Genome annotation

We annotated the genome with the Braker2 pipeline (Brůna, Hoff, Lomsadze, Stanke, & Borodovsky, 2020; v2.1.5, Hoff, Lomsadze, Borodovsky, & Stanke, 2019). Both RNAseq- and protein-derived hints were used to train GeneMark-ETP, which predictions were in turn used to train Augustus and obtain the final gene set. All protein data available for Arthropoda were downloaded from OrthoDB (v10, Kriventseva et al. 2019, https://v100.orthodb.org/download/odb10_arthropoda_fasta.tar.gz, last accessed 22.07.2020) and aligned using ProtHint. This dataset contains 2.6 million sequences and encompasses 170 species, including 40 of the same order (Hymenoptera) and 17 of the same family (Formicidae). RNAseq data produced above was aligned against the hard-masked genome using STAR (v2.7.2, Dobin et al., 2013), and secondary alignments were removed with SAMtools (v1.10, Li et al., 2009). After the Braker2 run, protein sequences of all gene models not supported by at least one hint were blasted against the Uniprot database (UniProt Consortium, 2019) and all models without any hit on Aculeata (wasps, bees and ants) were discarded from the final gene set.