^{1,2}, Anders Nordgaard

^{2,3}, Birgitta Rasmusson

^{2}, Ricky Ansell

^{2,4}, and Peter Rådström

^{1}

The principal component was used instead of only the manual grading scale, since the scale could be biased and the intra-relationships between the three measures are not included in the manual scale. Using the principal component includes the relationships between the three measures, and the shrinking is a compromise between a measure based on data and one based on common knowledge. Terms describing this procedure are “data-driven” or “model-assisted” grading.

In the current study, different DNA polymerase–buffer systems were compared with respect to forensic DNA profile quality using analysis of variance of the FI, as defined above. The measurements were performed in such a way that a design can be identified, although not balanced. The default case is two replicates for each analysis of each sample. With the existing data, we can compare the population means of the FI and make pair-wise comparisons using Ampli*Taq* Gold as the reference system. Upon investigating the data further, it was found to be beneficial to apply a logarithmic transformation of the values of FI to make them more normally and homoscedastically distributed. The resulting least significant difference between sample means is then exponentiated so that it can be interpreted on the original scale, and the result is a least significant ratio between two sample means.

We investigated the interpretation of the values of FI from a sample using two replicate measurements with each of the DNA polymerase–buffer systems. It is clear that a larger ratio between sample means from such a set of measurements means that the two polymerase systems differ and, in particular, that one system is better than the other. To find an approximation of this ratio, we used the samples for which the FI was obtained (at least once) for each of the polymerase systems studied. A one-way analysis of variance was performed on each of these samples (logarithmic values) and a pooled sample variance was calculated. This estimated variance was then used to calculate the least difference in the means of logarithms of profile indices required to be able to state that a particular system has a higher population FI than the control system. Finally, this least significant difference is transformed into a least significant ratio (LSR) of the geometric means of the original profile indices (FIgm). More specifically, if *x*_{1} and *x*_{2} are the two replicate FI values obtained with the polymerase system being investigated, and *y*_{1} and *y*_{2} are the corresponding indices obtained with the reference system, the following inequality should hold if the chemistry of the polymerase system investigated is significantly better than that of the reference system

where LSR is the estimated least significant ratio at a specified level of significance. We suggest a family significance level of 10% for the three systems compared with Ampli

*Taq*Gold, resulting in an LSR of 1.985.

FI cannot be calculated for blank electropherograms/DNA profiles. Therefore, it was defined as 0.05 in these cases. This is the lowest value obtainable, given by a profile with only one detected peak at the detection limit (50 rfu).

**Results**

**Success rate of routine DNA profiling of crime scene saliva stains**

Twelve percent (232 of 1936) of the investigated saliva crime scene samples produced blank electropherograms/DNA profiles using standard Ampli*Taq* Gold DNA polymerase. 19.2 percent of samples from cigarette butts (174 of 907 samples) and 2.6% of samples from swabs from bottles, cans, and foodstuffs (22 of 843 samples) produced blank DNA profiles.

**Analysis of standardized mock crime scene saliva samples using a forensic singleplex real-time PCR assay**

Five of the nine DNA polymerases (Bio-X-Act Short, Ex*Taq* Hot Start, KAPA2G Robust, Omni*Taq,* and PicoMaxx High Fidelity) produced average amplification efficiencies around the ideal 1.0 (0.93–1.12) (Table 1). The other four (Ampli*Taq* Gold, r*Tth, Taq,* and *Tth*) gave higher values, from 1.26 to 1.46, and also showed higher standard deviations. The linear ranges of amplification differed considerably between polymerases, from 3.3 log units for Bio-X-Act Short and PicoMaxx High Fidelity, down to 1.3 log units for Ampli*Taq* Gold (Figure 1, A and B). The detection limit for the least sensitive polymerases, r*Tth* and *Tth,* was almost 20× higher than the limit for the most sensitive polymerase, Bio-X-Act Short (Table 1; Figure 1, A and D). Based on these results, Bio-X-Act Short, Ex*Taq* Hot Start, and PicoMaxx High Fidelity were deemed the most robust. All the analyzed negative amplification controls were blank.