Loading the front page.
Compounds, proteins, tissues and the outcomes they have been measured against, in one graph. Every finding shows the paper behind it and the design of that study, and its grade is computed from both — so weak evidence reads as weak.
Three readings of the same paths: from a substance, from a result, or from a place in the body.
What the literature reports, where it goes, and the molecule itself.
A measured endpoint, and everything reported to move it.
Where the graph reaches, and what has a transporter route there.
Every node is keyed by an external ontology. No serial integers, no internal names.
Every edge is one of 18 relations, and no others exist. That is what lets two compounds be compared instead of merely described.
Declared, not yet used
The vocabulary is fixed before the evidence arrives, not extended to fit it.
A grade is derived from the study behind the claim, not asserted by whoever wrote it. These are the inputs it is derived from.
| A | Well established | Replicated human trial evidence, consistent direction |
|---|---|---|
| B | Likely | Human evidence, limited replication or design limits |
| C | Suggestive | Weak human evidence, or strong evidence in a mismatched population |
| D | Preliminary | Non-human or in vitro only |
| F | Speculative | Mechanistic inference, or no evidence found |
The two channels are independent. Grade is drawn as line weight, opacity and dash; class is drawn as shape. Neither is a colour, so both survive greyscale. The full key.
A summary written for a model rather than a person: what the graph holds, what the counts mean, and which of them must never be added together.
Served now, at /llms.txt.
Traversal over 127,548 precomputed, degree-weighted paths. Findings and inferences are separable in every result.
Programmatic access in development.
The whole graph, with provenance columns intact, under the licences of its upstream sources.
Bulk export in development.