65 pair returned as 61,63 ; 59,65 returned as 61,63
and no result found for the 63,67 pair.
DNA profile error rate now down to 4 per cent - official
That is 4 in 100 forensic DNA profiles as used in the UK consisting
of 10 markers and 20 datapoints.
Most recent data from journal article
International Journal of Legal Medicine (2004 ) 118: p83-89
http://www.springerlink.com/app/home/contribution.asp?wasp=847987d4f86447faa3e8be5a4106a539&referrer=parent&backto=issue,4,13;journal,10,54;linkingpublicationresults,1:101167,1
or
http://tinyurl.com/82rk9
July 2005 I find it is now available on the web as
http://medweb.unimuenster.de/institute/remed/spurenkommission/Information/IJLM_GEDNAP_II.pdf
http://medweb.uni-muenster.de/institute/remed/spurenkommission/Information/IJLM_Rand_etal.pdf
http://medweb.uni-muenster.de/institute/remed/gednap/Information/IJLM_GEDNAP_II.pdf
The GEDNAP blind trial concept part 2.
Data for year 2002 which is an improvement
2001, 5 per cent erroneous profiles.
2000 , 7 per cent erroneous profiles.
This data is anonymised so the good and bad are lumped
together - considering the bad do not know they are
bad (or at best, in disagreement with the consensus).
These are the results of GEDNAP ( German DNA Profiling
Group ) blind trials of testing samples at 136 labs in 30
European countries.
First thing to note this was blind trial ( not
double blind ) so all participants knew in advance and
could process immediately after calibration, not post-pub Friday
afternoon processing / interpretation/ transciption etc.
Much of the 'improvement' is because:
Many incorrect results in previous GEDNAP trials had been due to specific types
of body-fluid stains so those "have since been discontinued because of
the inconsistency of the amount of DNA present " from the more recent
trials, despite those sorts of stains being commonly found forensically.
Then the intractable human error:-
"most of the errors were made each time by a very few number of
laboratories and of course compound errors such as interchange
of two samples caused a disproportionate number of errors relative to
the one mistake made when sampling the wrong test stain. However,
this is not taken into consideration when calculating the error rate."
( these specific errors included or excluded in the calculated error rate? )
"the most common type of error has always been transcriptional errors
followed by incorrect interpretation due to failing to recognise
an error, these types of human error are to some extent unavoidable
under any prevailing circumstances."
I repeat this is results from labs knowing they were
being tested and presumably on their best behaviour.
Then the more technical/ systemic problems producing mistyping because of
stutter, 'long alleles', unrecognised rare alleles etc.
Interesting examples where the automated processing
has failed to differentiate and requiring human 'correction' - their term.
The accompanying printoff examples show just how
subjective these can be. This is all from ideal
provided test-sample stains.
Table 1
Results expressed as single loci because there is no
standard set used in all Europe.
Results for Gednap 20 & 21 which show 0.7 per cent on 13,868 single loci tests gives
an
error rate of 1 in 11 expressed for CODIS profiles, 1 in 14 for UK FSS 10 loci profiles. After that result they moved the
goal posts. They dropped the stains such as "cigarette butts and mixtures
of body fluids as well as hairs". Because these samples were
producing disproportionately more errors.
Beggars belief doesn't it - just how often, forensically,
do they find 50/50 blood/blood traces, victim and offender bleeding
the same amount ?.
Taking the most recent, moved goal post, GEDNAP results and expressing
in terms of 13 loci CODIS profiles then 1 in 19 is erroneous by 1 locus
which is all that is needed to produce an erroneous profile
19 correct out of 20 datapoints is fine in 4 out of 100
DNA profiles if you are identifying body parts, say,
from a small pool of possibles
but not for the usual forensic purposes of implied uniqueness.
Also, from FSI 143 (2004) 47-52
http://www.sciencedirect.com/science?_ob=ArticleURL&_udi=B6T6W-4C2R3WD-1&_user=10&_handle=V-WA-A-W-AW-MsSAYWA-UUA-U-AAWCUVYYCW-AAWWZWEZCW-WCUAABZEE-AW-U&_fmt=summary&_coverDate=06%2F30%2F2004&_rdoc=4&_orig=browse&_srch=%23toc%235041%232004%23998569998%23503752!&_cdi=5041&view=c&_acct=C000050221&_version=1&_urlVersion=0&_userid=10&md5=9c0b7e5969279f0689686c46d6198bd1
or http://tinyurl.com/75dbu
False Homozygosities comparing 2 DNA profiling kits with divided samples taken from
2055 individuals showing 15 errors comparing SGM plus results to Powerplex 16 results
on 5 loci , so error rate of 0.7 per cent or 1 in 140.
Person SGM Powerplex
vWA
1 15,15 15,17
2 17,17 17,18
3 15,15 15,18
4 16,16 16,18
5 18,18 18,21
D8S1179
6 12,12 12,16
7 14,14 10,14
8 13,13 13,18
FGA
9 22,25 25,25
10 23,23 21,23
D18S51
11 14,16 14,14
12 10,10 10,18
13 15,16 15,15
So false homozygosity slippage by as much as 8 alleles.
And one error on D5S818 comparing Profiler and Powerplex
10,11 11,11
10,12 12,12
The tally for errors on the cross comparison
of Powerflex and SGM+ on (2055 minus 254 )
samples for the 8 common to both were
VWA ; 5
D8 ; 3
FGA ; 2
D18 ; 3
D21 ; 0
D16 ; 0
D3 ; 0
THO1 ; 0
Until further evidence emerges for FH errors
concerning D19, D2 used in the UK and
D5,D13,D7,TPOX,CSF1PO for CODIS (for more than
254 samples) then
these are the revised figures using the average
for those 8 to represent the others
SGM+ ; 10/8 * 13/(2055-254) = 0.009
CODIS ; 13/8 * 13/(2055-254) = 0.012
Reference sample DNA Profile error rates
SGM+ fundamental single kit error rate = 0.9 percent
CODIS fundamental single kit error rate = 1.2 prcent
This is a systemic failing and for normal single kit only
processing, sets the other bound for error rates. So error
rates from such studies are
In FSS 10 loci terms between 1 in 14 to 1 in 140 wrong
Confirmed , near enough, by FSS study published in
FSI 112 (2000) 151-161
Comparing just SGM and SGM plus and they have the
arrogance or niaivety to state "This is a rare event"
The false result turns up in the SGM flavour that is currently
used, not the older version. Concerning one HUMFIBRA result " The crime scene stain and
reference samples were designated as 19,26 with SGM wheras the SGM Plus results
showed only a 19 allele". Consistent error as it was repeated, 167 samples, so 1 in 167 is the
corresponding figure so in no way can be considered rare when they
also have the arrogance to quote 1 in billions for false match figures.
In USA, CODIS 13 loci terms, 1 in 11 to 1 in 140 wrong.
My profile apparently has 3 homozygous pairs - is that true or
an artifice ?. If I could afford it I would go to
an independent lab and have it tested with something
other than SGM kit. It is my conjecture that my profile
may falsely match with other people in the UK with
even more likelihood because they also have apparent
homozygosities at D8(13,13),D21(29,29),D16(12,12)
when in fact we are all different on 10 loci
but SGM falsely registers these false homozygosities.
As far as I know no-one has analysed and published the
results of checking for co-ancestry and independence i.e.
say inheriting a 17 on D2 does not predispose to inheriting
a 15 on D18 say , by checking hundreds of thousands of such profiles.
Similar to people of one background ,having blue eyes are likely
to have blonde hair and someone from another backgrond with black hair
are more likely to have dark eyes.
A good review of
DNA profile technical problems with lab printout examples
In depth grounds
for questioning reliability of DNA evidence
Some further exploration of
DNA profile problems
DNA Evidence: is it safe
to convict?
DNA
Evidence: science or smoke and mirrors?
I can be pretty certain my following critique of DNA
profiling will not be published in the likes
of Forensic Science International.
Proponents of mass DNA profiling like to trot out large
googol type numbers giving the probability of 2 people having the same DNA
profile, seemingly derived from little more than, some sort of product rule of
number of markers and the number of possible sites on these markers. They use
some pretty impenetrable statistics to show there is no aliassing between these
DNA markers. That is, they are of the opinion that the inheritance of these
marker sites is independent. Saying, if you inherit one set of "numbers" on one
marker then you are NOT predisposed to inherit a given set of "numbers" on
another site. I would rather rely on figures derived from real life.
I tried getting full details from Professor
Chaseling but she did not reply to my email enquiries of 20 Jan, 2002 and 14
April ,2002. In Australia a Prof Janet Chaseling of Griffith University did an
experiment taking DNA samples from the likes of politicians and blood donors;
reported in the Sydney Morning Herald 22/04/2000.
The
Sydney article no longer available from the original site
or
