PyamilySeq: exposing the fragility of conventional gene (re)clustering and prokaryotic pangenomic inference methods

Pangenomics has become a central framework for exploring microbial diversity and evolution, enabling researchers to distinguish genes that define shared biological function from those that drive adaptation. However, this relies on clustering genes by sequence similarity, a process that is far less deterministic than often assumed. This study introduces PyamilySeq, a transparent and flexible toolkit designed to diagnose and quantify hidden biases within gene clustering and pangenome inference methodologies. Using PyamilySeq, we can see how clustering thresholds (often hard-coded and poorly documented) and paralog handling can substantially alter gene family composition. Surprisingly, even parameters unrelated to clustering, such as decimal precision (0.8 versus 0.80), output selection, and even CPU and memory allocation, can alter gene family assignments, challenging the assumption that identical clustering thresholds yield consistent results. Furthermore, tools often fail to report biologically meaningful or representative sequences for gene families, undermining downstream analyses. These findings reveal systematic fragilities in gene clustering and pangenome construction and highlight that pangenomics is not merely a data-driven task but a methodological one, where transparency, reproducibility, and interpretability are as critical as biological insight. This work calls for a re-evaluation of how pangenomes are constructed and compared, and advocates for methodologies that make their assumptions explicit and their results verifiable.