We first discuss the permutation property for one SNP. Consider a nuclear family with two parents, m affected and n unaffected siblings. Let πΊππ be the number of allele 1 transmitted from parents, and let πΊπππ be the number of non-transmitted allele 1 from parents, to affected sib i , where i=1,2,…,m Let πΊ ′ ππ be the number of allele 1 transmitted from parents, and let πΊ ′ πππ be the number of non-transmitted allele 1 from parents, to unaffected sib j , where j=1,2,…,n Table A.1 shows the transmitted and non-transmitted statistics for the affected and unaffected sibs for the nuclear family. Since the total number of allele 1 in parents is fixed, πΊπ1 + πΊππ1 = πΊπ2 + πΊππ2 = β― = πΊππ + πΊπππ = πΊ ′ π1 + πΊ ′ ππ1 = πΊ ′ π2 + πΊ ′ ππ2 = β― = πΊ ′ ππ + πΊ ′ πππ If m=n πΊπ1 − πΊ ′ π1 = πΊ ′ ππ1 − πΊππ1 πΊπ2 − πΊ ′ π2 = πΊ ′ ππ2 − πΊππ2 . . . πΊππ − πΊ ′ ππ = πΊ ′ πππ − πΊπππ π n π n ′ ′ ∴ ∑ πΊππ − ∑ πΊ ππ = ∑ πΊ πππ − ∑ πΊπππ − − − − − (1) π=1 π π=1 π=1 m π=1 π m ∑ πΊππ − ∑ πΊ ′ ππ = ∑ πΊ ′ πππ − ∑ πΊπππ − − − − − (2) π=1 π=1 π=1 π=1 We define ππ = number of triads (two parents and one affected offspring) and ππ = number of discordant sib pair (one affected and one unaffected sibs) in the nuclear family. Let ππ be the sum of difference in allele counts between the transmitted and the nontransmitted alleles over all triads. Let ππ be the sum of difference in allele counts between the affected and unaffected sibs over all discordant sib pairs. π Therefore , ππ = ∑(πΊππ − πΊπππ ) π=1 π π If π = π, ππ = ∑ πΊππ − ∑ πΊ ′ ππ π=1 π π=1 π If π ≠ π, ππ = π ∑ πΊππ − π ∑ πΊ ′ ππ π=1 π=1 The PDT statistic for this nuclear family is calculated as 1 ππ +ππ (ππ + ππ ) After permuting the transmitted and non-transmitted alleles, the statistics for ππ and ππ can be calculated as: m π(π) = ∑(πΊπππ − πΊππ ) i=1 π n π(π) = ∑ πΊπππ − ∑ πΊ ′ πππ π=1 j=1 We investigate the relationship between ππ + ππ and π(π) + π(π) in the following three cases. (Case 1) The numbers of affected and unaffected sibs are equal. (m=n) π m ππ = ∑(πΊππ − πΊπππ ) = − ∑(πΊπππ − πΊππ ) = −π(π) π=1 π i=1 π π π ππ = ∑ πΊππ − ∑ πΊ ′ ππ = ∑ πΊππ − ∑ πΊ ′ ππ π=1 π=1 π π=1 π=1 π n n ′ π(π) = ∑ πΊπππ − ∑ πΊ πππ = ∑ πΊπππ − ∑ πΊ ′ πππ π=1 π j=1 π ππ = ∑ πΊππ − ∑ πΊ ′ ππ π=1 π=1 π=1 j=1 n π π n ππππ (1) ∑ πΊ ′ ′ πππ − ∑ πΊπππ = − (∑ πΊπππ − ∑ πΊ πππ ) = π=1 π=1 π=1 j=1 = −π(π) ∴ ππ + ππ = −(π(π) + π(π) ) (Case 2) The number of the affected sibs is greater than the number of the unaffected sibs. (m>n) In this case, ππ and π(π) have the same form as in case 1. Therefore, we only discuss ππ and π(π) here. π π ππ = π ∑ πΊππ − π ∑ πΊ ′ ππ π=1 π=1 π π π(π) = π ∑ πΊπππ − π ∑ πΊ ′ πππ π=1 π=1 π π ππ = π ∑ πΊππ − π ∑ πΊ ′ ππ π=1 π=1 π π π π = π ∑ πΊππ − π ∑ πΊ ′ ππ + π ∑ πΊππ − (π − π) ∑ πΊ ′ ππ π=1 π=1 π π=π+1 π π=1 π π ′ = π (∑ πΊππ − ∑ πΊ ππ ) + ∑ (πΊππ − πΊ π=1 π=1 π π ′ π1 ) + … + π=π+1 ∑ (πΊππ − πΊ ′ ππ ) π=π+1 π π = π (∑ πΊππ − ∑ πΊ ′ ππ ) + ∑ ∑(πΊππ − πΊ ′ ππ ) π=1 π=1 π π=π+1 π=1 π π π ′ = π (∑ πΊ πππ − ∑ πΊπππ ) + ∑ ∑(πΊ ′ πππ − πΊπππ ) − − − πππππππππ π‘π (1) π=1 π π=1 π=π+1 π=1 π π(π) = π ∑ πΊπππ − π ∑ πΊ ′ πππ π=1 π=1 π π π π ′ = π ∑ πΊπππ − π ∑ πΊ πππ + π ∑ πΊπππ − (π − π) ∑ πΊ ′ πππ π=1 π π=1 π=π+1 π π π=1 π = π (∑ πΊπππ − ∑ πΊ ′ πππ ) + ∑ (πΊπππ − πΊ ′ ππ1 ) + … + ∑ (πΊπππ − πΊ ′ πππ ) π=1 π=1 π=π+1 π=π+1 π π π π ′ = π (∑ πΊπππ − ∑ πΊ πππ ) + ∑ ∑(πΊπππ − πΊ ′ πππ ) π=1 π π=1 π=π+1 π=1 π π π ππ = π (∑ πΊ ′ πππ − ∑ πΊπππ ) + ∑ ∑(πΊ ′ πππ − πΊπππ ) π=1 π=1 π π=π+1 π=1 π π π ′ = − {π (∑ πΊπππ − ∑ πΊ πππ ) + ∑ ∑(πΊπππ − πΊ ′ πππ )} = −π(π) π=1 π=1 π=π+1 π=1 ∴ ππ + ππ = −(π(π) + π(π) ) (Case 3) The number of the affected sibs is less than the number of the unaffected sibs. (m<n) In this case, ππ and π(π) have the same form as in case 1. Therefore, we only discuss ππ and π(π) here. π π ππ = π ∑ πΊππ − π ∑ πΊ ′ ππ π=1 π=1 π π π(π) = π ∑ πΊπππ − π ∑ πΊ ′ πππ π=1 π π=1 π ππ = ∑ πΊππ − π ∑ πΊ ′ ππ π=1 π=1 π π π π ′ = π ∑ πΊππ − π ∑ πΊ ππ + (π − π) ∑ πΊππ − π ∑ πΊ ′ ππ π=1 π π=1 π π=1 π π=π+1 π = π (∑ πΊππ − ∑ πΊ ′ ππ ) + ∑(πΊππ − πΊ ′ ππ+1 ) + … + ∑(πΊππ − πΊ ′ ππ ) π=1 π=1 π=1 π=1 π π π π ′ = π (∑ πΊππ − ∑ πΊ ππ ) + ∑ ∑ (πΊππ − πΊ ′ ππ ) π=1 π=1 π π=1 π=π+1 π π π = π (∑ πΊ ′ πππ − ∑ πΊπππ ) + ∑ ∑ (πΊ ′ πππ − πΊπππ ) − −πππππππππ π‘π (2) π=1 π=1 π π=1 π=π+1 π π(π) = π ∑ πΊπππ − π ∑ πΊ ′ πππ π=1 π=1 π π π π = π ∑ πΊπππ − π ∑ πΊ ′ πππ + (π − π) ∑ πΊπππ − π ∑ πΊ ′ πππ π=1 π=1 π π=1 π π=π+1 π π ′ = π (∑ πΊπππ − ∑ πΊ πππ ) + ∑(πΊπππ − πΊ πππ+1 ) + … + ∑(πΊπππ − πΊ ′ πππ ) π=1 π=1 π π ′ π=1 π=1 π π = π (∑ πΊπππ − ∑ πΊ ′ πππ ) + ∑ ∑ (πΊπππ − πΊ ′ πππ ) π=1 π=1 π π=1 π=π+1 π π π ′ ππ = π (∑ πΊ πππ − ∑ πΊπππ ) + ∑ ∑ (πΊ ′ πππ − πΊπππ ) π=1 π=1 π π=1 π=π+1 π π π = − {π (∑ πΊπππ − ∑ πΊ ′ πππ ) + ∑ ∑ (πΊπππ − πΊ ′ πππ )} = −π(π) π=1 π=1 π=1 π=π+1 ∴ ππ + ππ = −(π(π) + π(π) ) Because 1 ππ +ππ is a constant, in Cases 1,2and3. 1 ππ +ππ (ππ + ππ ) = − 1 ππ +ππ (π(π) + π(π) ) is satisfied Therefore, if we permute the transmitted and non-transmitted alleles simultaneously from both parents to all the children, it simply results in a sign change in the PDT statistic. The argument is still true when we have different numbers of affected and unaffected siblings. Since there can be LD among SNPs, we need to permute haplotypes to preserve the LD structures among SNPs when analyzing multiple SNPs. We then discuss the relationship between the permutation of haplotypes and the single-SNP PDT statistics for two SNPs. π We use ( ) to denote a haplotype for two SNPs, where a is the allele at the first π SNP, and b is the allele at the second SNP. Also consider a nuclear family with two parents, m affected and n unaffected siblings. 1 Let π»ππ be the number of haplotype ( ) transmitted from parents, and let π»πππ be 1 1 the number of non-transmitted haplotype ( ) from parents, to affected sib i , where 1 i=1,2,…,m 1 Let π» ′ ππ be the number of haplotype ( ) transmitted from parents, and let π» ′ πππ 1 1 be the number of non-transmitted haplotype ( ) from parents, to unaffected sib j , 1 where j=1,2,…,n 1 Table A.2 shows the transmitted and non-transmitted statistics for haplotype ( ) for 1 affected and unaffected sibs for the nuclear family. 1 Since the number of haplotype ( ) in parents is fixed, π»π1 + π»ππ1 = π»π2 + π»ππ2 = 1 β― = π»ππ + π»πππ = π» ′ π1 + π» ′ ππ1 = π» ′ π2 + π» ′ ππ2 = β― = π» ′ ππ + π» ′ πππ If m=n π»π1 − π» ′ π1 = π» ′ ππ1 − π»ππ1 π»π2 − π» ′ π2 = π» ′ ππ2 − π»ππ2 . . . π»ππ − π» ′ ππ = π» ′ πππ − π»πππ π n π n ′ ′ ∴ ∑ π»ππ − ∑ π» ππ = ∑ π» πππ − ∑ π»πππ − − − − − (1) π=1 π π=1 m π=1 m π=1 π ∑ π»ππ − ∑ π» ′ ππ = ∑ π» ′ πππ − ∑ π»πππ − − − − − (2) π=1 π=1 π=1 π=1 Let ππ be the sum of the difference in the haplotype counts between the transmitted and the non-transmitted haplotypes overall all triads. Similarly, let ππ be the sum of the difference in haplotype counts between the affected and unaffected sibs over all discordant sib pairs. π Therefore , ππ = ∑(π»ππ − π»πππ ) π=1 π π If m = n, ππ = ∑ π»ππ − ∑ π» ′ ππ π=1 π=1 π π If m ≠ n, ππ = π ∑ π»ππ − π ∑ π» ′ ππ π=1 π=1 1 The PDT statistic for haplotype ( ) for this nuclear family is calculated as π·(1) = 1 1 1 ππ +ππ (ππ + ππ ) After permuting the transmitted and non-transmitted haplotypes, then m π(π) = ∑(π»πππ − π»ππ ) i=1 π n π(π) = ∑ π»πππ − ∑ π» ′ πππ π=1 π(1) = 1 j=1 1 (π + π(π) ) ππ + ππ (π) Using the similar procedures for the single-SNP situation, we can show that π·(1) = 1 −π(1) 1 1 2 2 The same procedure can also be applied to haplotypes ( ) , ( ) , ( ) and the same 2 1 2 conclusion can be made. π Define π·(π) is the PDT statistic for haplotype ( ) at two SNPs π π π π(π) is the PDT statistics for haplotype ( ) after permuting the transmitted π π and non-transmitted haplotypes , where a=1,2 and b=1,2 Let the PDT statistics for allele 1 at the first and second SNPs be π·1 and D1′ , respectively. Similarly, let the PDT statistics for allele 2 at the first and second SNPs be π·2 and D2′ , respectively. It can be shown that π·1 = π·(1) + π·(1) = −π(1) − π(1) = − (π(1) + π(1) ) = −π·(1) 1 2 1 2 1 2 π·2 = π·(2) + π·(2) = −π(2) − π(2) = − (π(2) + π(2) ) = −π·(2) 1 2 1 2 1 2 π·1′ = π·(1) + π·(2) = −π(1) − π(2) = − (π(1) + π(2) ) = −π·(1′ ) 1 1 1 1 1 1 π·2′ = π·(1) + π·(2) = −π(1) − π(2) = − (π(1) + π(2) ) = −π·(2′ ) 2 2 2 2 2 2 Thus, if we permute the transmitted and non-transmitted haplotypes, it simply results in simultaneous sign changes in the PDT statistics for the two SNPs. We then check if the results are still true when a haplotype has recombination. Assume we consider a nuclear family with one affected and one unaffected siblings and only the affected sib has a haplotype with recombination from the mother. We only consider the case where the genotypes at the two SNPs in the mother are both heterozygous. If at least one SNP is homozygous, the observation of haplotype in the affected sib is the same as the situation of no recombination. Then the same proof from the condition of no recombination can be applied. 1 2 Therefore, we consider the condition that the mother has haplotype ( ) ( ). Then, 1 2 we would like to show that π·(1) + π·(1) = −π(1) − π(1) is still true. 1 2 1 2 π Let ππ΄ππ be the number of haplotype ( ) transmitted from parents and πππ΄ππ be π π the number of non-transmitted haplotype ( ) from parents to the affected child. π π Let ππππ be the number of haplotype ( ) transmitted from parents and πππππ be π π the number of non-transmitted haplotype ( ) from parents, to the unaffected child. π Table A.3 summarizes the transmitted and non-transmitted statistics for haplotype π ( ) for the affected and unaffected siblings. π 1 For haplotype ( ) 1 1 If there is no recombination in the mother, then the number of haplotype ( ) in 1 parents is fixed, ππ΄11 + πππ΄11 = ππ11 + πππ11 1 However, since haplotype ( ) has recombination in the mother and only the affected 1 child has the haplotype with recombination from the mother, the number of haplotype 1 1 ( ) in the affected child is the number of haplotype ( ) in the unaffected child 1 1 minus 1. Therefore, ππ΄11 + πππ΄11 + 1 = ππ11 + πππ11 ∴ ππ΄11 − ππ11 = πππ11 − πππ΄11 − 1 − − − − −(1) Let ππ 1 be the difference in the haplotype counts between the transmitted and non( ) 1 transmitted haplotypes. Similarly, let ππ 1 be the difference in the haplotype counts ( ) 1 between the affected and the unaffected siblings. Therefore , ππ 1 = ππ΄11 − πππ΄11 , ππ 1 = ππ΄11 − ππ11 ( ) 1 ( ) 1 In this case, nT =1 and nS =1 1 The PDT statistic for haplotype ( ) for this nuclear family is calculated as π·(1) = 1 1 1 2 (ππ 1 + ππ 1 ) ( ) 1 ( ) 1 After permuting the transmitted and non-transmitted labels, then ππ 1 = πππ΄11 − ππ΄11 ( ) 1 ππ 1 = πππ΄11 − πππ11 ( ) 1 Therefore, ππ 1 = ππ΄11 − πππ΄11 = −(πππ΄11 − ππ΄11 ) = −ππ 1 ( ) 1 ( ) 1 ππ 1 = ππ΄11 − ππ11 = πππ11 − πππ΄11 − 1 = −ππ 1 − 1 ( ) 1 ( ) 1 − − − ππ¦ (1) 1 For haplotype ( ) 2 1 Similar to haplotype ( ), we can show that ππ΄12 + πππ΄12 = ππ12 + πππ12 + 1 1 ∴ ππ΄12 − ππ12 = πππ12 − πππ΄12 + 1 − − − − −(2) Also define ππ 1 = ππ΄12 − πππ΄12 , ππ 1 = ππ΄12 − ππ12 ( ) 2 ( ) 2 1 The PDT statistic for haplotype ( ) for this nuclear family is calculated as π·(1) = 2 2 1 2 (ππ 1 + ππ 1 ) ( ) 2 ( ) 2 After permuting the transmitted and non-transmitted haplotypes, then ππ 1 = πππ΄12 − ππ΄12 ( ) 2 ππ 1 = πππ΄12 − πππ12 ( ) 2 Therefore, ππ 1 = ππ΄12 − πππ΄12 = −(πππ΄12 − ππ΄12 ) = −ππ 1 ( ) 2 ( ) 2 ππ 1 = ππ΄12 − ππ12 = πππ12 − πππ΄12 + 1 = 1 − ππ 1 ( ) 2 ∴ π·(1) + π·(1) = 1 2 = ( ) 2 − − − ππ¦ (2) 1 (π + ππ 1 + ππ 1 + ππ 1 ) ( ) ( ) ( ) nT + nS π(11) 1 2 2 1 (−ππ 1 − ππ 1 − 1 – ππ 1 + 1 − ππ 1 ) ( ) ( ) ( ) ( ) nT + nS 1 1 2 2 =− 1 (π + ππ 1 + ππ 1 + ππ 1 ) = −π(1) − π(1) ( ) ( ) ( ) nT + nS π(11) 1 2 1 2 2 The argument is still true when we have different numbers of affected and unaffected siblings. The procedure can also be extended to multiple SNPs and it can be shown that permuting transmitted and non-transmitted haplotypes at multiple SNPs results in simultaneous sign changes for all of the SNPs. Figures Figure 1.A. Independent haplotypes are labeled with different colors. Haplotypes transmitted from the mother to the affected sib have a crossover between the two SNPs. Tables Table A.1 Transmitted Non Transmitted Affected Affected Affected Unaffected Unaffected Unaffected sib 1 sib 2 sib m sib 1 sib 2 sib n πΊπ1 πΊπ2 … πΊππ πΊ ′ π1 πΊ ′ π2 … πΊ ′ ππ πΊππ1 πΊππ2 … πΊπππ πΊ ′ ππ1 πΊ ′ ππ2 … πΊ ′ πππ Affected Affected Affected Unaffected Unaffected π»π1 π»π2 … π»ππ π» ′ π1 π» ′ π2 … π» ′ ππ π»ππ1 π»ππ2 … π»πππ π» ′ ππ1 π» ′ ππ2 … π» ′ πππ Table A.2 Transmitted Non Transmitted Table A.3 Affected Transmitted Non Transmitted Unaffected ππ΄ππ ππππ πππ΄ππ πππππ Unaffected