INSTRUCTIONS â DELETE AFTER READING
21/09/2015
Examination of a Novel Method to Allow for a Range in the Number of Contributors to a DNA Profile: A Key Area of Subjectivity in DNA Evidence Interpretation
S. Cooper, C. McGovern, L. Russell, D. Abarno, D. Taylor, JA. Bright, J. Buckleton
Complex Mixture Interpretation
Increased sensitivity + expanded evidence types = more and more and more mixtures Criticism of divergent approaches to interpretation
2
Does introducing probabilistic software improve consistency in reporting?
9
1
21/09/2015
Collaborative Study- Case 1 2p 1:1 High t
SJ Cooper, CE McGovern, JA Bright, D Taylor, JS Buckleton Investigating a common approach to DNA profile interpretation using probabilistic software FSI: Genetics, 2015. 16: 121-131.
4
Case 1 – LR to Complainant μ=10.3612 σ=0.0210
Log(LR)
8
6
4
2
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Participant number 5
Collaborative Study – Case 2 Lower t, increased ambiguity below ST
6
2
21/09/2015 Log(LR) 16
14
12
8
10
6
4
2
0 1 2 3 4 5 6 7 8 10 11 12
16
9
Participant number
Case 2 – LR to Person of Interest μ=14.28 σ=0.03
13 14
14 15 16 17
12
18
N=2 N=3
19
μ=8.92 σ=0.74
20
Log(LR)
10
8
6
4
2
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Participant number 7
Collaborative Study - Case 3 Complex mixture, using replicates Replicate 2 1
8
Case 3 - Using Multiple Amplifications 14
μ=11.32 σ=0.74 12
Log (LR)
10
8
6
4
2
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Participant number
9
3
21/09/2015
Collaborative Study Conclusions Greatly improved evidence evaluation can be achieved by implementation of the same probabilistic software
When estimating N is unambiguous it is possible to achieve standardization Assigning an estimated N remains a key point of divergence PCR replicates help improve confidence and consistency Study highlighted the need for further work in addressing uncertainty in the estimation of N
Can this be done statistically? 10
An Extension to STRmix™
D Taylor, J-A. Bright and J.S. Buckleton. Interpreting forensic DNA profiling evidence without specifying the number of contributors. FSI: Genetics, 2014. 13: 269-280
STRmix™ uses an iterative re-sampling process to generate the weights for genotype combinations MCMC (Monte Carlo Markov Chain) samples from a multi-dimensional probability space
11
How does it work in theory? The number of contributor N becomes a dimension of the probability space Performs MCMC in multiple probability spaces – each defined by a different number of contributors N
Can now enter a contributor ‘range’ rather than a single estimate
S jw| HPr( N w| H ) j Pr( 1) S Z Pr( LR j S j'w| HPr( N w| H ) j ' Pr( 2 )S Z Pr( j' n
LR
n
1
n
n'
n'
j
j
j'
j'
| N n , H1 )
j
n'
2
j'
| N n' , H 2 ) 12
4
21/09/2015
How does it perform in practice? Ground Truth Known Profiles – Low t 1P T75 N 1 2 1 2 1 2 1 2 1 2
2 3 4 5
Mr 100 49:51 100 64:36 100 62:38 100 49:51 100 39:61
1P T75
Zn 0.998 0.002 0.921 0.079 0.038 0.962 0.995 0.005 0.752
28 27
Log(LR)
Run 1
26 25 24 23 22
0.248
1
2
3 Run
4
MPN
Stratified
N=1
2p 4:1 T250 N 2 3 2 3 2 3 2 3 2 3
2 3 4 5
Mr 20:80 73:14:13 80:20 11:78:11 21:79 14:79:7 20:80 8:16:76 80:20 16:78:6
Zn 0.244 0.756 0.999 0.001 0.991 0.009 0.198 0.802 0.976 0.024
4
4 3
46:14:14:26 3.017E-13 9:64:27 0.999
4
28:53:11:8 2.086E-04
2
Mr 27:9:64 64:21:9:6 64:9:27 48:36:8:8 28:63:9
Zn 0.999 2.775E-05 0.993 0.007 1.000
26 25 24 23 22
3p 8:3:1 T650, LR to minor
1
2
3 Run
4
Stratified
N=2
25 MPN
24
Log(LR)
3
N 3 4 3 4 3
2p 4:1 T250, LR to minor 28 27
3p 8:3:1 T650 Run 1
5
Log(LR)
Run 1
5
23 22 21 1
2
3
4
Run MPN
13
Stratified
N=3
Case Scenarios – 2p 10:1 Z 2/Z 3
14
Case Scenarios – 2p 10:1 Z 2/Z 3 Case 1 Original Input
2 3 4 5
N 2 3 2 3 2 3 2 3 2 3
Mr 9:91 8:92:0 91:9 3:94:3 91:9 0:6:94 91:9 1:98:1 8:92 2:96:2
Original Input, LR to minor
Zn 0.442 0.558 0.893 0.107 0.997 0.003 0.991 0.009 0.999 1.84E-05
35 30
Log(LR)
Run 1
25 20 15 1
2 3 4 5
N 2 3 2 3 2 3 2 3 2 3
Mr 9:91 7:87:6 92:8 88:7:5 91:9 6:6:88 9:91 8:5:87 9:91 7:88:6
Zn 0.820 0.180 0.991 0.009 0.991 0.009 0.997 0.003 0.986 0.014
3 4 5
N 2 3 2 3 2 3 2 3 2 3
Mr 91:9 86:7:7 9:91 7:6:87 92:8 87:7:7 91:9 7:7:86 9:91 87:7:7
5
with ADI, LR to minor
Zn 3.82E-07 0.999 8.76E-07 0.999 0.012 0.988 1.11E-09 1 0.002 0.998
30 25 20 15 1
Increased Stutter, LR to minor
MPN
35
2
3 Run Stratified
4
5
Ground N=2 Truth
30
Log(LR)
2
4
N=2 truth Ground 35
Case 1 Inc Stutter Run 1
3 Run Stratified
Log(LR)
Run 1
2
MPN
Case 1 with ADI
25 20 15 10 1
2 MPN
3 Stratified
Run
4 N=2
5 N=3 15
5
21/09/2015
Case Scenarios – Degradation Z 2/Z 3
16
Case Scenarios – Degradation Z 2/Z 3 LR to minor 2P T1100 Run
N
Mr
Zn
1
2
89:11
0.993
3
3
4
5
1:12:87
0.007
12
2
10:90
0.080
10
3
3:9:89
0.920
2
11:89
0.004
3
87:1:12
0.996
2
89:11
0.806
4
3
3:91:6
0.194
2
2
11:89
2.43E-05
3
87:12:1
0.999
Log(LR)
2
14
8 6
0 1
2
MPN
3 Run
Stratified
4
N=2
5
N=3
17
Case 2 – Collaborative Study 16
14
16
Collaborative Study, Case 2 N
1
Log(LR)
102 3
8 4
Mr
15 Zn
2
71:29
0.999
3
9:22:69
2.98E-05
2
70:30
0.786
3
5:26:69
0.213
2
72:28
0.995
3
27:3:70
0.004
2
28:72
0.923
3
67:26:7
0.076
5
2
6 6
3
72:28
0.699
69:10:21
0.301
2
29:71
0.920
3
68:15:17
0.079
7 4
2
72:28
0.944
7:26:66
0.055
8
2
29:71
0.957
3
14:17:69
0.042
3
14
Log(LR)
12 Run
13
12
11
10 1
2
3
4
MPN
Stratified
N=2
5
13
15
6
7
8
Run
2
N=3
0 1
2
3
4
5
6
7
8
9
10
11
12
Participant number
14
16
17
18
19
20
18
6
21/09/2015
Translating theory into practice… Most mixtures would not warrant this approach. Real life applications limited to complex, higher order mixtures with ambiguity underpinning the estimation of the number of contributors
Ongoing development… Addressing run-to-run variability Replication and optimising run parameters Extending study to expanded contributor ranges Challenging case examples 19
Acknowledgements Stuart Cooper and Laura Russell, ESR, Auckland, NZ Damien Abarno, FSSA, Adelaide, Australia Duncan Taylor, Jo-Anne Bright, John Buckleton, STRmix™ developers
20
7
Examination of a Novel Method to Allow for a Range in the Number of Contributors to a DNA Profile: A Key Area of Subjectivity in DNA Evidence Interpretation
S. Cooper, C. McGovern, L. Russell, D. Abarno, D. Taylor, JA. Bright, J. Buckleton
Complex Mixture Interpretation
Increased sensitivity + expanded evidence types = more and more and more mixtures Criticism of divergent approaches to interpretation
2
Does introducing probabilistic software improve consistency in reporting?
9
1
21/09/2015
Collaborative Study- Case 1 2p 1:1 High t
SJ Cooper, CE McGovern, JA Bright, D Taylor, JS Buckleton Investigating a common approach to DNA profile interpretation using probabilistic software FSI: Genetics, 2015. 16: 121-131.
4
Case 1 – LR to Complainant μ=10.3612 σ=0.0210
Log(LR)
8
6
4
2
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Participant number 5
Collaborative Study – Case 2 Lower t, increased ambiguity below ST
6
2
21/09/2015 Log(LR) 16
14
12
8
10
6
4
2
0 1 2 3 4 5 6 7 8 10 11 12
16
9
Participant number
Case 2 – LR to Person of Interest μ=14.28 σ=0.03
13 14
14 15 16 17
12
18
N=2 N=3
19
μ=8.92 σ=0.74
20
Log(LR)
10
8
6
4
2
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Participant number 7
Collaborative Study - Case 3 Complex mixture, using replicates Replicate 2 1
8
Case 3 - Using Multiple Amplifications 14
μ=11.32 σ=0.74 12
Log (LR)
10
8
6
4
2
0 1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Participant number
9
3
21/09/2015
Collaborative Study Conclusions Greatly improved evidence evaluation can be achieved by implementation of the same probabilistic software
When estimating N is unambiguous it is possible to achieve standardization Assigning an estimated N remains a key point of divergence PCR replicates help improve confidence and consistency Study highlighted the need for further work in addressing uncertainty in the estimation of N
Can this be done statistically? 10
An Extension to STRmix™
D Taylor, J-A. Bright and J.S. Buckleton. Interpreting forensic DNA profiling evidence without specifying the number of contributors. FSI: Genetics, 2014. 13: 269-280
STRmix™ uses an iterative re-sampling process to generate the weights for genotype combinations MCMC (Monte Carlo Markov Chain) samples from a multi-dimensional probability space
11
How does it work in theory? The number of contributor N becomes a dimension of the probability space Performs MCMC in multiple probability spaces – each defined by a different number of contributors N
Can now enter a contributor ‘range’ rather than a single estimate
S jw| HPr( N w| H ) j Pr( 1) S Z Pr( LR j S j'w| HPr( N w| H ) j ' Pr( 2 )S Z Pr( j' n
LR
n
1
n
n'
n'
j
j
j'
j'
| N n , H1 )
j
n'
2
j'
| N n' , H 2 ) 12
4
21/09/2015
How does it perform in practice? Ground Truth Known Profiles – Low t 1P T75 N 1 2 1 2 1 2 1 2 1 2
2 3 4 5
Mr 100 49:51 100 64:36 100 62:38 100 49:51 100 39:61
1P T75
Zn 0.998 0.002 0.921 0.079 0.038 0.962 0.995 0.005 0.752
28 27
Log(LR)
Run 1
26 25 24 23 22
0.248
1
2
3 Run
4
MPN
Stratified
N=1
2p 4:1 T250 N 2 3 2 3 2 3 2 3 2 3
2 3 4 5
Mr 20:80 73:14:13 80:20 11:78:11 21:79 14:79:7 20:80 8:16:76 80:20 16:78:6
Zn 0.244 0.756 0.999 0.001 0.991 0.009 0.198 0.802 0.976 0.024
4
4 3
46:14:14:26 3.017E-13 9:64:27 0.999
4
28:53:11:8 2.086E-04
2
Mr 27:9:64 64:21:9:6 64:9:27 48:36:8:8 28:63:9
Zn 0.999 2.775E-05 0.993 0.007 1.000
26 25 24 23 22
3p 8:3:1 T650, LR to minor
1
2
3 Run
4
Stratified
N=2
25 MPN
24
Log(LR)
3
N 3 4 3 4 3
2p 4:1 T250, LR to minor 28 27
3p 8:3:1 T650 Run 1
5
Log(LR)
Run 1
5
23 22 21 1
2
3
4
Run MPN
13
Stratified
N=3
Case Scenarios – 2p 10:1 Z 2/Z 3
14
Case Scenarios – 2p 10:1 Z 2/Z 3 Case 1 Original Input
2 3 4 5
N 2 3 2 3 2 3 2 3 2 3
Mr 9:91 8:92:0 91:9 3:94:3 91:9 0:6:94 91:9 1:98:1 8:92 2:96:2
Original Input, LR to minor
Zn 0.442 0.558 0.893 0.107 0.997 0.003 0.991 0.009 0.999 1.84E-05
35 30
Log(LR)
Run 1
25 20 15 1
2 3 4 5
N 2 3 2 3 2 3 2 3 2 3
Mr 9:91 7:87:6 92:8 88:7:5 91:9 6:6:88 9:91 8:5:87 9:91 7:88:6
Zn 0.820 0.180 0.991 0.009 0.991 0.009 0.997 0.003 0.986 0.014
3 4 5
N 2 3 2 3 2 3 2 3 2 3
Mr 91:9 86:7:7 9:91 7:6:87 92:8 87:7:7 91:9 7:7:86 9:91 87:7:7
5
with ADI, LR to minor
Zn 3.82E-07 0.999 8.76E-07 0.999 0.012 0.988 1.11E-09 1 0.002 0.998
30 25 20 15 1
Increased Stutter, LR to minor
MPN
35
2
3 Run Stratified
4
5
Ground N=2 Truth
30
Log(LR)
2
4
N=2 truth Ground 35
Case 1 Inc Stutter Run 1
3 Run Stratified
Log(LR)
Run 1
2
MPN
Case 1 with ADI
25 20 15 10 1
2 MPN
3 Stratified
Run
4 N=2
5 N=3 15
5
21/09/2015
Case Scenarios – Degradation Z 2/Z 3
16
Case Scenarios – Degradation Z 2/Z 3 LR to minor 2P T1100 Run
N
Mr
Zn
1
2
89:11
0.993
3
3
4
5
1:12:87
0.007
12
2
10:90
0.080
10
3
3:9:89
0.920
2
11:89
0.004
3
87:1:12
0.996
2
89:11
0.806
4
3
3:91:6
0.194
2
2
11:89
2.43E-05
3
87:12:1
0.999
Log(LR)
2
14
8 6
0 1
2
MPN
3 Run
Stratified
4
N=2
5
N=3
17
Case 2 – Collaborative Study 16
14
16
Collaborative Study, Case 2 N
1
Log(LR)
102 3
8 4
Mr
15 Zn
2
71:29
0.999
3
9:22:69
2.98E-05
2
70:30
0.786
3
5:26:69
0.213
2
72:28
0.995
3
27:3:70
0.004
2
28:72
0.923
3
67:26:7
0.076
5
2
6 6
3
72:28
0.699
69:10:21
0.301
2
29:71
0.920
3
68:15:17
0.079
7 4
2
72:28
0.944
7:26:66
0.055
8
2
29:71
0.957
3
14:17:69
0.042
3
14
Log(LR)
12 Run
13
12
11
10 1
2
3
4
MPN
Stratified
N=2
5
13
15
6
7
8
Run
2
N=3
0 1
2
3
4
5
6
7
8
9
10
11
12
Participant number
14
16
17
18
19
20
18
6
21/09/2015
Translating theory into practice… Most mixtures would not warrant this approach. Real life applications limited to complex, higher order mixtures with ambiguity underpinning the estimation of the number of contributors
Ongoing development… Addressing run-to-run variability Replication and optimising run parameters Extending study to expanded contributor ranges Challenging case examples 19
Acknowledgements Stuart Cooper and Laura Russell, ESR, Auckland, NZ Damien Abarno, FSSA, Adelaide, Australia Duncan Taylor, Jo-Anne Bright, John Buckleton, STRmix™ developers
20
7
