Raw reads were pre-processed by removing the adaptors and low-quality reads using BBMap. The filtered reads were normalized for depth based on kmer counts using BBNorm function. De novo transcriptomes were generated using both Trinity and velvet-oases. CD-HIT-EST was used to merge the resulting two de novo transcriptomes and reduce the transcript redundancy to 90% similarity and unique genes were predicted using transdecoder. Raw reads can be found in GenBank PRJNA912133.