We think the future of chemistry is Agents and Humans working together on the same data.
Most data lakes hold everything and understand nothing. ChemGraph is built on how chemical data actually relates: a structure to the assays it was tested in, an assay to the target it measures, a target to the pathways it starts. Your data lands in that model, not in a pile.
Agents query the graph directly over the Model Context Protocol, with the same evidence behind every answer that a person sees.
Your own pipelines and tools read and write the same graph over a standard API.
People explore it visually, like the view above: every link opens up to show the result, prediction or pathway behind it.
Add a new prediction method or data type and it becomes a new kind of link in the graph, available everywhere at once: to agents, to the API and in the viewer.
Add your own compounds and results and they sit alongside the public data under the same relationships, so every existing view and prediction draws on them straight away.
ChemGraph treats your organisation's data sources as a graph. Each kind of relationship has a resolver that knows where its data lives and how to fetch it, so a question becomes a walk across the graph rather than a bespoke query. A new resolver adds a new relationship; a new source plugs into an existing one. That is what lets the graph grow in both directions without being rebuilt.
As a starting point, ChemGraph ingests public data from 2 sources to the volumes below. Your own data adds to them.
8.7k
compounds
2.6M
assay results
1.6k
assays
599
adverse outcome pathways
ChemGraph puts compounds, assays, targets and adverse outcome pathways in one graph. No black-box risk scores: each link traces back to a measured result, the similar compounds a prediction was borrowed from, or the AOP-Wiki relationship it follows.
Where a compound has been tested, you see the result: which assays, how potent, and whether activity only shows up where cells are already dying.
Where it hasn’t, the nearest tested structures vote, weighted by similarity. You see who voted and how they were tested.
Hits are mapped onto adverse outcome pathways, from the molecular initiating event through to the outcome, so the gaps in the evidence are as visible as the hits.