Title: Strategies and calculations in DNA kinship cases
1Strategies and calculations in DNA kinship cases
- Charles Brenner
- consulting in forensic mathematics
- because I thought of it
2Outline
- Kinship method
- Likelihood ratios
- Paternity, Avuncular
- Mutation
- old way, suggested new way
- DNAVIEW demonstration
3Kinship method
- Genetic evidence
- Likelihood ratio
- Kinship program
- ref Brenner, CH Symbolic Kinship Program,
Genetics 145535-542, 1997 Feb
4Likelihood ratio
- Kinship I (Basic)
- Paternity
- avuncular
- Vs. exclusion
5What a likelihood ratio is
- Compares two explanations for data
- The heart of forensic mathematics
- http//dna-view.com
- cbrenner_at_uclink.berkeley.edu
6Likelihood ratio for being French
- Data Subject speaks 100 French words in 1 hour
- Explanation 1 subject is French
- 20 event
- Explanation 2 subject is not French
- 1 event
- LR20 Data is 20 times more characteristic of
French person
7Paternity why likelihood ratio?
8Likelihood ratio for paternity (PI)
- PI X/Y, where
- XP(genetic types manfather)
- YP(genetic types man not father)
- Interpretations
- Odds favoring paternity over non-paternity
assuming all other evidence is equally divided - Evidence is PI times more characteristic of
paternity
9What the exclusion method is
- Considers only one hypothesis
- which it assumes may be disproven by some data
sets - -- an artificial assumption at best (what about
mutations? Laboratory error?) - and completely useless in many situations
- siblingship
- uncle
10Paternity how likelihood ratio?
11Likelihood ratio for Paternity (PI)
- Data MotherPS, ChildPQ, ManRQ
- explanation 1 man is father
- (2ps)(2qs)/4 event
- explanation 2 not father his Q is coincidence
- (2ps)(2qs)(q/2) event
- LR1/(2q)
- If q1/20, data 10 times more characteristic of
father explanation
PS
RQ
PQ
PS
RQ
PQ
12Avuncular index (Is the man an uncle?)
13Other kinship cases
- Kinship program
- missing person
- null alleles
14Missing person kinship case
Explanation 1
Explanation 2
PS
PQ
PS
Mom
Body
PQ
PQ
PQ
Bro
Missing child
Body
Bro pq, Body pq Mom ps Dad
Bro pq, Child Mom pq Dad
LR(1q)/8pq -- e.g. about 14 if pq0.1
15Null allele
Explanation 1
Explanation 2
PS
PS
Q
Q-
P
P-
- LR o / (2op)(oq), where onull allele
frequency - e.g LR1/7 if pq0.25 and o0.01
16PI when possible mutation
- Concept
- Old method
- Str data
- New method
17Mutation analysis concept
- Data MotherPS, ChildPQ, ManRT
- Likelihood ratio analysis -- compute probability
of data assuming - Explanation 1 Paternity plus mutation
- need model of mutation
- Explanation 2 Nonpaternity real father ? Q
- 2 usually better explanation.
- LR
18Paternity case one exclusion
- PI20000 (combined, 4 loci)
- PI1/500 (PS, PQ, RT locus)
- overall PI 40.
- Conclusion probably paternity mutation
19Paternity case two exclusions
- PI2000 (combined, 3 loci)
- PI1/500 (PS, PQ, RT locus)
- PI1/200 (PS, PQ, YZ locus)
- overall PI 1/50.
- Conclusion probably non-paternity
20Mutation (old method)
- Data MotherPS, ChildPQ, ManRT
- Explanation 1 Paternity T or R ? N
- man transmits T (or R -- doesnt matter)
- ? chance T mutates
- ?? chance it ends up as Q
- Explanation 2 Nonpaternity real father ? Q
- sperm is Q
21Old mutation formula
- LR ?
- ?rate of mutations/meosis, e.g. 1/1000.
- Correct on average
- Much too low for small changes
- Much too high for big changes
22Old mutation formula is
- simple, but badly inaccurate
- suggests the 2 exclusion rule
- rule probably more accurate than the formula
- rule adequate for RFLPs
- rule not adequate for STRs
- need a new formula
23Mutation model (old formula)
Population fragment size distribution used
as mutation product size distribution
Allele of
Paternal allele
alleged father
of child
24Mutation model (more realistic)
mutation product size distribution
Population fragment size distribution
Allele of
Paternal allele
alleged father
of child
25Reasonable mutation model, STRs
Suggested model m ? /2 if s ? 1 m ? /20
if s ? 2 etc. where ? total (paternal)
mutation rate. c.f. Brinkmann et al
26Mutation LR (new, for STRs)
- Data MotherPS, ChildPN, ManRÑ
- Explanation 1 Paternity RÑ ? N
- 50 chance transmit Ñ
- ? chance Ñ mutates
- m chance Ñ ends up as N, assuming mutation
- m 0.5 if N, Ñ are 1 step apart
- m0.05 if 2 steps, etc
- Explanation 2 Real father ? N
- LR ?/(4q) (assuming single step)
- qallele frequency of the paternal allele N
27STR mutation rates
D12S1090 43/5272 paternal THO1 1/316
maternal
28Mutation reference
- http//dna-view.com/mudisc.htm
29Kinship II (advanced)
- More than two scenarios
- Three
- Many
- Strategies
- Whom to test
- Example
- Whether effective result is likely
30More than two scenarios
31Three scenarios --
- Father?
- Uncle?
- Unrelated?
32Father/Uncle/Unrelated analysis
Example LR for tested man being father, vs
uncle, is 53
33Likelihood ratios are transitive
- means that if explanation father is s times
better than uncle - and uncle is t times better than unrelated
- then father explains data st times better than
unrelated.
34Summary
- Likelihood ratios are the way to quantify
evidence - Mutation calculation must be changed for STRs
- Kinship
- All kinship problems have an explicit solution
- Multiple scenarios
- Triple ratio for three scenarios
- Lattice approach for the most complicated
situations
35Thanks
- Prof Antonio Alonso GEP-ISFH
- Audience for sitting patiently