HaploGrep 2 is a web application that communicates through a REST API with the web server. Thus, all computation intensive tasks are executed directly on the server. The haplogroup classification itself is based on pre-calculated phylogenetic weights that correspond to the occurrence per position in Phylotree and reflecting the mutational stability of a variant. In the updated classification algorithm, the weights are now scaled from 1 to 10 in a non-linear way (see Supplementary Table S1). Thus, the rare occurrences of variants in Phylotree will no longer influence the classification toward those haplogroups as much as in the previous version. Once the data is imported, the haplogroup classification is started automatically. Optimizations within the code led to a 20-fold speed-up compared to HaploGrep 1. By storing only the 50 highest ranked haplogroups per sample the memory consumption could be reduced significantly.
Furthermore, new dissimilarity metrics for the mtDNA haplogroup classification were introduced. In addition to the already implemented Kulczynski distance (1 (link)), the Jaccard index, the Hamming distance and the Kimura 2-parameter distance were included (24 ) (see Supplementary Table S2 and 3 for performance comparison). Further major improvements included a check for artificial recombination (25 (link)) and a check for systematic artefacts and for rare or potential phantom mutations (26 (link)). For detecting artificial recombination, we apply two different strategies: the first strategy, proposed by Kong et al. (27 (link)), counts the remaining variants that were not assigned to the resulting best haplogroup, and tests whether these variants could be assigned to another haplogroup. For this step, mutational hotspots are excluded (e.g. 315.1C or 16519). The second recombination strategy assumes prior knowledge about the specific placement of the fragments of the polymerase chain reaction products (amplicons). With this information in hand, a check comparing the profiles relative to the fragment ranges can be executed. The user-defined fragments are generated, and the profiles split accordingly. If the distance of both haplogroup fragments exceeds five phylogenetic nodes, the sample is listed as potentially contaminated.
Furthermore, new dissimilarity metrics for the mtDNA haplogroup classification were introduced. In addition to the already implemented Kulczynski distance (1 (link)), the Jaccard index, the Hamming distance and the Kimura 2-parameter distance were included (24 ) (see Supplementary Table S2 and 3 for performance comparison). Further major improvements included a check for artificial recombination (25 (link)) and a check for systematic artefacts and for rare or potential phantom mutations (26 (link)). For detecting artificial recombination, we apply two different strategies: the first strategy, proposed by Kong et al. (27 (link)), counts the remaining variants that were not assigned to the resulting best haplogroup, and tests whether these variants could be assigned to another haplogroup. For this step, mutational hotspots are excluded (e.g. 315.1C or 16519). The second recombination strategy assumes prior knowledge about the specific placement of the fragments of the polymerase chain reaction products (amplicons). With this information in hand, a check comparing the profiles relative to the fragment ranges can be executed. The user-defined fragments are generated, and the profiles split accordingly. If the distance of both haplogroup fragments exceeds five phylogenetic nodes, the sample is listed as potentially contaminated.
