Benchmarking genome recovery and ecological representation across aquatic communities of contrasting complexity

Abstract Genome-resolved metagenomics links microbial taxa to metabolic potential, but the ecological interpretation of reconstructed genomes depends on which taxa and genomic features are recovered. How sequencing inputs and reconstruction workflows influence this representation across communities of contrasting complexity remains insufficiently resolved. We generated 252 metagenomic assemblies spanning short- and long-read input volumes, assemblers, and polishing modes in coastal water and mangrove sediment communities, and paired each assembly with five binners. Three findings emerged. First, the relative contributions of tool choice and data volume differed among assembly strategies, with algorithmic factors often explaining comparable or greater variation in recovery scores than sequencing input. Second, conventional performance gains did not consistently translate into improved taxonomic representation or genomic feature retention: putative gene dropout was generally highest in short-read, intermediate in short-read-first hybrid, and lowest in long-read-first representatives of shared genome clusters, particularly for mobile genetic elements, and discrepancies between MAG-derived and read-based taxonomic profiles persisted, especially in sediments and at finer taxonomic ranks. Third, although recovery frequency was positively correlated with abundance, high-abundance taxa still frequently failed to yield MAGs, particularly in complex sediments. Genome recovery was thus jointly shaped by sequencing input and workflow choice, and technical gains did not reliably translate into ecological fidelity. Recovery bias is therefore likely to be underestimated by conventional performance metrics, and our results indicate that such bias is more pronounced in the complex sediment communities than in the water communities sampled here.