Concatenation, Conflict, and Complexity: Genealogical Heterogeneity Mimics Substitutional Heterogeneity for Nucleotide Model Selection

Phylogenetic conflict-particularly hidden gene-tree discordance generated by incomplete lineage sorting (ILS)-is pervasive in multilocus and phylogenomic datasets, yet its consequences for nucleotide substitution model selection remain poorly understood. Modern molecular studies increasingly collect and concatenate large sets of independent loci sampled across distant and often poorly characterized regions of the genome, creating significant potential for heterogeneity when analyzed in combination. Here, we examine whether intra-alignment genealogical conflict can influence standard model selection procedures to favor parameter-rich substitution models even when sequences evolve under a simple substitution process. Through a series of in silico case studies, we simulated sequence evolution under the simplest rate-homogeneous Jukes-Cantor (JC69) model and generated concatenated alignments as mosaics of multiple loci, each evolving on its own gene tree drawn under the multispecies coalescent. Conflict was increased by manipulating conditions expected to elevate ILS and gene-tree heterogeneity and embedding progressively more hidden genealogies within alignments while holding total alignment length constant. Despite all data being generated under JC69, model selection frequently favored more complex models, with varying sensitivity depending on the number of taxa, the expected amount of conflict, and the specific selection criterion applied. A dominant pattern was frequent inclusion of among-site rate variation parameters (+ G4 and/or + I), and under extreme conflict, model selection increasingly favored richer substitution models (e.g., SYM, GTR). Broadly, our results showed that hidden conflict can manifest as substitutional and rate heterogeneity, driving selection procedures to compensate with additional parameters in concatenated analyses under high conflict. Broadly, our study contributes to a greater understanding and appreciation of the challenges in modeling molecular evolution in the era of multilocus phylogenetics.