PIA : more accurate taxonomic assignment of metagenomic data demonstrated on sedaDNA from the North sea

Assigning metagenomic reads to taxa presents significant challenges. Existing approaches address some issues, but are mostly limited to metabarcoding or optimized for microbial data. We present PIA (Phylogenetic Intersection Analysis): a taxonomic binner that works from standard BLAST output while mitigating key effects of incomplete databases. Benchmarking against MEGAN using sedaDNA suggests that, while PIA is less sensitive, it can be more accurate. We use known sequences to estimate the accuracy of PIA at up to 96% when the real organism is not represented in the database. For ancient DNA, where taxa of interest are frequently over-represented domesticates or absent, poorly-known organisms, more accurate assignment is critical, even at the expense of sensitivity. PIA offers an approach to objectively filter out false positive hits without the need to manually remove taxa and so make presuppositions about past environments and their palaeoecologies