The Bacteroidota methyltransferase BF2170 methylated 15 of 170 tested carboxyl drugs and raised bezafibrate permeability ~20-fold in Caco-2 assays; plasma exposure evidence is limited to mono-colonized mice.

An engineered conjugative plasmid blocked over 99.9% of resistance-plasmid acquisition in vitro and gut colonization in mice (n=3 groups); complex human microbiomes remain untested.

Feasible in laboratory systems (rated 62% likely, conditionally), but clinical safety, efficacy, and defenses such as plasmid replacement or chromosomal resistance remain unestablished.

A CDC-reactor design using clinical IncX1 and IncU plasmids tracks transfer, loss, replacement, and resistance simultaneously; the predicted spatial transfer regimes await actual testing.

Fiber loss suppresses butyrate producers in animal and human studies, but no trial has tested ultra-processed diets directly, so attribution rests on observational data plus borrowed mechanisms.

LTP stimuli recruited IRSp53 to assemble Arc capsids and release vesicles whose Arc mRNA lowered AMPA receptors in cultured rodent neurons; in vivo validation is absent.

RNAi knockdown of the chrnal-E receptor increased sleep in Cassiopea (preprint; small samples, no rescue), hinting that cholinergic wake control predates the central nervous system.

HTRA1 showed opposite stress regulation in male versus female nucleus accumbens across human postmortem tissue and mouse experiments (preprint); human causality is unproven.

A NEJM review finds 6.5–12.9% of mothers affected with workable screening and treatment options, while its strongest prevention claim rests on one small observational study (n=78).

Resampling across ABCD, HCP, and UK Biobank shows tiny effects (median |r|≈0.01) requiring thousands of participants; better phenotype measurement is an underused lever to cut that demand.

A convergent intrathecal antibody response to one EBV epitope marked a minority of MS patients and validated at 99.36% specificity in an independent cohort (n=909); causality is unresolved.

Under gene-held-out splits, reported codon-model gains shrank from +2.3–14.3 points to roughly +2 or reversed, and a memorization canary beat every neural model; datasets and code are public (CodonBench).

A multi-model workflow matched human reviewers at review scale, supporting supervised acceleration rather than autonomous reviews, with independent replication still pending.

Fine-tuning improved retrospective antimalarial screening (two replicates, no prospective validation), and the best model depends on ranking by overall ROC-AUC versus top-1% enrichment.

The governance layer delivers deterministic, auditable screening decisions, but biological hazard detection remains proxy-based and partly circular, with no deployed-effectiveness evidence.

A preprint maps training on model output to drift and model merging to recombination with simulation-verified closed forms; LLM-scale evidence covers two model families with LoRA adapters.

A pan-cancer latent-diffusion model improved reconstruction (PCC 0.31 versus 0.24 for baselines) within tested platforms; inferred genes should be kept distinct from measured evidence.

A 1B-parameter model trained on 23 million spatial microenvironments simulated tissue responses and produced one combination (CXCL10 plus anti-INHBC) validated only in vitro.

Six preregistered experiments (N=1,554) found a computational intimacy model tracking human judgments at r≈0.99 against a 0.997 noise ceiling, tested only in US vignette contexts.

Figure-design guidance separates the evidence-backed endosomal-release step from formulation-specific claims about tumor accumulation, receptor identity, and apoptosis.

Enformer applied to 34 hominin and ape haplotypes predicted 37,592 lineage-specific regulatory elements with 7 of 9 validated in luciferase assays (R=0.81); most loci await mechanistic explanation.

Auanema rhodense discards about 61% of its germline genome at the 12-cell stage at precise 29-bp palindromic motifs (preprint); the eliminating nuclease and its evolutionary cause remain unknown.

The method mapped over 25,000 crossovers across 3,842 haploid mouse sperm nuclei and confirmed elevated recombination in Fancm knockouts (n=3 per genotype, subtelomeric bias noted).

Flagella construction and operation consume a substantial share of cellular energy; whether selection acts on heterogeneous motility distributions rather than averages is plausible but undemonstrated.

Across 3,158 dicot species in 11 ecosystems, growth traded off with longevity (R²=0.22, P<0.0001), with 78% of variance unexplained and tropical sites missing.

Repeated stimulation triggered a 4,907-gene stress program versus 789 genes after a single drop, but a 10-minute sampling confound leaves the plant-memory interpretation unproven.

Hydrothermal networks produced autocatalytic peptides and weak reversibility, but the universal period-five claim rests on analogy with two networks and no statistical testing (reviewed quality 30%).

Cell and mouse evidence supports mitophagy activation, but no randomized human trial has demonstrated improved survival or durable functional benefit.

Cohorts associate 2–3 cups daily with modestly lower mortality while acute caffeine raises blood pressure about 8/6 mm Hg in hypertensive people; the population benefit remains causally untested.

Thirty-three Sicilian centenarians carried ~50-fold higher 1918 H1N1 titers than septuagenarians yet showed zero SARS-CoV-2 cross-reactivity, ruling out antibody cross-protection as their resilience mechanism.

A Navigation Guide review of 25 studies across 12 countries rated prenatal pyrethroid exposure as sufficient evidence of neurodevelopmental harm at moderate quality, while childhood evidence stays limited.

A US spatial study (N=21,851) linked PM2.5 to sperm DNA fragmentation (estimate 0.73, p<0.0001), but the cross-sectional design and a larger age effect keep causal inference cautious.

Animal and cell studies consistently show inflammatory signaling and oxidative stress, while human causal evidence remains sparse, leaving dose–response translation as the central gap.

A narrative review connects UV photochemistry, DNA repair failure, and immune modulation to tumor formation, without original data or a systematic search protocol.

In UK Biobank (n=190,275), measured omega-3 levels looked protective (aHR 0.86) while genetically predicted levels raised risk (OR 1.11), suggesting levels mark metabolic health rather than cause benefit.

Tirzepatide outperformed GLP-1 drugs on HbA1c (up to 2.4%) and weight (up to 22%) per an independent 21-trial meta-analysis, while the reviewed synthesis itself lacks PRISMA flow and bias appraisal.

A bottom-up estimate puts tissue-specific cost near 41 kcal/day, tenfold below common 300–500 kcal surplus advice, though no primary data underpin the estimate and a key parameter comes from infants and animals.

In a 20-year drought-exclusion experiment, primed trees held transpiration resistance (0.75 versus 0.46) and recovered within 143 days at the cost of ~26.5% smaller crowns, from a single plot pair.

Carollia perspicillata phase-locked calls to low-noise dips of 4-Hz noise in the lab (13 of 14 bats) and the field, and frontal-auditory inactivation disrupted timing in all 8 tested bats.

Solid-state nanopores held a liquid pathway at -5 to -20 °C, stretching DNA translocation from 0.026 to 10.33 ms per base, though the liquid state is inferred electrically rather than visualized.

No published N-body grid yet quantifies Earth-analogue survival against binary separation and eccentricity; survival likely collapses at periastron rather than mean separation, pending new simulations.