ProPaths reads a protein's entire literature and builds its interactome the way an expert curator would. Every interaction comes with its mechanism, direction, effects, kinetics, and the pathways it belongs to.
Every interaction is typed by kind and placed in a pathway ontology that keeps biological systems distinct, with each partner where the literature shows it functions and has an effect.
Most resources tell you these two proteins are associated. Here is what ProPaths holds instead.
Five stages, run per protein, from the primary literature to a traceable graph.
The primary literature on a protein, through a query built for interaction biology, taking full text wherever it exists. No later stage can recover biology that was never read.
Every paper is read for the interactions it reports: which protein acts on which, by what mechanism, in what direction, with what measured result. This pass is deliberately generous, and is judged on how little it misses.
A second pass argues against the first and rejects most of what it produced. Speculation in a discussion section is not a finding, and anything the evidence does not carry is dropped rather than kept at a low score.
What survives becomes a graph rather than a list, placed in the pathway ontology at every level it belongs to. A partner sits in every pathway the literature puts it in, not one.
Every paper touching a pair is read together into distinct claims, each carrying its mechanism, its measurements with the assays that produced them, and the papers behind it. Papers that disagree stay distinct.
Twenty minutes of compute. A few dollars. One interactome.
The interactome is a tool, not just a view. Everything in the demo is also a clean read API and a local MCP server, so an agent works the graph the way a curator does. It gets the map first, then pulls full mechanism and papers for the one edge it needs. Sized for a model's context window, not a browser, and it speaks MCP, so it drops into Claude in one line.
Read the full API reference, or get the MCP server on GitHub.
For write access, higher limits, or a hosted MCP, get in touch.
The public demo is free and no account is required. Click the link below to get started.
If you have a question, or want your own protein read, get in touch.