Paper Review — verify claims with raw data

Visual Summary — key signals recovered from the paper

Concise evidence-backed critique (visual-first, explanation second)

• Scale & data: The study analysed ~4.54M proteins from 508 UniProt reference proteomes, producing ~154k OGs and N0 HOGs and using OrthoFinder and Count for HOG construction and ancestral inference — all code/data and a web app are provided for inspection Diwan et al. methods/data
• Main claims reproduced: bursts of gene/function gains at eukaryotic root, Metazoa, Embryophyta and vertebrate nodes; parallel functional enrichment at terrestrialization and multicellularity transitions — results are consistent with the authors' reported enrichments and with expectation from comparative genomics across deep nodes Diwan et al. main results
• Robustness steps taken: authors compared Wagner parsimony variants (g1,g0.5), ML and phylostratigraphy and found Jaccard similarity >0.9 between parsimony and ML, and sampling subsampling (40–90%) showing near-robust HOG node assignments once sampling reaches ~70% (356 spp) — good practice though sensitivity to key clades remains Diwan et al. robustness analyses

Major strengths

1. Comprehensive, integrated pipeline: OrthoFinder→MAFFT/IQ-TREE→paralog filtering→species tree→HOGs→Count ancestral states and multiple functional databases (GO/Pfam/KEGG/Reactome) — transparent and standard tools, enabling reproducibility Diwan et al. pipeline
2. Function-centric HOG counting: enrichment tests counting HOGs carrying a GO term (rather than raw gene counts) reduces bias from large gene families and emphasizes independent gene-family contributions.
3. Clear web resource and code release (GitHub + interactive app) — crucial for reuse and validation.

Key limitations, blindspots and risks (critical)

• Annotation bias: mapping functions depends on UniProt/Swiss-Prot/TrEMBL annotations (August 2021 extract used). Uneven curation means model-organism-rich clades will drive enrichments; authors acknowledge this but some cluster-level inferences (e.g., precise pathway percentages) remain sensitive to annotation density Diwan et al. annotation caveat
• Orthology and paralog handling: OrthoFinder is high-performing, but orthology inference errors (split/merge OGs, misrooted gene trees) can move inferred gain nodes; the authors attempt paralog pruning but complex duplications (e.g., lineage-specific expansions) remain a source of uncertainty (see Natsidis et al. 2021 on orthology systematic errors) Diwan et al. orthology handling
• Dependence on species sampling & tree topology: bursts of gains at a node can be exaggerated by undersampling sibling lineages; authors did subsampling tests, but sparse representation of particular phyla (e.g., tardigrades, some onychophorans) means some transitions may be missed or temporally shifted.
• Function vs mechanism: enrichment of semantically-similar GO clusters demonstrates convergence in function-level solutions, but these are not evidence that identical molecular mechanisms evolved — authors do note different HOGs supply these functions, which is correct; readers must not conflate GO-level convergence with identical genetic innovations.
• Prokaryote HGT effects: horizontal gene transfer can confound presence/absence reconstructions for prokaryotes; the authors acknowledge this, and large prokaryotic gains (e.g., cyanobacteria) need careful interpretation.

Suggested robustness checks / follow-ups (practical)

1. Re-run key analyses with alternative orthology methods (OMA, eggNOG) and with gene-tree–species-tree reconciliation to quantify node shifts for high-interest HOGs.
2. Filter GO annotations by evidence code (exclude purely electronic/IEA-only annotations) for a sensitivity test of enrichment signals.
3. Complement presence/absence with expression or proteomic evidence where available (model clades) to prioritize candidate transition genes for lab validation.
4. Explicit HGT-aware reconstructions for bacterial nodes (e.g., SimPhyNI-like simulation nulls) to test whether prokaryotic bursts are vertical or transfer-mediated.

Where this paper moves the field

It operationalizes a top-down, function-centred phylogenomic workflow at biodiversity scale, making explicit: (a) how to map functions to HOG-based gene-family origins; (b) how to detect convergent enrichments at phenotypic transitions; (c) how to provide a searchable resource for placing new genomes into an evolutionary functional map — thus directly enabling phylo-informed annotation and hypothesis generation for biodiversity genomics Diwan et al. web app and resource

Run deeper (one-click): start an AI science agent to re-run targeted analyses

Data & code access (important)

Data and scripts are provided on GitHub and the web resource: GitHub repo and funcevol.russelllab.org — indispensable for reproduction and reanalysis (authors declare availability) Diwan et al. availability

Conclusion (short)

Strong, well-engineered study producing a practical resource and defensible, interpretable high-level conclusions about bursts of functional innovation and repeated functional convergence across major transitions; conclusions are robust at the level of broad GO-clusters and nodes but sensitive to annotation biases, orthology errors and taxon sampling for fine-scale claims — follow-up tests and alternative-orthology cross-checks are recommended before embracing single-gene evolutionary timing claims.