kamino: fast proteome-wide variant calling for amino acid phylogenomics

Key results

1. 01

   Shownkamino built alignments much faster than the BUSCO marker pipeline on the four 120-isolate datasets, but used more memory because it holds a large graph in memory.

   Prokaryotic taxa (Mycobacterium, Enterobacteriaceae): one to two orders of magnitude faster. Eukaryotic taxa (Hypocreales, Insecta): two to three orders of magnitude faster. The gap depended on which BUSCO marker set was used. Runs used 4 CPUs; BUSCO time is the combined wall-clock time of busco v6.0.0 and BUSCO_phylogenomics. · Assessing kamino efficiency and scalability; BUSCO analyses

2. 02

   SuggestedIn Mycobacterium, Enterobacteriaceae and Hypocreales, trees grouped by inference method rather than by alignment source, so kamino and BUSCO alignments carried broadly similar signal. In Insecta, kamino trees clustered apart and differed at deep nodes.

   For each taxon, nine trees (kamino, BUSCO broad and BUSCO taxa alignments, each run with BioNJ, FastTree and IQ-TREE) were compared by cluster information and Robinson-Foulds distances. No kamino topology matched any BUSCO topology. In Insecta, only kamino trees recovered Condylognatha (Thysanoptera + Hemiptera) and only BUSCO trees recovered Palaeoptera (Odonata + Ephemeroptera), a group whose monophyly is disputed. · Comparing outputs of kamino and a marker-based…; Quality assessments and phylogenetic analyses

3. 03

   Shownkamino alignments were less saturated than BUSCO alignments in all four comparison taxa. Except in Insecta, where BUSCO broad was comparatively short, they were about as long as the reduced BUSCO set's alignments, and all alignments had similarly low Pythia difficulty.

   The saturation slope (patristic vs uncorrected distances, 0 to 1, higher means less saturated) favoured kamino in all four datasets. It was nearly twice as high for kamino in the two eukaryotic datasets (Hypocreales, Insecta). Pythia scores ranged from 0 to 0.06 for all alignments. BUSCO taxa (taxon-specific markers) gave much larger alignments, and BUSCO broad was comparatively short for Insecta. · Comparing outputs of kamino and a marker-based…

4. 04

   Shownkamino produced useful alignments up to phylum rank for eukaryotes. Prokaryotic alignments shrank sharply at higher ranks, which mostly restricts prokaryotic use to genera and families.

   Tested on 40-proteome datasets from 93 eukaryotic (NCBI) and 269 prokaryotic (GTDB) taxa. Eukaryotic Pythia scores were below 0.1 except for Aves and Magnoliopsida. Prokaryotic scores reached 0.8 at higher ranks, and alignments scoring above 0.2 were always the shortest. Missing data rose from about 5% (genus) to 10% (phylum). Lowering -f from 0.85 to 0.75 made most alignments about 1.5- to 2-fold longer. · Assessing the diversity limits of kamino; Discussion

5. 05

   ShownRuntime and memory grew roughly linearly with the number of proteomes, more steeply for eukaryotes. Extra threads helped mainly up to 4 CPUs.

   Each of the four taxa was rerun at 40, 80, 120, 160 and 200 isolates on 4 CPUs. The 120-isolate sets were rerun with 1 to 12 CPUs: runtime dropped sharply from 1 to 4 CPUs, with modest gains beyond that. Hardware: Intel Xeon Platinum 8358, 2.60 GHz, 48 GB RAM. The authors note that memory may limit runs on hundreds of large eukaryotic proteomes. · Assessing kamino efficiency and scalability; Quality assessments and phylogenetic analyses