3 Person example #1
Victim Vaginal Swab
Victim and Consensual Partner
“Who Did It”
The Scenario
• Victim hosted a party at her place
• She had sex with her boyfriend prior to guests
arriving
• She drank too much and went bed before all
guests left
• She “dreamed” that someone had sex with
her – boyfriend was still there
The Scenario
• She woke up and ID’d Suspect leaving her
room
• Suspect admitted he was in the room, but it
was just to grab an X-box game he had left in
there
The Egram
Sample Interpretation
• Consistent with 3 persons
• Appears to have high- middle- and low-level
contributors
• Perhaps the low level contributor drops out?
– Some minor alleles <300
– Such as our good friend FGA
• X and Y somewhat balanced
– Will we be lucky and have Victim as the minor?
The Data Table
Match to Victim
• Victim “fits” every where but FGA
• She has a 22 there that is missing from the
mixture
• She is the low-level contributor
Match to Consensual
•
•
•
•
All alleles of Consensual present
He is 20, 23 at D2 (6 Alleles)
So he is high-level contributor
(State assumption in report)
Match to Suspect
• All alleles of Suspect are found
How many would assume
1. Victim only
2. Consensual only
3. Victim and
Consensual
4. No one
My approach
• I assume both V and C, and will communicate
that in the report
• Because V is minor, I’m not concerned about
alleles <300, as I know who they belong to
• I have a pretty good idea that C is the major
(or at least the majorish)
• I’m interested in the in-between contributor
My approach
• I know we already did the match to the
Suspect, but we’ll hide his profile before we
start looking at the mixture
• I already decided the mixture is interpretable
as a three person mixture, and I have lots of
ways to deal with the stat
Any conclusions about Suspect?
1.
2.
3.
4.
5.
Match
Included
Inconclusive
Excluded
None of the above Profile cannot be
interpreted
The Deconvolution
• We’ll “hide” the Suspect’s type from the table
– No peeking!
• Remember to put the calculator into 3 person
mode (Not a trivial step)
• Let’s go highest to lowest in terms of number
of alleles in our interpretation
• We’ll apply both Victim and Consensual and
try to deconvolute Unknown type
• I have to do those 4 things now
The Egram
The Deconvolution
• Start with D2 (six alleles) – And nothing fits!
The Deconvolution
• Recalculate for stutter at D2
• Still didn’t help – but I’m smarter than the
software so I’m over ruling it
The Deconvolution
• Why am I going to over rule the computer and
make the 17 go with the 24?
• They are both <300 rfu
– They are both in the Danger Zone
– No requirement for a phr to be >50% if I’m in the
Danger Zone
The Deconvolution
• So if I just lower the phr filter to 40%, they go
together
The Deconvolution
• I can now deconvolute that locus and send the
results to my new table
• In fact, we get 4 loci with only one option for
the Unknown type that auto-graph
• So far:
– V = 6%
– C = 68%
– U = 27%
The Deconvolution
The Deconvolution
• D18 – This is really hard to see, so I apologize –
13, 15 is best (15, 15 P = 14%) (14, 15 phr 59%)
The Deconvolution
• D16 – A homozygote 9 (that’s the only
Unknown allele) means 11% Unknown rather
than the 20+% we’ve been seeing
The Deconvolution
• D16
The Deconvolution
• D16 – Welcome to the “Not Excluded” option
• Even by the “eyeball” method, you can see
that a 9, 11 works well for Unknown
The Deconvolution
• D16 (not excluded) option
• Circular or double sharing
The Egram
The Deconvolution
• D16 – Test the “not excluded” option by
raising PHR to 70%
• It’s the only thing left – best fit
• But since no P, it can’t graph
– (Maybe some day – we’ve got some ideas…)
The Deconvolution
•
•
•
•
D5
V = 11, 13
C = 8, 12 (Major)
U= ? (mid level)
The Deconvolution
• D5, U = ? (mid level) – But lots of choices
The Deconvolution
• D5 – mid-level has averaged 27% so far
?
The Deconvolution
• D5 set mP to 5% to get rid of the tiny bits
• Also set PHR to 60% to get rid of the option
with the Unknown type being the 89% major
contributor
The Deconvolution
• D5 now only 2 choices
• I could live with this using the obligate
function, but…. Where’s the fun in that?
• At 70% PHR, the Unknown must be 12, 12
– That P of 26% is very nice compared to 27% avg
– PHRs for the 2 heterozygous knowns are still OK
The Deconvolution
• D7
• V = 7, 10
C = 8, 10
• Pretty sure U has a 9
The Deconvolution
• D7 – 9 is in sutter, so correct for it
• We’ve learned 50% correction is OK
The Deconvolution
• D7 – What did stutter correction do for us?
• We’ve learned from testing that the “true”
combination generally benefits the most from
stutter correction
• The “false” combination may not show
improvement
• It can be rather subtle
• (I’m talking about PHR and P)
The Deconvolution
• D7 – Before correction
7,10(phr=98; p=07) 8,10(phr=98; p=79) 9(p=15)
7,10(phr=76; p=06) 8,10(phr=76; p=68) 9,10(phr 76; p=25)
• D7 – After correction
7,10(phr=93; p=04) 8,10(phr=93; p=83) (p=13)
7,10(phr=81; p=04) 8,10(phr=81; p=72) 9,10(phr 81; p=24)
The Deconvolution
• D7 – So the stutter correction generally
improved the bottom option, and (slightly)
“unimproved” the middle option
• At 80% phr, the top (not excluded) option
went away
• (I admit the “not excluded” takes some
practice samples to get comfortable with)
The Deconvolution
• D7 – My final answer
• Because the bottom answer improved the
PHR fairly well and kept the P (24%) of the
Unknown right where it’s been (27% avg), I
pick 9, 10
– I know that both choices have good PHR
– I know that I picked the one with the 2nd best PHR
– But, I can’t figure out why the enzyme would
“miss” half of the Unknown DNA (I didn’t choose
the one with P = 13%)
The Deconvolution
• Where we stand so far:
The Egram
The Deconvolution
• THO1 – V is 7, 9 E is 6, 8
• 7, 7 would work great for a mid-level
The Deconvolution
• In fact, at 70% PHR, the Unknown
is a 7, 7 (25% of total DNA)
• Great PHR for the other two
– 100% and 95%
The Deconvolution
• D13
• Again, why would I pick an option where ½ the
Unknown DNA just disappeared?
• This tells me he’s not a homozygote here
The Deconvolution
• CSF – Kind of a mess to start with
The Deconvolution
• CSF – But V is the low level, not 39%
The Deconvolution
• CSF – Look carefully at these options
• The top one says Consensual and Unknown
together combine to account for 61% meaning V is 39% - doesn’t make fit
• The bottom one says Victim and Unknown
combine to account for 39%
The Deconvolution
• CSF – Quite a few choices, so I really have to
study the peak pattern and the known profiles
The Deconvolution
• CSF – This tells me that the Victim is totally
masked
The Deconvolution
• CSF
• In fact, V’s 6% is probably lost in the expected
variation of the other 2 contributors
• Therefore, I will temporarily “un-apply” the
Victim as a required genotype, and treat this
as a 2 person locus
The Deconvolution
• CSF
• As a 2 person locus, 11, 12 isn’t bad
The Deconvolution
• CSF
• Then correcting for 50% stutter
pretty much makes it perfect
The Deconvolution
• CSF
• So I’ll go back to a three person mixture and
go with an 11, 12
The Egram
The Deconvolution
• D8
• V = 14, 14
C = 13, 14
The Deconvolution
• D8 – Change to two person locus as before,
Victim + Unknown around 30% so far
• So could Unknown be a 14, 14 (only 14%)?
• But the 14 Allele is 8642 rfu (blown out)
The Deconvolution
• D8 – Since 14 allele saturated the detector, we
don’t know what the real height is
• A bunch of 14 could still be out there
somewhere
• So even though a 14, 14 looks like a low
proportion, that’s my call
• (You should probably have an upper limit!)
• In our current world I’d have to really argue to
use this injection – this is an old, old sample
The Deconvolution
• TPOX
• V = 8, 8
C = 8, 11
The Deconvolution
• TPOX
• 1% for V + U – NO
• 99% for C + U – Maybe, but V is pretty good at
D16, D18 and D2 so probably not
The Deconvolution
• TPOX
• We can “test” by raising PHR
• Only 11, 11 remains
The Deconvolution
•
•
•
•
D3
V = 15, 16
E = 15, 16
Finally an easy one
15, 15 (12%) and 16, 16 (1%) don’t work – 15, 16
The Deconvolution
• That leaves FGA
• We know V drops out here
• So – In the following order:
– “Un-apply” V reference (we know the 22 dropped)
– “Ignore” the 20 allele (it has nothing to do with
the other two alleles)
– Switch to 2 person calculator
– (You can always “consider” an allele again)
The Deconvolution
•
•
•
•
FGA
Might seem easy – it’s not 1:1
But, this is worst case scenario for stutter!
24 is very tall, and stutter at FGA = 17%
The Deconvolution
•
•
•
•
FGA
So correct for stutter at FGA
Use the 50% we know fits mixtures
23, 24 for Unknown is 34%, Consensual is 66%
The Deconvolution
• We now have a full single source Unknown
profile – no Anys, no Obligates
FGA
Amel
Victim dropped out
2 people share type
Not Excluded
Not Excluded
3 people share type
2 people share type
CSF1PO
D3S1358
2 people share type
D8S1179
The Deconvolution
• Graph looks good, but some things don’t
graph
The Stat
• We can’t really use the option from the stat
window cause the software remembers a
mess
– Some loci were converted to 2 contributors
– We ignored an allele at FGA
– Etc…
The Stat
• But, who cares?
• We have a single source profile
• 1 in 124 Quintillion
How many of you would…
1. Say I was nuts to do
that deconvolution?
2. Agree and be OK as
my Tech Reviewer?
3. Want to go talk to my
Tech Leader about my
“interpretation?
33%
1
33%
2
33%
3
0 of 30
30
Countdown
How many think…
1. I was conservative?
2. I was the opposite of
conservative?
3. I got the right answer?
33%
1
33%
2
33%
3
0 of 30
30
Countdown
Match to Suspect
• I call that an exclusion
• Is it still an “anti-conservative” interpretation
if I excluded the guy?
Now What?
• Well, because we excluded this suspect, they
sent in another reference from some other
guy that was at the party
• This first guy always said that he went in her
room, but it was only to get his shoes that he
left in there
• So now we’ll bring in Suspect #2’s reference
and see how it fits
New Data Table
• To cut down on open windows, we’ll do a new
table of the original mixture, the Unknown
type and the new reference of Suspect #2
Match to Suspect #2
• Suspect #2 matches the Unknown profile we
deconvoluted from the mixture
A Much Faster Way
•
•
•
•
Apply all three references, V, C, and S
Set for 3 people
Auto call references
View call report
A Much Faster Way
• Doesn’t graph well – Click through the loci
• There are 7 with zero support (no combinations)
A Much Faster Way
• Get on the phone to the investigators and tell
them that although all of Suspect’s DNA is
found in the mixture, it just doesn’t fit right.
• Can you please send in another reference?
• Then apply that new reference to the mixture
and see how it fits
A Much Faster Way
• Still doesn’t graph every where, but we know
that (shared types, not excluded, etc)
• But support at every locus (D2 and FGA special)
Things to Consider
• It takes more than just the presence of alleles
to be an inclusion
• The inclusion should be supported by all PHR
and P expectations – it’s about the math
• I know we’re supposed to look at the mixture
without applying the reference, but…
Things to Consider
• If you apply all references and nothing fits,
you have an exclusion
• If you apply the references and it all fits, you
just did a “real time LR” (sort of)
• But most importantly:
Things to Consider
• Even though we “peeked” at the reference of
Suspect #1, it played no role whatsoever in
the final interpretation and stat
• If we would have stopped when we saw the
result of the first comparison of Suspect #1 to
the mixture and just done a CPI, we would
have just included an innocent man