eggNOG-mapper v2: Functional Annotation, Orthology Assignments, and Domain Prediction at the Metagenomic Scale

Key results

1. ShownAlthough the underlying databases doubled in size, v2 annotates queries faster than v1, and database optimizations for faster, parallel queries combined with a new in-memory mode that curbs slow disk I/O yield a much larger average speedup.

   Annotation rate (annotated queries per second) up 16% on average versus previous versions; all changes together raise annotation rates by 608% on average versus eggNOG-mapper v1 (Figure 1B; 20 sets of 1–100 random proteomes, 10 CPUs, 80GB RAM). · Coverage and performance improvements; Figure legends

2. ShownMoving to eggNOG v5 (4.4 million orthologous groups and over twice as many organisms as the prior release) raised annotation coverage and phylogenetic resolution when a human gut metagenomic gene catalog subsample was re-annotated.

   Re-annotating 1.75 million randomly subsampled catalog proteins: coverage up 3.23% versus eggNOG-mapper v1 (56,569 newly annotated proteins); alignment scores for queries significantly better than earlier versions (Wilcoxon W = 1.2E+12, p-value < 2.2E−16). · Coverage and performance improvements

3. ShownPFAM domains are by default inherited from inferred orthologs at almost no extra cost but with some misassignments; an optional realignment of queries to ortholog domains removes false positives and closely matches full de novo PFAM annotation at low cost.

   Using full de novo realignment as the reference, ortholog transfer scores F1 89.7% and realignment refinement F1 98.9% (Figure 1E); tests used 25 random sets of 10–100,000 Progenomes v2 proteins on 30 CPUs and 240GB RAM; realign vs full de novo runtimes are in Figure 1F. · Protein domain annotations; Figure legends

4. SuggestedTiming benchmarks indicate v2 annotates genomes faster than Prokka, most clearly on large metagenomic datasets, and that the MMseqs2 search mode runs faster than the default Diamond mode with comparable sensitivity but higher memory needs.

   Figure 1C: average minutes to annotate genomes, v2 vs Prokka (20 sets of 1–100 random genomes, 10 CPUs, 80GB RAM); Figure 1D: MMseqs2 (-s 3,5,7) vs Diamond sensitive mode on 25 random Progenomes v2 protein sets (30 CPUs, 240GB RAM). No numeric timings or sensitivity values are given. · Coverage and performance improvements; Sequence Mapping modes; Figure legends

5. ClaimedNew workflow stages add Prodigal ORF calling for prokaryotic contigs, taxon-restricted custom databases and annotation scopes, first-pass pairwise ortholog reports against every eggNOG v5 genome, and functionally decorated GFF plus multi-source annotation tables.

   Feature descriptions only: Prodigal normal, anonymous, or training modes with custom translation tables; ortholog reports filterable by target taxa and one-to-one or many-to-many type; sources include KEGG, Gene Ontology, EC numbers, BiGG, CAZy, COG and eggNOG OGs, with per-source coverage plotted (Supplementary Figure 1); no accuracy benchmarks. · ORF prediction; Adjusting Taxonomic Scopes; Orthology reports; Annotation sources