The Toxicity Oracle — experimental
Type any drug — on or off our catalog — pick an organ system, and the Oracle reasons over live PubMed and drug-class basic science to grade how plausible a given toxicity is. It is a hypothesis-generating instrument: a way to ask “could this drug do that?” and see the evidence the question rests on.
Experimental — hypothesis-generating only. The Oracle reasons over an automated PubMed review and drug-class basic science to grade the plausibility of a toxicity. It is not medical advice, not a clinical decision tool, and may be wrong. Verify every claim against primary literature.
AKI, electrolytes, glomerular & tubular injury
How it works
Three steps from a free-text drug name to a graded, cited verdict.
Targeted PubMed search
The drug and chosen organ system are translated into a structured PubMed query. The Oracle pulls the result count and the most relevant articles so you can see exactly what literature exists.
Drug-class reasoning
The retrieved evidence is combined with mechanistic priors — the drug's class, its molecular target, and how related agents behave — to reason about whether the toxicity is biologically coherent.
Graded verdict
The Oracle returns a plausibility grade (from Established through Plausible to Unlikely), a plain-language summary, the reasoning chain, and the citations — never a number without the evidence behind it.
- · Absence of evidence is not evidence of absence — a low result count for a new agent often reflects how new it is, not its safety.
- · Class-level reasoning can mislead when a drug breaks from its class; verdicts are priors, not proof.
- · PubMed indexing lags, abstracts can be ambiguous, and the Oracle cannot read full text or weigh study quality the way a human can.
- · It generates hypotheses to investigate — it does not establish causation or guide individual patient care.
The Toxicity Oracle is an experimental, hypothesis-generating research aid built on automated PubMed review and basic-science reasoning. Its output can be incomplete or wrong. It is not a diagnostic device, not a substitute for clinical judgement, and must not be used to make decisions about real patients. Always verify against primary literature and qualified clinicians.