Scientific evidence, scope, and limits

BGPT helps scientific AI inspect claims, supporting evidence, uncertainty, and next experiments.

Updated August 15, 2026

Featured evidence questions

The wavefunction and physical reality

Quantum mechanics clearly uses the wavefunction successfully, while its status as a physical object remains interpretation-dependent.

Dark energy: measured effect, unknown cause

Accelerated expansion is strongly measured, but the physical mechanism remains unresolved and model-dependent.

Long reads link resistance genes to hosts

Long-read and hybrid assemblies can improve ARG–plasmid–host linkage, although mobility and risk still require validation.

Liquid biopsy limitations

Performance varies with disease stage, sample handling, assay definitions, and cross-platform agreement, limiting universal interpretation.

Paper reviews with evidence ratings

Host–microbiome metabolism of a plant toxin in bees

In vivo metabolite patterns, strain assays, proteomics, and GH3 validation support microbial prunasin clearance; final cyanide handling remains incompletely measured.

Hidden microalgae in coral reef systems

Plastid-marker reanalysis reveals coral-associated spatial patterns, but marker bias and cross-study integration limit abundance and causal claims.

Single-cell microbial and viral associations in protists

Cell-resolved genomes and viral transcripts identify candidate ecological hubs, while ingestion and transient co-isolation remain alternative explanations.

T6SS phospholipase effector and plant reproduction

Biochemistry, lipid assays, bacterial competition, and plant phenotypes implicate TleB, but native secretion and the complete signaling pathway remain uncertain.

Starship transposons and fungal virulence evolution

Comparative genomics links Starships to plastic regions, mobile cargo, and NLP6 function, while broad causal claims rely mainly on association plus one deletion test.

BRCA2 stabilization of RAD51 filaments

Cryo-EM, binding assays, mutagenesis, and cross-linking support a TR2 brace across RAD51 protomers, with local-resolution limits at the peptide site.

Visual deprivation and intrinsic neuronal plasticity

Electrophysiology, Kv2.1 localization, pharmacology, and reversibility support impaired LTP-IE in deprived L5 neurons; generality beyond the model is limited.

eIF5A in activated CD8 T cells

Genetic disruption, nascent proteomics, and RNA sequencing support selective translation effects, while GC7 shows broader pharmacological consequences.

Quality-factor limits on biological information carriers

The screening framework usefully treats spectral separation and persistence as separate requirements, but several biological inputs remain estimated rather than directly measured.

Immunometabolism and cancer therapy

The narrative review maps nutrient competition and metabolite signaling across immune states, but lacks systematic-search methods and quantitative synthesis.

Influenza antibody breadth evolution

Deep mutational scanning and BLI support nested constraints across H1, H3, and influenza B binding, with in vivo maturation still untested.

Cortical GABA during human motor learning

Learning-specific MRS GABA changes and early-performance correlations support a candidate state marker, although MRS cannot isolate synaptic causality.

Photochemical risks of expanded genetic alphabets

Photophysical measurements and light-dependent cell effects support singlet-oxygen risk, while ROS specificity and wider biological generality remain limited.

Partial CD45 exclusion in T-cell contacts

Live-cell imaging links partial CD45 exclusion with signaling at sub-micron contacts, while causal exclusion-fraction manipulation remains needed.

Kleptoplast donor effects in sea slugs

Donor identity and feeding history predict photosynthetic persistence in Elysia crispata; molecular repair mechanisms were not directly measured.

Ectomycorrhizal fungi, nitrogen, and Scots pine photosynthesis

Axenic experiments show altered nitrogen and water economy without guaranteed photosynthetic stimulation; incomplete fungal nitrogen accounting limits causal inference.

Killifish adaptation to crude-oil toxicity

Common-garden population comparisons connect resistance with AHR-module transcription and cardiac outcomes, but expression evidence does not establish pathway necessity.

Stage-resolved egg-production genetics in chickens

Stage-specific variance mapping identifies a CNNM2 direction switch, while causal variants and regulatory effects remain unresolved association signals.

Alzheimer’s disease risk genes in Drosophila

Broad fly phenotyping separates structure, function, and resilience modules, but translation to human Alzheimer’s mechanisms remains model-dependent and preprint-stage.

Falsifiable experiments and analysis plans

Spatial scalp maps and hair-follicle lineage disruption Atlas data can identify niche-addressed inflammatory hypotheses, but causal lineage derailment requires independent validation and tissue perturbation. Spatially selective competition and resident diversity Localized resource use may suppress invaders while preserving residents, but unused niches and diversity costs remain testable ecological predictions. Extremophile adaptation and DNA physical properties Temperature-linked sequence motifs may reflect DNA mechanics, although current evidence is genomic correlation rather than direct physical causation. RNAquarium segment-discordance checks Independent datasets, pipeline perturbations, contiguous coverage, and read-pair support can distinguish genuine segment history from assembly or classification mixing. Shrinkage versus false positives in gene-set classification Comparing filtered and unfiltered SRBCT gene sets across permutations and cross-validation splits can test whether shrinkage alone explains very low false-positive counts. Infection, RNA editing, and later tumor exhaustion ADAR1-linked editing changes and tumor exhaustion programs are separately supported, while a long-term causal connection needs longitudinal perturbation. Calcineurin inhibition and nonlinear T-cell fate Feedback models predict dose-dependent NFAT, cytokine, and apoptosis responses that can be tested with graded inhibitor exposure and time-resolved readouts. Muscle thyroid signaling during weight regain The proposed D2/D3 mechanism is a plausible hypothesis, but human intracellular T3 markers must precede and mediate preferential fat recovery to support causality. HSV-1 interferon evasion as a distributed system Viral proteins target multiple signaling and metabolic gates, while cell type, species, and metabolic state may determine the dominant bottleneck. Mouse defense battery robustness The proposed anxiety–panic factor structure is useful for screening, but replication across strains, threats, and new drug classes is required.

The Evidence Layer for Scientific AI

Ask questions across claims linked to experiments, exact results, limitations, and sources.Know what the science actually supports before you trust the answer.

Explore by Goal