We then examined the effect of uncertain allele designations by randomly designating some alleles of B as uncertain, first with Pr(unc) = 0.4 and then Pr(unc) = 0.8. In both conditions, at each locus and in each replicate a Poisson mean one number of alleles not in the profile of B was also designated as uncertain, with types

randomly selected according to frequencies in the UK Caucasian database. For all these simulated profiles, one-contributor hypotheses were compared, B under Hp and X under Hd. Next two-contributor CSPs were simulated, based on the profiles of A and C. Two conditions were simulated, both used PrA(D) = 0.2, while PrC(D) was initially 0.8 and then 0.6. Dropin was not simulated. For shared alleles the dropout probability was the product of the dropout probabilities for each ZD1839 molecular weight contributor having that allele. Two-contributor hypotheses were compared, with each of A and C in turn taking the role of Q, while the other was treated as unknown in the analysis. Additionally one-contributor-plus-dropin hypotheses were compared, only for A playing the role of Q ( Table 3). Three-contributor CSPs were then simulated under three conditions, with dropout

probabilities for Donors A, B and C as shown in Table 3. Dropin was included as for the one-contributor simulations. Three-contributor hypotheses were compared, with A playing the role of Q and the other two contributors being treated as unknown. We used a CSP from an

actual crime investigation, consisting of five replicates: two using standard SGM+ profiling and three generated using an LCN protocol with 34 PCR cycles (Table 4). This example MLN8237 was submitted to us for likeLTD analysis, and as is typical only limited information about the profiling protocol was provided by the profiling lab. These details are not required by likeLTD because it estimates the unknown parameters from the CSP allele designations. We re-sampled the five actual replicates to generate simulated profiles with up to eight replicates, consisting of standard replicates only, sensitive replicates only, or both. Six distinct Sclareol alleles were observed at locus D8, but no more than three replicated alleles were observed at any locus. Three-contributor hypotheses were compared, with all contributors unknown under Hd, and no dropin ( Table 3). For the good-template experiments (500 pg), Fig. 1 (left) shows that the ltLR equals the IMP for all numbers of replicates (one through eight). This is the expected result, and the exercise shows that in this simple setting there is no deterioration in the quality of the computed LR for large numbers of replicates. Low DNA template (60 pg) generates an ltLR about 1.6 bans below the IMP for one replicate, but the gap is very small for two replicates and is negligible for larger numbers of replicates. For very low DNA template (15 pg) the ltLR is just under 6 bans for a single replicate, about 6 bans below the IMP.