Review Papers Claim by Claim

Visual review: Tissue Tropism & Transmission Ecology predict virulence (Brierley et al., 2019)

Critical appraisal (concise, evidence-linked)

1. Data collation strength: Comprehensive species list (214 human‑infective RNA virus species as of 2015), systematic literature searches, transparent coding rules for virulence (binary plus supplemental ranks). This breadth is valuable for broad comparative inference but necessarily simplifies complex, strain- and context-dependent virulence into a two-category variable Data collation & virulence coding
2. Statistical approach — pros and cons: Tree-based methods (classification tree + random forest) are appropriate for mixed, categorical predictors and small-n situations; random forest improves generalisability and handles 'large p, small n' but loses simple effect sizes and can mask phylogenetic non-independence. Authors acknowledged taxonomic family likely proxies for conserved molecular traits not encoded in predictors Machine learning methods & caveats
3. Key findings and biological plausibility: Neural/systemic/renal tropism and direct-contact or respiratory transmission correlate with higher predicted virulence; limited human transmissibility (0Main results summary
4. Limitations and blind spots:
   • Binary virulence coding collapses strain-level/host-population variation (authors note examples like dengue, Lassa) — may misclassify viruses with rare severe outcomes.
   • Tissue tropism is often incompletely known or biased by sampling — many viruses detected only in blood ('viraemia' category) limiting tropism resolution.
   • Taxonomic family as top predictor indicates substantial unmodelled confounding (genomic/proteomic determinants, vector species differences).
   • Comparative non-independence (phylogeny) not modelled explicitly — could inflate apparent trait associations tied to clades.
   • Model utility for newly discovered viruses limited because tropism often unknown early; authors suggest receptor/sequence-based inference as future direction.
   All these limitations are acknowledged in the paper and test robustness analyses (sensitivity subsets, alternative virulence codings) gave similar directions but cannot eliminate the listed issues Limitations & sensitivity analyses
5. Reproducibility & transparency: Authors provide data and R script via Figshare (DOI in paper). Methods use standard R packages (rpart, randomForest, missForest). This is good for reproducibility; remaining challenges are subjective trait codings and literature-derived data quality Data & code availability

What would change the paper's conclusions?

• High-quality strain-level virulence datasets showing opposite associations (e.g., multiple high-virulence strains in respiratory/gastrointestinal-tropic viruses) would weaken tissue-tropism claims.
• Explicit phylogenetic comparative models demonstrating family-level signal fully explains trait associations would reduce causal interpretation of ecology driving virulence.
• New reservoir/host data indicating tropism is misassigned for many viruses (blood-only detection vs true organ infection) would alter partial-dependence patterns.

Confidence, novelty, and usefulness (concise)

Confidence in reported model results: moderate — models robust on provided dataset and sensitivity checks, but conclusions constrained by trait-coding granularity and phylogenetic confounding Model robustness claims

Concrete, prioritized next steps (testable)

1. Integrate phylogenetic comparative methods (e.g., phylogenetic generalized linear mixed models) to partition clade-level effects from ecological traits.
2. Generate strain-/lineage-level virulence labels where possible (CFRs, hospitalisation rates) and re-run models to test if species-level binary coding biases results.
3. Develop sequence-based predictors of tissue tropism (receptor usage signatures, accessory protein motifs) so that early genomic data can feed predictive models for novel viruses (authors suggested this direction).
4. Use mechanistic experiments (cell/tissue organoids, receptor mapping) to validate predicted links between tropism and severe pathology for prioritized virus families.

Concise verdict