Methodological pitfalls in plant pangenome gene family identification may lead to biased evolutionary inferences

Pangenome-level gene family identification often applies sequence similarity clustering without phylogenetic or synteny information, which risks biologically misleading evolutionary inferences. Using five transcription factor families (bHLH, MYB, NAC, WRKY, MADS-box) across 401 rice pangenome accessions, we compared clustering strategies: OrthoFinder alone, cd-hit alone, MMseqs2 alone, and OrthoFinder-informed refinement by cd-hit or MMseqs2. Methods solely based on sequence similarity merged distinct orthogroups and generated fewer orthogroups than approaches incorporating graph-based orthology. Conflicting cluster assignments, measured against OrthoFinder, varied strongly among families, from approximately 14% in MADS-box to approximately 57% in MYB, and were associated with protein length differences. Core, shell, and cloud gene classifications shifted substantially depending on the method, especially in MYB, NAC, and WRKY families. Critically, Ka/Ks distributions for core genes were highly method-sensitive, with orthology-aware methods yielding more convergent and less variable estimates of selective pressure, whereas noncore gene estimates remained robust. These findings demonstrate that neglecting graph-based orthogroup inference inflates methodological artifacts. We recommend a two-step strategy: initial graph-based orthogroup delineation followed by sequence similarity refinement to balance evolutionary accuracy and resolution in pangenome-scale gene family studies.