24
24 April 2012 Searching DNA databases with complex DNA profiles: the SmartRank project Hinda Haned [email protected]

Searching DNA databases with complex DNA profiles: the SmartRank project

Embed Size (px)

DESCRIPTION

The presentation explains the principles of DNA database searching via a likelihood ratio model.

Citation preview

Page 1: Searching DNA databases with complex DNA  profiles: the SmartRank  project

24 April 2012

Searching DNA databaseswith complex DNA profiles: the SmartRank project

Hinda [email protected]

Page 2: Searching DNA databases with complex DNA  profiles: the SmartRank  project

DNA mixtures

I A large number of traces recovered from crime stains aremixtures

I Up to two distinct alleles per contributor

I As few as a single allele: allele sharing

Searching DNA databases with complex profiles — November 2013 1

Page 3: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Mixture analysis: a difficult task

I What genotypes created themixture?

• Individual 1 : 13/15• Individual 2 : 14/17

Or• Individual 1 : 13/17• Individual 2 : 14/15

Or• Individual 1 : 13/15• Individual 2 : 14/17• Individual 3 : 15/15

Large number of potential genotypes consistent with the mixture.

Searching DNA databases with complex profiles — November 2013 2

Page 4: Searching DNA databases with complex DNA  profiles: the SmartRank  project

ENFSI recommendation

If possible, mixed DNA-profiles should be interpreted and designatedinto their contributing DNA-profiles. Mixed profiles from (known)victims and (unknown) donors sometimes can be resolved becausethe alleles of the DNA-profile of the victim can be subtracted fromthe mixed profile. The remaining alleles must belong to the unknowndonor.

Searching DNA databases with complex profiles — November 2013 3

Page 5: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Basic deconvolution method: Binary model

I Manual method for the resolution of two-person mixtures. Itrelies upon the experience of the expert along with theapplication of a number of numerical guidelines

Searching DNA databases with complex profiles — November 2013 4

Page 6: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Two person-mixture

I Significant differences in contribution

I Major and Minor profiles can be deduced

I Drop-out from the minor is deemed possible

Searching DNA databases with complex profiles — November 2013 5

Page 7: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Two person-mixture

Major 13,13 14,17 11,11 16,18Minor 14,15 16,16 9,9 19,20

Searching DNA databases with complex profiles — November 2013 6

Page 8: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Complex mixtures

I At least three people involved

I One major (victim), possibly two minors

I Drop-out is deemed possible

Searching DNA databases with complex profiles — November 2013 7

Page 9: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Complex mixtures

I Binary model cannot be applied to high order mixturesI Consequences for data base search:

• Mixed profiles cannot be fully exploited• A person of interest could be in the database, but the

genotype of the minors cannot be deduced

I Perform search with all alleles or with ‘required alleles’ thatmay have come from the person of interest. However:

• Searching with mixtures: increased risk of spurious associations• Risk of reporting a ‘numerical match’: inconsistency with ratio

of contribution in the questioned sample

Solution: Likelihood ratio framework

Searching DNA databases with complex profiles — November 2013 8

Page 10: Searching DNA databases with complex DNA  profiles: the SmartRank  project

The likelihood ratio framework

I Development of a likelihood-ratio model enabling theinterpretation of complex DNA samples:

• DNA mixtures: multiple donors• Low template DNA samples: allele drop-out (missing allele)

and allele drop-in (spurious allele)

I Haned et al, Forensic Sci. Int. Genet. 2012

I Gill & Haned, Forensic Sci. Int. Genet. 2013

Searching DNA databases with complex profiles — November 2013 9

Page 11: Searching DNA databases with complex DNA  profiles: the SmartRank  project

The likelihood ratio framework

LR =Probability of the evidence under prosecution hypothesis

Probability of the evidence under defense hypothesis

I LR = 1: evidence is neutral

I LR > 1: the evidence suppports the prosecution hypothesis

I LR < 1: the evidence suppports the defense hypothesis

Searching DNA databases with complex profiles — November 2013 10

Page 12: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Likelihood ratios and mixtures

I Two alternative hypotheses:• Prosecution hypothesis: the victim, the suspect and one

unknown are the donors• Defense hypothesis: the victim and two unknowns are the

donors

The probability of the evidence is 125,000 more likely if theprosecution hypothesis is true than if the defense hypothesis is true.

Searching DNA databases with complex profiles — November 2013 11

Page 13: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Requirements for LR calculations

I Formulate hypotheses:• Hp: Victim, the individual in the database and one unknown

are the donors• Hd: Victim, and two unrelated unknowns are the donors

I Evaluation of drop-in and drop-out levels

Searching DNA databases with complex profiles — November 2013 12

Page 14: Searching DNA databases with complex DNA  profiles: the SmartRank  project

LRs and database search

Searching DNA databases with complex profiles — November 2013 13

Page 15: Searching DNA databases with complex DNA  profiles: the SmartRank  project

LRs and database search

Searching DNA databases with complex profiles — November 2013 14

Page 16: Searching DNA databases with complex DNA  profiles: the SmartRank  project

LRs and database search

Searching DNA databases with complex profiles — November 2013 15

Page 17: Searching DNA databases with complex DNA  profiles: the SmartRank  project

LRs and database search

Searching DNA databases with complex profiles — November 2013 16

Page 18: Searching DNA databases with complex DNA  profiles: the SmartRank  project

LR distribution for ranked genotypes

Searching DNA databases with complex profiles — November 2013 17

Page 19: Searching DNA databases with complex DNA  profiles: the SmartRank  project

LR distribution for ranked genotypes

Person of interest outlier with

LR >> 109

Searching DNA databases with complex profiles — November 2013 18

Page 20: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Feasibility study: experimental set-up

Searching DNA databases with complex profiles — November 2013 19

Page 21: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Successful extractions rates

Two-person mixtures: % of profiles with a given rankBins for ranks High drop-out Moderate drop-out Low drop-out1 5 58 941-50 11 94 1001-100 11 95 100

Three-person mixtures: % of profiles with a given rank1 0 19 301-50 0 62 941-100 0 72 99

Searching DNA databases with complex profiles — November 2013 20

Page 22: Searching DNA databases with complex DNA  profiles: the SmartRank  project

False positives

False positives for two-person mixturesHigh drop-out Moderate drop-out Low drop-out

Min. 0 11 6Max. 234 1103 473

False positives for three-person mixturesMin. 0 4 124Max. 16 1722 2406

Searching DNA databases with complex profiles — November 2013 21

Page 23: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Limitations

Although complex mixtures can be searched in the database, bearin mind that the true perpetrator may not be in the database

I Hp: Victim, the individual in the database and one unknownare the donors

I Hd: Victim, and two unrelated unknowns are the donors

Probabilistic model ≡ data intelligence

Searching DNA databases with complex profiles — November 2013 22

Page 24: Searching DNA databases with complex DNA  profiles: the SmartRank  project

Future work

I Implementation guidelines:

• Extraction efficiency: to be increased by improving LR-modelparameters

• Robustness of the model vs. partiality of the profiles

I User-friendly free software:• SmartRank software project• Supported by ENFSI Monopoly 2013 project (starting in 2015)• To be distributed for free on-line

Searching DNA databases with complex profiles — November 2013 23