Stitch or Cluster? A Comparison of Alternative Phylogenomic Dataset Assembly Strategies for Blenny fish.

Phylogenomics has revolutionized the way we infer evolutionary relationships. Several bioinformatic pipelines have been developed for assembling phylogenomic datasets, in which orthology inference is a key step. Here, we compared two alternative strategies for assembling phylogenomic datasets: sequence clustering (OrthoFinder) and a new similarity-based approach that enriches a predefined set of loci (Patchwork). We downloaded publicly available genomic data for the fish family Blenniidae, which comprises a heterogeneous set of source data (genome skimming, transcriptomes, genomes) obtained by various sequencing technologies (Illumina short reads, long Nanopore reads, 454 pyrosequencing). These data are characterized by diverse levels of sequencing depth, read length, and per-base accuracy, representing a typical scenario of data reuse for phylogenomic purposes. All data types, regardless of accuracy and sequencing depth were suitable to phylogenetically place species, even at estimated sequencing depths ~1.6x, but 454 data produced extremely long branches. For assembling our phylogenomic datasets, Patchwork outperformed OrthoFinder because it generated fewer but taxonomically more complete multiple sequence alignments. Our study is the first to test the evolutionary relationships among combtooth blenny fish with genome-scale data, which was previously studied with multi-locus datasets. We also explore two alternative approaches to combine marker-rich phylogenomic data with taxonomically broad multi-locus markers obtained by Sanger sequencing, supporting that simple data concatenation does not necessarily outperform phylogenomic constraints on multi-locus datasets.