Towards Commercialization: Preliminary Developmental Validation of a High Resolution Melt Curve Mixture Prediction Assay and SVM Tool, Virginia, 2020-2022

In the current study, roughly 170 single source samples and 32 two-person mixture samples were tested using both the integrated Quantiplex®-high resolution melt (HRM) assay and Quantifiler™ Trio-HRM assay, then the entire HRM datasets were exported for prediction modeling using both linear discriminate analysis (LDA) and support vector machine (SVM) algorithms in R Studio software. For proof-of-concept, only 8 different genotypes, including a genotype of "mixture", were represented (for each locus) in testing. A portion of the samples tested were used to "train" the software and the remaining sample data was used as unknowns (or "validation") samples for prediction. When samples were tested in the Quantiplex®-HRM assay, an overall accuracy of 87.88 percent was exhibited, correctly classifying 87.5 percent of single source samples as such and 90 percent of mixture samples. Similarly, when samples were tested in the Quantifiler™ Trio-HRM assay an overall accuracy of 79.2 percent was exhibited, with 89.2 percent of single source samples accurately classifying and 43.8 percent of mixtures accurately classifying. Additionally, quantification values obtained from the integrated assays as well as the quality metrics such as the slope, R2, and y-intercept, were not significantly different than those obtained in the standard assays.