Check Your Claims Before Reviewers Do

Miolane's Geomstats library shows a delayed-impact citation curve typical of infrastructure scientists (3 citations in 2021 → 116 in 2022), and her 2026 organoid persistent-homology work detected loop topology above surrogate nulls in 14 of 18 datasets — a finding that could become a measurement standard for organoid intelligence if H2 size-thresholds replicate.

Limitations: h-index 10–13 with modest per-paper impact; many outputs are preprints or workshop papers; theoretical claims rest on linearization and surrogate assumptions not yet independently replicated.

Proposed falsification: Re-run organoid persistent homology with a null preserving higher-order correlations across all 18 datasets.

Van Kooyk (h-index 101, ~39,000 citations) co-discovered DC-SIGN, a field-creating Cell publication that seeded glyco-immunology, and her 2000–2002 spike of 12,000+ citations shows how deep mechanistic follow-through on one molecule outperforms broad shallow portfolios.

Limitations: frequent reliance on in vitro/ex vivo human models with limited in vivo validation; small histology cohorts; one BBB study rated 6/10 for reproducibility.

Proposed falsification: Humanized DC-SIGN+ DC transfer into BLT mice with AZN-D1 blockade to test in vivo HIV dissemination claims.

Thewissen (h-index 46, ~6,827 citations, 218 works) published multiple first-author Nature/Science papers that redefined whale origins, though his citation mass rests on roughly five decisive fossil specimens rather than publication volume.

Limitations: reliance on fragmentary fossils, inherently interpretive functional morphology, and limited pre-registration typical of the field.

Proposed falsification: Bayesian total-evidence reanalysis of published character matrices to test whale–artiodactyl relationship stability.

Wong (h-index 32, 5,128 citations) delivered repeated firsts — logical CAR control, orthogonal split-Cas13 (CRISTAL), m5C-saRNA CAR T with polycistronic gating — with rigorous preclinical controls, though small NSG cohorts, some unequal dosing in survival comparisons, and declared biotech equity conflicts temper the preclinical-to-clinic story.

Proposed falsification: Cross-validate saRNA-BiTE in humanized immune-reconstituted mice for off-target T-cell exhaustion or cytokine toxicity.

Frangakis (h-index 48, ~9,613 citations) helped launch cellular cryo-ET, refuted the textbook 30-nm chromatin fiber, and now reports megabase-scale bacterial DNA conduits — a genuinely novel HGT mechanism whose causal genetics remain unproven and whose flagship work is preprint-stage.

Proposed falsification: Correlative cryo-FLM/cryo-ET on intact A. baumannii to test whether ~2 nm conduit filaments persist without ghosting artifacts.

Pos (h-index 43, 126 works) defined the AcrB peristaltic efflux mechanism (Science 2006) and, late-career, challenges orthodoxy with conduit-mediated megabase HGT — strong controls and transparent counterevidence, but single-strain scope, no biogenesis gene identified, and a micelle-accessibility confound in vitro.

Proposed falsification: Tn-seq identification of the conduit biogenesis factor in a conduit-defective background.

Shin's FALCON achieved 84-fold spleen:liver selectivity by tuning compositional ratios within fixed lipid sets — a counterintuitive result suggesting ratio optimization, not new lipid chemistry, may drive organ tropism; confidence is moderate because the h-index 9 / 676-citation profile carries OpenAlex identity ambiguity.

Proposed falsification: Test whether 84-fold and 111-fold selectivity ratios persist outside C57BL/6 mice.

Hänelt (h-index 23, 56 works) combines single-molecule FRET, EPR, and cryo-ET in a coherent membrane-transport program; her 2026 conduit discovery, if confirmed, opens a new axis of bacterial genetics, but the mechanism is incomplete — no conduit-forming gene identified, single strain pair, unreviewed.

Proposed falsification: CRISPRi sweep of envelope-remodeling genes to causally link conduits to 1.1 Mbp recombination.

The recurring epistemic gap across Zhou-linked mechanisms (HUWE1–DNA-PKcs, ZGRF1, EXO1/MiDAS, METTL3) is the missing catalytic-dead rescue: nearly every claim is built on depletion plus inhibitors, which conflates a protein's enzymatic role with off-target and scaffolding effects; LYTAR vaccine attenuation and HUWE1 neddylation carry the strongest causal support, while TCGA survival links are explicitly correlative. Citation metrics (h-index 125) come from a likely mismatched materials-chemist profile and cannot be attributed.

Proposed falsification: A five-mutant CRISPR knock-in panel with a standardized DR-GFP/NHEJ/proteomics battery.

Poe (h-index 27, 3,369 citations, 82 works) contributes solid empirical REM-hippocampal plasticity work in top journals; her citation profile is review-driven (~21% from one locus coeruleus review), and flagship correlational claims rest on n=8 subsamples with drug manipulations that are not REM-specific.

Proposed falsification: Powered, pre-registered, multi-strain replication of the theta-ratio/reversal-learning correlation.

Huang (h-index 73, 389 works, ~20,900 citations) repeatedly reframed fields — outer-membrane mechanics, path-dependent microbiome recovery, phage precision editing, agentic-AI microbiome discovery — with large-scale sampling (2,876 fecal samples; >1M transposon mutants), defined falsification criteria, and open code/data.

Proposed falsification: Reciprocal cross-fostering with metagenomic SNV tracking to split transmission- from selection-driven microbiome recovery.

Xie's 2026 2×2 experiment (n=116) shows large affective effects (η²≈0.14–0.16) from dynamically LLM-generated humor but small retention effects (η²≈0.046), suggesting humor acts primarily as an engagement lever with cognitive benefits contingent — though only immediate retention was tested, no pre-registration was stated, and OpenAlex identity attribution is ambiguous.

Proposed falsification: Pre-registered replication with a 2-week delayed retention test and humor→cognitive-load→retention mediation analysis.

Wilfling (h-index 21, ~3,894 citations, ERC-funded) delivered molecular-resolution in situ structure of an autophagic condensate with a single-residue liquid-to-glass mechanism — genuinely new — though tomogram counts are small (n=4/7), residues 1–37 are unresolved, and generality beyond yeast is untested.

Proposed falsification: Cryo-ET time course at endogenous Ape1 expression to test whether liquid-to-glass transition occurs physiologically.

Baker (h-index 57, >20,400 citations) co-defined CPR/DPANN and Asgard archaea — reshaping tree-of-life and eukaryogenesis debates — and pioneered reference-free tetra-ESOM binning before it became mainstream, though several flagship works are large consortium papers where his specific conceptual contribution is harder to isolate.

Proposed falsification: Cross-validate tetra-ESOM bin purity on long-read HiFi metagenomes against Hi-C strain links.

Zhu (h-index 40, ~5,944 citations, 166 works) authored community benchmarks (CAFA3, CAMI2) and 2026 tools — VACmap's variant-aware chaining and SCGBinner's SCG-guided contrastive learning — with large benchmarks, deposited code, and negative results reported, though SCGBinner's flagship 71% HQ-MAG gain rests on a single 250 Gbp soil sample.

Proposed falsification: Held-out ENA soil samples with blinded AMBER scoring to test SCGBinner's cross-site generality.

Weis (h-index 57, ~10,700 citations, 194 works) pioneered host-initiated bleaching mechanisms, lectin-glycan recognition, and the Aiptasia model system, and publicly quantified how culture-media nutrient excess (~6,000–15,000× seawater nitrate) confounds symbiosis comparisons — a model of field-level self-correction, though replicate counts (n=3–5) and single clonal lines limit some mechanistic studies.

Proposed falsification: NanoSIMS isotopic tracing across multiple host genotypes to test whether cross-generational provisioning replicates beyond the H2 line.

Srivastava (h-index 20, ~4,931 citations, 77 works, R35-funded) built the Hofstenia acoel model system, sequenced key early-animal genomes, and proposed the falsifiable "tiled brain" hypothesis reframing brain centralization as condensation of distributed computational units — with explicit sample sizes, open data/code, though neural tile-computation claims remain behavioral-inference-based.

Proposed falsification: Single-tile holographic inhibition during prey-tracking to directly map tile-level computation.

Beets (true h-index 29 across a merged record; a disambiguation fragment shows h-index 0) has a first-author Science paper (vasopressin/oxytocin signaling in learning), a peptidergic connectome resource, and a 2026 K2P mechanism preprint with powered imaging cohorts (n=10–51) — but K+ sufficiency is untested, some statistics are missing, and findings are restricted to one neuron.

Proposed falsification: Targeted K+ extrusion with voltage clamping to test K+ sufficiency in AVA.

Bousios (62 works, ~1,271 citations, h-index 15) shows how lineage-specific TE expertise (ATHILA motifs) enabled cross-clade synthetic genomics — the Geum alpha-omega centromere-architecture transition discovery — with explicit statistics and open code, though flagship outputs are computational with no direct functional tests and largely middle-author on the largest-impact works.

Proposed falsification: CRISPR-edited ATHILA elements with scrambled junction motifs to test centromeric targeting.

González-Rosa (h-index 21, 2,606 citations, 49 works) created the field-standard zebrafish cryoinjury scar-regression model (470+ citations), and his Tnni3k work quietly falsifies his own earlier polyploidization narrative for that locus — evidence claims are empirically constrained — though small n (4–16/group), non-blinded analyses, and untested pkmb epistasis weaken the preprint.

Proposed falsification: Tnni3kOE × pkmb overexpression epistasis test to identify the principal fibrosis effector.

Beye identified csd as the honeybee primary sex-determiner (Cell 2003) and traced its evolution by tra duplication (Nature 2008) — paradigm-setting work in a non-model organism — though his metrics are partially conflated with a namesake X-ray physicist, illustrating author-disambiguation as the weakest link in bibliometrics.

Proposed falsification: Reciprocal CRISPR domain-swaps of csd hypervariable repeats to test the recognition-allele model.

Lee's h-index diverges >30-fold between databases for the same ORCID person (2 vs 21) — database provenance is itself a scientific variable; his genuinely strong per-paper record includes two Cell papers and a 725-citation flagship, but contributions are largely collaborative middle-authorship with no documented independent PI role.

Proposed falsification: Germ-free vs conventionalized ICI-resistant mice with targeted metabolite knockdown to isolate the bacterial metabolite's causal contribution.

Pitel (career h-index 40, 437 works; flagship Science 2014 and Nature Genetics 2013 papers) designs experiments that can disconfirm her own epigenetic-inheritance hypotheses — SNP-covariate falsification tests and null GO enrichment reported openly — a hallmark of self-correcting science in a field rife with overclaimed transgenerational effects.

Proposed falsification: Matched sperm and blood WGBS from epi+/epi− birds to test germline transmission vs genetic confounding.

Johnson (127 works, h-index 33, ~3,750 citations) unifies APOE4 Alzheimer's risk as a glial lipid-handling and inflammatory lipid-signaling defect (COX-2/prostaglandin flux, Plin2 storage) rather than mere amyloid accumulation — mechanistically deep multi-omics, but most claims rest on immortalized cell lines and mouse models with limited direct human validation, and AP-MS proteomics used only two replicates.

Proposed falsification: APOE4-TR/Plin2-KO × 5xFAD crosses with ARIA-like microbleed quantification post-anti-amyloid antibody.

Kanazawa (h-index ~82–90, >21,000 citations) discovered the DRPLA CAG expansion, the anti-GQ1b autoimmune biomarker, and GluR2 RNA editing in ALS, and unusually published rigorous null-result rTMS trials with realistic sham controls — though the rTMS trial lacks equivalence margins and some unusual findings were left unexplained.

Proposed falsification: Bayesian hierarchical reanalysis with pre-registered equivalence margins to quantify evidence for the rTMS null.

Bielekova (h-index 47, ~9,656 citations, 178 works) shows a rare full arc — from a falsified altered-peptide-ligand hypothesis to building the open-data infrastructure (GitHub, Shiny, GEO) that lets the community falsify her own biomarker claims; the 2025 n=394 longitudinal CSF proteomics study is openly shared but correlational and not yet externally validated, with industry CRADA entanglement flagged.

Proposed falsification: Prospective validation of the baseline CSF biomarker panel in an independent multi-ethnic cohort.

Sleimen-Malkoun (h-index 18, 40 works) has ~13% of lifetime citations from one multiscale-entropy methods paper — her real asset is a measurement toolkit, not a single discovery; her 2026 gerophysics pivot to falsifiable tests makes the next empirical paper the decisive datum for judging her ceiling.

Proposed falsification: Pre-registered perturbation dose-response testing a monotonic complexity–resilience curve against frailty staging.

Hipólito (h-index 20, 77 works, ~1,409 citations) followed a deliberate causal-chain strategy — first establishing accumbens KOR recruitment in pain-induced negative affect (Neuron 2019), then engineering intranasal liposomal NorBNI to target exactly that mechanism — but brain drug exposure is inferred rather than measured, Von Frey groups are underpowered (~4/group), and the scope is preclinical-only.

Proposed falsification: Direct microdialysis/autoradiography measurement of brain NorBNI concentration after intranasal dosing.

Zerjal (h-index 31–34, ~9,900 citations) co-authored a field-defining CNV atlas and the famous Mongol Y-lineage legacy finding, then pivoted to among the largest methylome-scale animal epigenetic-inheritance designs — with mirror-mating controls, FDR control, and fully open reproducible code/data (Zenodo, INRAE git).

Proposed falsification: Matched sperm and blood methylomes from 100 epi+ and 100 epi− G2 birds to test germline transmission.

Field (h=11, ~449 citations, 73 works) contributed underexplored metascience angles — replication target selection as a decision process, open science for qualitative research — with a practice-what-you-preach registered report, though her citation curve peaks during the 2020–2021 reform era and her portfolio is commentary-heavy with limited independent empirical output.

Proposed falsification: Retrospective cohort comparing registered-report vs conventional-paper citation trajectories on identical topics.

Raab (OpenAlex h-index 56, ~12,996 citations, 643 works) pairs a first-authored Lancet myeloma seminar and landmark-trial co-authorship with the Split-flow pipeline reframing inter-method demultiplexing discordance as an information-bearing quality gradient — with orthogonal TCR validation and an open Nextflow pipeline, though heuristic thresholds, incomplete donor genotype ground truth, and preprint status are caveats.

Proposed falsification: Spike-in ground-truth benchmark pooling four genotyped cell lines across viability and depth gradients.

Blundell (OpenAlex h-index 114, ~63k citations, 730 works; MODELLER and pkCSM among the most-cited tools in the field) sits at the top of computational structural biology; a conflicting low-metric record (h=3, 6 papers) is almost certainly disambiguation failure, and naive metric aggregation would catastrophically underestimate him.

Proposed falsification: Prospective benchmarking of HANSEN's Target Priority Score in mycobacterial surrogate systems.

Kagaya (h-index 13, 1,021 citations, 42 works) delivers first-of-kind tools — CryoZeta's diffusion-based cryo-EM integration (RNA TM-score 0.678 vs AF3's 0.383) and PathFold's sequence-only folding-pathway prediction — though PathFold uses circular validation against the same SMD paradigm used for training and lacks a data-availability statement; the queried Semantic Scholar fragment (h=0) is a misattribution.

Proposed falsification: Strict temporal holdout on post-cutoff EMDB deposits to test CryoZeta's training-distribution overfitting.

Kasai (h-index 37–42, ~128–192 works, ~5,700–6,800 citations) made first discoveries in kdr resistance (Ae. albopictus, novel Vssc alleles, CYP9M10 duplication) with causal gene-editing tests, and his 2026 finding that phylogenetic variance is ~86× smaller than species-specific variance challenges routine phylogenetic imputation in vector surveillance — though the PGLMM relies on COI-only phylogeny.

Proposed falsification: Re-fit the PGLMM with a genome-wide phylogeny to test whether COI limitation explains the decoupling.

Sen (h-index ~16–19, ~951 citations, 75 works) shows that warming and labile food synergistically — not independently — boost Arctic sediment oxygen demand (up to ~1030% above ambient), a finding implying carbon-burial forecasts treating temperature and food as independent factors overestimate Arctic seafloor storage; small ambient denominators and occasional n=3 replicate the flagged caveats.

Proposed falsification: Repeat the non-significant shelf result across three consecutive years with standardized algae additions.

Pittelkow (h-index 9, 429 citations, 67 works) applied Bayesian model averaging (BF_BMA) to FDA approval packages, exposing a ~40-fold evidential-strength spread across psychotropic classes (ADHD median BF 1820.4 vs antidepressants 94.2) — converting "FDA-approved" from a binary into graded evidential currency, though metrics rely on senior-collaborator co-authorship and FDA document extraction carries transcription risk.

Proposed falsification: Apply the BF_BMA pipeline to post-2015 oncology accelerated approvals to test generalizability.

Saccon (h-index 12, ~406 citations, 24 works) demonstrated that a 45-descriptor XGBoost baseline matches billion-parameter protein language models on most peptide developability tasks, and quantified his own model's shortcut learning (MW bias ω=0.72 vs 0.32 for reality) — a caution against inflated DL claims, though there is no wet-lab validation and pooled heterogeneous affinity labels (Kd/Ki/IC50/EC50) are a caveat.

Proposed falsification: Retrain with MW-residualized labels to confirm whether apparent skill is a molecular-weight shortcut.

Clee (h-index 36, ~6,869 citations, 79 works) co-authored the ABCA1/Tangier disease discovery — one of the defining gene identifications in lipid biology — yet ~40% of her lifetime citations trace to 1999 alone, showing how a single landmark can dominate metrics while her independent scientific identity lives in less-cited first-author and review work.

Proposed falsification: Pancreas-specific insulin receptor knockouts crossed with hyperinsulinemic models to test the hyperinsulinemia–obesity mechanism.

Jäschke (h-index 44–46, ~5,600–6,700 citations, 190–270 works) built a coherent method-generating chain — first ribozyme catalyzing carbon–carbon bond formation, NAD captureSeq, NIR light-up aptamers enabling live-cell STED RNA imaging — with LC-MS and crystal-structure validation, though in-cell imaging cohorts are small (n=30–50) and validation loops are self-citation-heavy.

Proposed falsification: Five-lab blind cross-benchmarking of SiRiuS turn-on fold and STED resolution.

Chen (h-index 28, ~3,391 citations, 117 works; first-author Cell influenza NA antibody paper, 440 citations) shows the field-wide benchmark-vs-independent gap in disguise: AMPBAN reaches 0.86 MCC on benchmarks but 0.69 on independent data with no wet-lab validation — a database-bias problem, not an author-specific failure, though the leakage-removal protocol and open code are strong.

Proposed falsification: Blinded MIC testing of 50 low-similarity top-ranked predictions vs random and sequence-only picks.