How Kelda Verifies Health Data
Every claim on Kelda traces to a specific database entry, a named researcher, or a published study. Here is exactly how we do it.
Our Data Sources
Kelda queries the same peer-reviewed molecular databases used by researchers and clinicians worldwide. Every number on a Kelda page comes from one of these six sources.
CTD
Comparative Toxicogenomics Database
What it is: Chemical-gene-disease relationships, therapeutic evidence, marker/mechanism associations.
We use it for: Drug-nutrient depletion mechanisms, gene interaction counts, disease associations.
ChEMBL
European Molecular Biology Laboratory
What it is: Molecular mechanisms of action, binding targets, potency measurements, clinical trial phases.
We use it for: Mechanism detail, compound comparisons, target identification.
FAERS
FDA Adverse Event Reporting System
What it is: Real-world safety reports from patients, healthcare providers, and manufacturers.
We use it for: Safety data, adverse event counts, serious event percentages.
PubMed
National Library of Medicine
What it is: 35 million+ biomedical research articles from peer-reviewed journals worldwide.
We use it for: Expert quotes with researcher name, journal, year, and verifiable PMID.
PharmGKB
Pharmacogenomics Knowledge Base
What it is: Gene-drug-disease relationships, dosing guidelines based on genetic variants.
We use it for: Genotype-dependent dosing (e.g., CYP2C19 variants for omeprazole).
USDA FoodData Central
U.S. Department of Agriculture
What it is: Nutrient composition for thousands of foods, including branded and survey data.
We use it for: Food source tables with mg per serving for depleted nutrients.
The Knowledge Graph
Raw database rows aren't useful on their own. Kelda pre-compiles 212 million rows of molecular data into a structured knowledge graph that connects compounds, genes, diseases, symptoms, and pathways in a single queryable network.
124,000+
Entities
Compounds, genes, symptoms, diseases, pathways, biomarkers, mechanisms
2.8 million
Edges
Directed connections between entities (treats, modulates, causes, binds)
212 million
Source rows
Pre-compiled from molecular databases into queryable profiles
7
Entity types
Compounds (81K), genes (28K), symptoms (8K), diseases (3K), pathways (2.8K), biomarkers (159), mechanisms
When you look up omeprazole on Kelda, you aren't reading a manually written article. You're reading a pre-compiled profile that synthesizes 4,642 PubMed articles, 395 randomized controlled trials, 108 gene interactions, and 122,780 FAERS safety reports — all traced back to their source databases.
Optimal vs Normal Ranges
Standard lab reference ranges are designed to detect disease — they tell you when something is clinically abnormal. But they don't tell you when something is suboptimal.
Kelda uses functional medicine optimal ranges alongside standard lab ranges, because symptoms often appear well before a lab value is officially “out of range.”
Example: Ferritin
Lab “normal” range: 20–200 ng/mL
Optimal range: 70–150 ng/mL
A ferritin of 32 ng/mL is “normal” by lab standards — but research shows that fatigue, hair loss, and restless legs often appear below 70. The optimal range catches these early depletions that lab “normal” misses.
Every biomarker page on Kelda shows both ranges side by side, so you can see where your results fall on both scales. We cite the clinical research behind each optimal range.
How We Maintain Accuracy
Every statistic cites its source database by name and count — not "studies show" but "395 RCTs across 360,638 patients in CTD."
Expert quotes link to real PubMed articles with verifiable PMIDs that anyone can look up.
Content is validated against strict type contracts and accuracy checks before publishing.
Pages are updated when source databases release new data, with a visible "Sources verified as of" date.
What We Don't Do
Transparency means being honest about what we are and what we aren't. We believe this honesty is more trustworthy than fabricated credentials.
We don't have a medical advisory board or review panel. We rely on the databases themselves as the authority.
We don't fabricate researcher names, credentials, or institutional affiliations.
We don't use vague phrases like "studies show" without citing the specific database, researcher, and count.
We attribute every claim to a named database or published researcher — nothing is unsourced.
This is health information, not medical advice. We always recommend discussing findings with your healthcare provider.
Source Attribution Model
Every content page on Kelda displays a standardized attribution block at the top. This is what it includes:
Based on research by Huan et al., European Journal of Pharmaceutical Sciences (2025). Data sourced from CTD, ChEMBL, FAERS. How we verify this data →
Sources verified as of April 2026
Primary researcher — the named author from a PubMed study relevant to the page topic, with journal name and publication year.
Database sources — the specific databases queried for this page (CTD, ChEMBL, FAERS, PharmGKB, USDA, PubMed).
Verification date — when the source data was last confirmed to be current.
Methodology link — every attribution block links back to this page, so readers can always check how the data was verified.
Questions about our data?
We're committed to transparency. If you spot an error or have questions about how we source our data, we want to hear from you.
Contact ushello@kelda.health
Educational tool — always consult your healthcare provider.