Page 262 - Applied Probability
P. 262

11. Radiation Hybrid Mapping
                                                                                            249
                              Prob. 3 are posterior probabilities calculated under various approximations
                                               L β (γ)+R β (γ)
                                            "
                                                         dγ in formula (11.20). The first approxima-
                              of the integrals
                                              e
                              tion is

                                                                             γ
                                                                       γ
                                                  L β (γ)+R β (γ)
                                                                    L β (ˆ)+R β (ˆ)
                                                 e
                                                                ∝ e
                                                            dγ
                                                                              ,
                                                                                         (11.22)
                              where ˆ γ is the maximum likelihood estimate and the log prior function
                              R β (γ) is taken as 0. The second approximation uses the actual log prior
                              function in (11.22) and replaces the maximum likelihood estimate by the
                              posterior mode. The third approximation is just the Laplace approximation
                              (11.21). For the numbers shown in Table 11.2, all 360 =  6!  orders were
                                                                                   2
                              included in the denominator of (11.20).
                                The three posterior probabilities displayed in Table 11.2 evidently agree
                              well. Except for one minor reversal for the Laplace approximation, the 10
                              listed orders have the same ranks. These posterior probability ranks are
                              roughly similar to the ranks based on minimum obligate breaks.
                                   TABLE 11.2. Best Locus Orders for Diploid Radiation Hybrid Data
                                        Orders         Prob. 1   Prob. 2  Prob. 3   Breaks
                                  1   2  3   4  5  6    .36114    .35690   .34569      52
                                  1   2  3   5  4  6    .32028    .33051   .32845      51
                                  2   3  4   5  6  1    .16736    .16301   .16601      51
                                  2   3  5   4  6  1    .14554    .14451   .15400      51
                                  1   2  3   6  5  4    .00244    .00222   .00233      54
                                  1   4  5   6  3  2    .00136    .00119   .00128      54
                                  1   3  2   4  5  6    .00054    .00045   .00053      56
                                  1   2  3   5  6  4    .00038    .00036   .00054      54
                                  1   3  2   5  4  6    .00024    .00022   .00029      55
                                  1   4  6   5  3  2    .00021    .00019   .00029      54
                                The failure of Table 11.2 to identify a decisively best order reflects uncer-
                              tainties in placing locus 1 to the right or to the left of the major cluster of
                              loci and in reversing loci 4 and 5 in this cluster. The maximum likelihood
                              odds for pair reversals under the best identified order are given in the dia-
                              gram below. It is interesting that the odds for inverting loci 1 and 2 provide
                              no hint of the overall ambiguity in ordering locus 1. Clearly, caution should
                              be exercised in interpreting pairwise inversion odds.

                                         Pairwise Inversion Odds for the Best Order

                                           5.8×10 9  6.7×10 2  1.7×10 13  1.1  1.5×10 4
                                         1 —— 2 —— 3 —— 4 —— 5 —— 6
                                Maximum likelihood estimates of the interlocus distances under the best
                              order are as follows:
   257   258   259   260   261   262   263   264   265   266   267