Author : Joenilo II E. Paduhilao

Summary : Pangenome graphs have been constructed to analyze the genetic diversity of wild and cultivated rice; however, their applications have been limited to graph characterization and variant calling, with little justification provided for software selection. This study aimed to extend the application of pangenome graphs by estimating genomic gains and losses between wild and cultivated rice through parsimonious inference of graph-encoded variants. To support this objective, a pipeline was developed centered on an appropriate graph builder identified by comparing three widely used software. Initial attempts with Minigraph revealed inaccurate and biased variant calling. In contrast, PanGenome Graph Builder (PGGB) achieved perfect recall and was adopted for final graph construction. The resulting pangenome graph contained approximately 400 Mb of non-reference sequences. Core sequences represented only 25% of the graph and encoded genes linked to core biological functions, while the variable category comprised 75% of the graph and contained over 50% repeats. The unmasked fraction of the variable category encoded genes with adaptation-related functions, including members of the protein kinase and NBS-LRR families. While INDEL-based parsimonious inference revealed a balanced distribution of gains and losses across internal branches, mapping SVs showed that the common ancestors of both Asian and African rice lineages experienced more losses than gains. In contrast, gain amplification on external branches was driven by repeats classified in the private category. Removing these elements restored a balanced pattern between gains and losses. These findings demonstrate the potential of pangenome graphs for inferring rice genome evolution and highlight the importance of software selection in pangenome construction.