Fig. 1: Genie architecture and parameters.

a Genie encoding process: the input data format is the uncompressed, binary MPEG-G record format that can store unaligned as well as aligned genomic data. FASTQ/BAM data must be transcoded to MPEG-G records before starting the encoding process. The records are regrouped into access units based on their alignment properties. Nucleotide sequences, record identifiers, and quality scores are then encoded into descriptor subsequences. To improve the efficiency of entropy encoding, a sequence transformation, splitting of symbols into subsymbols, and a subsymbol-level transformation are applied. Then the data is binarized into a stream of bits and compressed with CABAC. Finally, the compressed bitstreams and decoding parameters (collected during encoding) are wrapped into MPEG-G data structures. Optionally, these can be encapsulated into a container file together with external data (e.g. metadata or datasets encoded by third-party MPEG-G compliant software). We refer the reader to the methods section for a detailed description of all transformations. b Parameter optimization for the first subsequence of some selected descriptors (listed in Supplementary Table 1). Shown is the normalized compressed size, ordered from the worst (i.e., optimization progress of 0%) to the best (i.e., optimization progress of 100%) set of parameters found.